Evalueringsrapport om den tyske ordning: Federal Highway Research Institute, Section U4 "Driver Training, Driver Improvement", Bergisch Gladbach, den 31. maj 2010

Tilhører sager:

Aktører:


    Evaluation.pdf

    https://www.ft.dk/samling/20131/beslutningsforslag/B23/bilag/1/1304711.pdf

    Evaluation of the novice driver training models
    “Accompanied driving from 17” and “Voluntary further
    training seminars for holders of probationary driving
    licences”. Results up to November 2009.
    Georg Willmes-Lenz, Frank Prücher, Heidrun Großmann
    Third report on AP project F1100-4408016 “Evaluation of novice driver training models”
    Federal Highway Research Institute, Section U4 “Driver Training, Driver Improvement”
    Bergisch Gladbach, 31.05.2010 (expanded version)
    Retsudvalget 2013-14
    B 23 Bilag 1
    Offentligt
    2
    Contents
    1. Introduction 3
    2. Practical implementation of the VFT model: Results of the process evaluation 5
    2.1 Objective 5
    2.2 Method 5
    2.3 Results 5
    2.4 Significance of the results 7
    3. Road safety effectiveness of the VFT model: Results of the summative evaluation 8
    3.1 Objective 8
    3.2 Method 8
    3.3 Results 9
    3.4 Significance of the results 11
    4. Participation figures and road safety effectiveness of the VFT model: Evaluation of
    VZR and ZFER data records 11
    4.1 Objective 11
    4.2 Method 11
    4.3 Results 12
    4.4 Significance of the results 14
    5. Practical implementation of the AD17 model: Results of the process evaluation 15
    5.1 Objective 15
    5.2 Method 16
    5.3 Results 16
    5.4 Significance of the results 23
    6. Road safety effectiveness of the AD17 model: Results of the summative evaluation 23
    6.1 Objective 23
    6.3 Studies presented to date 23
    6.3 Results 24
    6.4 Significance of the results 26
    7. Conclusion 27
    8. Bibliography 29
    Annex 30
    Schade, F.-D., Heinzmann, H.-J. (2009). Summative Evaluation of Accompanied Driving from 17.
    Special evaluation: First evaluation results on the basis of self-reported driving behaviour.
    3
    1. Introduction
    The present report is a compilation of the evaluation results obtained by the end of 2009 in
    respect of the experimental novice driver training models “Voluntary further training semi-
    nars for holders of probationary driving licences” (here: “VFT model”) and “Accompanied
    driving from 17” (here: “AD17 model”). These models were initially introduced on a trial ba-
    sis in 2003 and 2005, respectively, and are to be evaluated and tested with regard to their
    road safety effectiveness before the government makes a decision on permanent integration
    into the driver licensing system.
    Alongside the question of road safety effectiveness (summative evaluation), the studies are
    also to analyse the experience gained from practical implementation of the individual con-
    cepts (process/formative evaluation). Whereas the safety impact is of direct significance for
    the decision on permanent adoption of the models, the results of the process evaluation are
    important independently of this decision for considerations of the possibilities for concept
    optimisation.
    The evaluation studies conducted by the Federal Highway Research Institute (BASt) were
    spread over a total of six sub-projects. These sub-projects are listed in Tab. 1, together with
    an indication of the precise subject of evaluation and the methodical approach. Complete
    and conclusive results are available from five of these sub-projects.
    In the case of the summative evaluation of the AD17 model, it is currently only possible to
    present partial results, due to the specific project schedule dictated by the analysis design.
    These results are available in the form of an intermediate report dated 31.07.2007 (Schade
    et al., 2007) and a special evaluation dated 30.11.2009 (Schade & Heinzmann, 2009). With its
    analysis of the safety-related behaviour of novice drivers during their first year of independ-
    ent driving on the basis of self-reported accident involvement and traffic offences, the latter
    report covers an important aspect of the overall project objective. Given the conclusive char-
    acter of the partial results from this special evaluation, it is already possible to make a sound
    assessment of the road safety effectiveness of the AD17 model at the present juncture.
    Statements on the road safety effectiveness of the AD17 model are also to be found in an
    evaluation of the University of Giessen, which investigates the impact of the experimental
    scheme of accompanied driving as initially implemented at regional level in the federal state
    of Lower Saxony (Stiensmeier-Pelster, 2008). The corresponding results have been incorpo-
    rated into the appropriate section of the present report.
    The Federal Highway Research Institute (BASt) intends to publish the research results from
    the individual evaluation projects in the form of scientific reports (M series “People and
    Safety”), so as to make the findings available in their full scope for the pending traffic policy
    discussions and for the further professional treatment of issues concerning safety-related
    novice driver preparation. It is expected that the reports on the evaluation projects which
    have already been completed will be published in the first months of the coming year. The
    publication of the final research report on the road safety effectiveness of the AD17 model,
    on the other hand, will not be possible until the corresponding behaviour-relevant data have
    been acquired and evaluated in full, presumably in autumn 2010. For this reason, and in the
    interest of early and comprehensive information, the special evaluation already completed in
    advance on 30.11.2009 has been appended to the present report in its full version.
    4
    Project Completion Evaluation topic Method
    FE 82.264/2004
    Evaluation of voluntary
    further training seminars
    for novice drivers (VFT).
    Formative evaluation.
    Project supervision:
    Dr. Hartmut Kerwien,
    Bielefeld
    End 2009 Practical
    implementation of
    the VFT model
    Analysis of concept-adequate
    implementation of the model
    on the basis of surveys and
    seminar observations
    FE 82.307/2006
    Evaluation of voluntary
    further training seminars
    for novice drivers (VFT).
    Analysis of effectiveness.
    Project supervision:
    Centre for Evaluation and
    Methods (ZEM), Bonn
    End 2009 Road safety
    effectiveness of the
    VFT model
    Evaluation of model-related
    changes in road safety atti-
    tudes with the aid of psycho-
    logical attitude measurements
    within the framework of a
    pre-post comparison with a
    control group
    FE 89.226/2009
    Evaluation of voluntary
    further training seminars
    for novice drivers (VFT),
    based on VZR data records.
    Project supervision:
    Federal Motor Transport
    Authority, Statistics Dept.,
    Group 31, Flensburg
    End 2009 Road safety
    effectiveness of the
    VFT model
    Determination of the scope of
    use of the model and compari-
    son of the driving behaviour of
    VFT participants and non-
    participants of similar age and
    experience on the basis of ZFER
    and VZR data records
    FE 82.298/2005
    Accompanied driving from
    17. Process evaluation.
    Project supervision:
    Institute for Empirical
    Sociology (IfeS), Nürnberg
    End 2009 Practical
    implementation of
    the AD17 model
    Analysis of concept-adequate
    implementation of the model
    and general practical experi-
    ence, based on the questioning
    of participating novice drivers
    and accompanists, as well as
    document analyses
    FE 89.221/2009
    Designing of accompanied
    driving practice.
    Project supervision:
    Institute for Empirical
    Sociology (IfeS), Nürnberg
    End 2009 Practical
    implementation of
    the AD17 model
    Deeper analysis of the inter-
    action between novice drivers
    and their accompanists on the
    basis of surveys
    FE 82.316/2006
    Accompanied driving from
    17. Summative evaluation.
    Project supervision:
    Federal Motor Transport
    Authority, Statistics Dept.,
    Group 31, Flensburg
    30.09.2010
    Partial
    results:
    31.07.2007
    30.11.2009
    Road safety
    effectiveness of the
    AD17 model
    Evaluation of the model-
    specific reduction in accident
    and traffic offence risks on the
    basis of a comparison of the
    rates of accident involvement
    and traffic offences between
    AD17 participants and conven-
    tionally trained novice drivers
    Tab. 1: BASt evaluation projects addressing the novice driver models “Voluntary further
    training seminars for novice drivers” and “Accompanied driving from 17”
    5
    2. Practical implementation of the VFT model: Results of the process
    evaluation
    2.1 Objective
    “Voluntary further training seminars for novice drivers” (VFT model) were introduced on a trial basis
    by way of the Novice Driver Further Training Ordinance (Fahranfängerfortbildungsverordnung,
    FreiFortbV) of 16th May 2003. In §6 of this ordinance, it is specified that the Federal Highway Re-
    search Institute (BASt) is to evaluate the model to assess the “effectiveness with regard to road
    safety”. The individual topics and procedures for the evaluation were discussed in advance of the
    introduction between the BASt and protagonists with practical interests, e.g. the German Road Safe-
    ty Council (DVR), the driving instructors and the German motorists' association ADAC, and agreed as
    follows: A first evaluation stage was to address questions concerning the functioning of the model,
    participation patterns, participant motivation and acceptance (process evaluation). Subsequently,
    the impact of the model approach on road safety was to be investigated as a second stage (summa-
    tive evaluation).
    The principal objective of the VFT model is to influence youth-specific attitudes and risk-related be-
    haviour by way of educative measures. The intervention is aimed at modification of the young driv-
    ers' value systems, attitudes and convictions to the benefit of road safety. The seminar leaders and
    moderators are expected to realise the affective, i.e. attitude-related learning objectives through
    forms of interaction which demand active contributions by the participants. This is to be achieved
    within the framework of group discussions, observed driving practice and practical safety training.
    The objective of the formative evaluation is to describe and analyse the practical implementation of
    the model under the aspect of concept-adequate realisation.
    2.2 Method
    The empirical analysis of practical implementation was founded on three study approaches:
    - Partially structured observations, interviews and written questionnaires to acquire the assess-
    ments and experience of the VFT seminar leaders and the moderators of the module “Practical
    safety training”.
    - Questionnaire survey of the VFT participants to acquire their view of the VFT programme mod-
    ules.
    - Structured descriptions of the seminar realisation on the basis of participative observations and
    interviews conducted by specially trained survey researchers.
    2.3 Results
    2.3.1 Questioning of seminar leaders and moderators
    The seminar leaders and safety training moderators shared the opinion that the limited response to
    the model was in part due to the voluntary nature of VFT participation. Its awareness level was seen
    to be inadequate and the costs deterred potential participants. The shortening of the probationary
    period was perceived as the predominant motivation for the participation of the novice drivers.
    2.3.2 Questioning of participants
    The questionnaires completed by the participants confirmed that the prime motivation for participa-
    tion was the shortening of the probationary period, followed by an expectation of improved driving
    6
    competence. Female participants replied far more often than their male counterparts that they
    wanted to learn to drive more safely and overcome driving anxieties. Participants under extended
    probation were even more distinctly attracted by the opportunity to shorten the probationary pe-
    riod, and were significantly less interested in learning to drive safely or improve their driving skills.
    The seminar module “Practical safety training” received the best marks in all assessment categories.
    The participants reported that the practical safety training had brought them the greatest benefits,
    that it had been more fun than the remaining modules, that it had offered the most valuable con-
    tents, that it had best met their expectations, and that it had been most interesting.
    In the opinion of the participants, the seminar leaders spoke too much about their own experiences
    during the group discussions. Otherwise, the seminar leaders and moderators received positive as-
    sessments.
    2.3.3 Seminar observations
    The seminar observations served to assess the degree of goal accomplishment of the individual pro-
    gramme modules. The first group discussion achieved the best result with an assessed goal accom-
    plishment of 75 per cent. It was already indicated during these first discussions, however, that the
    linking of the individual seminar modules was problematical. This applied above all to elaboration of
    the group's wishes concerning the observed driving practice and the practical safety training. A fur-
    ther problem was identified in time management. At the beginning of the first group discussion, in
    particular, considerable time was spent on organisational questions.
    Compared to the first group discussion, the second and third group discussions displayed a number
    of realisation deficits. The degree of goal accomplishment for the second module was assessed at 62
    per cent by the observers, and the third module was deemed to be only marginally better with a
    degree of goal accomplishment of 64 per cent.
    During the second group discussions, it was recognised that problems arose with regard to commu-
    nication of the psychological topics “Driving motives” and “Emotions”. The aspects of (a) the impact
    of emotions on attentiveness, (b) the correlation between emotions and competitive behaviour, and
    (c) the impairment of personal safety by emotions and time pressures, in particular, could not be
    conveyed adequately in numerous discussions. The interlinking seminar elements were also ne-
    glected somewhat during the second group discussion, and there were again certain problems with
    time management.
    The seminar leaders devoted the attention of the third group discussion above all to the topic of
    “Avoiding driving under the influence of alcohol or drugs”, and this was for the most part also treated
    adequately in the sense of the model concept. Nevertheless, it was conspicuous that the participants
    were not always able to formulate corresponding strategies to avoid driving under the influence of
    alcohol or drugs. It was often not possible to convince the participants to maintain behaviour pat-
    terns recognised as sensible in the future.
    Generally, it was shown that a review of previous group discussions was for the seminar leaders less
    important than the retrospective contemplation of practical elements.
    The objectives of the module “Observed driving practice” were considered achieved to an extent of
    almost 70 per cent (with the exception of the element “Modern driving”, which the observers rated
    with a degree of goal accomplishment of 58 per cent). The realisation of the exercise component was
    essentially satisfactory, although only few participants specified actual exercise situations. On the
    other hand, this aspect was frequently linked with the topic of ecological driving, which involved
    7
    above all the provision of tips on a fuel-saving driving style. Primarily safety-relevant driving tips were
    given less emphasis.
    The practical safety training achieved a degree of goal accomplishment of only 64 per cent, the sec-
    ond-lowest value of all modules. The whole thematic element “Braking” was handled in an extremely
    diverse manner, and in some seminars deviated significantly from the specifications of the handbook.
    The concept adequacy of the element “Driving through curves” was criticised. The exercises were
    generally performed very hastily and were seldom clearly differentiated; in many seminars, the
    speed specifications were not observed.
    VFT module Degree of goal
    accomplishment
    1st group discussion 75%
    2nd group discussion 62%
    3rd group discussion 64%
    Observed driving practice 69%
    Practical safety training 64%
    Tab. 2: Degrees of goal accomplishment in the modules of the VFT model
    2.4 Significance of the results
    It can be seen from the results that the novice drivers valued above all those benefits of the seminars
    which corresponded to their expectations and interests: The shortening of their probationary period
    and time spent on practical driving exercises.
    At the same time, it is evident that the seminar leaders experienced problems with the application of
    active, attitude-building training forms, and that it was the seminar element dealing with the topic of
    emotions which caused them the greatest difficulties.
    This result also corresponds to international experience. An analysis of Swedish novice driver training
    courses which followed an explicitly attitude-building concept, for example, revealed that the par-
    ticipants were later of the opinion that the course had above all served to improve their practical
    driving skills (cf. EU project ADVANCED, 2002).
    The evaluation report criticises the overall excessive number of specified goals for the individual
    seminar modules. This diversity of goals is at the same time a source of time management problems.
    It is suggested that the broad spectrum of learning objectives be narrowed down significantly and
    formed into a hierarchy of training goals, as a basis for effective implementation by the seminar
    leaders and moderators. In addition, thought should be given to a more streamlined timeframe.
    The recommendation from an expert point of view is to subject the VFT seminars to a thorough re-
    view in the light of the aforementioned points, referring both to the concept of the model and its
    concept-adequate implementation.
    8
    3. Road safety effectiveness of the VFT model: Results of the summative
    evaluation
    3.1 Objective
    The summative evaluation served to analyse the model-related effectiveness of the VFT model with
    regard to improvements in road safety. The criterion of accident involvement, however, was dis-
    counted as a possible measure of road safety effectiveness in the context of a survey study, as it was
    clearly foreseeable that the necessary sample size would not be attained.
    The objective of the VFT model is to reduce the risk of accident for young novice drivers by way of a
    positive influencing of road safety attitudes. The criterion chosen to verify the effectiveness of the
    model, therefore, was the change in attitudes achieved through VFT participation.
    3.2 Method
    On the basis of the goals formulated in the VFT model handbooks, 11 road-safety-relevant attitudes
    were identified and subsequently applied within the framework of the evaluation.
    The displayed attitudes were measured using reliability-tested attitude scales and a number of indi-
    vidual evaluation items on four occasions: Before VFT participation, shortly afterwards, and on two
    further occasions during the course of the subsequent year.1
    A reference sample of novice drivers
    who had not attended VFT seminars – parallelised in respect of gender, driving experience and
    school education background – was analysed at the same intervals using the same instruments.
    It proved particularly difficult to gain the cooperation of an adequate number of study participants;
    recruitment was only successful thanks to the active support of driving instructors who specifically
    approached novice drivers, and through the provision of financial incentives for participation. On the
    basis of an initial sample of over 1,000 persons, it was in the end possible to take the data from ap-
    proximately 300 persons (experimental group and control group) into account in the final evalua-
    tions.
    To verify the effectiveness of the model, the following comparative analyses were performed to
    evaluate the determined attitudes of the VFT participants (experimental group) and non-participants
    (control group):
    - Evaluation of the changes in displayed attitudes over the course of the measurements
    - Comparison of the displayed attitudes of VFT participants and non-participants on each meas-
    urement occasion
    - Combined evaluation of the changes between measurements and the differences between the
    experimental and control groups.
    The combined evaluation is necessary, as it is only on this basis that a statement can be made as to
    whether changes in the attitudes displayed by the experimental group – either positive or negative –
    can be attributed to the model intervention. A positive attitude change in the experimental group,
    for example, can only be interpreted as an intervention effect if no corresponding attitude change is
    displayed in the control group – given otherwise identical framework conditions.
    1
    The following overall evaluation takes into account only the results of the first three measurements, as no adequate data-
    base was obtained to permit analysis of the fourth survey results.
    9
    The differences in attitudes were in each case tested with regard to their statistical significance, and
    the corresponding effect size was determined. The influences of the individual distances driven and
    further confounding variables were checked during the evaluation.2
    It can be assumed that the model is effective if, after VFT participation, the attitudes of the partici-
    pants develop to the benefit of road safety to a significantly greater extent than those of the drivers
    in the control group. Even in the case of overall negative changes in attitude, a model effect can still
    be assumed if the VFT participants display a more favourable development than non-participants, i.e.
    if the deterioration in attitudes in the experimental group is less pronounced than in the control
    group.
    On this basis, the following three questions or hypotheses were considered:
    (1) Pre-post comparison: Do the VFT participants display improved attitudes after participation?
    (2) Absolute comparison of VFT participants and non-participants: Do the VFT participants display
    more favourable attitudes than the control group after participation?
    (3) Relative comparison between VFT participants and non-participants: Do the VFT participants
    display more favourable road-safety-relevant attitudes than the control group over the whole
    course of the measurements (irrespective of generally declining values)?
    3.3 Results
    The individual safety-relevant attitudes and competence indicators listed in Tab. 3 were evaluated.
    For the majority of the attitudes and competence indicators considered, no evidence of influence
    attributable to VFT participation was found.
    Partial indications of a positive effect of VFT participation were revealed merely in the attitude do-
    mains risk-taking disposition and hazard awareness, but could not be substantiated in the overall
    evaluation.
    Risk-taking disposition: Viewed over the whole course of the measurements, both the experimental
    group and the control group displayed a perceptible increase in risk-taking disposition. In the isolated
    evaluation of the third set of measurements, it was to be seen that the risk-taking disposition of the
    VFT participants was at this time slightly, but nevertheless significantly less distinct than in the con-
    trol group. This could be interpreted as a (medium-term) effect of the model. The extent by which
    the increase in risk-taking disposition is lessened due to VFT participation, however, is so small that it
    is not possible to assume a significant influence on risk-taking disposition.
    Hazard awareness was already more pronounced in the experimental group than in the control
    group before participation in VFT seminars. At the time of the second survey, i.e. immediately after
    VFT participation, on the other hand, no differences were to be determined between the groups. The
    results of the third measurement, finally, again showed a significantly greater hazard awareness
    among the VFT participants than in the control group. This difference, however, could be attributable
    not only to the VFT participation in the sense of a medium-term effect, but also to the differences
    which already existed at the time of the first measurements. Nevertheless, as the statistical effect
    2
    The post-hoc tests took the form of either t-tests for independent samples or variance analyses for independent samples
    (ANOVA) with Bonferroni tests. The measure for the effect size was Cohen's d in the case of the t-tests and partial eta-
    squared (η
    2
    ) in the case of the variance analyses. To evaluate the differences in the development of displayed attitudes
    between the experimental and control groups, variance analysis with repeated measurements was used.
    10
    size at the time of the third survey was found to be significantly greater than at the time of the first
    survey, model effectiveness can be assumed for this attitude domain.
    The VFT participants also displayed a significant change in their self-assessment of safety-relevant
    driving competence over the course of the measurements. There was a significant increase between
    the first and third surveys. As this same phenomenon was observed in the control group, however, it
    is not justified to conclude that the model is effective in this respect.
    In all other attitude and competence constructs taken as criteria to verify the effectiveness of the
    VFT seminars, no hypothesis-relevant results were recorded.
    Hypothesis
    Attitude scales/items (1) (2) (3)
    1. Personal importance of driving (Scale B) n.s. n.s. n.s.
    2. Realistic assessment of traffic demands (Scale C) n.s. n.s. n.s.
    3. Readiness to observe traffic rules (Scale D) n.s. n.s. n.s.
    4. Risk-taking disposition/risky behaviour (Scale E) n.s. ≤ .05
    1
    n.s.
    5. Adequate self-assessment with regard to driving ability in different complex driving
    situations (Scale F)
    n.s. n.s. n.s.
    6. Reflection of own driving behaviour (Scale G) n.s. n.s. n.s.
    6.1. Reflection of driving behaviour with regard to vehicle control (Scale G1) n.s. n.s. n.s.
    6.2. Reflection of driving behaviour with regard to incorrectly observed traffic situations
    (Scale G2)
    n.s. n.s. n.s.
    6.3. Reflection of driving behaviour with regard to influences arising from accompanying
    circumstances (Scale G3)
    n.s. n.s. n.s.
    6.4. Reflection of driving behaviour with regard to influences arising from moods and emotions
    (Scale G4)
    n.s. n.s. n.s.
    7. Hazard awareness (Scale H) n.s. ≤ .05
    1
    n.s.
    8. Risk avoidance (Scale I) n.s. n.s. (–)
    9. Locus of control (Scale J) n.s. n.s. (–)
    10. Assessment of own safe driving behaviour (Scale K) n.s. n.s. n.s.
    11. 1. Critical assessment of general driving ability (Item L1) n.s. n.s. n.s.
    2. General admission of driving competence deficits (Item L2) n.s. n.s. n.s.
    3. General assessment of driving competence with reference to safe driving (Item L3) ≤ .001 n.s. n.s.
    Hypotheses:
    (1) = VFT participants display improved attitudes after participation.
    (2) = VFT participants display more favourable attitudes than the control group after participation.
    (3) = VFT participants display more favourable attitudes than the control group over the whole course of the measurements
    n.s = Result not significant, hypothesis must be rejected; ≤.05 = Hypothesis confirmed with p ≤ .05; ≤ .001 Hypothesis con-
    firmed with p ≤ .001; (–) = Hypothesis was not tested
    1
    Referring to significant differences between participants and non-participants at the time of the third measurement.
    Tab. 3: Results of hypothesis testing in the summative evaluation of the VFT model
    11
    3.4 Significance of the results
    Viewed on the basis of the results of the present evaluation, the VFT model has failed to develop
    relevant effectiveness with regard to the whole spectrum of attitudes addressed by the programme.
    An attitude effectiveness is only revealed partially and to a minor extent at certain points: The results
    point to effects for particular individual aspects in only two of eleven attitude domains.
    Given the present results, it is not possible to attribute a sustained preventive effectiveness to the
    VFT model in the sense of an improvement in road-safety-relevant attitudes. Against this back-
    ground, it is rather inconceivable that the VFT model will be able to achieve more than just marginal
    changes in the driving behaviour of novice drivers.
    The evaluation results are not essentially different to those obtained in conjunction with the compa-
    rable preventive programme “Jugend fährt sicher” ("Young people driving safely”) and already pre-
    sented in the past (cf. Schulz et al., 1995).
    The repeated failure to supply proof for the effectiveness of attitude-related supplementary training
    offers for novice drivers in the form of VFT seminars or the earlier “Jugend fährt sicher” courses
    should give cause for a fundamental reappraisal of the existing concepts for attitude-related inter-
    vention addressing novice drivers. At the same time, it should be kept in mind that, alongside driving
    experience deficits, age- and novice-typical attitudes are decisive factors contributing to the above-
    average accident risk of young and novice drivers, and that the development of effective prevention
    concepts in this field is of major importance for road safety.
    4. Participation figures and road safety effectiveness of the VFT model:
    Evaluation of VZR and ZFER data records
    4.1 Objective
    The objective of this study was to determine the scope of utilisation of the model and to assess the
    road safety effectiveness of the VFT model through a comparison of the safety-relevant driving be-
    haviour of VFT participants and non-participants of the same age and with similar driving experience3
    on the basis of corresponding data records retrieved from the Central Register of Traffic Offenders
    (VZR) at the Federal Motor Transport Authority (KBA) in Flensburg.
    4.2 Method
    The number of drivers participating in the VFT model was determined on the basis of data retrieved
    from the Central Register of Driving Licences (ZFER), which is similarly maintained by the Federal
    Motor Transport Authority in Flensburg. The attributes gender, age, federal state of the responsible
    licensing authority, and probation status before participation were recorded in each case.
    Data on relevant traffic behaviour were retrieved from VZR data records for both VFT participants
    and non-participants – parallelised according to gender, age, federal state and duration of the proba-
    tionary period before the start of the present observation.
    The VFT participants formed the experimental group (E), further sub-divided into persons with a two-
    year probationary period (E2) and those with an extended probationary period (E4). The survey cov-
    3
    Measured by the duration of driving licence possession
    12
    ered all VFT participants who could be contacted (full sample). Under the given basic scheduling, it
    was possible to include practically the whole population of participants from the year 2007.
    Novice drivers for whom no VFT participation had been recorded since the granting of a class B or BE
    driving licence were assigned to the control group (C), with a corresponding sub-division into C2 and
    C4 in accordance with their probation status. The persons actually forming the control groups were
    selected by way of stratified random sampling.
    On the basis of personal identification data, the individual records held in the Central Register of
    Traffic Offenders were retrieved for all persons assigned to the experimental (E) and control (C)
    groups (N = 11,780) on two dates, namely on 18.01.2009 (to forestall the possible deletion of older
    offences from the register) and on 02.08.2009. The register numbers of unambiguous query results
    were listed and the automated VZR statistics database was searched for entries between 2007 and
    the survey date in 2009 on the basis of these register numbers. All data records retrieved in this way
    were listed and subsequently processed to ensure unambiguous assignment to the sample. From this
    data pool, which comprised all VZR records pertaining to E and C group members, a subset was
    formed containing the entries which referred to traffic offences which had been committed during
    the observation period, but at the same time also entered in the register at the latest 18 months
    after the start of the probationary period. These VZR records were subjected to thorough plausibility
    checks and categorised dichotomously according to their references to the catalogue of traffic of-
    fences and/or accident indicators.
    4.3 Results
    4.3.1 Scope of participation in the VFT model
    Ever since introduction of the VFT model in 2004, participation has remained at a distinctly low level
    of significantly less than one per cent of the persons granted a class B/BE driving licence. Tab. 4 pro-
    vides an overview of the participation in the years 2004 to 2008, compiled on the basis of an analysis
    of amendment notices submitted to the Central Register of Driving Licences at the Federal Motor
    Transport Authority by the local licensing authorities. Due to the limited participation, however, no
    mentionable road safety effect can be assumed for the model as a whole, irrespective of whether a
    significant safety effectiveness is determined in connection with VFT participation.
    Year VFT participants,
    total, N
    VFT participants with normal
    probationary period, N (%)
    VFT participants with extended
    probationary period, N (%)
    2004 498 426 (86 %) 72 (14 %)
    2005 1461 1094 (75 % ) 367 (25 %)
    2006 2310 1647 (71 %) 663 (29 %)
    2007 2433 1690 (69 %) 743 (31 %)
    2008 1826 1052 (58 %) 774 (42 %)
    Tab. 4: VFT participants in the years 2004 to 2008, categorised according to their probation status
    before participation
    4.3.2 Safety-relevant driving behaviour
    The safety-relevant driving behaviour of VFT participants (E) and non-participants (C) is compared in
    Tab. 5 on the basis of the assignment to different traffic offence categories in the central register.
    13
    For the VFT participants, the records pertaining to relevant traffic behaviour were analysed for an
    observation period of one year, starting at the end of the shortened probationary period subsequent
    to VFT participation. In the case of non-participants, the equivalent period of time was considered.
    Column E shows the ratio of VFT participants with records of the specified conspicuous behaviour to
    their non-participating counterparts. Contrary to expectations, the novice drivers were not found to
    drive more safely and with greater respect for traffic rules after VFT participation. Instead, the group
    of VFT participants displayed conspicuous behaviour consistently – and in some cases considerably –
    more frequently than the group of non-participants.
    Column F indicates the tested statistical significance of the differences between VFT participants and
    non-participants.
    A B C D E F
    Comparison Indicator E group C group Ratio
    E to C
    Significance
    Freq. Records Freq. Records
    Abs. % Abs. %
    E2-1 vs. C2-1 Accident 1,239 35 2.82 1,239 19 1.53 1.84 *
    Endangering 1,239 66 5.33 1,239 19 1.53 3.47 ***
    Alcohol/drugs 1,239 7 0.56 1,239 4 0.32 1.75 n.s.
    Speeding 1,239 68 5.49 1,239 18 1.45 3.78 ***
    Total 1,239 119 9.60 1,239 49 3.95 2.43 ***
    E2-2 vs. C2-2 Accident 355 12 3.38 355 6 1.69 2.00 n.s.
    Endangering 355 24 6.76 355 5 1.41 4.80 ***
    Alcohol/drugs 355 2 0.56 355 1 0.28 2.00 n.s.
    Speeding 355 33 9.30 355 7 1.97 4.71 ***
    Total 355 46 12.96 355 16 4.51 2.88 ***
    E4-1 vs. C4-1 Accident 525 29 5.52 525 21 4.00 1.38 n.s.
    Endangering 525 51 9.71 525 34 6.48 1.50 *
    Alcohol/drugs 525 7 1.33 525 5 0.95 1.40 n.s.
    Speeding 525 66 12.57 525 36 6.86 1.83 **
    Total 525 105 20.00 525 72 13.71 1.46 **
    E4-2 vs. C4-2 Accident 123 6 4.88 123 5 4.07 1.20 n.s.
    Endangering 123 6 4.88 123 3 2.44 2.00 n.s.
    Alcohol/drugs 123 0 0.00 123 2 1.63 0.00 n.s.
    Speeding 123 11 8.94 123 8 6.50 1.38 n.s.
    Total 123 19 15.45 123 16 13.01 1.19 n.s.
    Compared groups:
    E2-1: VFT participants with two-year probationary period, VFT participation in first year of probation; Start of observation: Earlier end of
    probationary period after 12 months (end of probation)
    C2-1: Non-participants with two-year probationary period; Start of observation: Beginning of second year of probation
    E2-2: VFT participants with two-year probationary period, VFT participation in second year of probation; Start of observation: Earlier end of
    probationary period after VFT participation (end of probation)
    C2-2: Non-participants with two-year probationary period; Start of observation: Timing analogous to E2-2
    E4-1: VFT participants with four-year probationary period, VFT participation in first to third year of probation; Start of observation: Earlier
    end of probationary period after 36 months (end of probation)
    C4-1: Non-participants with four-year probationary period; Start of observation: Beginning of fourth year of probation
    E4-2: VFT participants with four-year probationary period, VFT participation in fourth year of probation; Start of observation: Earlier end of
    probationary period after VFT participation (end of probation)
    C4-2: Non-participants with extended probationary period; Start of observation: Timing analogous to E4-2
    Indicators:
    A “Accident”: Culpable accident
    B “Endangering”: Record of an endangering of road traffic
    C “Alcohol/drugs”: Record of an alcohol- or drugs-related offence
    D “Speeding”: Record of a speeding offence
    E “Total”: Record of any traffic offence
    Significance:
    Difference between E and C groups, Fisher’s exact test, unidirectional hypothesis: n.s. = not significant, * = p ≤ .05, ** p ≤ .01, *** p ≤ .001
    14
    Tab. 5: VZR records pertaining to members of all compared E and C groups
    The novice drivers who attended a voluntary further training seminar before the final year of an ex-
    tended probationary period (E4-1) are most frequently represented in the records of the Central
    Register of Traffic Offences across all indicators. The sub-groups with the fewest VZR records were
    those of non-participants who were driving under a two-year probationary period at the time of
    sampling (C2-1 and C2-2).
    Particularly remarkable here is the greater frequency of VZR records of the VFT participants com-
    pared to the non-participants. This applies to all sub-groups and with one exception (alcohol/other
    drugs) also for all indicators. The frequency of conspicuous behaviour among VFT participants is be-
    tween 1.35 times (two-year probationary period, culpable accidents) and 4.04 times (two-year pro-
    bationary period, speeding offences) higher than in the case of non-participants.
    Tab. 6 summarises the indicators of conspicuous behaviour for the groups of drivers during a regular
    (two-year) or extended (four-year) probationary period.
    A B C D E F
    Comparison Indicator E group C group Ratio
    E to C
    Significance
    Freq. Records Freq. Records
    Abs. % Abs. %
    E2 vs. C2 Accident 1,594 47 2.95 1,594 25 1.57 1.88 **
    Endangering 1,594 90 5.65 1,594 24 1.51 3.75 ***
    Alcohol/drugs 1,594 9 0.56 1,594 5 0.31 1.80 n.s.
    Speeding 1,594 101 6.34 1,594 25 1.57 4.04 ***
    Total 1,594 165 10.35 1,594 65 4.08 2.54 ***
    E4 vs. C4 Accident 648 35 5.40 648 26 4.01 1.35 n.s.
    Endangering 648 57 8.80 648 37 5.71 1.54 *
    Alcohol/drugs 648 7 1.08 648 7 1.08 1.00 n.s.
    Speeding 648 77 11.88 648 44 6.79 1.75 **
    Total 648 124 19.14 648 88 13.58 1.41 **
    Compared groups:
    E2: VFT participants with two-year probationary period (E2-1 plus E2-2)
    C2: Non-participants with two-year probationary period (C2-1 plus C2-2)
    E4: VFT participants with four-year probationary period (E4-1 plus E4-2)
    C4: Non-participants with four-year probationary period (C4-1 plus C4-2)
    Indicators:
    A “Accident”: Culpable accident
    B “Endangering”: Record of an endangering of road traffic
    C “Alcohol/drugs”: Record of an alcohol- or drugs-related offence
    D “Speeding”: Record of a speeding offence
    E “Total”: Record of any traffic offence
    Significance:
    Difference between E and C groups, Fisher’s exact test, unidirectional hypothesis: n.s. = not significant, * = p ≤ .05, ** p ≤ .01, *** p ≤ .001
    Tab. 6: VZR records of E and C groups, summarised according to probation status
    4.4 Significance of the results
    The results of the analysis of data records retrieved from the Central Register of Driving Licences
    (ZFER) and the Central Register of Traffic Offenders (VZR) at the Federal Motor Transport Authority
    (KBA) reveal, firstly, the limited reach of the VFT model, which attracts the participation of less than
    one per cent of all novice drivers, and secondly, the unexpectedly poorer safety-relevant driving be-
    haviour of VFT participants after attendance at a voluntary seminar, compared to other novice driv-
    ers of the same age and with similar driving experience who have not attended a VFT seminar.
    15
    The present data evaluation does not permit VFT participation to be viewed as causal for the poor
    driving behaviour of the VFT participants, and it is thus unjustified to attribute directly counter-
    productive road safety effects to the VFT model. It is possible that other factors are responsible for
    the difference in driving behaviour. Thought is here to be given first to the probation rules, which
    were (for the most part) still applicable for the non-participants during the parallelised observation
    period, whereas the group of VFT participants was no longer subject to such rules, having benefited
    from a shortening of the probationary period on account of the seminar participation. Insofar as the
    probation factor is indeed causal for the different driving behaviour, the VFT model could be attrib-
    uted at least an indirect counter-productive road safety effect, as the shortening of the probationary
    period is after all a consequence of the VFT model. Further causal factors which could be assumed to
    explain the differences in driving behaviour may be found in the deviating characteristics of the sam-
    ple groups due to self-selection effects, e.g. the conceivable circumstance that the VFT seminars are
    attended especially by those novice drivers with a tendency to conspicuous driving behaviour, as a
    means to achieve a shortening of the probationary period. Before asserting such an assumption,
    however, it must be clarified why no differences in (poor) driving behaviour are to be seen between
    the VFT participants driving under a regular two-year probationary period and those whose proba-
    tionary period has already been extended to four years due to convictions for traffic offences. The
    latter, after all, would normally be expected to display at least a more distinct tendency to conspicu-
    ous behaviour than the novice drivers subject to a regular probationary period.
    The questions referring to possible causes for the poorer driving behaviour of the VFT participants
    cannot be answered on the basis of the present data. To this end, more detailed studies with a
    broader database are necessary.
    The results seem to indicate that it is not justified to continue promoting VFT participation through
    the incentive of a shortened probationary period. The removal of this incentive would have a preven-
    tive effect in respect of both aforementioned assumed causes for the poorer driving behaviour,
    namely the curtailing of the generally protective influence of the probationary period and the attrac-
    tiveness of VFT participation for drivers with a tendency to conspicuous driving behaviour.
    It is true that the results supply no proof of a causal correlation between VFT participation and
    poorer driving behaviour, but there is similarly no evidence that any positive road safety effects are
    generated by VFT participation, and thus no basis for a recommendation to maintain the model, for
    example with removal of the incentive of a shorter probationary period.
    5. Practical implementation of the AD17 model: Results of the process
    evaluation
    5.1 Objective
    The purpose of the process evaluation was to analyse the pilot scheme realised to test the model
    approach of “Accompanied driving from 17” at national level, in order to clarify the conditions for a
    broader and reliable routine implementation of the model, and at the same time to provide sound
    empirical data for an assessment and possible further development of the model concept. The issues
    placed in the foreground concerned the ready accessibility of the model, its practicability and the
    safety-relevant aspects of its implementation.
    16
    5.2 Method
    Within the framework of the process evaluation, 3,780 participants in the model selected at random
    from the Central Register of Driving Licences (ZFER) at the Federal Motor Transport Authority (KBA)
    were contacted on up to 4 occasions and asked to complete an online or paper-based questionnaire
    on their everyday practical experience of the model. In one instance, similar questionnaires were
    also sent to 1,735 of their accompanists.
    Certain sections of the differentiated questionnaires were repeated on each occasion. In this way, in
    addition to the cross-sectional evaluations referring to particular moments in time or durations of
    participation, it was possible to identify developments over the whole accompaniment phase at both
    the aggregate level of the AD17 population as a group (trend studies) and at the individual level of
    single participants (panel studies).
    The key data on the scope and quality of driving practice within the framework of the accompanied
    driving model were acquired on the basis of weekly reports. The subjects were asked to provide in-
    formation pertaining the relevant evaluation parameters retrospectively for the last seven days be-
    fore the date of the questionnaire (“report week”). This procedure guarantees minimal memory
    losses, valid data and – given the repeated questioning of the novice drivers at different points of
    their individual accompaniment phase – a true representation of the practical implementation of the
    model over the whole possible duration of 12 months.
    5.3 Results
    The results of the process evaluation presented in the following are to be viewed against the back-
    ground of rapid acceptance of the AD17 model in Germany and a response far exceeding the original
    expectations. By January 2008, all the federal states had adopted the experimental model. During
    the course of 2007, the model became the dominant form of driver training among 18-year-old nov-
    ice drivers, who can be estimated to represent approx. 40 per cent of the learner drivers obtaining a
    class B/BE driving licence (cf. Fig. 1). The proportion of AD17 participants, referred to all those ob-
    taining a class B/BE driving licence, was already 25 per cent in 2007, and a further significant increase
    in this figure to 35 per cent was then recorded in 2008. The rapid and comprehensive propagation of
    the AD17 model is not only an indicator of the high degree of acceptance of the model among novice
    drivers and their parents, but also evidence of the practicability of the underlying approach, as is also
    supported by numerous results of the process evaluation.
    Socio-structural characteristics
    The AD17 model participants were found to display certain socio-structural differences compared to
    a representative sample of all 17-year-old youths in Germany – like 18-year-old novice drivers in
    general, but here now somewhat more distinctively. At the time of the sampling in 2007, more par-
    ticipants were from families with a better educational or economic background.
    Motivation to participate in the model
    Participation in the accompanied driving model was overwhelmingly a decision of the young drivers
    themselves. Over and above that, parents and – to a slightly lesser extent – age cohorts were impor-
    tant sources of encouragement. The motivation for participation was determined above all by per-
    sonal interest in the ability to drive. Safety considerations were similarly one strong motive. Assis-
    tance and support for their children and everyday practical concerns, for example lower vehicle in-
    17
    surance premiums, were the most widespread motives in the parent population, closely followed by
    safety aspects. The reasons stated by participants and their parents are thus complemental and bear
    witness to the relevance and good practicability of accompanied driving in this age-specific phase of
    the life cycle.
    0
    5.000
    10.000
    15.000
    20.000
    25.000
    30.000
    35.000
    40.000
    45.000
    Jan 04 Apr 04 Jul 04 Okt 04 Jan 05 Apr 05 Jul 05 Okt 05 Jan 06 Apr 06 Jul 06 Okt 06 Jan 07 Apr 07 Jul 07 Okt 07 Jan 08
    nach Teilnahme
    am BF17
    Summe
    herkömmlicher
    Führerscheinerwerb
    Legend:
    Unfilled markers on dashed line: Persons obtaining a conventional driving licence; trend line: 4th order polynomial
    Filled markers on solid line: AD17 participants; trend line: 4th order polynomial
    Solid line without markers: Total; trend line: 3rd order polynomial
    Fig. 1: Development of the monthly numbers of persons between the ages of 18 years and 18 years
    and 3 months receiving a card driving licence for vehicle class B/BE in Germany, divided into
    those who participated in the AD17 model and those who obtained their driving licence in
    the conventional manner (Source: Schade et al. 2008)
    Accompanists
    The novice drivers generally had no difficulty finding a suitable accompanist. In most cases, two ac-
    companists were specified on the driving test certificate. Over the whole duration of the accompani-
    ment phase, parents dominated as the most frequent accompanists. In almost two-thirds of cases,
    the accompanist was the driver's own mother, belonged to the 40- to 49-year age group and lived in
    the same household as the novice driver. The person first mentioned as the most frequent accompa-
    nist generally remained the same over the whole course of the accompaniment phase. Any further
    accompanists were predominately also members of the novice driver's own family.
    18
    Preparation for participation in the model
    Only 2.5% of the participants and 3.8% of their most frequent accompanists visited an optional pre-
    paratory course in advance of the participation in the AD17 model. The new model was instead
    rather a topic for informal discussion and exchange.
    Actual duration of the accompaniment phase
    The majority of the young drivers was not able to utilise the maximum accompaniment duration of
    12 months. An average of five months passed between the candidate's 17th birthday and their re-
    ceiving a driving test certificate. The late granting of a driving licence was due either to delayed
    commencement of driver training or to a longer duration of training. The average duration of actual
    accompanied driving was between seven and eight months, with the quarter of participants with the
    shortest accompaniment phase specifying a duration of accompanied driving practice up to five
    months, and the quarter of participants with the longest accompaniment phase a duration of ac-
    companied driving practice of eleven or twelve months.
    Days of driving mobility
    The participants drove a vehicle with their accompanist on an average of 3.1 days during their first
    report week. The average number of days of driving mobility decreased slightly with increasing dura-
    tion of the accompaniment phase.
    Distances driven
    The novice drivers reported an average daily distance driven (median) of 9.3 km (arithmetic mean:
    13.2 km). These figures also take into account those days on which no driving was done. The median
    of daily driving from Monday to Friday was 7.8 km (arithmetic mean: 11.7 km). Longer distances
    were driven on Saturdays and Sundays (median: 8.5 km, arithmetic mean: 17.0 km) than during the
    week.
    The median of the average daily distance driven by mobile participants was 24.0 km (arithmetic
    mean: 32.4 km). Mobile novice drivers recorded a daily median of 20.8 km driving from Monday to
    Friday (arithmetic mean: 28.5 km), and a daily median of 25.0 km driving on the two days at week-
    ends (arithmetic mean: 41.0 km).
    Accumulating the figures, an average weekly distance driven (median) of 65.0 km (arithmetic mean:
    92.2 km) was documented for the whole group of model participants. Mobile participants drove on
    average 71.7 km per week (median; arithmetic mean: 102.3 km).
    Accordingly, AD17 participants drove on average 318.5 km per month (median: 260.4 km). This extra-
    polated monthly distance driven remained essentially constant over the course of the accompani-
    ment phase. Participants with an average accompaniment duration of eight months recorded
    approx. 2,400 km of driving practice over this period. Driving practice of the order of 3,800 km could
    thus be extrapolated for an accompaniment duration of twelve months.
    The comparison of driving mobility patterns between the 17-year-old model participants and “nor-
    mal” 18-year-old novice drivers reveals significant differences: Whereas accompanied driving is de-
    voted to practice, the development of individual driving mobility is shifted to the focus of independ-
    ent driving from the age of 18 onwards, reflected, for example, in a higher proportion of night-time
    leisure mobility at weekends.
    19
    Driving time
    On all weekdays, most accompanied journeys lasted merely up to 15 minutes, though on several
    days there was only little margin to those cases which reported a driving time of up to 30 minutes.
    Longer periods of driving were to be noted above all on Saturdays and Sundays. Due to the many
    days without any accompanied driving at all, half of all model participants – including the immobile
    participants – recorded an overall average daily driving time of no more than 9.6 minutes (median).
    Counting only the days on which they actually drove, half of the mobile participants reported a driv-
    ing time of up to 25.0 minutes, while the other half was on the road for even longer.
    Journey purpose
    The four dominant journey purposes or destinations for accompanied driving were private journeys
    (family, visits), household errands, leisure purposes and journeys to either school or the place of
    training/employment. This shows how participation in the accompanied driving model was embed-
    ded in the everyday activities of the novice drivers and their families.
    Passengers
    For each day from Monday to Thursday, more than two-thirds of the mobile model participants re-
    ported that they had driven with no other passengers beside the accompanist, whereas this applied
    to only around half of the drivers on Sundays. The more frequent “private” journeys at the weekend
    were reflected in a higher proportion of further adult or also younger passengers on those days.
    Roads used
    Throughout the model period, journeys within built-up areas dominated. Cross-country roads and
    motorways were used most frequently on Sundays.
    Interaction between novice driver and accompanist
    An analysis of the results regarding the interaction between novice driver and accompanist provided
    evidence of an appropriate interpretation and exercising of the assigned roles, on the part of both
    the novice drivers and the accompanists, in the sense of a constructive partnership to promote the
    acquisition of practical driving competence.
    During their journeys, the novice drivers perceived their accompanists as attentive passengers and
    reliable partners for verbal exchanges. The aspect of enhanced safety was not placed in the fore-
    ground by the novice drivers. In most cases, there was no explicit preparatory or evaluative discus-
    sion of the accompanied driving session. Approx. three in ten participants reported calming verbal
    intervention on the part of the accompanist (e.g. the advice to take a break or drive more slowly).
    Nevertheless, the participants were fully aware of their own responsibility as drivers. The support of
    the accompanist with regard to hazard perception did not intervene in the drivers' independent deci-
    sion-making responsibility. Situation-specific intervention by the accompanist, e.g. remarks concern-
    ing excessive speed, point to the generally protective character of accompanist behaviour.
    With increasing duration of the accompaniment phase, the approval for the presence of the accom-
    panist declined. The young drivers no longer acknowledged the attentiveness of the accompanist to
    the same extent as at the beginning, and the latter's support in terms of alertness to hazards, critical
    feedback and cautionary explanation was similarly only perceived to a lesser degree. Conversely, the
    participants' emphasis of their own responsibility as drivers increased. These developments all corre-
    20
    spond to the accordingly increased driving experience of the model participants as the accompani-
    ment phase progresses.
    The majority of participants and accompanists detected no influence on family relationships through
    the joint efforts embodied in the accompanied driving model. Positive effects, where at all signifi-
    cant, were seen by accompanying fathers.
    The role played by the accompanists apparently exerted influence also on their own safety-related
    driving behaviour. A considerable proportion of the accompanists stated that their participation in
    the accompanied driving model had led to them refreshing their own knowledge of traffic rules, pay-
    ing more attention to road safety issues, and themselves driving with a greater safety awareness. In
    this respect, the accompanied driving model assumes the character of a multi-generational road
    safety instrument.
    Subjective indicators of experience-building
    Already during the second wave of questioning, the model participants indicated a feeling of in-
    creased driving experience. Those with the most actual driving practice were praised significantly
    more often by their accompanists for their improved vehicle control and more foresighted, com-
    posed and confident driving.
    The impact of greater driving experience was also revealed in the fact that those participants with
    more driving practice described a feeling of uncertainty in difficult traffic situations less often than
    their counterparts with little driving practice. A comparison between former and active model par-
    ticipants on the basis of the latter's' initial questionnaires supplied indication that a period of up to
    three months of accompanied driving does not yet lead to a perception of growing driving experi-
    ence. In the trend results, an increase in driving experience was expressed in the significant decline in
    the proportion of novice drivers who felt unsafe in certain traffic situations over the course of time.
    Vehicles used
    The vehicles used within the framework of accompanied driving were on average 6.7 years old, had
    clocked a mileage of slightly more than 90,000 km and were most frequently rated with an engine
    power between 51 and 80 kW (69 to 109 hp).
    Compared to the vehicles of other 18-year-old novice drivers, the vehicles driven by model partici-
    pants were newer and powered by larger engines. This fact indicates the use of vehicles already pre-
    sent in the family of the participant, and thus vehicles which are more modern and equipped with
    more safety features that those typically driven by novice drivers.
    Traffic offences and accident involvement
    The few self-reported traffic offences leading to a fine being imposed on the model participant are a
    good indicator for the general law abidance of the AD17 driver population.
    According to their own information, the model participants were only very seldom involved in acci-
    dents. The limited scope of the process evaluation sample, however, does not permit generalisation
    of these results.
    21
    The data received from the federal states on conspicuous driving behaviour within the framework of
    the accompanied driving model, which were similarly incorporated into the process evaluation, also
    provided a clear indication of faithful abidance by the law and a low accident prevalence.
    Recommendation of the AD17 model to others
    Almost all the young drivers questioned replied that they would recommend participation in the
    accompanied driving model to others, the decisive reason mentioned being the possibility to build up
    more extensive practical experience before commencing independent driving.
    Conclusions for practical implementation
    The process evaluation of the nationwide experimental introduction of the model “Accompanied
    driving from 17” identifies the structures which have developed through this new approach to novice
    driver preparation in the context of the life realities of the young participants and their families.
    Against the background of the empirical results, the following conclusions and recommendations can
    be formulated:
    - Suitability of the model approach
    The AD17 model has demonstrated its suitability in respect of the central criteria of accessibility,
    practicability and safe implementation. Scope for optimisation is revealed in various individual as-
    pects.
    - Better utilisation of the additional opportunities for preparation
    Those young persons who wish to drive independently from the age of 18 years should be encour-
    aged to make the fullest possible use of the additional opportunities for preparation offered within
    the framework of the AD17 model. This means above all early commencement of their driver training
    and correspondingly early completion of the driving test, preferably already before their 17th birth-
    day. Furthermore, it calls for resolute exploitation of the opportunity for driving practice within the
    given private framework, for which – as the results of the evaluation show – the prerequisites are
    usually good, particularly with regard to the motivation of those involved.
    The conditions for better utilisation of the opportunities for preparation offered by the AD17 model
    can generally be improved by all measures serving to raise the level of information and further pro-
    mote motivation. In this connection, it is recommended that the instrument of accompanied driving
    be made a central topic of the active safety-relevant communication geared to the target group of
    prospective drivers and their parents. Further thought should be given, in particular, to closer coop-
    eration with schools. The use of foreign-language information media is similarly to be considered as a
    means to address corresponding target groups.
    - Optimisation of time management with regard to driver training
    Professional public relations work could bring further positive influence to bear on opinion formation
    with regard to both participation and support in the role of accompanist. Better knowledge of the
    possibilities for early granting of a driving licence could also serve to improve the young person's
    time management well in advance of his or her 17th birthday by encouraging an earlier commence-
    ment of the accompaniment phase.
    22
    - Number of accompanists entered in the driving test certificate
    As the potential availability of an accompanist is improved with an increasing number of accompa-
    nists entered in the driving test certificate, novice drivers should be recommended to nominate a
    sufficient number of persons for this role.
    - Preparation and accompanying support for participants: School road safety programmes,
    preparatory events, handbook and Internet
    School workshops and integration of the model into school road safety programmes are suitable
    means to provide corresponding information at an early stage and to enable young people to fit the
    extended period for development of driving experience into the schedules of their youth-specific life
    cycle.
    A mandatory introductory event for novice drivers and accompanists, e.g. in the form of an informa-
    tion evening, could prove a barrier to AD17 participation. The recommendation is consequently a
    voluntary offer of an introductory event and the parallel development of informative and creative
    accompanying media for novice drivers and accompanists, which could be conveyed to the target
    groups via various channels, such as driving schools, insurance companies, road safety campaigns
    and the Internet.
    The concept for a standard information package, serving as a handbook for the potential participants
    and their parents, should also be discussed. Given the particular affinity of the youth target group for
    the medium Internet, it appears expedient to design and maintain an Internet platform for the AD17
    model. Information presented in a visually appealing form (layout, integration of videos, etc.) could
    explain the scientific background and methodical purpose of the extended period of novice driver
    preparation – i.e. the correlations between driving practice, competence-building and accident risk –
    in a readily understandable fashion. At the same time, it could provide tips on the openings for early
    entry into the scheme of accompanied driving, advice on optimum realisation of diverse driving
    situations, and a forum for discussion and exchanges of the experience gained with the model, to
    mention just a few of the possibilities.
    - Offers of pedagogical support
    There is currently still little discussion of promotion for the accompanied driving model in the form of
    professional pedagogical support through the driving schools and driver safety training schemes. The
    AD17 model has introduced a completely new field into the existing system of novice driver prepara-
    tion, characterised by new participants, the consumption of considerable human and time resources,
    and new forms of activity. This new field of activity is closely associated with the established forms of
    professional driver training and is geared to the same objective, namely improvement of the initial
    practical competence of novice drivers. The participants in the accompanied driving model – accom-
    panists and novice drivers – should thus be consulted to determine the level of fundamental interest
    in offers of constructive professional support, as a basis for further consideration on the part of the
    relevant providers.
    - Further research
    It is suggested that a research project be installed to investigate the potential differences between
    AD17 participants and non-participants. The data collected through a survey addressing these two
    groups would permit a differentiated study of the motivation determining the learner driver's deci-
    sion for or against participation in the AD17 model.
    23
    The process evaluation of the pilot scheme considers a large sample of novice drivers and an exten-
    sive pool of corresponding background data on everyday practical implementation of the model. At
    the beginning of the evaluation, each novice driver consented to the retrieval of personal data re-
    cords from the Central Register of Traffic Offenders (VZR). Combination of these two data sources
    could permit deeper insights in the sustained effects of participation in the model even after the
    novice driver reaches adulthood.
    5.4 Significance of the results
    The considerable expansion of practical novice driver preparation in the context of the “Accompa-
    nied driving from 17” model has led to structural changes in the system of driver training in Ger-
    many. Nevertheless, it would appear that the possibilities are yet to be exploited to the full with re-
    gard to the duration of the accompaniment phase and the scope of actual driving practice. These
    aspects are recommended as topics for future optimisation efforts, for which purpose a basis may be
    found in the diverse results of the process evaluation.
    Alongside more effective utilisation of the inherent potential of the model, namely longer-term de-
    velopment of practical driving experience, it would also be expedient to promote meaningful inter-
    actions between the AD17 model approach and other similarly targeted measures within the frame-
    work of an integrated system of novice driver preparation in Germany.
    6. Road safety effectiveness of the AD17 model: Results of the summative
    evaluation
    6.1 Objective
    The principal objective for the summative evaluation of the AD17 model is to verify the effectiveness
    of the underlying approach of a greater scope of driving practice within the framework of novice
    driver preparation as a means to achieve improved driving and traffic competence, subsequently
    reflected in a reduced risk of accident involvement and a reduced likelihood of committing traffic
    offences at the start of an independent driving career. To determine the directly causal impact of the
    model, a comparison of participating and non-participating novice drivers must also consider and
    take into account the possibly distorting influences of model-independent factors. Furthermore, sec-
    ondary effects of the model may become significant in other areas beyond the development of driv-
    ing competence, above all in connection with altered mobility structures and the correspondingly
    changed accident situation – for example increased practice mobility during the accompaniment
    phase or greater demand for an early start to independent driving mobility already at the age of 18
    years.
    6.2 Studies presented to date
    Results referring to the road safety effectiveness of the AD17 model have been presented in three
    studies to date. Two of these studies are sub-project reports with intermediate results from the BASt
    evaluation of the AD17 model, processed by the Department for Statistics at the Federal Motor
    Transport Authority (KBA). The third study refers to the experimental scheme implemented at re-
    gional level in the federal state of Lower Saxony before introduction of the national model.
    24
    The fundamental approach common to all three studies is a comparison of safe driving behaviour
    between novice drivers who have taken advantage of the option of accompanied driving and those
    trained in the conventional manner (driver training exclusively in a driving school) during the initial
    phase of independent driving. The studies differ, nevertheless, in respect of several relevant para-
    meters, for example the period under observation, the data used to depict driving behaviour, and
    their consideration of confounding influences which could distort the measured impact of the model.
    The following table provides an overview of the aforementioned studies.
    6.3 Results
    In all three studies, a two-figure percentage reduction in the rates of accident involvement and traffic
    offences was determined as attributable to accompanied driving.
    The study conducted in Lower Saxony (Stiensmeier-Pelster, 2007) compared 4,454 AD17 participants
    and 2,421 conventionally trained novice drivers over their first 18 months of independent driving on
    the basis of data records referring to traffic offences or accidents in the Central Register of Traffic
    Offenders (VZR). This revealed that the AD17 participants committed 22.5% fewer traffic offences
    and were involved in 28.5% fewer accidents. A second objective of the study was to analyse the ef-
    fect of a longer or shorter period of actual practice during the maximum one-year accompaniment
    phase. The AD17 participants with an accompaniment phase of more than 6 months were found to
    have committed 23.1% fewer traffic offences and were involved in 57.0% fewer accidents than those
    whose accompaniment phase had been less than 6 months. All the aforementioned differences were
    determined to be statistically significant.
    Study Objectives Data evaluated Observed period after
    commencement of
    independent driving
    Stiensmeier-Pelster,
    2007 (evaluation study
    commissioned by the
    State Minister of Trans-
    port of Lower Saxony)
    Comparison of frequency
    of traffic offences and
    accident involvement
    between AD17 partici-
    pants and conventionally
    trained novice drivers
    Comparison of AD17
    participants with longer
    and shorter accompani-
    ment phases
    Participants in an ex-
    perimental scheme in the
    federal state of Lower
    Saxony before national
    implementation of the
    AD17 model
    Data records from the
    Central Register of Traffic
    Offenders (VZR)
    Months 1 to 18
    Schade et al., 2007 (study
    contributing to BASt
    project 82.0316/2006
    “Summative evaluation
    of accompanied driving”)
    Comparison of frequency
    of traffic offences and
    accident involvement
    between AD17 partici-
    pants and conventionally
    trained novice drivers
    Participants in the na-
    tionally implemented
    AD17 model from several
    federal states
    Data records from the
    Central Register of Traffic
    Offenders (VZR)
    Months 1 to 3
    Schade & Heinzmann,
    2009 (study contributing
    to BASt project
    82.0316/2006 “Summa-
    tive evaluation of ac-
    companied driving”)
    Comparison of frequency
    of traffic offences and
    accident involvement
    between AD17 partici-
    pants and conventionally
    trained novice drivers
    Participants in the na-
    tionally implemented
    AD17 model from several
    federal states
    Self-reported accidents
    and traffic offences
    Months 1 to 12
    25
    Comparison of AD17
    participants with greater
    and lesser actual driving
    practice during the ac-
    companiment phase
    above a defined rele-
    vance threshold
    Tab. 7: Studies presented up to the end of 2009 with results pertaining to the road safety effec-
    tiveness of the AD17 model
    For the first intermediate report within the framework of the BASt evaluation project (Schade et al.,
    2007), two groups of 7,500 persons each were selected at random to compare the safe driving be-
    haviour of AD17 participants and conventionally trained novice drivers, again taking data records
    referring to traffic offences or accidents in the Central Register of Traffic Offenders (VZR) as indica-
    tors. External scheduling requirements limited the observed period to the first three months of inde-
    pendent driving. The results pointed to an approx. 20% higher risk of traffic offences and a 30%
    higher risk of accident for the conventionally trained novice drivers compared to the AD17 partici-
    pants. As the period of observation was still deemed too short, however, these results could not yet
    be validated statistically and were thus initially interpreted by the authors merely as a trend in favour
    of the safety effectiveness of the AD17 model.
    The third study (Schade & Heinzmann, 2009) analysed the driving behaviour of a total of over 18,000
    eighteen-year-old drivers in the first year of their independent driving of a motor vehicle. Two sam-
    ples of drivers of the same age were selected at random from the Central Register of Driving Licences
    (ZFER): Firstly, AD17 novice drivers who had completed at least a 3-month phase of accompanied
    driving, and secondly, “normal” novice drivers who had obtained a conventional driving licence.
    The evaluation takes into account all self-reported accident involvement and traffic offences above a
    defined relevance threshold (accidents: police called to record the accident, estimated damage of at
    least €1,200, injury to persons; traffic offences: all offences punished with a fine of more than €25).
    On the basis of these data, the rates of significant accident involvement and significant traffic of-
    fences were determined both on a time-related basis, i.e. per 1,000 drivers and year, and according
    to the scope of actual driving practice, i.e. per million kilometres driven.
    Compared to the group of conventionally trained novice drivers, the AD17 participants were seen to
    be involved in 19 per cent fewer significant accidents and committed 18 per cent fewer significant
    traffic offences per 1,000 novice drivers and year. On the basis of actual driving practice (kilometres
    driven), the figure for significant accident involvement was even 23 per cent lower in the AD17
    group, while significant traffic offences were reduced by 22 per cent.
    In the course of a more detailed evaluation, the possible influence of various model-independent
    factors was analysed, in order to identify spurious correlations or effects which could mask those of
    accompanied driving – e.g. an uneven distribution of male and female drivers in the analysis groups.
    The confounding variables considered were the factors gender, school education background, place
    of residence, positive parental role model and vehicle availability, alongside the combinations nega-
    tive parental role model/lower education background and higher education background/limited ve-
    hicle availability. After taking into account the influences of these factors, adjusted values were de-
    rived for the difference between the AD17 group and the group of conventionally trained novice
    drivers. In all cases, these remaining differences were shown to be statistically either very or highly
    significant (cf. Tab. 8).
    26
    Besides the verification of significantly lower rates of accident involvement and traffic offences
    among novice drivers who had taken part in the accompanied driving scheme, the study also re-
    vealed indications that the extent of actual driving practice during the accompaniment phase corre-
    lates positively with later driving behaviour in the sense of a “dose-response relationship”.
    6.4 Significance of the results
    The evidence for a two-figure percentage reduction in the rates of accident risk and traffic offences
    among novice drivers during the initial phase of independent driving, as to be found in the various
    (partial) evaluations presented to date, documents a considerable road safety effectiveness of the
    accompanied driving model, and indeed a scope of safety relevance far beyond that anticipated in
    the context of novice driver measures to date.
    Statistical variable AD17 group Conventional
    driver training
    Difference
    between
    AD17 and
    conventional
    driver training
    Difference be-
    tween AD17 and
    conventional
    driver training
    taking into ac-
    count confound-
    ing variables
    Rate of accident involvement
    per 1,000 drivers and year
    89.4 110.5 -19% -17%***
    Rate of traffic offences
    per 1,000 drivers and year
    66.6 81.6 -18% -15%**
    Rate of accident involvement
    per million kilometres driven
    10.87 14.07 -23% -22%***
    Rate of traffic offences
    per million kilometres driven
    8.10 10.38 -22% -20%***
    Legend: n.s. = Difference to parity (1.0) not significant; * = statistically significant (p ≤ 0.05);
    ** = statistically very significant (p ≤ 0.01); *** = statistically highly significant (p ≤ 0.001);
    Tab. 8 : Rates of significant accident involvement and traffic offences in the analysis groups per
    1,000 drivers and year and per million kilometres driven
    In Germany, empirical proof of model-specific road safety effectiveness has so far only been fur-
    nished with regard to the probationary driving licence introduced in 1986. Compared to the accom-
    panied driving model, however, the extent of the attested safety gain attributable to the instrument
    of a probationary driving licence is considerably narrower: On the basis of the evaluation results, a
    model-specific reduction in road safety risks of 5 per cent is assumed, but that applicable exclusively
    to male novice drivers in urban contexts (cf. Meewes & Weißbrodt, 1990).
    The present findings on the road safety effectiveness of the accompanied driving model stem from
    different, in many respects independent studies on the basis of extensive samples. The validity of the
    result tendencies which have emerged in all the (partial) evaluations to date, namely the indication
    of a significant road safety effectiveness of the accompanied driving model, is supported in particular
    by the careful consideration given to the influences of confounding variables in the special evaluation
    27
    presented recently by Schade and Heinzmann (2009), and by the critical methodical appraisal of the
    results obtained (cf. the original report contained in the annex to this evaluation).
    This notwithstanding, the scientific findings presented to date have still not been able to clarify all
    questions relating to the effectiveness of the accompanied driving model. It remains essentially
    open, for example, whether and to what extent the risk-reducing effects of the accompanied driving
    model are preserved beyond the first year of independent driving, and likewise whether secondary
    effects lead to a parallel increase in risk within the framework of changed mobility structures.
    7. Conclusion
    The overall evaluation of the VFT model was unable to confirm the road safety effectiveness of this
    approach. No improvements were achieved in the vast majority of the road safety attitude domains
    addressed by the VFT model. Given the lack of attitude effectiveness, it is also not possible to assume
    that the VFT model exerts an effective influence on behaviour.
    It is especially disturbing to learn that participants in the VFT model display a significantly greater risk
    of accident involvement and traffic offences compared to novice drivers of the same age and with
    similar driving experience who have not attended a VFT seminar. Empirical clarification of the origin
    of this negative result will require more detailed studies on the basis of a more extensive set of data.
    Possible explanations, insofar as a directly causal effect of the VFT participation is excluded, can most
    reasonably be assumed to lie in the curtailed effectiveness of the probationary licence rules in the
    case of VFT participants (participation is honoured with a shortening of the probationary period by
    up to one year) and self-selection effects in conjunction with VFT participation (model may attract
    above all those novice drivers with a tendency to conspicuous driving behaviour, as a means to
    achieve a shortening of the probationary period). In view of the significantly poorer driving behaviour
    of the VFT participants, and against the background of the aforementioned plausibility considera-
    tions regarding the underlying causes, however, it seems expedient to already now remove the in-
    centive of a shorter probationary period in case of VFT participation.
    The evaluation of practical implementation of the VFT model revealed need for further development
    at several points. This refers to the quality of the active, attitude-building training forms to be applied
    by the seminar leaders and moderators, as well as questions concerning optimised seminar organisa-
    tion in the interest of training quality. It is here recommended that the seminar concept be subjected
    to a thorough review, and that the conditions for concept-adequate implementation be improved.
    The evaluation results published to date with regard to the road safety effectiveness of the AD17
    model show evidence of a model-related two-figure percentage reduction in the rates of accident
    risk and traffic offences during the initial phase of independent driving from the age of 18 years. A
    special intermediate report presented on 30.11.2009 within the framework of the ongoing BASt
    evaluation revealed a 22 per cent lower accident risk and 20 per cent fewer traffic offences for AD17
    participants in their first year of independent driving, in each case referred to the scope of kilometres
    driven. Further relevant factors influencing driving behaviour, e.g. gender, were also taken in ac-
    count, so as to permit a statement on the solely model-related effectiveness. The figures document a
    considerable enhancement of novice driving competence as a result of the period of accompanied
    driving. Even so, the findings have still not been able to provide a conclusive answer to all questions
    28
    relating to the effectiveness of the AD17 model. It remains to be clarified, whether and to what ex-
    tent the model of accompanied driving leads to changes in mobility patterns (mobility during the
    accompaniment phase, increased demand for driving licences at an earlier age), and which additional
    risk loads this may entail. Results addressing these questions and likewise the road safety gains from
    accompanied driving during the second year of independent driving practice can only be presented
    with the final results of the BASt evaluation in autumn 2010.
    The AD17 model has already attracted a rapid and widespread response throughout Germany during
    its trial phase. This testifies to the high level of acceptance among novice drivers and their parents,
    and is at the same time indication of the practicability of the underlying approach. This is supported
    comprehensively by the diverse results of the process evaluation.
    The considerable expansion of practical novice driver preparation in the context of the accompanied
    driving model has led to structural changes in the system of driver training in Germany. Nevertheless,
    it would appear that the possibilities are yet to be exploited to the full with regard to the duration of
    the accompaniment phase and the scope of actual driving practice, and these aspects can be rec-
    ommended as topics for future optimisation efforts.
    The results of the evaluation projects addressing the experimental novice driver training models
    “Voluntary further training seminars for holders of probationary driving licences” and “Accompanied
    driving from 17” are thus now available as a basis for assessment of these models and for the pend-
    ing discussions on further development of the system of novice driver improvement in Germany.
    They are to be integrated directly into a BASt project which is aimed at elaboration of a framework
    concept, and thus the technical and professional foundations for the further development of novice
    driver preparation, in cooperation with experts representing scientific research, traffic politics and all
    those concerned with practical implementation.
    29
    8. Bibliography
    EU-Project ADVANCED (2002). Description and analysis of post licence driver and rider training. Final
    report. Brussels: CIECA.
    Funk, W. et al. (2009). Begleitetes Fahren ab 17 Jahre – Prozessevaluation des bundesweiten Modell-
    versuchs. Abschlussbericht zum Forschungsprojekt FE 82.298/2005 der Bundesanstalt für Straßenwe-
    sen, Bergisch Gladbach.
    Funk, W. et al. (2009). Gestaltung der Übungspraxis im Begleiteten Fahren. Abschlussbericht zum
    Forschungsprojekt FE 89.221/2009 der Bundesanstalt für Straßenwesen, Bergisch Gladbach.
    Heinzmann. H.-J., Schade, F.-D. (2009). Überprüfung der Verkehrsbewährung von FSF-Teilnehmern
    auf der Grundlage von VZR-Daten. Zwischenbericht zum Forschungsprojekt FE 89.226/2009 der Bun-
    desanstalt für Straßenwesen, Bergisch Gladbach.
    Kerwien, H. (2009). Absenkung des Fahranfängerrisikos durch freiwilliges Dazulernen. Formative Eva-
    luation des Modells “Freiwillige Fortbildungsseminare für Fahrerlaubnisinhaber auf Probe” (FSF).
    Abschlussbericht zum Forschungsprojekt FE 82.264/2004 der Bundesanstalt für Straßenwesen, Ber-
    gisch Gladbach.
    Meewes, V. & Weissbrodt, G. (1992). Führerschein auf Probe – Auswirkungen auf die Verkehrssi-
    cherheit (Schriftenreihe der Bundesanstalt für Straßenwesen, Reihe “Unfall- und Sicherheitsfor-
    schung Straßenverkehr”, Heft 87). Bremerhaven: Wirtschaftsverlag NW.
    Schade, F.-D., Heinzmann, H.-J. (2007). Summative Evaluation des Begleiteten Fahrens ab 17. Zwi-
    schenbericht. Forschungsprojekt FE 82.0316/2006 der Bundesanstalt für Straßenwesen. Juli 2007,
    Kraftfahrt-Bundesamt, Abteilung Statistik, Flensburg.
    Schade, F.-D., Heinzmann, H.-J. (2009). Summative Evaluation des Begleiteten Fahrens ab 17. Son-
    derauswertung: Erste Evaluationsergebnisse auf Basis der selbstberichteten Verkehrsauffälligkeit.
    Forschungsprojekt FE 82.0316/2006 der Bundesanstalt für Straßenwesen. November 2009, Kraft-
    fahrt-Bundesamt, Abteilung Statistik, Flensburg.
    Schulz, S.-O., Henning, H.J. & Chaselon, F. (1995). Jugend fährt sicher. Wirksamkeit des Modellver-
    suchs. Abschlussbericht zum Forschungsprojekt FP.2.9124 der Bundesanstalt für Straßenwesen. Ber-
    gisch Gladbach.
    Sindern, E.M. & Rudinger, G. (2009). Evaluation der Freiwilligen Fortbildungsseminare für Fahranfän-
    ger (FSF). Wirksamkeitsuntersuchung. Abschlussbericht zum Forschungsprojekt FE 82.307 der Bun-
    desanstalt für Straßenwesen, Bergisch Gladbach.
    Stiensmeier-Pelster, J. (2008). Abschlussbericht zum Niedersächsischen Modellversuch “Begleitetes
    Fahren ab 17”. Giessen: Justus-Liebig-Universität, Fachbereich 06.
    30
    Annex
    Summative Evaluation of
    “Accompanied Driving from 17”
    Special evaluation:
    First evaluation results
    on the basis of self-reported driving behaviour
    Research Project FE 82.0316/2006
    of the Federal Highway Research Institute
    Franz-Dieter Schade
    Hans-Jürgen Heinzmann
    Federal Motor Transport Authority,
    Dept. for Statistics Flensburg, 30.11.2009
    Contents
    Foreword.........................................................5
    1 Objectives..................................................5
    2 Method ......................................................8
    3 Results obtained with regard to self-
    reported driving behaviour.......................9
    4 Critical methodical evaluation...............11
    4.1 Are the analysis results statistically
    significant?............................................ 11
    4.2 Are the AD17 and cDL analysis
    groups strictly comparable and can
    the results be generalised?................... 11
    4.3 Do any “external” variables exert
    distorting influences?........................... 11
    4.4 Are the results distorted by non-
    cooperating participants?.................... 18
    4.5 Does the model possibly achieve
    merely a temporary effect? ................. 19
    4.6 What evidence supports a (causal)
    effect of the AD17 model? ................... 20
    4.7 Are factors known which could
    further enhance the AD17 effect?....... 23
    4.8 Is self-reported behaviour a
    sufficiently valid basis for
    evaluation?............................................ 24
    4.9 Does the knowledge of participation
    in a road safety study exert a
    systematic positive influence on
    driving behaviour?............................... 24
    5 Summary and conclusion ......................25
    Bibliography .................................................26
    Abbreviations:
    AD17 Accompanied driving from
    the age of 17
    AD17 group Analysed sample of persons
    who obtained their driving
    licence under the AD17
    model
    cDL group Analysed sample of persons
    who obtained their driving
    licence in the conventional
    manner (“cDL”)
    VZR Verkehrszentralregister –
    Central Register of Traffic
    Offenders
    ZFER Zentrales Fahrerlaubnis-
    register – Central Register
    of Driving Licences
    5
    Foreword
    The following exposition, which builds upon the
    method-oriented intermediate report on the present
    project “Summative Evaluation of Accompanied
    Driving from 17” of 31.07.2008, similarly possesses
    the status of merely an intermediate report. It is
    planned to present a final report in autumn 2010,
    as soon as all envisaged sources of data have
    been utilised and evaluated.
    The results presented here can nevertheless be
    considered final and conclusive in the sense that
    they cover the whole aspect of self-reported behav-
    iour on the part of the survey participants. In other
    words, the data source “Participant survey”, with all
    responses with regard to traffic accidents and traf-
    fic offences during the first year of independent
    driving, was available in its entirety as a basis for
    the present evaluation. The final report will thus
    also contain no fundamentally different findings.
    This notwithstanding, the results presented here
    may still be supplemented or relativised by those of
    two outstanding study tasks: Firstly an analysis of
    the data records pertaining to the survey partici-
    pants which are held in the Central Register of
    Traffic Offenders (VZR), and secondly an analysis
    of VZR data records pertaining to selected driver
    groups which were retrieved independently of the
    survey – and without the knowledge of the persons
    concerned – and subsequently processed anony-
    mously1
    . These analyses require a longer period of
    time, because the corresponding VZR entries only
    become available for consideration with a delay of
    up to twelve months after the traffic offence or ac-
    cident.
    An analysis of the VZR data records is an impor-
    tant and indispensable element of the evaluation.
    The first of the two analyses serves to safeguard
    the findings against the influences of subjective
    memory effects or even falsification tendencies, the
    second to take into account possible distortions
    arising from the participants’ knowledge of the
    study objectives. The overall evaluation can only be
    viewed as verified after presentation of these two
    additional analyses. The statements of the present
    report2
    are thus to be read under this important
    proviso.
    1
    By replacing names and reference numbers with a random
    project number
    2
    The authors would like to thank Mr. Willmes-Lenz (BASt)
    and Ms. Bremer (KBA) for their reviewing of the initial draft
    of 30.09.09 and for their valuable comments and amend-
    ments.
    1 Objectives
    There are three possible effects which could con-
    ceivably be observed following implementation of
    the model of “Accompanied driving from 17”
    (AD17). These effects are to be taken into account
    in an evaluation. For the sake of maximum clarity,
    they are here to be described separately, although
    the individual effects will to a certain extent be su-
    perimposed in practice:
    Hypothetical effect 1: Model expands the at-risk
    population
    The introduction of new possibilities generally leads
    to an increased demand, because new target
    groups are addressed. Persons who would other-
    wise have obtained a driving licence later, or possi-
    bly not at all, may be especially receptive for the
    new AD17 model. Consequently, the number of 18-
    year-old drivers increases, and with it the number
    of traffic accidents involving this age group (Fig. 1,
    right-hand block compared to left-hand block).
    Hypothetical effect 2: Model leads to internal differ-
    entiation into “good” and “poor” risks3
    The introduction of new possibilities, and thus of
    additional selection options, generally leads to dif-
    ferentiation within the target group. Persons repre-
    senting so-called protective factors – e.g. female
    and higher-level school education – and a corre-
    spondingly lower accident risk may be concen-
    trated in the group of “AD17 drivers”. Those with a
    higher accident risk, accordingly, then form the
    bulk of the group which obtains a driving licence in
    the conventional manner (“cDL drivers”). The total
    number of accidents remains unaffected by the in-
    ternal differentiation (Fig. 2).
    In the chosen example, the number of accidents
    involving AD17 drivers is 10,000 less than to be
    expected after division into two equal groups, that
    for the cDL drivers correspondingly 10,000 more.
    Hypothetical effect 3: Model itself serves as a pro-
    tective factor
    Certain components of the AD17 model exert posi-
    tive (causal) influences on the participating drivers
    and thus lower their accident risk (Fig. 3). The ac-
    cident situation of the cDL drivers remains un-
    changed and corresponds to their proportion of the
    overall pool of drivers (in the chosen example:
    50,000 accidents).
    3
    Terminology of the insurance branch
    6
    Fig. 1: Models expands the at-risk population (figures for illustration only)
    Fig. 2: Model leads to internal differentiation into good and poor risks (figures for illustration only)
    -------- Situation without AD17 --------
    Total population of 18-year-olds
    100,000 accidents per year
    Non-
    drivers
    40%
    Drivers
    60%
    -------- Situation with AD17 --------
    Total population of 18-year-olds
    140,000 accidents per year
    Non-
    drivers
    20%
    Drivers
    80%
    -------- Situation without AD17 --------
    Total population of 18-year-olds
    100,000 accidents per year
    Non-
    drivers
    40%
    Drivers
    60%
    -------- Situation with AD17 --------
    Total population of 18-year-olds
    60,000
    accidents per
    year
    Non-
    drivers
    40%
    cDL dri-
    vers
    30%
    AD17 dri-
    vers
    30%
    40,000
    accidents per
    year
    7
    Fig. 3: Model serves (causally) as a protective, i.e. risk-reducing factor (figures for illustration only)
    In practice, all three effects may be found superim-
    posed. One of the challenges for an evaluation is
    thus to distinguish the aforementioned effects. The
    mere fact of a reduction in the per capita risk of
    AD17 drivers compared to cDL drivers cannot be
    taken as evidence for Effect 2 or 3. The objectives
    for the present evaluation refer exclusively to the
    Effects 2 and 34
    . Effect 3, and the underlying ques-
    tion as to the risk-reducing potential of accompa-
    nied driving, is here placed in the foreground.
    4
    An evaluation of Effect 1 is not subject of the present report.
    A first statement on this effect was already to be found in
    the intermediate report of 31.07.2008 (Schade, Heinzmann
    & Feddersen, 2008, p. 33). It was there noted that the de-
    mand for driving licences among 17 to 18-year-olds in-
    creased by only a few percentage points in 2006 and 2007.
    More recent statistics on the numbers of probationary driv-
    ing licences, however, indicate moderate growth in the age
    group under 20 years for driving licence classes B and BE
    (and those classes which include these classes), namely by
    4.6% in 2005, a further 6.3% in 2006, and again by 5.1% in
    2007. In 2008, by which time all the German states had fi-
    nally adopted the AD17 model as a pilot scheme, the growth
    then stagnated (-0.5%). It can be derived from these figures
    that the demand for driving licences in this age group has
    increased by the order of around 20% in total since 2004.
    This conclusion, however, remains provisional and must be
    verified in detail, underpinned with further data and referred
    to demographic developments in the context of the final re-
    port.
    -------- Situation with AD17 --------
    Total population of 18-year-olds
    50,000
    accidents per
    year
    AD17 dri-
    vers
    30%
    40,000
    accidents per
    year
    Non-
    drivers
    40%
    cDL dri-
    vers
    30%
    Total population of 18-year-olds
    100,000 accidents per year
    Non-
    drivers
    40%
    Drivers
    60%
    -------- Situation without AD17 --------
    8
    2 Method5
    The persons asked to participate in the survey un-
    derlying the present evaluation were selected at
    random from the Central Register of Driving Li-
    cences (ZFER) at the Federal Motor Transport Au-
    thority (KBA) in Flensburg. This random sample in-
    cluded both persons who had just obtained a driv-
    ing licence on the basis of the AD17 model (AD17
    group: approx. 20,000 persons) and persons of the
    same age who had obtained a driving licence in the
    conventional manner (cDL group: approx. 40,000
    persons). Participation in the online survey was
    voluntary; in the end, properly completed question-
    naires were received from almost 9,000 partici-
    pants in the AD17 group and almost 10,000 partici-
    pants in the cDL group.
    The participants completed an initial questionnaire
    on average 6 to 7 months after commencement of
    independent driving after their 18th birthday, and
    then a final questionnaire after on average a further
    7 months. In individual cases in which the period
    between these two survey dates was particularly
    long, an additional intermediate questionnaire was
    sent to the participant. Except where explicitly men-
    tioned otherwise, the study considers also those
    persons who terminated their participation prema-
    turely after completing an initial or intermediate
    questionnaire. In these cases, the applicable period
    of observation is the period up to receipt of the last
    properly completed questionnaire.
    The purpose of the questionnaire was to acquire
    details of all evaluation-relevant driving behaviour
    during the applicable period, namely all forms of
    accident involvement, irrespective of the attribut-
    able portion of blame6
    , and all forms of punished
    traffic offences.
    For the evaluation of this self-reported behaviour,
    “significance thresholds” were defined to exclude
    trivial cases and to establish a common set of crite-
    ria for all groups. The significance threshold for ac-
    cidents is damage of “significant value” (assumed
    to mean €1,200 or more, in line with legal practice
    5
    For details, see intermediate report of 31.07.2008
    6
    The aspect of blame for accidents was not addressed, as
    this could firstly detract from the willingness to give honest
    replies, or even to participate in general, and secondly the
    validity of such information remains questionable even when
    given by the most honest survey participants. Furthermore,
    the 18-year-old driver is the person chiefly to blame in the
    vast majority of cases, as can be seen from the accident
    statistics of the Federal Statistical Office (2009, p. 143):
    73% of the 18 to 20-year-old male drivers and 67% of the
    female drivers bore the main blame for accidents with in-
    jured persons in 2008. As the proportions quoted in the
    aforementioned statistics drop rapidly with increasing age, it
    can be extrapolated that the proportion of cases in which
    the main blame is attributable to the 18-year-old drivers un-
    der review here (i.e. without consideration of the 19 and 20-
    year-olds) will be much higher still.
    to date with regard to § 315c of the German Crimi-
    nal Code, StGB) or injury to persons. It remains ir-
    relevant whether the damage or injury is incurred
    by the survey participant or another involved per-
    son. A further indicator for the exceeding of a cer-
    tain relevance threshold is deemed to be the re-
    cording of an accident by the police, as is recom-
    mended and practised in case of a suspicion of
    driving under the influence of alcohol, for example.
    Following inspection of the standard federal cata-
    logue of traffic offences and punishments, the rele-
    vance threshold for traffic offences was set at a
    fine of €25, as offences punishable by fines above
    this amount are only rarely to be considered minor
    infringements (meaning parking offences or the
    like).
    Which is the decisive criterion for the evaluation:
    The frequency of relevant incidents referred to time
    and the applicable population – here per 1,000
    drivers and year – or the frequency referred to the
    extent of driving practice – here per million kilo-
    metres driven?
    Schade & Heinzmann (2008, p. 17) discuss the
    conditions under which a time-based criterion (per
    1,000 drivers and year) is more appropriate than a
    distance-based criterion (per million kilometres). In
    administrative contexts, especially where the pri-
    vate driving licence holder is the subject of traffic
    policy or other individual decisions, they plead
    against the use of kilometre-based assessment.
    After all, it is for good reason that the law grants no
    “mileage bonus” with regard to traffic offences and
    accidents, for example (cf. also Holte, 2006).
    On the other hand, one of the objectives of accom-
    panied driving from the age of 17 is to promote the
    development of driving competence. As such com-
    petence is to be operationalised as the frequency
    of errors relative to the scope of driving practice,
    however, the frequency of accidents and traffic of-
    fences per million kilometres driven should be con-
    sidered as a second criterion alongside the time-
    based assessment.
    The statistical calculations of this evaluation em-
    ploy the so-called Poisson regression method. This
    form of analysis permits the simultaneous influ-
    ences of different events on a count variable – here
    the number of traffic offences and accidents within
    a given period of time – to be determined sepa-
    rately and assessed individually in respect of their
    statistical significance. In doing so, this method
    also takes into account the varying numbers of
    persons in the different groups and any differences
    in the lengths of periods under observation.
    9
    3 Results obtained with regard
    to self-reported driving behav-
    iour
    The survey served to analyse self-reported acci-
    dents and traffic offences which were deemed to lie
    above a specified significance threshold. Tab. 1
    summarises the cases covered by this analysis.
    Scope of sample N = 18,762
    Reply medium Online questionnaire
    Analysis groups
    AD17 group (8,785) and
    cDL group (9,977)
    Analysis period
    Evaluation of the data of all re-
    turned questionnaires
    Tab. 1: Analysis conditions and number of cases
    More than 18,000 persons took part in the survey
    and together reported on more than 18,000 years
    of driving experience and almost 150 million kilo-
    metres driven. Together, they reported 1,372 sig-
    nificant traffic offences and 1,852 cases of in-
    volvement in significant traffic accidents (Tab. 2).
    0
    20
    40
    60
    80
    100
    120
    Gruppe hFS Gruppe BF17 Gruppe hFS Gruppe BF17
    Unfallbeteiligungen Verkehrsverstöße
    Rate
    der
    Ereignisse
    pro
    1000
    Personen
    und
    Jahr
    Fig. 4: Rates of significant accident involvement and traf-
    fic offences in the analysis groups per 1,000 driv-
    ers and year
    0
    2
    4
    6
    8
    10
    12
    14
    16
    Gruppe hFS Gruppe BF17 Gruppe hFS Gruppe BF17
    Unfallbeteiligungen Verkehrsverstöße
    Rate
    der
    Ereignisse
    pro
    Millionen
    Pkw-Kilometer
    Fig. 5: Rates of significant accident involvement and traf-
    fic offences in the analysis groups per million
    kilometres driven
    AD17 group cDL group
    Number of survey participants 8,785 9,977
    Total number of years under observation 8,855 9,589
    Total kilometres driven (in millions) 72.84 75.33
    Reported cases of significant accident involvement 792 1,060
    Reported cases of significant traffic offences 590 782
    Tab. 2: Numbers of survey participants, significant accidents and traffic offences by analysis group
    Compared to the group of cDL drivers, the rate of
    significant accident involvement among AD17 driv-
    ers is seen to be 19% lower per 1,000 drivers and
    year, and even 23% lower per million kilometres
    driven (Tab. 3; see also Fig. 4 and Fig. 5).
    The differences in the rates of significant traffic of-
    fences are only marginally reduced: Among AD17
    drivers, the figure per 1,000 drivers and year is
    18% lower, and that per million kilometres driven
    22% lower than for the cDL drivers. In other words,
    the expected effect of the AD17 model is reflected
    clearly in all indicators.
    10
    Statistical variable AD17 group cDL group Difference
    between AD17
    and cDL
    Rate of accident involvement
    per 1,000 drivers and year
    89.4 110.5 -19%
    Rate of traffic offences
    per 1,000 drivers and year
    66.6 81.6 -18%
    Rate of accident involvement
    per million kilometres driven
    10.87 14.07 -23%
    Rate of traffic offences
    per million kilometres driven
    8.10 10.38 -22%
    Tab. 3: Rates of significant accident involvement and traffic offences in the analysis groups per 1,000 drivers and year
    and per million kilometres driven
    11
    4 Critical methodical evaluation
    In general, as is also the case here, a diversity of
    objections can be expressed to question purely de-
    scriptive results. A critical and scientific methodical
    evaluation is thus imperative. Such evaluation is
    the objective of the discussion in this section.
    4.1 Are the analysis results statisti-
    cally significant?
    The aforementioned values were determined on
    the basis of a limited sample and are by nature
    subject to a sampling error. Are the established dif-
    ferences nevertheless statistically significant, or
    could they be interpreted by critics as a conse-
    quence of a sampling error?
    An analysis of statistical significance within the
    framework of the so-called Poisson regression
    produces values of less than one per cent7
    as the
    probability of a sampling error for each of the four
    results8
    . The “null hypothesis”, namely that there
    are no differences between the groups (or that the
    AD17 group even performs less well than the group
    of novice drivers holding conventional driving li-
    cences), can be rejected with a probability of error
    of less than one per cent. The alternative hypothe-
    sis that the AD17 group performs better, on the
    other hand, can be assumed to be true.
    4.2 Are the AD17 and cDL analysis
    groups strictly comparable and
    can the results be generalised?
    The question addresses the uncertainty as to
    whether the compared analysis groups are really
    comparable, such that the established differences
    can be attributed clearly to participation in the
    AD17 model and are not a result of “side effects”.
    At the same time, it is to be asked whether the re-
    sults obtained are suitable for generalisation.
    In the course of preliminary analyses (Schade,
    Heinzmann & Feddersen, 2008), it was determined
    that the distortions in the two samples, insofar as
    they exist at all, are only minor, meaning that the
    samples can be considered representative for the
    federal states participating in the AD17 model9
    .
    7
    In all cases p < 0.001 for the unidirectional hypothesis that
    the rates of accident involvement and traffic offences in the
    AD17 group are lower than those of the cDL group
    8
    Differences in accident involvement and traffic offences re-
    ferred to both the number of drivers and the kilometres
    driven
    9
    At the time of sampling in mid-2007: Bavaria, Berlin, Bran-
    denburg, Bremen, Hamburg, Lower Saxony, North Rhine-
    Westphalia, Rhineland-Palatinate, Saarland, Saxony,
    Schleswig-Holstein
    Even so, there are small differences which are to
    be taken into account in the further evaluation.
    Otherwise, the groups were of the same age during
    the period under observation. The period of obser-
    vation for both groups began immediately with their
    commencement of independent driving, and fur-
    thermore during the same calendar period, mean-
    ing that they were subject to the same traffic condi-
    tions and even the same seasonal influences. All
    seasons of the year are covered.
    The fact that the number of participants and the
    mean duration of observation differ slightly be-
    tween the two groups is taken into full account in
    the methodology, as it applies a doubly relativised
    measure for comparison: The number of relevant
    occurrences (accidents, traffic offences) per 1,000
    persons and year.
    4.3 Do any “external” variables exert
    distorting influences?
    Another possible objection is that the AD17 and
    cDL groups may differ in further factors besides the
    model by which they obtained a driving licence, and
    that these factors could have a protective effect,
    i.e. they could already reduce the risks of accident
    involvement or traffic offences. Such factors which
    are known from traffic research include female
    gender and a higher level of school education. The
    objection refers to the differentiation into “good”
    and “poor” risks described in Section 1 as Effect 2.
    Could it be the case that the AD17 model attracts
    more female drivers and persons with a higher
    level of school education, and that this group al-
    ready displays a lower risk merely on the basis of
    these and possibly further protective factors, rather
    than any positive influences for the young drivers
    being attributable to the particular experience of the
    AD17 model?
    It is scarcely possible to conclusively invalidate ob-
    jections of this kind unless all potential protective
    factors are known. On the other hand, the objection
    can only refer to those factors which are firstly
    proven to be protective, and secondly occur more
    frequently in the AD17 group than in the cDL group.
    It is to be considered in the following, whether or
    not these conditions are met.
    The preliminary analyses for the intermediate re-
    port revealed a tendency in the direction of more
    female drivers in the AD17 group than in the cDL
    group, and likewise more residents of rural areas
    and more persons who had obtained or were pre-
    paring for at least an advanced school-leaving cer-
    tificate (in the following summarised as persons
    with a higher school education background). In ad-
    dition, a difference was determined between the
    groups with regard to the driving behaviour of a pa-
    rental role model: AD17 drivers replied slightly
    12
    more frequently that their parents were correct in
    their abiding by road traffic rules10
    . This could be a
    further protective factor.
    In this context, it is expedient not only to investigate
    objections that protective factors could possibly es-
    tablish a spurious correlation between the AD17
    model and a reduced frequency of accident in-
    volvement or traffic offences. Attention should also
    be paid to the opposite case: An external variable
    with an “anti-protective” effect, namely the some-
    what greater vehicle availability in the AD17
    group11
    , leads to underestimation of the causal in-
    fluence of the AD17 model on road safety. After all,
    only those licence holders with actual opportunities
    to drive are able to commit traffic offences or be
    involved in accidents. Where AD17 drivers have
    more frequent opportunities for accident involve-
    ment, however, this overlays the possible accident-
    reducing influence of the AD17 model.
    To adequately estimate the degree of causal corre-
    lation between the AD17 model and traffic behav-
    iour, it is thus necessary to take into account “ex-
    ternal variables” which could overlay and neutralise
    any causal effects.
    Proof of whether or not the specified factors display
    the expected protective and anti-protective correla-
    tions with accidents and traffic offences is obtained
    by way of regression calculations. The results are
    shown in Tab. 4.
    Whereas the factors education background, place
    of residence12
    and parental role model – contrary to
    expectations – are shown to have no or only minor
    influence on the rate of novice driver accident in-
    volvement and traffic offences, the factor gender is
    highly significant: Female novice drivers display a
    22% lower accident rate than male novice drivers,
    and even a 50% lower frequency of traffic offences.
    As was expected, a higher vehicle availability in-
    creases the rates of accident involvement and traf-
    10
    A positive parental role model is assumed for the purposes
    of the following analyses if the survey participant confirmed
    conscientious driving behaviour on the part of both parents,
    insofar as data were given (survey question: “My fa-
    ther/mother is very correct in abiding by road traffic rules“;
    this statement is “true” or “absolutely true”). Approx. 45% of
    novice drivers gave such a positive assessment of their par-
    ents.
    11
    Unlimited vehicle availability over the period of observation
    is assumed where the participant replies unanimously in all
    questionnaires completed that he or she is the owner of the
    vehicle used or at least its sole user (applicable in 37% of
    cases).
    12
    An evaluation on the basis of place of residence (as op-
    posed to the usual reference to the place of accidents) was
    recently presented by Holz-Rau & Scheiner (2009). Accord-
    ing to their results, town-dwellers face a slightly greater risk
    of accidents involving minor injury than the residents of rural
    areas, but a considerably reduced risk of a traffic accident
    resulting in serious or fatal injuries.
    fic offences. Alongside vehicle availability per se,
    the combinations with education background are
    also relevant in respect of driving behaviour:
    Where a higher education background coincides
    with limited vehicle availability, the rates of both ac-
    cident involvement and traffic offences are reduced
    significantly, and that beyond the extent which is
    already to be expected from the simple combina-
    tion of factors (such multiplication effects are
    known in statistical analysis as interaction effects).
    The analysis shows, therefore, that at least the
    slightly higher proportion of female drivers in the
    AD17 group could have contributed to the positive
    result of the AD17 model. In this respect, the objec-
    tion of result distortion under review here must be
    taken seriously. The highly significant interaction
    between vehicle availability and education back-
    ground, on the other hand, is not to be assigned
    critical importance, as there is no difference be-
    tween the AD17 and cDL analysis groups in terms
    of this specific combination of factors – limited ve-
    hicle availability and higher education background
    – irrespective of the slight difference in vehicle
    availability which is revealed when this factor is
    viewed in isolation.
    At the same time, the analysis points to the consid-
    erable influence of the factor vehicle availability,
    which could well mask any strong causal effects if
    it is not taken into account explicitly.
    Time-based comparison
    In the following, regression analysis is used to ver-
    ify whether the differences in accident involvement
    and traffic offences between the two analysis
    groups remain valid when the influences deter-
    mined in Tab. 4 are taken into account. Questions
    of particular interest are: Could the differences be-
    tween the analysis groups presented in Tab. 3
    possibly be explained merely by the slightly differ-
    ent gender composition? Does the slight difference
    in vehicle availability between the groups mask any
    strong causal effect?
    13
    Factor Ratios of accident involvement and traffic offences
    compared to the relevant reference group
    Accident involvement per
    1,000 drivers and year
    Traffic offences per
    1,000 drivers and year
    Gender:
    Female (versus male)
    0.78 *** 0.50 ***
    Education background:
    Advanced school certificate (versus lower)
    0.91 n.s. 1.00 n.s.
    Place of residence:
    Rural areas (versus others)
    1.02 n.s. 0.88 *
    Parental role model:
    Positive (versus negative)
    1.00 n.s. 0.98 n.s.
    Vehicle availability:
    Unlimited (versus limited)
    1.95 *** 2.38 ***
    Combination negative parental role model /
    lower education background (versus others)
    1.18 n.s. 1.31 *
    Combination higher education background /
    limited vehicle availability (versus others)
    0.68 *** 0.59 ***
    Legend: n.s. = Difference to parity (1.0) not significant; * = statistically significant (p ≤ 0.05); ** = statistically very significant
    (p ≤ 0.01); *** = statistically highly significant (p ≤ 0.001)
    Interpretation example: The ratio of the accident involvement of female novice drivers compared to their male counterparts is 0.78.
    This means that the accident risk for female drivers is lower by 22%.
    Tab. 4: Rates of accident involvement and traffic offences as dependent on selected external factors (time-based com-
    parison)
    Factor Ratios of accident involvement and traffic offences
    compared to the relevant reference group
    Accident involvement per
    1,000 drivers and year
    Traffic offences per
    1,000 drivers and year
    Gender:
    Female (versus male)
    0.78 *** 0.50 ***
    Education background:
    Advanced school certificate (versus lower)
    0.92 n.s. 1.02 n.s.
    Place of residence:
    Rural areas (versus others)
    1.03 n.s. 0.88 n.s.
    Parental role model:
    Positive (versus negative)
    1.01 n.s. 0.99 n.s.
    Vehicle availability:
    Unlimited (versus limited)
    1.97 *** 2.40 ***
    Combination negative parental role model /
    lower education background (versus others)
    1.18 n.s. 1.32 *
    Combination high education background /
    limited vehicle availability (versus others)
    0.68 *** 0.59 ***
    Driving licence model:
    AD17 (versus conventional model)
    0.83 *** 0.85 **
    Legend: n.s. = Difference to parity (1.0) not significant; * = statistically significant (p ≤ 0.05); ** = statistically very significant
    (p ≤ 0.01); *** = statistically highly significant (p ≤ 0.001)
    Interpretation example: The ratio of the accident involvement of AD17 drivers compared to those obtaining a driving licence in the
    conventional manner is 0.83. This means that the accident risk for AD17 drivers is lower by 17%.
    14
    Tab. 5: Influence of the driving licence model on the rates of accident involvement and traffic offences, taking into ac-
    count further factors (time-based comparison)
    These questions are addressed by Tab. 5. The
    sole objective here is to ascertain whether and to
    what extent the originally established impact of the
    AD17 model remains valid when the objection of
    distorting “external variables” is taken into account.
    The bottom row of the table shows a significant re-
    duction in accident involvement by 17% (originally
    19%; see Tab. 3) for the AD17 group compared to
    the cDL group, and a reduction of 15% (originally
    18%) in respect of traffic offences.
    This shows that the AD17 model has indeed re-
    sulted in a slight internal risk differentiation in the
    sense of the effect described in Section 1, namely
    both in favour of the AD17 group (more female
    novice drivers) and to its detriment (more novice
    drivers with unlimited vehicle availability). The two
    opposing effects do not neutralise each other com-
    pletely, however, with the result that the originally
    established differences between the groups are re-
    duced by two to three percentage points.
    The remaining road safety gain of 15 to 17% from
    the AD17 model can now be attributed to the spe-
    cific influence of “AD17 experience” (assuming that
    there are no further, unknown external variables to
    be considered).
    The following graphs (Fig. 6 and Fig. 7) illustrate
    the differences between the analysis groups if only
    the protective factor gender is taken into account.
    Subsequently, a further pair of graphs demon-
    strates the considerable “anti-protective” influence
    of vehicle availability on accident involvement and
    traffic offences (Fig. 8 and Fig. 9).
    0
    20
    40
    60
    80
    100
    120
    140
    hFSe BF17 hFSe BF17
    männlich weiblich
    Untersuchungsgruppen
    Rate
    der
    Unfallverwicklungen
    pro
    1000
    Personen
    und
    Jahr
    Fig. 6: Rate of accident involvement per 1,000 novice
    drivers and year in the first year of independent
    driving, differentiated by driving licence model
    (conventional/AD17) and gender
    0
    20
    40
    60
    80
    100
    120
    140
    hFSe BF17 hFSe BF17
    männlich weiblich
    Untersuchungsgruppen
    Rate
    der
    Verkehrsverstöße
    pro
    1000
    Personen
    und
    Jahr
    Fig. 7: Rate of traffic offences per 1,000 novice drivers
    and year in the first year of independent driving,
    differentiated by driving licence model (conven-
    tional/AD17) and gender
    0
    20
    40
    60
    80
    100
    120
    140
    160
    180
    hFSe BF17 hFSe BF17
    uneingeschränkte Pkw-Verfügbarkeit eingeschränkte Pkw-Verfügbarkeit
    Untersuchungsgruppen
    Rate
    der
    Unfallverwicklungen
    pro
    1000
    Personen
    und
    Jahr
    Fig. 8: Rate of accident involvement per 1,000 novice
    drivers and year in the first year of independent
    driving, differentiated by driving licence model
    (conventional/AD17) and vehicle availability
    15
    0
    20
    40
    60
    80
    100
    120
    140
    hFSe BF17 hFSe BF17
    uneingeschränkte Pkw-Verfügbarkeit eingeschränkte Pkw-Verfügbarkeit
    Untersuchungsgruppen
    Rate
    der
    Verkehrsverstöße
    pro
    1000
    Personen
    und
    Jahr
    Fig. 9: Rate of traffic offences per 1,000 novice drivers
    and year in the first year of independent driving,
    differentiated by driving licence model (conven-
    tional/AD17) and vehicle availability
    How great is the influence which can be attributed
    to the further differences between the AD17 and
    cDL groups which were determined in the course
    of the preliminary analyses?
    A series of further minor differences is by nature
    such that the effects on accident involvement and
    traffic offences, insofar as they are manifested at
    all, would rather tend to diminish the positive im-
    pact of the AD17 model. Consequently, they are
    irrelevant for the objection of a falsely positive
    AD17 effect: AD17 drivers tend to report less prior
    practice with other vehicles, emphasise more fre-
    quently the important of a high level of mobility,
    mention more frequently the availability of several
    vehicles in their household and the availability of a
    vehicle with an engine power of more than 50 kW,
    and report more frequently a weekly distance
    driven of more than 200 km and a driving time of
    more than four hours. All these differences imply
    an increased risk of accident involvement or traffic
    offences for the AD17 group and thus cannot be
    taken as a basis for the expressed criticism of a
    spurious correlation. It is rather the case that they
    promote underestimation of any AD17 effect13
    .
    Two of the important differences revealed by the
    preliminary analyses, however, could still be rele-
    vant: It is less often the case that AD17 drivers use
    either older vehicles or vehicles in a poor technical
    condition. This circumstance could have helped the
    AD17 group to a lower rate of accident involve-
    ment. To investigate this hypothesis, the replies of
    the two analysis groups with regard to the nature of
    any accident involvement were scrutinised more
    13
    Given the limited sample sizes, it is not possible to incorpo-
    rate all these factors into a regression analysis, and in this
    way to separate their particular influence so as to retain an
    unfalsified AD17 effect: The frequency matrix contains zero
    assignments, which are impermissible for the calculation.
    closely. Of the 1,335 cases of accident involvement
    reported by the cDL group, 10 instances were at-
    tributed to “technical defects or maintenance defi-
    cits” (0.7%); among the 1,036 cases of accident
    involvement reported by the AD17 group, this
    cause was mentioned 9 times (0.9%). Technical
    causes thus played only a very minor role in the
    accidents, and there was furthermore practically no
    difference between the two groups. The better
    condition of the vehicles used by the AD17 group is
    thus unsuitable as an explanation for their lower
    accident rate.
    Kilometre-based comparison
    Would the results perhaps be better if the rates of
    accident involvement and traffic offences were to
    be compared on the basis of kilometres driven?
    The general survey data show that the annual dis-
    tances driven by participants in the AD17 group
    were on average 4.7% higher than in the cDL
    group (8,226 km compared to 7,856 km; derived
    from Tab. 2). Increased exposure to traffic can
    hardly serve to explain a reduced risk of accident
    involvement and traffic offences, and a kilometre-
    based evaluation of driving behaviour is thus
    unlikely to call the effectiveness of the AD17 model
    into question. Nevertheless, this objection is still to
    be investigated.
    Tab. 6 shows the results of Poisson regression
    analyses of kilometre-based accident involvement
    and traffic offences. From these figures, it can be
    determined firstly (see the bottom row of the table),
    that a kilometre-based comparison, taking into ac-
    count the same control variables, actually places
    greater emphasis on the AD17 effect, now showing
    a reduction in the accident rate by 22% (previously
    17%) and a reduction in traffic offences by 20%
    (previously 15%).
    Secondly, as expected, it becomes clear that a
    consideration of accident involvement and traffic
    offences relative to the kilometres driven dilutes the
    influence of vehicle availability, and furthermore the
    influence of gender. The (greater) vehicle availabil-
    ity and the (male) gender of the drivers thus play no
    significant role for the numbers of accidents per
    million kilometres. For the numbers of traffic of-
    fences per million kilometres, however, they are
    still of considerable, albeit slightly reduced impor-
    tance14
    (see Fig. 10 and Fig. 11).
    14
    The factor of vehicle availability, as defined here (see foot-
    note 11), comprises two essential elements, both of which
    are related to the absence of a social corrective: On the one
    hand, a high degree of availability means that the driver
    feels less restricted in his/her impulses to use the vehicle
    and probably drives more kilometres as a result. Secondly,
    as the driver is usually also the owner of the vehicle, and as
    such less obliged to account to others for driving behaviour
    (e.g. notices of fines are received directly), he/she may be
    16
    less hesitant to risk traffic offences.
    The kilometre-based comparison neutralises the first con-
    tributory element of vehicle availability, namely the amount
    of driving done. The accepted risk of convictions for traffic
    offences, on the other hand, remains unaffected by the new
    perspective. The remaining effect of vehicle availability
    shown in Tab. 6 presumably reflects this second compo-
    nent.
    17
    Factor Ratios of accident involvement and traffic offences
    compared to the relevant reference group
    Accident involvement Traffic offences
    per million
    kilometres
    per year
    (cf. Tab. 5)
    per million
    kilometres
    per year
    (cf. Tab. 5)
    Gender:
    Female (versus male)
    0.98 n.s. 0.78 *** 0.63 *** 0.50 ***
    Education background:
    Advanced school certificate (versus lower)
    1.17 n.s. 0.92 n.s. 1.30 ** 1.02 n.s.
    Place of residence:
    Rural areas (versus others)
    0.95 n.s. 1.03 n.s. 0.81 ** 0.88 n.s.
    Parental role model:
    Positive (versus negative)
    1.00 n.s. 1.01 n.s. 0.97 n.s. 0.99 n.s.
    Vehicle availability:
    Unlimited (versus limited)
    1.01 n.s. 1.97 *** 1.24 ** 2.40 ***
    Combination negative parental role model /
    lower education background (versus others)
    1.17 n.s. 1.18 n.s. 1.31 *. 1.32 *
    Combination high education background /
    limited vehicle availability (versus others)
    0.92 n.s. 0.68 *** 0.79 * 0.59 ***
    Driving licence model:
    AD17 (versus conventional model)
    0.78 *** 0.83 *** 0.80 *** 0.85 **
    Legend: n.s. = Difference to parity (1.0) not significant; * = statistically significant (p ≤ 0.05); ** = statistically very significant
    (p ≤ 0.01); *** = statistically highly significant (p ≤ 0.001)
    Tab. 6: Influence of the driving licence model on the kilometre-based rate of accident involvement and traffic offences,
    taking into account further factors and compared to the time-based calculation in Tab. 5
    0
    2
    4
    6
    8
    10
    12
    14
    16
    Gruppe hFS Gruppe BF17 Gruppe hFS Gruppe BF17
    uneingeschränkte Pkw-Verfügbarkeit eingeschränkte Pkw-Verfügbarkeit
    Rate
    der
    Unfallbeteiligungen
    pro
    Mio.
    Pkw-Kilometer
    Fig. 10: Rate of accident involvement per million kilo-
    metres driven in the first year of independent driv-
    ing, differentiated by driving licence model (con-
    ventional/AD17) and vehicle availability
    0
    20
    40
    60
    80
    100
    120
    140
    Gruppe hFS Gruppe BF17 Gruppe hFS Gruppe BF17
    uneingeschränkte Pkw-Verfügbarkeit eingeschränkte Pkw-Verfügbarkeit
    Rate
    der
    Verkehrsverstöße
    pro
    Mio.
    Pkw-Kilometer
    Fig. 11: Rate of traffic offences per million kilometres
    driven in the first year of independent driving, dif-
    ferentiated by driving licence model (conven-
    tional/AD17) and vehicle availability
    18
    4.4 Are the results distorted by non-
    cooperating participants?
    The above analyses were based on the replies of
    all survey participants, i.e. also those who termi-
    nated their cooperation prematurely after returning
    an initial or intermediate questionnaire. Would the
    results have been more valid without consideration
    of these non-cooperating survey participants, de-
    spite the correspondingly diminished sample size?
    To answer this possible objection, the analysis for
    Tab. 5 was repeated without the data of those per-
    sons who terminated their survey participation
    prematurely. The analysis conditions and adjusted
    numbers of cases are to be seen in Tab. 7.
    Scope of sample N = 12,532
    Reply medium Online questionnaire
    Analysis groups
    AD17 group (6,021) and
    cDL group (6,511);
    limited to participants who com-
    pleted a final questionnaire
    Analysis period
    Evaluation of the data of all re-
    turned questionnaires
    Tab. 7: Analysis conditions and number of cases
    Factor Ratios of accident involvement and traffic offences
    compared to the relevant reference group
    Accident involvement
    per year
    Traffic offences per year
    Adjusted
    sample
    Values in
    Tab. 5
    Adjusted
    sample
    Values in
    Tab. 5
    Gender:
    Female (versus male)
    0.79 *** 0.78 *** 0.51 *** 0.50 ***
    Education background:
    Advanced school certificate (versus lower)
    0.93 n.s. 0.92 n.s. 0.92 n.s. 1.02 n.s.
    Place of residence:
    Rural areas (versus others)
    1.06 n.s. 1.03 n.s. 0.92 n.s. 0.88 n.s.
    Parental role model:
    Positive (versus negative)
    0.98 n.s. 1.01 n.s. 0.98 n.s. 0.99 n.s.
    Vehicle availability:
    Unlimited (versus limited)
    2.01 *** 1.97 *** 2.30 *** 2.40 ***
    Combination negative parental role model /
    lower education background (versus others)
    1.09 n.s. 1.18 n.s. 1.32 * 1.32 *
    Combination higher education background /
    limited vehicle availability (versus others)
    0.71 ** 0.68 *** 0.66 ** 0.59 ***
    Driving licence model:
    AD17 (versus conventional model)
    0.83 *** 0.83 *** 0.83 ** 0.85 **
    Legend: n.s. = Difference to parity (1.0) not significant; * = statistically significant (p ≤ 0.05); ** = statistically very significant
    (p ≤ 0.01); *** = statistically highly significant (p ≤ 0.001)
    Interpretation example (row “Gender”): The ratio of the accident involvement of female novice drivers compared to their male coun-
    terparts is 0.79. This means that the accident risk for female drivers is lower by 21%.
    Tab. 8: Influence of the driving licence model on the rates of accident involvement and traffic offences, taking into ac-
    count further factors and excluding those persons who terminated their survey participation before completing a
    final questionnaire (compared to the corresponding values of the unadjusted sample from Tab. 5)
    19
    The results in Tab. 8 (grey columns) display only
    marginal changes compared to the values of Tab.
    5, and furthermore changes with no effect for the
    previously drawn conclusions. It can be seen, in
    particular, that the inclusion of those persons who
    terminated their survey participation prematurely in
    the analyses of Section 4.3 does not lead to an
    overestimation of the AD17 effects – rather the op-
    posite is the case (see last row of table).
    4.5 Does the model possibly achieve
    merely a temporary effect?
    Another objection addresses the possibility that, on
    account of the much longer period of parental ac-
    companiment in the AD17 group, the positive ef-
    fects also last somewhat longer15
    than the other-
    wise equally positive effects of conventional driver
    training, even though this difference may become
    less and less distinct after a few weeks or months.
    Is the proven AD17 effect thus real exclusively or
    predominantly at the beginning of independent driv-
    ing and does it then fade rapidly?
    To be able to investigate this question, the self-
    reported accidents and traffic offences of all par-
    ticipants who took full part in the survey and also
    completed a final questionnaire were assigned to
    an early and a later analysis period: Firstly the pe-
    riod up to completion of the initial questionnaire (on
    average six to seven months after the commence-
    ment of independent driving), and secondly the pe-
    riod between the initial and final questionnaires (on
    average approx. seven months). In the following –
    for the sake of simplicity – these periods are de-
    scribed as the first half and second half of the ob-
    servation period. Tab. 7 indicates the cases which
    were deemed relevant here.
    First of all, it was determined by way of Poisson re-
    gression16
    , whether the rates of accident involve-
    ment and traffic offences actually differ significantly
    between the two periods. This produced a highly
    interesting result (see last row of Tab. 9): While the
    rate of traffic accidents fell significantly by on aver-
    age 19% between the first and second half of the
    observation period, the rate of traffic offences rose
    significantly by on average 29% between the two
    periods. Although very important for road safety re-
    search, this result is not a subject for the discus-
    sion within the framework of this report and thus
    cannot be followed up at this juncture.
    15
    The previous accompanists (parents) continue to travel in
    the passenger seat for several weeks in the mind of the
    young driver.
    16
    The factor “Place of residence”, which had not displayed
    particular significance in any of the previous regression
    analyses, was no longer included for the subsequent analy-
    ses, so as not to burden the sample size unnecessarily.
    The relevant question for the present analysis re-
    fers solely to whether or not the protective influ-
    ence of the AD17 model decreases in the second
    half of the observation period. From a statistical
    point of view, this should be reflected in the Pois-
    son regression in the form of a significant interac-
    tion effect between driving licence model and sec-
    tion of the observation period. Interaction effects,
    however, are revealed neither for accident involve-
    ment nor for traffic offences. In both cases, a re-
    gression model comprising solely main effects, in
    other words a model which assumes exclusively
    zero interactions, is shown to be fully compatible
    with the data (Pearson chi-square for the model
    “Traffic accidents” with 55 degrees of freedom:
    63.3, p = 0.207; Pearson chi-square for the model
    “Traffic offences” with 55 degrees of freedom: 47.8,
    p = 0.744).
    Viewing the rates of accident involvement and traf-
    fic offences (see Fig. 12 and Fig. 13), it is even
    possible to identify an – admittedly not significant –
    opposing tendency: The decrease in the rates of
    accident involvement and traffic offences in con-
    nection with the AD17 licence model actually ap-
    pears to be even more distinctive in the second half
    of the observation period than in the first half.
    The aforementioned objection of a possibly tempo-
    rary effect is thus not supported by the data. It can
    be assumed that the effect of the AD17 model is
    preserved over the whole observation period cover-
    ing the first approx. 14 months of independent driv-
    ing. Nevertheless, it is planned to return to this
    question once more at a later date and to perform
    a further more detailed analysis on the basis of
    data from the Central Register of Traffic Offenders.
    0
    20
    40
    60
    80
    100
    120
    hFSe BF17 hFSe BF17
    1. Hälfte 2. Hälfte
    Untersuchungsgruppen
    Rate
    der
    Unfallverwicklungen
    pro
    1000
    Personen
    und
    Jahr
    Fig. 12: Rate of accident involvement, differentiated by
    driving licence model (conventional/AD17) and
    section of observation period
    20
    0
    20
    40
    60
    80
    100
    hFSe BF17 hFSe BF17
    1. Hälfte 2. Hälfte
    Untersuchungsgruppen
    Rate
    der
    Verkehrsverstöße
    pro
    1000
    Personen
    und
    Jahr
    Fig. 13: Rate of traffic offences, differentiated by driving
    licence model (conventional/AD17) and section of
    observation period
    Factor Ratios of accident involvement and traffic offences
    compared to the relevant reference group
    Accident involvement per
    1,000 drivers and year
    Traffic offences per 1,000
    drivers and year
    Gender:
    Female (versus male)
    0.79 *** 0.51 ***
    Education background:
    Advanced school certificate (versus lower)
    0.93 n.s. 0.92 n.s.
    Parental role model:
    Positive (versus negative)
    0.98 n.s. 0.98 n.s.
    Vehicle availability:
    Unlimited (versus limited)
    2.01 *** 2.30 ***
    Combination negative parental role model /
    lower education background (versus others)
    1.09 n.s. 1.32 *
    Combination higher education background /
    limited vehicle availability (versus others)
    0.71 ** 0.66 **
    Driving licence model:
    AD17 (versus conventional model)
    0.83 *** 0.83 **
    Section of observation period:
    Second half (versus first half)
    0.81 *** 1.29 ***
    Legend: n.s. = Difference to parity (1.0) not significant; * = statistically significant (p ≤ 0.05); ** = statistically very significant
    (p ≤ 0.01); *** = statistically highly significant (p ≤ 0.001)
    Interpretation example: The ratio of accident involvement during the second half of the observation period compared to the first half
    is 0.79. This means that the accident risk is reduced by 19%.
    Tab. 9: Rates of accident involvement and traffic offences as dependent on the section of the observation period, taking
    into account further factors
    4.6 What evidence supports a (causal)
    effect of the AD17 model?
    Particularly sound evidence for the (causal) effect
    of a measure is often only to be obtained through
    verification of a so-called “dose-response relation-
    ship” and proof of the “specificity” of the effect. To
    this end, a cause of the effect is postulated, i.e. an
    “agent”, the intensity of which is to determine the
    degree of effect. In the case of the AD17 model,
    the “agent” is driving practice while accompanied
    by an adult driver. If this practice tends to zero,
    then it is no longer possible to expect a positive ef-
    fect from the model. The driving practice can be
    measured by the duration of the accompaniment
    phase as a number of months, though the more
    direct approach would be to determine the distance
    driven with accompaniment in kilometres.
    21
    The specificity of an effect describes how the posi-
    tive response to the underlying measure is wit-
    nessed predominantly in precisely the aspect of
    behaviour for which the measure was developed.
    In the case of the AD17 model, the principal effect
    should be achieved in terms of overall driving
    safety and not merely in a sphere of behaviour
    which is only loosely associated with safety (e.g.
    confident navigation in an unknown town or particu-
    lar skill when parking). A non-specific effect would
    give rise to doubt as to the underlying effect
    mechanism.
    Dose-response relationship:
    To permit statements on the dose-response rela-
    tionship, the AD17 group was analysed more
    closely. The corresponding analysis conditions and
    the number of cases are given in Tab. 10.
    Scope of sample N = 6,021
    Reply medium Online questionnaire
    Analysis group
    All participants of the AD17 group
    who completed a final question-
    naire
    Analysis time
    Evaluation of the data of all re-
    turned questionnaires
    Tab. 10: Analysis conditions and number of cases
    The sample of AD17 drivers was divided firstly on
    the basis of the duration of the accompaniment
    phase (up to 6 months17
    , over 6 up to 10 months,
    over 10 up to 12 months), and secondly according
    to actual driving practice during this period (up to
    500 km accompaniment, 501 to 1,000 km accom-
    paniment, more than 1,000 km accompaniment).
    The “dose-response relationship” here refers to the
    hypothesis that AD17 drivers reporting a longer ac-
    companiment phase, and in particular those with
    more driving practice in terms of kilometres, will
    display lower rates of accident involvement and
    traffic offences18
    , because the protective effect of
    the model should be greater under these condi-
    tions. Correspondingly, drivers with a less intensive
    accompaniment phase should display higher rates
    of accident involvement and traffic offences.
    To test this hypothesis, the intensity of the accom-
    paniment phase is taken into account in the re-
    17
    Due to the selection criteria applied, there are unfortunately
    very few cases of an accompaniment phase of less than 4
    months in the sample, meaning that no analysis of this
    group is possible, despite the fact that it could be particularly
    interesting for the questions addressed by the present
    study.
    18
    This is a so-called unidirectional hypothesis, meaning that
    the statistical testing is likewise unidirectional.
    gression analyses as a predictor, based on the
    data provided by the survey participants on the
    length of their accompaniment phase in months
    and their driving practice during this accompani-
    ment phase in kilometres. Due to the significantly
    reduced sample size – after all, the analysis views
    only the AD17 group, and there, in turn, only those
    persons who also completed a final questionnaire –
    it is only possible to consider the most important
    control variable, namely gender, in the regression
    analysis. Otherwise, certain frequency cells would
    remain empty.
    One particular methodical difficulty stands in the
    way of the direct approach. Driving practice during
    the accompaniment phase correlates positively
    with driving practice during the first year of inde-
    pendent driving: Evidence shows that those partici-
    pants who record the most kilometres during the
    accompaniment phase will also drive more later.
    This still applies if vehicle availability is taken into
    account. As more driving means a higher level of
    risk, a spurious correlation is manifested: Para-
    doxically, a more intensive accompaniment phase
    is linked with a later increased risk of accident in-
    volvement and traffic offences. It could thus be
    concluded that accompaniment is counter-
    productive. There is no simple remedy for this me-
    thodical difficulty (which results from the presence
    of a so-called “confounder”). The solution is to refer
    the numbers of accidents and traffic offences not to
    the period of a year, as in Section 4.3, but instead
    to the kilometres driven, so that differences in indi-
    vidual driving practice no longer influence the re-
    sults directly.
    The results of the regression analyses are shown
    in Tab. 11 and Tab. 12. While less driving practice
    during the accompaniment phase (see Tab. 11)
    leads to an increase in accident involvement and
    traffic offences, the duration of the accompaniment
    phase – at least for a minimum duration of 4
    months – displays no clear correlation (see Tab.
    12). Applying the defined significance threshold of
    1.0 per cent for the present analyses – which,
    given the importance of the study, represents the
    just still acceptable probability of error – only one
    effect remains for the evaluation of the results19
    :
    Driving practice of less than 500 kilometres during
    the accompaniment phase can be linked to an in-
    19
    The analysis suffers in that the sample size is too small to
    support rare occurrences and weak effects. If all reported
    accidents and traffic offences were to be considered, in-
    stead of only those exceeding the defined relevance thresh-
    old (see Section 2), the data set would be much larger
    (1,373 traffic offences instead of 425, and 784 traffic acci-
    dents instead of 590). This would raise the power of the sta-
    tistical test. In fact, the influence of driving practice would
    also achieve the demanded minimum significance of 1 per
    cent with regard to accidents under these more favourable
    conditions.
    22
    creased rate of traffic offences per million kilome-
    tres. A dose-response relationship can thus be de-
    termined with regard to the rate of traffic offences,
    but not with regard to the rate of accident involve-
    ment.
    Factor Ratios of accident involvement and traffic offences
    compared to the relevant reference group
    Accident involvement
    per million kilometres
    Traffic offences
    per million kilometres
    Driving practice:
    up to 500 km (versus over 1,000 km)
    1.21 * 1.47 ***
    Driving practice:
    500 to 1,000 km (versus over 1,000 km)
    1.25 * 1.14 n.s.
    Gender:
    Female (versus male)
    0.93 n.s. 0.62 ***
    Legend: n.s. = Difference to parity (1.0) not significant; * = statistically significant (p ≤ 0.05); ** = statistically very significant
    (p ≤ 0.01); *** = statistically highly significant (p ≤ 0.001); n.a. = significance test not applicable, as unidirectional hypothesis to be
    rejected
    Tab. 11: Influence of driving practice on the kilometre-based rates of accident involvement and traffic offences, taking
    into account gender
    Factor Ratios of accident involvement and traffic offences
    compared to the relevant reference group
    Accident involvement
    per million kilometres
    Traffic offences
    per million kilometres
    Duration of accompaniment phase:
    4 to 6 months (versus 10 to 12 months)
    0.95 n.a. 1.03 n.s.
    Duration of accompaniment phase:
    6 to 10 months (versus 10 to 12 months)
    1.13 n.s. 0.86 n.a.
    Gender:
    Female (versus male)
    0.93 n.s. 0.64 ***
    Legend: n.s. = Difference to parity (1.0) not significant; * = statistically significant (p ≤ 0.05); ** = statistically very significant
    (p ≤ 0.01); *** = statistically highly significant (p ≤ 0.001); n.a. = significance test not applicable, as unidirectional hypothesis to be
    rejected
    Tab. 12: Influence of the duration of the accompaniment phase on the kilometre-based rates of accident involvement and
    traffic offences, taking into account gender
    Specificity:
    In the aforementioned analyses, trivial accidents
    and traffic offences were explicitly excluded. This
    approach was chosen for methodical reasons, so
    as not to leave decisions on which events and cir-
    cumstances were to be considered worthy of re-
    porting to the subjective discretion of the survey
    participant, and instead to ensure the application of
    common criteria for all replies received.
    An analysis of trivial accidents and traffic offences
    (Tab. 13) shows no particular differences between
    the AD17 and cDL groups. This supports – along-
    side a diminished validity of data on trivial circum-
    stances – the assumption of a specific effect of the
    AD17 model, as there is little influence on triviali-
    ties.
    23
    AD17 group cDL group Difference
    between
    AD17 and cDL
    Rate of trivial accident involvement per
    1,000 drivers and year
    27.6 28.7 - 4%
    Rate of trivial traffic offences per 1,000
    drivers and year
    134.2 133.6 ± 0%
    Tab. 13: Rate of insignificant (trivial) accident involvement and traffic offences in the analysis groups
    In the following, it is to be determined whether the
    AD17 effect becomes more distinct with increasing
    severity of the accident involvement or traffic of-
    fence. To this end, separate calculations20
    are per-
    formed with modified significance thresholds.
    Tab. 14 indeed shows that the risk-reducing effect
    of the AD17 model is especially evident with regard
    to serious accidents and traffic offences.
    Indicator and
    significance threshold applied
    Difference
    between
    AD17 and
    cDL
    Accidents:
    All reported accidents -16%
    Only accidents with injury to persons,
    damage exceeding € 1,200 or re-
    cording by the police
    -19%
    Traffic offences:
    All reported traffic offences - 6%
    Those punishable by a fine over €15 -12%
    Those punishable by a fine over €25 -18%
    Those punishable by a fine from €40 -30%
    Tab. 14: Dependence of the rates of accident involvement
    and traffic offences per 1,000 drivers and year on
    the significance threshold applied (the thresholds
    applied elsewhere in the present study are high-
    lighted grey)
    The model is apparently effective above all in re-
    spect of accidents and traffic offences above a
    higher relevance threshold, and less so or not at all
    in respect of minor incidents such as parking in-
    fringements or “car park bumps” when manoeu-
    vring. This specificity is a further point which sup-
    ports a direct causal correlation between model
    and effect.
    20
    These calculations are not regression analyses taking into
    account control variables, but direct determination of the
    rate of accident involvement and traffic offences per 1,000
    drivers and year by the same method as was applied in Sec-
    tion 3.
    4.7 Are factors known which could
    further enhance the AD17 effect?
    It is possible that the AD17 model, as it is currently
    implemented, is still not optimal, meaning that the
    effect assessments presented here actually under-
    estimate the true potential of accompanied driving.
    Are further factors known, alongside the duration
    and distance of accompanied practice treated in
    Section 4.6, which could enhance the success of
    the AD17 model?
    One possibility which comes to mind is the gender
    of the accompanying passenger, especially as
    studies document the positive influence of female
    passengers on young drivers (Williams, 2003, pre-
    sents corresponding findings). The hypothesis that
    a female accompanist could enhance the effect of
    the AD17 model, and that this effect could also be
    dependent on the gender of the driver, was tested
    for the AD17 group on the basis of the present
    data. The analysis conditions and the number of
    cases corresponded to those of Section 4.6 (there
    see Tab. 10).
    The results of the Poisson regression analyses for
    accident involvement and traffic offences, each on
    the basis of both time and kilometres driven, are to
    be found in Tab. 15 (row highlighted grey): There is
    no evidence that the gender of the accompanying
    passenger during the accompaniment phase influ-
    ences the relevant driving behaviour of the partici-
    pant during the observation period – the main ef-
    fects referring to the accompanist are not signifi-
    cant in any of the four analyses. No mentionable
    interactions of the genders of the driver and ac-
    companying passenger occur (all four regression
    models display a good fit under the assumption of
    no interaction; p between 0.98 and 0.34). There is
    thus no combination of genders which is signifi-
    cantly more favourable than all others in the con-
    text of the AD17 model.
    24
    Factor Ratios of accident involvement and traffic offences
    compared to the relevant reference group
    Accident involvement Traffic offences
    per million
    kilometres
    per 1,000
    drivers and
    year
    per million
    kilometres
    per 1,000
    drivers and
    year
    Gender of driver:
    Female (versus male)
    0.95 n.s. 0.71 *** 0.65 *** 0.49 ***
    Gender of most frequent accompanist:
    Female (versus male)
    1.00 n.s. 1.01 n.s. 0.97 n.s. 0.97 n.s.
    Legend: n.s. = Difference to parity (1.0) not significant; * = statistically significant (p ≤ 0.05); ** = statistically very significant (p ≤
    0.01); *** = statistically highly significant (p ≤ 0.001)
    Tab. 15: Influence of the gender of the most frequent accompanying passenger during the accompaniment phase on the
    rates of accident involvement and traffic offences in the period under observation
    4.8 Is self-reported behaviour a suffi-
    ciently valid basis for evaluation?
    Objections are occasionally expressed with regard
    to the method of self-reported behaviour, especially
    where the subject – as here – refers to “negative
    behaviour” or even failure21
    . It is claimed that the
    method is insufficiently valid for an evaluation with
    far-reaching (legal) consequences.
    To be able to investigate this objection, the results
    presented here are to be complemented by an
    evaluation of the data records pertaining to the sur-
    vey participants in the Central Register of Traffic
    Offenders (VZR). In accordance with the project
    schedule, however, these data will not be available
    in full until spring 2010.
    4.9 Does the knowledge of participa-
    tion in a road safety study exert a
    systematic positive influence on
    driving behaviour?
    The objection that participants could be influenced
    by their knowledge of the study objective is justified
    in principle. However, there is initially no reason to
    assume that this objection does not apply equally
    to both analysis groups. Distortion of an equal ex-
    tent is unproblematic from the methodical point of
    view, because the present study is based solely on
    comparisons and not on absolute figures.
    Mentionable pretence on the part of the survey par-
    ticipants, as is indeed to be documented in studies
    with a short duration, seems to be reasonably ex-
    21
    Relevant studies (e.g. recently Staubach & Lüken, 2009),
    however, fail to support the assumption that the accident re-
    ports of involved persons are generally less useful.
    cluded given the observation period of more than a
    year.
    Moreover, the design of the evaluation study pro-
    vides for similar analyses and calculations to those
    presented here to be performed also for groups of
    drivers who have not been contacted and are con-
    sequently unaware of the study purpose: Two large
    samples of young drivers of both present groups
    have been drawn from the Central Register of Driv-
    ing Licences (ZFER) and corresponding data re-
    cords, if any, are to be retrieved from the register of
    traffic offenders. In accordance with the project
    schedule, however, these data will not be available
    until 2010.
    25
    5 Summary and conclusion
    Three questions are to be investigated, of which
    only the third is a core subject for the present
    evaluation:
    1. Does the AD17 model expand the at-risk popu-
    lation of 18-year-old drivers (with the probable
    consequence of increased accident figures; Ef-
    fect 1)?
    2. Does the AD17 model lead merely to internal
    differentiation into “good” and “poor” risks (Ef-
    fect 2)?
    3. Does the AD17 model itself have a directly pro-
    tective effect (Effect 3)?
    Method
    The present evaluation refers to self-reported in-
    stances of relevant driving behaviour – traffic of-
    fences and traffic accident involvement – from over
    18,000 eighteen-year-old drivers in the first year of
    their independent driving of a motor vehicle, repre-
    senting in total more than 18,000 years of “proba-
    tion” and almost 150 million kilometres of driving
    practice.
    Two samples comprising persons of the same age
    are compared: Drivers who had completed at least
    a 3-month phase of accompanied driving before
    the approximately one-year project observation
    phase (“AD17 drivers”), and “normal” novice driv-
    ers who had obtained a conventional driving li-
    cence shortly after their 18th birthday (“cDL driv-
    ers”). The members of both analysis groups were
    drawn at random from the records of the Central
    Register of Driving Licences (ZFER) and asked to
    participate in the study on a voluntary basis.
    Around 44% of the AD17 drivers and 25% of the
    cDL drivers actually completed the initial question-
    naire on the Internet (the final questionnaire, on
    average seven months later, was completed by
    slightly fewer participants).
    All forms of accident involvement are evaluated,
    irrespective of the portion of blame attributable to
    the participant, insofar as the police was called to
    record the accident, damage of at least €1,200 was
    estimated or injury to persons was reported (“sig-
    nificant accidents”). In addition, all traffic offences
    punished with a fine of more than €25 were taken
    into account (“significant traffic offences”). The
    rates of significant accident involvement and sig-
    nificant traffic offences were determined both on a
    time-related basis, i.e. per 1,000 drivers and year,
    and according to the scope of actual driving prac-
    tice, i.e. per million kilometres driven.
    Result
    The rate of significant accident involvement in the
    group of AD17 drivers is seen to be 19 per cent
    lower than that in the group of cDL drivers, while
    the rate of significant traffic offences is 18 per cent
    lower, both figures calculated per 1,000 novice
    drivers and year.
    When seeking to evaluate specifically the influence
    on driving competence, it is expedient to consider
    the results in relation to actual driving practice (per
    million kilometres driven): On this basis, the rate of
    significant accident involvement in the group of
    AD17 drivers is even 23 per cent lower than in the
    group of cDL drivers, and the rate of significant
    traffic offences 22 per cent lower.
    Critical methodical evaluation
    These results stand up to all critical methodical ob-
    jections – with minor limitations – and can thus be
    deemed to provide first proof, within the framework
    of the overall project, for the effectiveness of the
    AD17 model in the sense of causal effect 3:
    - The specified differences in accident involve-
    ment and traffic offences are statistically signifi-
    cant.
    - The random samples of young drivers, and like-
    wise the observation conditions, are compara-
    ble, and the results obtained in the eleven fed-
    eral states which were participating in the model
    at the time of sampling are suitable for generali-
    sation.
    - The objection that the model leads to internal
    risk differentiation between the two analysis
    groups (Effect 2), and that this is in part respon-
    sible for the determined differences (spurious
    correlation), cannot be invalidated conclusively:
    The slightly greater proportion of female drivers
    in the AD17 group does indeed improve the
    good result of the AD17 drivers by a further few
    percentage points (implies a greater AD17 effect
    than is actually the case).
    - At the same time, however, the greater vehicle
    availability in the AD17 group is an “external”
    variable which masks the actual causal effect of
    the AD17 model by a few percentage points (im-
    plies a lesser AD17 effect than is actually the
    case).
    - If both distorting influences are taken into ac-
    count, this leaves a reduction of 17 per cent in
    significant accident involvement22
    and 15 per
    cent with regard to significant traffic offences as
    the causal effect (Effect 3) in the time-based
    22
    As, for methodical reasons, the survey considers only acci-
    dent involvement and does not seek to clarify any attribut-
    able portion of blame, it can be expected that the reduction
    in culpable accidents would prove even more distinct.
    26
    analysis (per 1,000 novice drivers and year). In
    the kilometre-based comparison (per million
    kilometres driven), the reduction is correspond-
    ingly to 22 per cent for accident involvement and
    to 20 per cent for significant traffic offences.
    - The inclusion of data from those persons who
    terminated their survey participation prematurely,
    before completing a final questionnaire, does not
    distort the results in favour of a greater AD17 ef-
    fect (rather the opposite is the case).
    - The possible objection that the AD17 effect is
    merely a temporary consequence of the long
    and intensive accompaniment phase can be re-
    futed: The effect during the second six months
    of independent driving is at least as great as dur-
    ing the first six months (if at all different, then ac-
    tually greater).
    - Initial evidence of a “dose-response relationship”
    for the AD17 model could be found: Those par-
    ticipants who gather little driving practice during
    the accompaniment phase later display an in-
    creased risk with regard to traffic offences.
    There is similarly a tendency towards a corre-
    sponding correlation in respect of accident in-
    volvement. The length of the accompaniment
    phase, on the other hand, appears to play no
    role from a minimum duration of three months
    upwards.
    - It is possible to verify a certain specificity for the
    AD17 model: It has little or no effect on trivialities
    (e.g. “car park bumps” when manoeuvring or
    minor parking infringements). By contrast, the
    effectiveness of the AD17 model increases with
    the severity of accidents or traffic offences.
    - The latter findings are to be deemed aspects
    which further underpin the (causal) effectiveness
    of the AD17 model (in the sense of Effect 3).
    - No evidence was found to indicate that the effec-
    tiveness of the AD17 model is enhanced when a
    female passenger assumes the role of accom-
    panist.
    - Objections regarding the validity of self-reported
    behaviour and the possible influence of a par-
    ticipant's knowledge of the study objective are to
    be treated in further analysis steps on the basis
    of data retrieved from the Central Register of
    Traffic Offenders during 2010.
    Conclusion and outlook
    Following critical methodical appraisal of the data
    and valuations presented here, it is reasonable to
    speak of an overall positive effect of the AD17
    model with regard to road safety and compliance
    with traffic regulations. Even after taking into ac-
    count a number of influencing factors, it is possible
    to ascertain a reduction in the rate of accident in-
    volvement by 22 per cent and a reduction in traffic
    offences by 20 per cent per million kilometres
    driven.
    Nevertheless, an evaluation incorporating Effect 1
    (model expands the at-risk population) could still
    relativise this result: If it is shown, for example, that
    the introduction of the AD17 model leads to 20 per
    cent more 18-year-olds holding a driving licence,
    and consequently 20 per cent more 18-year-old
    drivers subject to an accident risk (Effect 1), then a
    20 per cent reduction in the accident rate in the first
    year (Effect 3) is just sufficient “compensation”. In
    this case, the net effect for road safety in the first
    year of independent driving would be zero. The
    AD17 model, consequently, must seek its justifica-
    tion in a road safety gain in subsequent years, as-
    suming that such gains exist. It is thus intended to
    extend the period of observation for the data re-
    cords to be retrieved from the Central Register of
    Traffic Offenders to cover a second year, and in
    this way to determine whether the AD17 model
    achieves a longer-term effect.
    Bibliography
    Holte, H. (2006): Unfälle per 1 Million Kilometer.
    Was ein fahrleistungsbezogener Risikowert
    besagt (und was nicht). Zeitschrift für Ver-
    kehrssicherheit, 115-116.
    Holz-Rau, C. & Scheiner, J. (2009): Verkehrssi-
    cherheit in Stadt und (Um-)Land. Zeitschrift für
    Verkehrssicherheit, 171-177.
    Schade, F.-D. & Heinzmann, H.-J. (2008): Alters-
    typisches Verkehrsrisiko. Berichte der Bundes-
    anstalt für Straßenwesen, Heft M 193, Bergisch
    Gladbach.
    Schade, F.-D., Heinzmann, H.-J. & Feddersen, S.
    (2008): Summative Evaluation des Begleiteten
    Fahrens ab 17. Projekt-Zwischenbericht an die
    Bundesanstalt für Straßenwesen, Kraftfahrt-
    Bundesamt, Flensburg.
    Statistisches Bundesamt (2009): Fachserie 8: Ver-
    kehr, Reihe 7: Verkehrsunfälle 2008. Wiesba-
    den.
    Staubach, M. & Lüken, P. (2009): Bewertung von
    Zeugenaussagen verunfallter Fahrzeugführer.
    Zeitschrift für Verkehrssicherheit, 129-134.
    Williams, A.F. (2003): Teenage drivers: Patterns of
    risk. Journal of Safety Research, 34, 5-15.