The GDE matrix explained: what the Goals for Driver Education framework says, and what the evidence shows

GDE stands for Goals for Driver Education. The framework was introduced in its present form within the EU-funded GADGET research project in 1999, and was first published internationally in a 2002 journal paper by Hatakka, Keskinen, Gregersen, Glad and Hernetkoski. Three of the five authors were at the University of Turku in Finland; the other two were at transport research institutes in Sweden and Norway.

The framework has two dimensions. The first is a four-level hierarchy of driver behaviour, running from vehicle manoeuvring at the bottom to goals for life and skills for living at the top. The second is three goals for training: knowledge and skills, risk-increasing factors, and self-evaluation. Four levels by three columns gives a grid of twelve cells.

The central idea is that basic skills are used under the guidance of higher-level goals and motives, so driver training should deal with those higher levels as well as with the basics. The 2002 paper puts it bluntly: if the motivational level fails to produce a safe way of driving, no amount of skill in traffic or in handling the car is high enough to make up for it.

• Purpose. To make the goals of driver training clearer, and to give a tool for evaluating driver-education methods along two dimensions at once.
• Status. The 2002 paper describes itself as an analytical review, not an empirical study. ERSO, the European Commission's road safety observatory, says there is little evidence so far that adding the higher levels makes training more effective, and that its effect on crash rates is not yet known.
• Ireland. The RSA's Learner Driver Training syllabus mentions the matrix once; the RSA's Driving Instructor's Handbook mentions it in its glossary and in one appendix.
• Name. The 2002 paper calls it the "GDE framework"; "GDE matrix" is the name used by the MERIT project, ERSO and the RSA, and it is the name this guide uses.

Describing driving as a hierarchy was not a new idea in 2002. Hatakka and colleagues point out that hierarchical approaches were already important in traffic psychology, citing among others Janssen (1979), Michon (1985, 1989), Ranney (1994) and Summala (1985). Earlier hierarchies, such as those of Mikkonen and Keskinen (1980), Rasmussen (1980) and van der Molen and Bötticher (1988), mainly described the performance side of driving. The GDE authors argue that a hierarchy can also take in the motivational and attitudinal side, and that while the idea of higher levels controlling lower ones is found in Janssen and Michon, those writers did not focus on how motives raise or lower risk.

The four-level hierarchy behind the matrix was developed by Keskinen (1996), building on a three-level hierarchy by Mikkonen and Keskinen (1980). Keskinen's paper, "Why do young drivers have more accidents?", came from a conference held in Köln in December 1994. His key addition was a fourth, top level, goals for life and skills for living, which links driving to how a person behaves in the rest of their life. The MERIT working paper says Keskinen's hierarchy has many similarities with Michon's, the main difference being its goal-oriented rather than behaviour-describing perspective.

The step from a hierarchy to a matrix came in the EU-funded GADGET project, which added knowledge and skills, risk-increasing factors and self-evaluation skills and linked them to each of the four levels. The 2002 paper credits its Table 1 to GADGET (1999), and its acknowledgements say the paper's content is mainly a result of that project. Its reference list gives the project's full name: Guarding Automobile Drivers through Guidance, Education and Technology. Molina and colleagues credit the original proposal to a 1999 chapter by Hatakka, Keskinen, Gregersen and Glad in a report from the project's third work package. The journal manuscript was first received in July 1999 and appeared in Transportation Research Part F in 2002.

Because of where it came from, Peräaho, Keskinen and Hatakka note, the framework is sometimes called the "Gadget-matrix". ERSO says it was later updated many times, citing a 2003 Swedish-Finnish literature review by Engström and colleagues. Two later EU projects, DAN (2000) and ADVANCED (2002), also relied on it. Kleisen, citing Keskinen, adds that it had already been used in several projects to analyse both driver training and driver testing before the first journal paper appeared.

Peräaho and colleagues report that the framework has been widely acknowledged in the European traffic research community as a fruitful starting point, and also that it has been criticised for lack of detail. ERSO calls it "best practice" because it was developed from knowledge of risk conditions and learning processes; that label describes how it was built, not the results of trials.

Later sources such as Peräaho et al., MERIT, ERSO and Molina et al. number the levels from 1 at the bottom to 4 at the top, and this guide does the same. Level names vary a little between sources (ERSO, for example, calls level 1 "Vehicle control"); we use the names from the 2002 paper. Peräaho and colleagues add a label for how general each level is: a specific task, a specific situation, a specific trip, and "global".

For teaching, the top two levels need not be kept strictly apart: Peräaho and colleagues say that although level 3 is about driving and level 4 about life outside traffic, it is "perhaps unnecessary" to differentiate much between them when it comes to practical teaching methods, and the same repertoire of methods can be applied to both.

The second dimension is three goals for training, and each of them applies at every level. The 2002 paper's column headings are "Knowledge and skills", "Risk-increasing factors" and "Self-evaluation"; other sources word them slightly differently, for example "self-assessment" in MERIT and ERSO.

Molina, García-Ros and Keskinen (2014) show how one topic can travel up each column, using examples of their own rather than official cell wording:

• Knowledge and skills, keeping a safe distance: control the car to keep one; know how it depends on road conditions; handle social pressure inside the car about it; and be aware of how your personal characteristics affect it.
• Risk-increasing factors, speeding: know its effect on car control; know its effect on adjusting to changing traffic; see its link with route planning; and see its link with your lifestyle.
• Self-evaluation, overtaking: know your strengths and weaknesses in the car-handling skills it needs; know your ability to overtake in different traffic situations; see how peer pressure affects your decision to overtake; and see how your lifestyle can affect it.

Peräaho and colleagues stress that the hard split into three columns is for theory only: in practice, teaching a skill, recognising its risks and evaluating yourself should alternate and support each other.

The table below paraphrases Table 1 of Hatakka et al. (2002), which the paper credits to the GADGET project (1999). As in the original, the highest level is the top row. The cells give examples, not a complete syllabus: the original heads them "Essential contents (examples)" and ends each list with "etc."

Scroll sideways to see all three columns →

The core assumption is that a higher level guides behaviour on the levels below it. Failure or success higher up changes the demands placed on skills lower down, and a driver's goals can either raise or lower risk. Peräaho and colleagues give a worked example of how that plays out:

This is why the 2002 paper concludes that if the motivational level fails to produce a safe strategy, no level of skill in traffic or in vehicle handling is high enough to compensate. The authors link this to earlier risk-motivation theories (Fuller, 1984; Näätänen and Summala, 1974). They also stress that driving is largely self-paced: the driver chooses the demands placed on their own skills, so good psychomotor skills alone are not enough for safe driving.

Peräaho et al. go further, calling the factors at the highest level the most important from a safety point of view, because the effect of any safety knowledge depends on whether and how the driver uses it. MERIT's working paper likewise calls level 4 the most basic and important part of driver education. Both are argued positions rather than measured findings.

For the wider psychology behind these ideas, see our overview of traffic psychology and road safety.

The GDE authors link much of young drivers' risk to the upper levels, though they hedge the claim. Peräaho and colleagues write that young drivers' crashes, especially young men's, follow "to an alarming extent" from problems that "may be attributed to" the upper half or right-hand corner of the framework, but they give no figures for this. Their reasons include the following.

• Motives, not just skill. The 2002 paper notes, citing Jessor, that young people often take risks not for the sake of risk but because the behaviour serves their developmental needs.
• Young men and women crash differently. The 2002 paper says speeding and impaired driving are typical of young male drivers and rare among young women, and puts this down to different motives (Laapotti and Keskinen, 1998). As Vlakveld reports that Finnish study of fatal crashes with a culpable driver aged 18–21, the difference is one of degree: in the loss-of-control crashes, speed was too high in 83% of the young men's cases and 40% of the young women's, and alcohol was involved in about half and 6% respectively; the results suggest the women's crashes more often came from poor vehicle handling (Vlakveld, 2011). Because nobody has suggested that women have better technical skills, the paper says gender effects are probably related to the highest levels.
• Company matters. In loss-of-control crashes, young men carried more passengers than young women, and their passengers were more likely to be friends.
• Passing the test did not predict safety. In a Japanese study (Renge, 1983), men who needed fewer driving-school hours, and men who did better in the driving test, were later involved in more crashes and violations; no such effect was found in women. In Finland, young men passed the test more easily but had more crashes after licensing (Keskinen et al., 1992). From this the authors argue that tests measuring maximal performance are not enough and should also cover driving style. Senserrick and Haworth describe the same pattern for male drivers in a passage on UK and European research into licensing tests, but the references for that passage include the 2002 paper itself, so it may not be independent support; they name no single study and give no effect sizes. We look at what the Irish test measures in Is Ireland's Driving Test Fit for Purpose?
• More skill can mean harder tasks. Citing Evans (1991), the paper says better technical skills are likely to lead drivers to take on harder tasks, such as driving faster or overtaking in heavier traffic, rather than simply to make them safer.
• Overconfidence. ERSO says young drivers, and young men in particular, tend to overestimate their skills and underestimate how complex traffic is, and so choose safety margins that are too small.

The GDE literature itself says skill and experience matter. The 2002 paper notes that novices' skills are not yet automatic, which leads to information overload and poorer observation, that experience has a strong effect on crash liability, and that more practice before licensing reduced novices' crash involvement in a Swedish evaluation (Gregersen et al., 2000), though two of the GDE authors later warned that this result may have been affected by self-selection, and that evaluations of accompanied driving in France and Norway found no safety benefit once drivers held a full licence (Keskinen and Hernetkoski, 2011). ERSO adds that handling a car can be learned in a few hours, while anticipating hazardous situations takes years of practice. MERIT points out that the only way to build routine is to drive, often in the first months with a licence, which is also the most dangerous period of a driving career. Isler and colleagues draw a line between deliberate risk-seekers and inexperienced drivers who take risks because they do not yet realise the dangers.

The GDE authors use skid training as a warning example. The 2002 paper says several attempts to improve safety only through lower-level skills, such as handling a car on slippery roads, failed to decrease accidents. It also says some groups benefit from slippery-road training while others get a negative effect from it, without saying which groups or by how much.

Figures from Norway illustrate the concern. After Norway made a skid-training course compulsory for new car drivers in 1979, Glad's (1988) evaluation found that total accidents among novice male drivers rose by 17% and accidents on slippery roads by 23%, compared with a control group that had no skid training (Shinar, 2017, p. 313; the same figures are reported by Katila et al., 2004, as cited by Rosenbloom et al., 2008). Glad (1988), a Norwegian-language report, is one of the two studies the 2002 paper cites for its point; the other, Christensen and Glad (1996), concerned a later mandatory course for drivers of heavy vehicles. Two of the 2002 paper's authors describe both evaluations as giving clear negative results and as based on an experimental design, which is unusual in driver-education research (Keskinen and Hernetkoski, 2011).

Finland drew a different lesson. It moved its skid training away from manoeuvring skills and towards anticipation and awareness of the risks of slippery conditions. Katila et al. (2004), as reported by Rosenbloom et al. (2008), then found that the training had no effect on slippery-road accidents for men or women, although trained drivers were more confident and less afraid of slippery conditions than untrained ones. That extra confidence did not lead to more slippery-road accidents, and Katila and colleagues argued that what matters is how skills are used. Rosenbloom and colleagues agree that saying more confidence will inevitably lead to more accidents is an oversimplification, while still seeing training that avoids overconfidence as a worthwhile goal.

• ERSO: courses focused on advanced vehicle control such as skidding are not advisable for novice training, because they may lead to over-confidence, and skid control needs extensive, regular practice that a one-day course cannot give.
• Isler et al.: when skid training is combined with insight and safety-skill training, positive outcomes can be expected.
• Peräaho et al.: any skills training should be combined with, or followed by, training in the risks of using those skills, and a skills exercise should not end in success. They list conditions under which overconfidence has occurred, including too much emphasis on manoeuvring and too little on risk awareness, and exercises not followed by discussion, but they cite no study for that list.
• The principle behind it: avoiding a critical situation matters more than handling it; a skilful driver can make an evasive manoeuvre, but a safe driver avoids getting into the emergency.

Our reading is that the evidence does not show skid training always increases crashes; it shows that the same skill can be used to drive more safely or to drive closer to the limit, which is the GDE argument in miniature. For the wider background see Risk Compensation.

The 2002 paper draws two main conclusions: motivational aspects need more emphasis in driver education, and teaching methods should be re-evaluated, with more use of active learning and self-reflection. Peräaho and colleagues turn that into more detailed guidance, summarised here.

Peräaho and colleagues reproduce example exercises from practitioners, taken from the ADVANCED and GADGET projects, and Isler and colleagues describe one more. They are illustrations, not a prescribed curriculum:

• A failed overtake (level 1). On a track, learners accelerate to about 50 km/h, steer past an obstacle and try to get back into lane, and the car starts to slide. The group then discusses what their chances would have been at 100 km/h.
• A following-distance demonstration (level 2). Learners estimate the usual safety margin at 50 km/h, two instructor cars drive at that margin, and the lead car brakes hard at a point the learners do not know in advance. The aim is to show how far a car travels in the critical "one second", and that the "quick reactions" young drivers often talk about do not exist.
• Trip-planning problem solving (level 3). A group discussion of learners' own experiences as passengers or drivers, then of the risky situations they identified, then a scenario to work through, such as a tired driver setting off on a Friday.
• A timed pressure course (level 3). Learners are told to finish a course as fast as possible, but the real aim is to show how hard it is to drive under pressure from a loud radio, questions, penalties and onlookers.
• A self-evaluation drive. In Isler and colleagues' New Zealand study, the young driver and a driver coach each rated five skills after an on-road drive, then discussed the differences until they agreed; the agreed rating was often lower than the participant's own, and the aim was better calibration.

A 2003 Swedish-Finnish review by Engström and colleagues, as summarised by Senserrick and Haworth, recommended that training treat self-evaluation as "essential but not automatic" and include small-group peer discussion of norms, lifestyles and life goals. It also concluded that methods which let young drivers actually experience risk and their own shortcomings are more effective, though the summary gives no data for this. Our page on learning from errors looks at why a practice exercise does not always need to end in success.

ERSO describes the EU HERMES project as an attempt to combine the GDE matrix with learner-centred teaching: instead of giving instructions, a coach asked questions that prompted learners to reflect on how they drove and how they could improve. MERIT's working paper calls coaching an optimal method for addressing attitudes to risk, which it places at level 4, and says coaching is designed to improve self-awareness, in line with the matrix's emphasis on self-evaluation at all four levels.

HERMES defines coaching as a learner-centred method that engages body, mind and emotions to build awareness and responsibility, with learner and coach on an equal footing. The RSA's Driving Instructor's Handbook describes it as shifting training from the ADI as the expert to the learner taking more responsibility for their own driving, in partnership with the ADI.

Bartl, who managed the HERMES project, argues that coaching is the way to put the matrix into practice and that, for higher-level issues, it is ahead of teaching methods; the report presents this as a position, not a tested result. The report also admits that a three- to five-day course is unlikely to turn someone into a very good coach, so other methods will probably continue to be used.

For practical coaching moves in real lessons, see Coaching, Not Instructing and our coaching methodology.

In 2002 it was explicitly a hypothesis: Hatakka and colleagues wrote that pre-licence education is hypothesised to be still too concentrated in the low-left boxes of the framework, while the upper-right boxes tend to be ignored and lack effective methods. Peräaho et al. describe basic driver education as mainly covering the lower-left diagonal half of the matrix, with the link to personality and motives "still in its infancy". MERIT's working paper and ERSO say training focuses on levels 1 and 2, without citing data at that point. ERSO also says, without giving a reference, that only a few trainers are able to apply the matrix in their work, because of lack of time and of background from their own training.

Molina, García-Ros and Keskinen point out that such claims rest mainly on expert opinion rather than empirical findings. MERIT's 2004 questionnaire, for example, went to 30 European countries (27 replied) and was answered by government bodies and/or national driving-school associations, who rated how far each cell of the matrix was covered in the obligatory training of driving instructors. Coverage leaned towards the lower-left of the matrix, though it varied widely between countries; Molina and colleagues read the results as showing self-evaluation especially neglected. Their own 2014 study is the largest survey we found:

Instructors rated, from 0 to 10, how far they thought each of twelve goals, one for each cell of the matrix, was achieved in training. The questionnaire did not mention the GDE. All four level averages were above 7, and self-evaluation was the lowest-rated column (7.34, against 7.88 for knowledge and skills and 7.83 for risk factors). The authors call the instructors' view "rather positive" overall and see the relative shortfall at the higher levels as a potential weakness of the Spanish curriculum. No significance test of the differences between levels is reported. The survey records instructors' opinions, not what learners actually learned, and the authors say their results apply only to Spanish instructors.

Two smaller sources point the same way. HERMES audits in Austria found instructors' communication focused mainly on the lower levels of the matrix, with no talk at all about personal beliefs and attitudes in theory lessons. In group interviews with 19 students for an Australian PhD thesis, Kleisen found that participants saw learning to drive as learning to handle the car, and did not mention the role of skills such as situational awareness.

The wider picture for driver training is sobering. ERSO cites several reviews concluding that formal pre-licence training does not reduce novice crash rates, and says formal training to pass the test does not lead to fewer crashes after licensing than learning informally with family or friends. Zhang and colleagues, citing literature reviews going back to the 1980s, say none has found evidence that driver education reduces teenagers' crashes on the open road. For post-licence education, Senserrick and Haworth report a Cochrane review of 24 randomised controlled trials (Ker et al., 2003) that found no sign that any one form was more effective than another, and Rosenbloom and colleagues cite a 2005 meta-analysis of 21 studies by Ker that found no definite proof that post-licence training prevents crashes. ERSO adds that advanced post-licence training has been shown to influence behaviour and attitudes, but evaluations of its effect on crash risk are inconclusive. Kleisen describes the effectiveness of driver education for safety as "as yet unproven".

The GDE authors argue that this does not settle the question. Hatakka and colleagues accept that criticism of formal training's limited effect on accidents is justified, but say it does not prove training useless, since no single measure can cut accidents radically. They add that a lack of clear results for one method does not show that training in general is ineffective, and that many studies compared formal with informal training without analysing what the formal training taught or how. Isler and colleagues cite Lonero (2008), who found that most programmes had received little formal evaluation, and add that crash numbers are a crude outcome because crashes are rare for individual drivers. They also report the view of Crick and McKenna, cited in Haworth et al. (2000), that concluding driver education does not work may be premature, because the lack of evidence may reflect weak evaluations or course content.

Scroll sideways to see the whole table →

ERSO summarises Isler's trial as showing better outcomes for the higher-order group. The paper's own results are more mixed. On the on-road assessment the gains were improvements within each group: after training, the groups did not differ significantly, and the vehicle-handling group's overall score improved too. The clearer difference between the groups was in self-reported attitudes to risky driving, which after training were significantly safer in the higher-order group than in the vehicle-handling group; for overall confidence, the change from before to after training also differed significantly between the groups. Whether the safer attitudes will lead to safer driving "has yet to be determined", the authors say. They also note that earlier evaluations of "insight" training, which aims to raise awareness of risk and reduce confidence, reported positive long-term safety effects; those results come from the studies they cite, not from their own trial.

A US simulator study by Zhang and colleagues (2018), which does not mention the GDE, found that training in hazard anticipation and attention improved the performance of "careful" drivers but not of "careless" ones. The authors suggest that, for careless drivers, education aimed at attitudes might work better than skills training, a conditional suggestion they did not test, and they note that their test scenarios resembled the training scenarios. Our reading is that this fits the GDE idea that motives and style can stop trained skills being used, but the link is ours, not the authors'.

• It lacks detail. Peräaho and colleagues acknowledge that the framework has been criticised for lack of detail, without saying by whom.
• It describes driving in general. The same authors say it is not entirely suitable for describing the behaviour of a particular driver.
• The columns have not shown up in data. In the Spanish instructor survey, a factor analysis split the twelve items into two factors, one made up mostly of level 3 and 4 items and the other of level 1 and 2 items (one level-2 self-evaluation item loaded with the higher levels), and neither factor separated the three columns. Because the two factors were highly correlated, the authors went on to treat the twelve items as a single scale. The authors do not conclude that the columns are invalid; they call for further analysis, and note that measuring everything in one survey may have affected the result. Peräaho et al. themselves say the split into columns serves theoretical purposes only.
• It frames safety negatively. Kleisen argues, as her own critique, that the framework is built from a negative angle, highlighting "risk increasing factors" and "risky tendencies". Citing Keskinen (2007), she adds that models should also describe how normal, trouble-free driving happens, which she sees as especially relevant because the GDE model is used in driver training.
• Awareness may not be enough. Zhang et al., citing Mayhew et al. (2014), link young drivers' skill deficits to the finding that programmes focused on awareness of risk-taking do not appear to produce lasting change in behaviour, a link they hedge with "probably".
• Where is the line on privacy? The 2002 paper itself notes that trying to change the highest level can be seen as intruding on a person's private life and values.

The RSA refers to the GDE matrix only briefly. The Learner Driver Training (LDT) syllabus says it emphasises self-assessment and self-reflection, and that research shows these are important skills in line with the higher levels of the GDE matrix. For more information it points to the final report of the EU GADGET project. It does not name the four levels, describe the columns or say which levels count as "higher".

The Driving Instructor's Handbook defines the GDE matrix in its glossary as organising driving behaviour into four hierarchical levels, and says it stresses learners' awareness of their own motivation, reflection on their role among other road users, and the ability to drive a vehicle. It does not name the four levels or the columns, and gives no source for the matrix. Its only other mention is in an appendix on RSA school programmes, which says that ADIs who consistently reinforce and build on those programmes' road safety messages will help learners to achieve "the goals associated with more effective learning as outlined in the GDE matrix". The handbook describes the LDT syllabus as a shift in emphasis from handling the vehicle to interacting with other road users, and says it is "now recognised" that too much emphasis was put on vehicle-control skills in the past, without giving a source.

What the RSA does build, in detail, is a practice of self-assessment and self-reflection:

• Definitions. Self-assessment is gathering information about your own driving, judging its quality and deciding what to improve; self-reflection is critically considering your previous actions, incidents and decisions.
• Why. The syllabus says drivers who are more aware of their strengths and weaknesses are likely to be safer, and that self-assessment helps guard against over-confidence, which it calls one of the major dangers for novice drivers; it gives no data for this.
• How. ADIs and sponsors should ask questions in a sequence, starting with self-assessment and moving on to self-reflection. The syllabus example after a roundabout runs: how well do you think you negotiated it, what makes you say that, why do you think that happened, and what will you do differently next time?
• Errors. When a learner makes an error such as following too closely, the syllabus asks whether it points to a lack of knowledge and understanding, a lack of skill, inappropriate attitudes, or a combination.
• Coaching. The syllabus places emphasis on coaching and instruction that turn learners into active participants, and the handbook calls encouraging self-analysis one of the ADI's most important roles.

The LDT syllabus is organised into four modules and five "main areas of driver competence". These are the RSA's own structures, and the syllabus does not map them onto the GDE levels or columns.

Essential Driver Training (EDT) is a programme of 12 one-hour lessons whose objectives came from an analysis of driving situations shown to contribute to collisions and deaths on Irish roads. The RSA says EDT is derived from the LDT syllabus, does not cover all of it, and complements rather than replaces ordinary lessons. The EDT guidelines ask learners to leave at least two weeks between lessons and recommend taking at least six months over the 12. EDT is compulsory: the RSA's learner booklet says that anyone whose first category B learner permit is dated on or after 4 April 2011 must take the course, and must complete it before taking the driving test. EDT learners keep a logbook with a self-analysis page, which the RSA suggests ADIs use to plan lessons, while respecting a learner's choice not to share it. The RSA expects EDT to have a positive effect on collision rates; that is a stated expectation, and neither RSA document presents an evaluation. Our EDT lessons guide covers the programme lesson by lesson.

How we apply this in our own lessons is set out in The Professional Standard, with practical coaching moves in Coaching, Not Instructing and our coaching methodology.