The Scottish Government

« Previous | Contents | Next »

Listen

EXTERNAL-TO-VEHICLE DRIVER DISTRACTION

CHAPTER TWO DRIVER DISTRACTION

2.1 Driver Distraction has been defined as 'a shift of attention away from stimuli critical to safe driving towards stimuli that are not related to safe driving' (Streff and Spradlin, 2000: 1). Two major distinctions tend to be made in the literature: between internal-to-vehicle distraction and external-to-vehicle distraction. Internal vehicle distraction includes everything inside the car (including the driver's own actions), so it covers everything from distraction from the car radio/CD player, distraction from other passengers, distraction from use of mobile phones (or other in-car communication system), and the driver 'distracting him/her self' (for example, by day dreaming).

2.2 By the same token, external-to-vehicle distraction includes everything outside the car, from weather conditions to billboards to children playing and so forth. Most (but not all) external-to-vehicle distraction will tend to be visual (as a result of the enclosed nature of the modern motor vehicle), and most internal-to-vehicle distraction will tend to be audible (although of course there are exceptions. It is arguable that, for example, police/ambulance sirens may distract drivers. Cf. Withington, 1998).

2.3 It should be noted that most of the studies discussed in this report originate from the United States. Studies in other countries will be noted where available. However, the fact that there are cultural differences between countries (for example, in this study, between the kinds of advertisements permitted, and their prevalence) should be borne in mind. Just because something has been demonstrated to be the case in one country, it does not necessarily follow that it is the case in another.

STUDIES OF DISTRACTION

2.4 There are a large number of ways to study driver distraction. However, most attempts are of four major kinds.

1: Laboratory studies.

2: Statistical/epidemiological studies
2:1 Correlation studies
2:2 Double blind experiments

3: Questionnaire/verbal discourse studies

4: Focus group/interview studies.

2.5 These are of course not mutually exclusive. For example, laboratory studies may be of the double blind variety, and there is obviously an overlap between discourse studies and focus groups. However, with the examples below, the differences should become obvious.

2.6 It should be noted that these studies break down into two main types: quantitative and qualitative (that is studies involving numbers versus studies involving discourse) with 1 and 2 being quantitative and 3 and 4 being qualitative. This of course refers to method of input of data, not necessarily that of storage and analysis. So for example, a questionnaire form will be analysed as quantitative data (assuming questions are of the 'yes' 'no' variety, or questions in which subjects are asked to rate things on a scale). However, it should not be forgotten that the input is still discourse, albeit discourse notated in the form of quantitative data.

2.7 The next four sections will describe the various methods of analysis, and discuss advantages and disadvantages of these methods.

Laboratory Studies

2.8 An example of a laboratory experiment would be an experiment conducted in a driving simulator in which (for example) simple driving tests were carried out and response times measured while a 'distracter' (which might be in visual or audio form) was introduced. Response times to the tests with and without the distracter could then be compared. If they were slower in the presence of the distracter then we might infer that the distracter was interfering with driving responses.

2.9 In terms of scientific practice it should be noted that laboratory experiments are generally considered to be the most reliable form of evidence. However the problem always remains of generalising from lab experiments to a real world situation. This is particularly the case when the experiments are more abstract studies of perception (for example noticing dots on a computer screen). This is known as the problem of ecological validity.

Statistical Studies

Correlation Studies

2.10 These are studies carried out 'in the field'. For example if we wished to study the effects of billboards on driver behaviour we could take a list of all the accidents in the general area and see if we can correlate them with areas with high numbers of billboards. Alternatively if we were studying the effects of mobile phones we could obtain mobile phone records and see if there was a correlation between the time that drivers were using their phones and when an accident took place.

2.11 It is important that controls are included in these types of study (so in the first example we should attempt to correlate accidents to billboards in an area with a low number of billboards as well, and compare the statistics of the two areas to show we have a genuine effect).

2.12 However even if a control is used, there are still two main problems with these sorts of studies. Firstly there is the difficulty of obtaining accurate statistical data. For example, in the mobile phone studies above, it is difficult to get data accurate enough to show that drivers were using their mobile phones at the actual time of the accident.

2.13 Secondly, and more profoundly, there is the problem that 'correlation does not imply causation'. Even if we found a perfect correlation between billboard numbers and accident rates, this does not prove that billboards cause accidents. There may be some hidden variable that links the two. For example, there is a correlation between accidents caused by SPADs (Signals Passed at Danger) on the UK Railways, and the presence of bananas in the supermarkets. But this does not mean one causes the other. Instead there is a hidden factor (sunlight/summer) linking the two. The more sunlight there is the more likely drivers are to SPAD, and the more likely there are to be bananas in the shops. In the same way, billboards may exist for a reason (for example, a greater number of shops, and, hence, a greater number of cars pulling out into the road), and it may be this reason that is the 'hidden' causal factor linking billboards and accidents.

Before and After Studies

2.14 Again using the example of a billboard, a before and after study would take accident data from an area before a billboard was installed and then after the billboard was installed. If the billboard was having an effect then it would be expected that the accident rate would go up. In an ideal world the billboard would then be removed and the rate would go down again. It should be stressed that before and after studies must have two forms of control. Firstly, another area, as similar as possible, must also have its accident rate analysed to show that the effect was really caused by the billboard. Secondly, the whole surrounding area must be studied to show there is no effect of 'accident migration' (the still controversial theory that lowering the accident rate in one area merely 'displaces' the accidents and leads to them taking place elsewhere).

2.15 Before and after studies are by far the best way of demonstrating a 'real world' effect. However, they are difficult to conduct (especially, for example, if one thinks of the effect of safety warning signs, mobile phones, or adverts on taxis). Moreover, experimental procedures in this area tend to be lax. Elvik (1997) performed a meta-analysis of 36 'before and after' studies concerning accident reduction strategies. He discovered that few of the studies had allowed for four aspects of traffic behaviour:

2.16 It is point 3 that particularly applies in this case. There are random statistical variations in data, and without adequate controls it is possible to mistake these fluctuations for genuine 'causal' effects. For example, in a two-year study, if a billboard was erected at the end of year 1, and traffic fatalities went up, it would be easy to conclude that the billboard 'caused' these fatalities. But it may well have been that the fatality rate for year 1 was unusually low, and that the increase was simply a 'regression to the mean' effect, not a causal one.

Questionnaire Data

2.17 These kinds of study usually consist of questionnaire data filled in by the police or some other official as to the causes of the accident. They may be based on an interview with the driver or the official's own guess as to what happened. It is important to note that there is no fixed methodology for questionnaire data and that, therefore, methodologies can vary. For example, in some instances drivers are interrogated by the police and specifically asked (for example) whether they were distracted. Sometimes however the police merely have a form to fill in, and do not tick the 'driver distracted' box unless the driver volunteers the information. It is therefore important to ensure that similar methodologies were used in data gathering before comparing datasets. Unfortunately information relating to this is not always freely available.

2.18 It is important to note that despite the fact that the method of analysis for this kind of Study (output) is quantitative, input is qualitative. That is, all that is being analysed is discourse (usually from the driver) which is logged on paper and then on a database. It is possible that the causal explanations produced by drivers for their own behaviour are influenced by social factors (for example, stress, anger, fear of prosecution) as much as by a desire to tell 'the truth'. For example, a private citizen writing in the NHTSA internet forum on driver distraction wrote: 'Today, I heard that 25% of accidents were caused by driver distraction. I question that because of a wreck I once had. I was busy lighting a cigarette when I crashed into someone. Of course, I chucked the cigarette immediately and when the police came, I told them I "just didn't see the guy." If the same thing happened when I was fiddling with GPS device or whatever, I probably wouldn't admit that either. I think the statistics about accident causes should be viewed with some suspicion.' (NHTSA, 2000: In Vision Technologies: Experience and Research (Other) 'Question Those Statistics'). On the basis of this quote, the idea that qualitative input to statistical databases is necessarily veridical must be questioned.

Focus Groups/Interview

2.19 This is perhaps the most obvious way of gaining information about driver distraction: to ask the drivers themselves why they become distracted. However, to the best of our knowledge no studies of this sort have been carried out in this field (with one exception: see section 4.42), perhaps due to fears that the information proffered will be 'subjective'. However it should be noted that interview (questionnaire) data are just as subjective (and perhaps more so, considering the situations in which they are obtained).

« Previous | Contents | Next »