Fear of Driving: What the Research Actually Says

When somebody rings a driving school and opens with "sorry, I think I'm a hopeless case", they almost always say it believing they are the only person in the county in that position. They are not.

Two population studies used the same cut-off and disagreed substantially. In New Zealand, 16% reported moderate to severe driving anxiety, with a further 52% reporting mild anxiety and only 31% none at all. In France, the same thresholds applied to a nationally structured sample of 5,000 adults produced 32.4%. strong

Taylor 2018, Transportation Research Part F 58, 70–79. 1,500 adults drawn at random from the New Zealand electoral register, 441 returned the survey (29.4%). The author states plainly that at that response rate "this cannot be considered to be a prevalence study", while remaining the largest population-based study of driving anxiety there is. Fort et al. 2023, Journal of Transport & Health 32. The gap between 16% and 32.4% is real and unresolved, so the honest answer is a range rather than a single figure.

Whichever figure you take, the conclusion is the same: this is not a fringe problem. By contrast, driving phobia in the diagnostic sense is rare, at roughly one person in a hundred. The orders of magnitude are completely different. What follows from that for driving lessons, nobody knows. No study has ever established who needs lessons and who needs treatment. We return to that near the end.

Becker et al. 2007, European Psychiatry 22(2), 69–74. Caveat: the 1.1% figure comes from a study of young women in Dresden, refers to occurrence at any point in life, and we have only seen it quoted by other authors rather than in the original table. Comparing it with a 0 to 10 self-rating is a comparison of orders of magnitude, not a precise ratio.

One finding from the New Zealand study deserves a sentence of its own: the groups at different anxiety levels did not differ in crashes or incidents over the previous twelve months. The more anxious respondents did more avoiding, held more negative thoughts about driving and felt less safe, but they were not crashing more. Being a nervous driver is not, in this data, the same as being a dangerous one. Very few of them had sought any help.

It is also worth turning one assumption on its head immediately. In the French sample, the older the respondent, the lower the reported anxiety. The relationship is weak (r = −0.195, so age explains a few per cent of the differences between people) but it runs consistently in the direction nobody expects. Men and women reported comparable levels up to about 35, with the gap opening only after that.

The most telling number for a driving school is a small one. Among those five thousand adults, 59 people had never taken a driving licence because they were afraid to drive (mean anxiety 7.39 out of 10, 69% of them women). A further 129 held a licence and never drove at all. These people almost never appear in any driving school's figures, because they never ring.

So what do people name instead? Two things nobody asks about:

• Panic attacks. In that same 1994 study, 53% named a panic attack as the primary reason, more than three times as often as a crash. 81% had experienced a panic attack at some point, though only 14% met the criteria for panic disorder.
• Criticism from whoever taught them before. In an Australian study of 1,600 people, criticism was named as often as knowing a crash victim. In a study of 190 anxious drivers, driving with someone who criticises your driving was rated as producing a high level of anxiety. The strongest confirmation comes from the structured French sample of 5,000: "driving with a person who criticises your driving" is the second most avoided driving situation of all, behind only driving while tired, and ahead of driving in fog.

Half of anxious drivers cannot name a starting point at all

The largest of these studies asked directly and published the full breakdown:

Stephens et al. 2022, n = 1,600 people reporting some level of driving anxiety (77% women, 75% licensed, 20% on a learner permit). weak This is an online survey of people who volunteered that they were anxious, and the starting point is recalled from memory, so answers are open to whatever explanation comes most easily.

That first row matters more than all the others put together. Half of anxious drivers cannot point to any event that started it. If you have been searching your own history for the moment that explains all this, and coming up empty, it does not mean you have missed something or that you are making a fuss. It means you are in the largest group.

A knock without injuries goes with lower anxiety, not higher

The studies above share a weakness: they recruited volunteers. The French sample of five thousand was structured to match the population, and allows four groups to be compared directly.

Fort et al. 2023, n = 5,000, F(3,4995) = 32.32, p < 0.0001. strong People who had been in a crash without injuries reported significantly lower anxiety than people who had never crashed at all. Data gathered at a single point in time, so it cannot say what causes what.

Read it carefully, but the direction is clear: a knock on its own is not enough to produce a fear of driving. There is an explanation the data does not rule out, and it changes the sense of the whole table: people who drive a lot have more minor collisions and tend to be less afraid. The relationship may run the other way round. What is certain is that the no-injury group is not more frightened. Only a crash with injuries lifts the average, and then only modestly.

The most cited reason is other drivers, not your own ability

Asked outright what contributed to their anxiety, respondents in that same sample ranked it: lack of confidence in other drivers' behaviour (1.88 on a 0 to 3 scale), then having heard about the dangers of driving through the media (1.40), then witnessing a crash (1.29), and only fourth, lack of confidence in their own ability (1.28). The gap between first place and the rest is clear; second to fourth are separated by hundredths of a point, so that ordering should not be read literally.

Fort et al. 2023, asked of the 3,907 respondents reporting any level of anxiety. These are reasons people attribute themselves, not measured causes.

This inverts the usual conversation in a driving school. The nervous learner is normally convinced the problem is them. The most commonly named reason is what they expect from everybody else. The practical consequence, and this is our inference rather than a finding, is that building skill alone will not fully address it. Reading other road users, and leaving yourself margin for their mistakes, needs to be worked on in its own right.

A cluster analysis, meaning a statistical search for groups of similar people, run on 85 anxious drivers found two distinct focuses of fear. The distinction is simple enough to draw out with a single question.

The question is: are you afraid of what the car and the traffic might do, or of how you are going to feel?

Taylor, Deane & Podd 2000, n = 85, volunteers recruited through the media. weak The split into two focuses of fear comes from that study. The "what we think helps" and "hurts" columns are our conclusion, not its findings: the authors tested no teaching method, and nobody has checked whether matching approach to type changes anything. As a way of opening a conversation it works. As a typology it is not established.

Separately, people whose anxiety began with criticism are more afraid of others' judgement and of having a panic attack than people whose anxiety began some other way, while being no different on fear of crashing. For that person, safety statistics and a talk about airbags miss what is actually raised. What lands instead is: nobody in the back seat, an explicit "I will stay silent through this manoeuvre", and no corrections in front of a third party.

Stephens et al. 2022, n = 1,600. The comparison is between groups with different starting points, not against people with no anxiety at all.

In a German sample of 810 people the most commonly named situations were bad weather (rain, fog, ice) at 19.8%, unfamiliar routes at 16.7%, dual carriageways and motorways at 14.4%, and heavy traffic at 14.2%.

Fischer, Schröder, Taylor & Heider 2021, European Journal of Psychological Assessment, online sample n = 810 (74% women, mean age 28). Important caveat: a convenience sample gathered through social media, medical facilities, university mailing lists and online self-help platforms, so it is not a population estimate. The ranking of situations is what holds, not the absolute values.

The structured French sample produces a very similar ordering despite asking in a completely different way. Most feared there were driving in fog (2.4 on a 0 to 3 scale), driving while tired (2.35) and driving in heavy rain (2.3). Weather and visibility come top in both, which is a reasonably strong signal given how differently the two samples were assembled.

But the most interesting figure is a different one. Total avoidance was reported by 3% of people. Partial avoidance, meaning "I drive, but only around my own area and on short trips", by 23.2%. That proportion comes from a volunteer sample and should not be transferred to the population. What holds is the direction: there are many times more people quietly avoiding than people who will not get in a car at all.

This is the quiet problem in the whole subject. Someone who cannot get into a car knows they have a problem. Someone who has driven the same route to work for ten years and goes around the ring road usually thinks they simply prefer it that way. In that same German sample, over half of those who admitted to fear had been afraid for at least three years.

It is also worth noticing what is absent from these lists: the quiet, familiar estate. Which is precisely where a nervous learner will ask to stay, and precisely where lessons quietly stop producing progress.

Not tunnel vision. Fewer glances in the mirrors

A study run in real traffic in an instrumented car found no tunnel vision. What actually happens under heavy mental load is the gaze concentrating on the road ahead while glances into the mirrors drop away. In a simulator study, time spent looking at the road rose from 82.2% to 86.2%, while the share going to the right door mirror fell from 7.1% to 5.6%. weak

Recarte & Nunes 2003, full text (n = 12, real traffic, instrumented car); Broadbent et al. 2023 (n = 21 in a fixed-base simulator, though eye data was captured from only 17, so the percentages rest on 17). The statistically significant drop applied to one of three mirrors; the rear-view and left mirrors did not reach significance. Lane position was not measured with any sensitive continuous metric and the authors flag that themselves as a gap. Participants had held licences for about five years on average, so they were not learners. Common caveat: both studies manipulated mental workload, not fear, and measured nobody's anxiety. That worry loads a driver in the same way as a mental task is our inference, not their finding.

The practical significance is considerable. An instructor checking "is he still holding the lane while I talk to him" is asking about something that study did not measure. The infraction count genuinely did not change, but the authors note their measures were too coarse to detect anything here. The question that tells an instructor more is: is he still looking?

Anxiety costs effort first, and causes errors only later

A meta-analysis of 58 studies across more than eight thousand people shows that anxiety degrades efficiency (reaction time, effort) while effectiveness (accuracy) does not significantly suffer. The deficits appear in inhibiting the wrong response and in switching attention, but not in refreshing information in working memory, and they grow when the load is high. strong, not about driving

Shi, Sharpe & Abbott 2019, 58 studies, N = 8,292, g = −0.58. We searched the full text: it contains no study conducted in a car. Transferring it to a driver is our assumption.

Transferred to the road, and nobody has tested this in a car, it would look like this: the first symptom is not an error, it is cost. You will get through that junction, only slower, later, with more effort and with nothing in reserve for whatever happens next. "I can do it, but it wipes me out" is not an excuse. It is exactly what the theory predicts.

More often than panic, you see excessive caution

On the road, anxiety usually does not look like panic. It looks like excessive caution. That does not mean panic never happens; for some people it does, and there is guidance on it further down. A systematic review of 29 papers links driving anxiety with over-cautious driving, negative feelings and avoidance. The authors themselves note that most conclusions rest on questionnaires and that little is known about behaviour in actual driving. weak

There is one study in which anxious people were actually put behind the wheel and assessed on a standard sheet by a professional instructor who did not know which group they were in. The result: the anxious group made more errors (38.8 against 30.6, d = 0.63), but the five most frequent errors were identical in both groups, with observation technique at junctions topping both lists. Anxiety changes the number of errors, not their kind.

Taylor, Deane & Podd 2007, Behaviour Research and Therapy 45(4), 805–818, table 1. t(98) = 3.52, p = 0.001. Important caveat about who these people were: all women, average age 43, with around twenty years of driving behind them, recruited through media advertising. These are experienced drivers who are afraid, not learners. The authors say the findings need repeating on more varied samples.

In that same table there is a detail that says more about excessive caution than anything else here. There is one category where the anxious group did better than the control group: following distance, the two-second rule (0.17 errors against 0.28). People who were frightened of driving kept a bigger gap to the car in front than people who were not. Their problem was not a shortage of caution.

This has an uncomfortable consequence for the test, though it comes from the marking sheet rather than from research: excessive caution is itself a marked fault. Progress is assessed on reasonable pace and the absence of undue hesitancy, and inadequate progress at junctions and roundabouts appears on the RSA's own list of common reasons for failing. A nervous candidate does not crash the car. They hesitate at junctions, drive well below the limit and pick up faults for progress and observation. Reassuring that person that "being careful never hurt anyone" ignores what the sheet looks like. Excessive caution is not safer, it is differently risky.

Being nervous is probably not the same as being alert

In a study measuring skin conductance, meaning how much the palms sweat, learner drivers responded physiologically to developing hazards rather less often than licensed drivers, not more. The difference did not reach statistical significance, and the licensed group was also significantly older, so experience and age cannot be separated here. weak

Chirles et al. 2021, n = 38 (20 learners, 18 licensed), a seated video task. Main effect p = 0.065, significant only for hazards that developed early in the clip (p = 0.017). No anxiety questionnaire was used.

Tension caused by an instructor or an examiner is probably not the same system as alertness to a hazard. But let us be explicit about what that study did not test: it measured neither anxiety nor the presence of an instructor, and only compared learners with licensed drivers. As far as anyone can say today, being tense does not by itself make you a more observant driver. It does not make you a worse one either.

• One study of a real test, on thirteen people. Candidates who failed had significantly higher anxiety and heart rate during the actual test, and their anxiety measured before it began sat at the very top of the scale, higher than practically everyone in the comparison norms. But the authors write plainly that the direction of causation cannot be established: the raised heart rate may have been the result of the faults committed. The sessions were not rotated, and in twenty years nobody has repeated that physiological measurement on a real test. weak
• Two randomised trials. In both, the intervention genuinely lowered pre-test anxiety, and in both the difference in pass rate was not statistically significant. In the first, outcome data survived for only half the participants (34 of 68), and the raw difference was large (29% failed against 53%), so that is more likely a lack of statistical power than evidence of no effect; a secondary analysis excluding four people who said they did not believe in placebos did come out significant. In the second, on seventy candidates selected precisely for high test anxiety, anxiety on the day fell and the pass rate did not move.
• Two attempts to raise anxiety experimentally, both in a simulator. In the first, anxiety was successfully raised and driving did not change. In the second the manipulation did not work at all, and when the change in anxiety was analysed regardless of group, it was associated with more wandering within the lane, but only where the visual load was high. The participants were young licensed drivers, not candidates on test.
• The closest analogue, a practical clinical exam scored by an examiner. In a systematic review, six of eight studies found no association at all between anxiety and performance.

Fairclough, Tattersall & Houston 2006 (n = 13); Schaefer & Enge 2024 (n = 68, outcome data for 34) and Biassoni et al. 2026 (n = 70, active control group, no difference in pass rate); Barnard & Chapman 2018 (two simulator studies, licensed participants); Martin & Naziruddin 2020 (review, 9 papers). Caveat on the last: it counts how many studies came out "yes" and how many "no" rather than pooling their results, and that method inherently favours a null finding.

Two things that look contradictory need reconciling here, because they are not. There is no evidence that how nervous you are predicts the test result; that claim has simply never been tested properly. There is, however, a specific behaviour that is common in nervous candidates and that an examiner marks regardless of where it came from: hesitating at junctions, driving well below the limit, failing to make progress. The sheet does not assess how you feel. It assesses what the car did. That is why the work goes into behaviour rather than into you being calm on the day, because nobody can guarantee the latter and nobody has shown it is the thing that decides the outcome.

A mock test result may tell you nothing about who will pass

The same study threw up something that, if it holds, changes what a mock test is for. Candidates who went on to fail made fewer minor faults in the mock (6.1 on average) than those who went on to pass (11.3). weak

Fairclough et al. 2006. The caveat the authors raise themselves: the mock was marked by the candidate's own instructor and the real test by a government examiner, so the difference may be an artefact of who was holding the clipboard. At thirteen people this is a signal, not proof.

The point of a mock test therefore lies not in the score but in the arousal. A mock with a familiar instructor who uses your first name and comments as you go does not reproduce what happens on the day. A mock with a stranger who says nothing and hands over the sheet only at the end reproduces it considerably better.

There is support for this from outside driving. A meta-analysis of fourteen studies of pressure training (sport and law enforcement) gives g = 0.67 for performance under pressure after one outlier is removed, and the most commonly used manipulation was precisely being judged by an observer, in 8 studies of 14. Novices did at least as well from it as experienced performers. strong, not about driving

Low et al. 2021, Sport, Exercise, and Performance Psychology 10(1), 14 studies, n = 394. Fundamental caveat: not one of those studies involves driving a car. The authors state plainly that none of the variables they examined, including number of sessions, was a significant moderator, and that all the novice studies were also short ones. Long runs of more than five sessions came out weakest (g = 0.42), but that is three studies with a confidence interval crossing zero, so treat it as a hint in favour of two or three proper mocks rather than a weekly ritual, not as proof.

The pass rate does not rise with later attempts

This part rests not on a study but on the complete register of every category B test conducted in a year, so there is no sampling problem. There is a different problem, addressed below: who ends up in which column.

DVSA register (Great Britain), financial year 2025/26, table DRT121D. strong This is a selection effect, not a causal one: the pool sitting a sixth test is simply a different pool from the one sitting a first. "Sixth or later" is one bucket for everything from the sixth attempt upwards, so 43.8% describes no particular attempt. British data; the Irish figures are below.

Read that precisely. The data does not say a later attempt is harder for you. It says that on average it is not easier: from the second through the fifth attempt the rate sits around half, the same as the first, and even at the sixth more than four in ten pass. The reflex of "just book another test, it'll come right next time" has no support in these numbers. A new date on its own changes nothing. This next part is our recommendation rather than a research finding: after a second failure, what makes sense is a diagnosis, meaning which specific aspects, which routes, which situations, and only then a date.

For Irish context, in the twelve months to June 2026 the category B pass rate was 52.4% nationally and 48.1% in County Dublin, against 57.4% in Meath and 54.9% in Kildare. The spread between counties is about nine percentage points. How much of that comes from harder traffic and how much from the fact that a different set of people sit the test in each county cannot be separated from this data. It is mainly worth knowing so that a failure does not get read as something personal.

CSO / Road Safety Authority, table ROA31, taken directly from the CSO data service and recalculated. No-shows: 9,469 nationally and 3,388 in Dublin, being 3.4% and 4.0% of tests delivered plus no-shows. Counting tests not conducted or abandoned in the denominator as well gives 3.2% and 3.6%.

The largest study of learner drivers ever conducted followed 42,851 people in Great Britain. The averages came out as 52 hours of professional tuition plus 19 hours of practice with friends and family. Among those who passed it was 67 hours in total (47 with an instructor and 20 privately). Practically everyone (99%) took some professional tuition, but only just over half (55%) practised privately. strong

Wells, Tong, Sexton, Grayson & Jones 2008, Cohort II: A Study of Learner and New Drivers, Road Safety Research Report No. 81, Department for Transport (full 223-page report). Caveats: British data now some years old, hours recalled by the learners themselves, and it covers learners in general rather than people who are afraid of driving.

Two things follow for anyone who is frightened. First, the number of hours rises with age, consistently, up to the 40 to 49 group. People aged 17 to 19 spent roughly half as long learning as people in their thirties and above. Women spent on average eighteen hours more with an instructor than men, while men did about three hours more private practice. If you are starting in your thirties or forties and it feels like slow going, that is not an impression. It is the norm.

Second, something that looks absurd until you think about it: the pass rate fell as the number of hours with an instructor rose. That does not mean lessons are harmful. It means the same thing as the retake table above: hours do not determine the outcome, the outcome determines the hours. People who find it harder take more lessons. Which is why there is no number of hours that guarantees a pass, and nobody honest will give you one.

One genuinely useful finding sits alongside it: the pass rate was higher for people who did any private practice at all than for those who did none, but barely rose with more private hours after that. The fact that you drive between lessons is worth more than exactly how much you drive.

An automatic removes one layer of load. A 1998 study found that in novices, changing gear in a manual measurably degraded their detection of road signs, while making no difference to experienced drivers. Nobody has tested whether an automatic helps people who are anxious, so that is our inference rather than a finding. The honest balance: an automatic removes an entire stream of work, at the cost of a licence restricted to automatics. For somebody drowning in workload that can be a good trade rather than a defeat. weak

Shinar, Meir & Ben-Shoham 1998. A single study, never repeated, and we could not obtain the full text so the sample size is unknown to us.

The problem with breathing as the main tool is subtle. If breathing is the thing that "gets you round the roundabout", then having got round it you will credit the breathing rather than yourself. Breathing to steady your hands before a specific action is fine. Breathing as a ritual for controlling anxiety becomes a prop, which we come back to below. Put briefly: if breathing helps you, breathe. Just do not build the plan around it, because the plan is supposed to consist of driving.

What works instead is cheaper: evidence you generated yourself. "You said you'd stall on that hill. You didn't. What did you do differently?" If you ask "was that alright? was that alright?" after every manoeuvre, that is reassurance-seeking, and answering it every time feeds it.

Read honestly, the original is an argument for lowering pressure on a difficult, not-yet-mastered skill. Precisely the opposite of the way it gets quoted.

None of which makes a ritual worthless. It makes it something that should be free, provided by your instructor, and described honestly: this is to settle your nervous system, not to make you drive better.

It is graded in vivo exposure: systematically driving where you avoid, in increasing difficulty. It is neither new nor impressive. It has the strongest evidence base of anything discussed here, and the source of that base needs stating immediately: it comes from treating phobias in general, not fear of driving. For fear of driving specifically, no method has a full-scale evidence base.

Wolitzky-Taylor et al. 2008, 33 randomised trials, 90 treatments, n = 1,193. strong, not about driving None of those trials concerns driving phobia: they are spiders, snakes, heights, claustrophobia, flying, dentists, injections.

Two things are worth taking from that table. First, the active ingredient, meaning the part that does the actual work, is exposure itself: treatments containing exposure beat psychotherapies without it (0.44) and that advantage holds up at follow-up. A real car on a real road beats imagining it and beats VR (0.38), though the advantage melts away after a few months (0.20). About conversation, videos or watching your instructor, that analysis says nothing, because those were not the comparisons. Second, and this is uncomfortable for the whole industry: placebo alone scored 0.57. Somebody feeling better after a chatty "confidence lesson" is not evidence that the method worked.

Exposure is not about waiting the fear out

The older version said: stay in the situation until the fear halves. The modern model says something different: the aim is not for anxiety to fall during the session, but for a specific prediction to be disproved. weak

Craske et al. 2014 and 2022, the inhibitory learning model, in which the old fear does not disappear but a new, competing piece of learning grows alongside it. Caveat: this is a well-argued theoretical framework, but there are few trials comparing it head to head with the older version on clinical outcomes. It is the best available model, not a demonstrated advantage.

It is worth separating two claims that easily blur together. That exposure works has strong evidence. How best to run it, meaning whether testing predictions beats waiting for the fear to subside, is currently the best available model rather than a demonstrated advantage. In practice this matters little, because both versions tell you to do the same thing: drive where you avoid. They differ in how you know the lesson went well.

A ladder, meaning a list of situations ordered from easiest to hardest, is the basic tool. Every rung looks the same:

Vary the context, do not perfect one route

The model predicts that variety, meaning different times of day, different roundabouts, different weather, is what protects against relapse. Perfecting one safe route produces a skill that transfers nowhere. That is a prediction of the model rather than a separately demonstrated result.

Name your props and agree when you give them back

Props in learning to drive look like this: the instructor's hand hovering over the wheel, only ever the quiet estate, never above 50, always the same route, mum in the back seat. In a randomised trial, exposure with "judiciously used" safety behaviours produced neither better nor worse outcomes than exposure without them, and the same acceptability. But the people allowed to use them regularly explained the good outcome by the prop itself, along the lines of "it didn't bite me because I had gloves on". strong, not about driving

Blakey et al. 2019, n = 60, spider phobia, four sessions, and in the "judicious use" group the number of permitted safety behaviours fell from two to zero over the course of treatment. The number of such explanations was significantly greater than zero, though there is nothing to compare it against, since the other group had no safety behaviours at all. A review by the same team recommends eliminating safety behaviours as quickly as the patient is willing. We found no study of safety behaviours in a driving context.

In practice: props are fine for getting somebody moving. They just need naming at the first lesson, with a written agreement about when they go. The sentence that signals trouble is "I only managed it because you were there".

Spread the learning out, but attack a specific fear in a block

Drivers who qualified through an intensive course reported an incident in their first two years significantly more often (43%) than those trained traditionally (26%), on almost identical total hours and with no difference in the number of test attempts. weak

de Craen & Vlakveld 2013, 35 intensive-course drivers against 351 traditionally trained, Netherlands. Caveats: small intensive group, incidents self-reported, observational design. This study is not about people with a fear of driving and does not measure how they learn. We bring it in as a pointer about spacing lessons out, not as a finding about anxiety.

How that squares with something that sounds like the opposite. In phobia treatment, one long, well-graded session was not found to give worse results than many short ones, and takes significantly less total time. That is a comparison between studies rather than a random allocation to one of two versions, so it says "no difference found", not "just as good". The two results do not conflict, because they are about different things. An intensive course for the whole licence goes with a worse subsequent record. One long exposure session in phobia treatment is not worse than many short ones. The practical rule: spread the driving lessons out, and attack a specific fear in a longer block, and only where the block genuinely climbs the ladder rather than circling a car park for two hours.

Odgers et al. 2022, 55 studies of single-session therapy and 30 of multi-session, n = 1,758 in total.

Silence during the manoeuvre

Research in real traffic shows that tasks requiring you to produce speech load a driver considerably more than tasks requiring you to receive it. The numbers are stark. Simply listening to a recording did not reduce rear-view mirror checks at all (Cohen's f = 0.14 and 0.19, neither significant). Having to reproduce what you had just heard, or do mental arithmetic, cut mirror checks with very large effects (f between 1.02 and 1.48). Which means a leading question mid-manoeuvre ("so what are you going to do at this junction?") costs far more than a short instruction. weak

Recarte & Nunes 2003, JEP: Applied 9(2), 119–137, tables 5 and 6 of the full text, which the APA publishes free at apa.org. Twelve drivers in an instrumented car in real traffic, so a small sample, but on-road rather than in a simulator. For Cohen's f, 0.10 is small, 0.25 medium and 0.40 large, so values above 1.0 are very large indeed.

This also applies to the popular exercise of having a learner describe aloud what they see. Three experiments by one group show that producing a commentary lengthens hazard response times, and in one of them reduces accuracy too, and that the cost does not disappear with practice. Commentary training is more complicated than it is usually presented: in a simulator, learners trained in commentary had fewer collisions and slowed and braked earlier, but in a later study by the same group, simply being exposed to commentary improved accuracy only at the first test and gave no advantage over an active control group. So commentary can be a good exercise, but it is a poor standing requirement.

Young, Chapman & Crundall 2014 (two experiments); Young, Crundall & Chapman 2017 (one study); Crundall et al. 2010 (simulator).

Constant commentary improves what you see in the lesson. Not necessarily what remains

In a classic study, feedback given after half the attempts, and faded out over time, produced 35% less error on a delayed test than feedback given after every attempt. On a test taken immediately there was no difference at all. strong, not about driving

Winstein & Schmidt 1990, experiment 2, n = 58 (two groups of 29). Fundamental caveat: a laboratory task on a lever, not driving. No study exists testing an instructor's feedback frequency in real driver training.

The logic is hard to ignore all the same: the driving test is precisely a delayed test with the help withdrawn. An instructor who talks non-stop sees a better pupil than that pupil is when left alone. The practical consequence, and this is our conclusion: every lesson should contain at least one deliberately silent stretch, assessed only afterwards.

The quality of the attention, not the quantity

Two large meta-analyses say the same thing from opposite directions. In the first, across 607 effect sizes, feedback improved performance moderately on average, but over 38% of individual effects were negative, and among the things that weakened its effectiveness was praise. In the second, across 435 studies, "good" and "wrong" gives d = 0.24, while feedback describing the task and the process gives d = 0.99. strong, not about driving

Kluger & DeNisi 1996; Wisniewski, Zierer & Hattie 2020. Both concern learning in general, not driving. In the first, the effect on physical tasks was smaller than on others, which the authors could not explain.

• "Good." or "No, wrong." is the weakest tier.
• "You were too close on the approach." is the middle tier.
• "You were too close on the approach because you started braking after the line rather than before it. On the next roundabout, pick your braking point at the second lamp post." is the strongest.

Incidentally, the feedback sandwich (praise, criticism, praise) has a surprisingly thin evidence base. The only experiment testing it directly involved 91 students doing maths problems and came out in its favour, though the authors themselves note it is one study in one condition. None of that concerns learning to drive, and praise in the Kluger and DeNisi analysis weakened the effect, because it moves attention from the task to the person.

The instructor's own nerves change the lesson plan

In a randomised experiment, inducing negative beliefs about exposure led people to build a less ambitious ladder, choose less anxiety-provoking tasks and actively try to reduce the other person's anxiety. weak

Farrell et al. 2013, n = 53, full text. Very important caveat: these were not therapists or even graduates. They were psychology undergraduates with a mean age of 19.9, recruited by email offering brief training, and the "client" was an actor. No client outcomes were measured, so the study shows an effect of beliefs on the practitioner's behaviour, not on what happened to anybody being treated.

Limitations aside, the conclusion is uncomfortably apt. If an instructor keeps putting off the dual carriageway, picks the quiet route "because she isn't ready yet" and reaches for the dual controls a fraction early, it is worth checking whether they are managing their own discomfort. A simple safeguard: write the route ladder down before the lesson, with the pupil, and commit to the next rung in advance. It is also worth naming the incentives: an instructor whose income depends on the lesson being pleasant has a built-in reason to under-dose. That is our observation, not a research finding.

A driving instructor is not a therapist, and an honest school should have a written point past which it stops selling lessons as the answer. The nearest authoritative standard is the UK clinical guideline on post-traumatic stress; Ireland has no published equivalent.

NICE guideline NG116 (a clinical guideline, not a study). A road collision is a non-combat trauma, so both EMDR recommendations apply directly.

And one figure worth hearing if you are thinking "I've been like this since the crash years ago, that's just how I am now". In a randomised trial of people an average of five years after their crash, 8 to 12 sessions of therapy produced a very large effect, and 67% no longer met the criteria for the disorder immediately after treatment, rising to 76% at three months. weak

Maercker et al. 2006. Forty-two people completed the study: 21 in therapy and 21 in a wait-list comparison group, so the 67% and 76% refer to twenty-one people. A small single-centre study, but randomised and conducted in exactly the population at issue.

It is also worth knowing that the first and so far only published randomised trial of treatment for fear of driving ran its exposure in cooperation with a driving school: partly in a simulator, partly in a driving school car with an instructor, and finally in the patient's own car with the therapist in the passenger seat. The dose was around 18 to 20 sessions of a hundred minutes, twice a week. Which is to say: "I had a few lessons and it didn't help" is not a trial of treatment.

Fischer, Heider, Taylor & Schröder 2021, n = 34, a pilot study with a wait-list comparison group, 88% women.

There is a visual-vestibular condition that produces a sensation of drifting or turning while driving, provoked very specifically by high speed, winding roads, going downhill and being overtaken. In the literature it is called motorist's disorientation syndrome, first described in 1985.

Two figures make the difference: a measurable vestibular abnormality or an exaggerated response to moving visual fields is present in 60 to 100% of such cases, while anxiety accompanies only 17 to 39%. weak

Dumas et al. 2026, a narrative review of eight publications. Caveat: the quoted frequency of 1 to 5% refers to patients at a specialist balance clinic and says nothing about how common this is among learner drivers, because nobody has measured that.

The warning sign is straightforward. If your unease follows a physical pattern (speed, bends, gradients, a lorry overtaking) rather than a social one (junctions, examiners, other people's judgement), and if there is dizziness or migraine in the background, that is a question for your GP, who will usually refer on to an ENT specialist or a neurologist. An eye test is worth doing at the same time. Piling on more motorway miles is provocation for this person, not therapy.

• Not one study exists of an intervention delivered by a driving instructor. Every controlled study of treatment for driving fear is clinician-led. The question every instructor actually has, "does what I do help?", has never been tested.
• Nobody has studied anxiety in learner drivers as a group of their own. Every figure about "driving anxiety" comes from people who already hold a licence.
• Nobody knows what order to tackle feared situations in. Worse, the structure of the standard avoidance scale separates weather avoidance from traffic avoidance as distinct dimensions, so it is probably not one ladder at all.
• Nobody knows how many lessons an anxious person needs. We know how many hours the average learner takes (52 with an instructor) and that the number rises with age, but nobody has ever counted it separately for people who are frightened. Which is precisely the question everybody who rings asks first.
• There is no Irish data at all on how common driving anxiety is, nor any guidance for instructors on it. Every figure in this article comes from New Zealand, Germany, Australia, France, Britain, the Netherlands or North America.
• There is nothing on repeated failure. Not one peer-reviewed study of what happens to a person's belief in themselves after a failed driving test.
• Nothing links reduced anxiety to fewer crashes after passing. Even if you settle a learner and get them through the test, there is no evidence they will be a safer driver afterwards.

That last point deserves its own sentence, because it is uncomfortable for the entire industry. A Cochrane review covering 24 randomised trials and close to 300,000 participants found no reduction in crashes from driver education (relative risk 0.98). That concerns training after licensing rather than learning before the test, but it marks the ceiling on what this field has ever demonstrated. strong

Ker et al. 2003. The authors themselves flag the poor methodological quality of the included trials and signs of publication bias.

The basis is a literature review across seven areas: the nature and frequency of driving anxiety, test anxiety, motor learning, the physiology of stress, treatment interventions, instructor behaviour, and differences between groups of learners. Every claim then went through a separate check whose job was to refute it: to establish whether the cited work exists, whether the numbers match digit for digit, and whether the conclusion had been stretched beyond what the study actually showed.

Of 105 claims checked, 30 did not survive in their original form, and a second check of the finished text produced fifty-odd further corrections. Among the errors caught: a completely invented article title attached to real authors' names, a meta-analysis quoted without the rebuttal published in the same journal, a secondary analysis presented as the main result, a sample size from one experiment attached to the result of another, a doubled treatment-group size, and a DOI that does not exist. All of those corrections are reflected above.

Several papers sat behind paywalls and at first we could work only from abstracts. Ten were later obtained in full and checked at source, including a government report covering 42,851 learner drivers whose file had stopped being served and had to be recovered from an archive. As a result several figures here, including the full breakdown of where anxiety starts and the effect size from the on-road assessment, come from the original tables rather than from other people's summaries. In two places, reading the original forced us to weaken an earlier sentence rather than strengthen it, and it stayed weakened.

Our own interest, to be clear about it. We write this as a driving school that sells lessons, including lessons to people who are frightened of driving. Everything said above about the absence of evaluation for nervous-driver courses applies to us too: nobody has checked whether what we do works. The section on when to stop selling lessons and refer somebody to a doctor is here for exactly that reason.

This is not medical advice. If you recognise panic attacks in yourself, avoidance lasting months, or returning images of a crash, the right person is your GP, not a driving school.