The Scottish Government

« Previous | Contents | Next »

Listen

EXTERNAL-TO-VEHICLE DRIVER DISTRACTION

CHAPTER SEVEN SPECIFIC STUDIES: BILLBOARDS

7.1 It may seem obvious that the distraction phenomena most regularly given as the primary causal factor in the various studies carried out (for example, the Tri-level study) would be the phenomena that have been most studied in the various experiments and studies that have been performed since. However this is not the case. Science is a social process, and in the study of safety, social and political pressures have a stronger influence than in many other fields of science. Experiments and studies take place when money is available for the research, and, therefore, experiments which are easy to carry out and might lead to definite alterations in safety legislation will be prioritised over more 'theoretical' experiments. So studies on whether children, or the police (or other accidents) function as distracters are few and far between. This is mainly because until recently the technological ability to create simulations of complex dynamic situations (such as a child running into the road) in a laboratory setting was not available (one of the very few attempts to model dynamic situations is Berkhout, 1979). Even now that this can be done it is difficult to set adequate experimental controls (part of the reason mobile phone experiments as described in section 4 are so popular is that they are presumably reasonably easy to do). The behaviour of children (or the police) is infinitely variable, and it would be very difficult to make strong inferences from even worthwhile experiments. Moreover even if the results did indicate that distraction could follow in these situations, it is difficult to see what specific safety improvements this would lead to.

7.2 More work has been done on billboards/safety signs etc., and it is on these features that this chapter will concentrate. Billboards are static (as opposed to dynamic) and standardised. In other words it is easier to carry out actual experiments and studies, and easier to make inferences as to safety events in the real world. Therefore billboards are a 'test case' for this sort of distraction. If billboards can be demonstrated to be distracters it seems possible that other features can be as well, and that the more complex experiments that would be needed to test this hypothesis should take place. On the other hand if the results from billboard studies are too ambiguous to draw definite conclusions from, then it is highly likely that in studies of more diffuse and complex phenomena the results would be even more ambiguous.

7.3 One thing should be repeated: it does not follow from the fact that billboards are not provided as explanations for crashes in police records that they therefore do not function as such. To quote Shoaf: 'The fundamental precept of advertising is its impact on the subconscious mind. Therefore if an accident has been directly or indirectly caused by the attention of a motorist being subconsciously attracted away from the highway by an advertising sign, he would not, of course, be in a position to testify that the presence of the sign was even a partial cause for the occurrence of the accident.' (Shoaf, 1955: 71). This is yet another reason why the qualitative studies of accident causation such as Treat et al (op cit.) should be treated with caution.

THE LEGAL POSITION

7.4 There are many legal guidelines for the placement of adverts and signs by the side of the road. For example, 'Road Geometry' (2001) states ''Types of advertisement likely to cause danger to road users, and are open to control on public safety grounds are those that: -

• Impair sight lines at corners, bends, junctions or accesses.
• Obstruct traffic signs or signals or are likely to distract road users because of their unusual nature
• Leave insufficient clearance for vehicles on the carriageway' (Department for Transport, 2001: 10/11) (italics added).

7.5 However, to quote Holohan (1979) 'Most cities have ordinances regulating the extent to which advertising can distract a driver's attention, yet they are often couched in ambiguous terminology and are based on policy makers' hunches rather than on actual safety evidence' (Holohan, 1979: 8). Moreover, Johannson discovered that most drivers have a poor recollection of any road signs they saw while travelling, suggesting that visual stimuli of this sort are not often brought to conscious attention. However, this would not affect the findings of experimental studies concerning involuntary (unconscious) saccades (Johanson and Rumar, 1966: Johannson and Backlund, 1970).

7.6 Before going on to ask about the actual safety evidence Holohan's next sentence should be borne in mind. 'On the whole it appears that there has been very little actual research on potential distracters in field settings. Most of the research that has been done was completed nearly two decades ago. Many of the results are equivocal…. and based on either purely correlational data or on data resulting from poorly designed field experiments.' (Holohan, 1979: 8). The situation has not improved much since 1979.

THE STUDIES

7.7 The first studies that will be looked at are correlational studies. A point that should be made about the correlational studies is that they tend to classify distracters as if objects within a classification are all roughly the same. For example (that, if taverns and gas stations are classification categories) all taverns and gas stations are roughly the same size, height, have roughly the same number of customers and so on. This may be justified in terms of gas stations and taverns (or on the other hand it may not), but it is doubtful whether this methodology is the best one for billboards, signs, and other visual distracters. Some signs are large, some are small. Some are illuminated, some are not. Some carry complex messages, others do not. This is important as the evidence (presented below) suggests that the effects of signs/other distracters are situational.

THE NORTH CAROLINA STUDY

7.8 In 1974, the Highway Safety Research Centre in North Carolina carried out a wordsearch on their electronic database containing over 200,000 accidents. A number of words and phrases pertaining to advertising, signs and billboards were selected, and then accident reports which contained these words were selected and analysed. Amongst the phrases used were 'sign', 'eyes off the road' 'billboard', and 'distracted'. However, the study failed to provide any hard evidence that this form of distraction was a major contributory factor to accidents. The main problem with this study, apart from the relatively unsystematic methodology, was the fact that the search is only as good as the qualitative data held on the database, which is prone to various biases as detailed in paragraph 2.18, above (Anonymous, 1974).

THE ADY STUDY

7.9 In 1967 Ronald Ady carried out one of the best studies into the issue of billboards and their effects on the accident rate. This was one of the best studies in this area because it was not merely a search through the databases (like the North Carolina study) or a correlation study (which cannot prove causation). Nor was it a laboratory study (with the concomitant problems of ecological validity). Instead, it was a simple before and after study with a control: still one of the best ways of analysing an effect.

7.10 Ady's study was an attempt to deduce whether Broadbent's 'filter' theory of attention or Hebb's 'arousal' theory was the best one to explain distraction and attention. He located nine advertising signs that met the criteria for 'distraction' he had set out. That is, they were large, (at least 15 feet by 50 feet), illuminated, and information rich (for example, a weather forecast or news information). Accident rates were studied monthly for one year before and one year after the signs were erected: however, adequate data was available for only three of the nine signs. An area of road on the same highway was used as a control (to allow for rate of traffic flow). The two year study period was used to allow for the Hawthorne effect: given that, in Hebb's theories, to begin with the advertising would function as a novel (and hence) arousing element, but that as the information was repeated, the level of arousal level would decrease. Ady stated that to the best of his knowledge there were no other variables that could account for change in accident rates.

7.11 Ady discovered that two out of the three signs did NOT produce alterations in accident rates. However, the third sign DID produce a level of alteration at the 5 % significance level.

7.12 Ady deduced from this that whereas billboards and signs did not necessarily produce a distracting event, in the right situation it was possible that they did. He pointed out that the third sign was placed on the corner of a sharp bend, and that in some ways it was the most conspicuous sign (with bright white lights, which were removed six months later after pressure from state highway officials). This bears out the point of Bahcall and Kowler, 1997: that it is not so much the stimulus in itself as the relationship between the stimulus and other aspects of the perceptual field that causes distraction.

7.13 Ady argued that although the results were not entirely clear, it seemed to be evidence in favour of Broadbent's theories. However it is also possible to see this as bearing out Holohan's idea (1979), that the real danger was of surprise (as in the theories of Berlyne) in a situation where the driver felt him/herself to be safe. One does not normally expect bright flashing fifty foot signs to be placed on sharp bends. It is also possible that the bend did not look as sharp as it actually was, and that the driver felt him/herself to be safe, and that the surprise of seeing the sign caused accidents.

7.14 As mentioned above this is one of the best studies performed in this field. However, it is open to criticisms. Firstly, there are the small numbers: the fact that only the third sign showed a statistical significance might be explained by 'regression to the mean' (that is, that in the year previously accidents rates had been abnormally low, and that this was merely a statistical 'recovery' to the normal rate): this is particularly important as it is not made clear whether the 'control' area for sign three also contained a sharp bend (Elivk, 1997). Moreover Ady fails to describe the road that precedes the bend. The Minnesota Highway Department has demonstrated that sharp bends following long, monotonous stretches of road are more dangerous than sharp bends following short lengths of road. This would therefore be evidence for 'highway hypnosis' (or even, if the billboard was 'absorbing', phototaxis), and would therefore tend to support Hebb's conclusions, not Broadbent's (Lauer and McMonagle, 1955).

7.15 Ady's use of all accident rates as opposed to merely fatal accidents might be criticised: generally non-fatal accidents are under-reported, and 'differences' in rates can be the result of different reporting strategies. Fatal accidents, on the other hand, are almost always reported, and treated in the same way. Moreover, Ady's report contains little 'raw data' (it only contains statistical data), and is somewhat short on detail. Nevertheless it is still one of the best studies available, and is possibly the best evidence for the hypothesis that billboard distraction is situational (that is, it will not occur 'automatically' but only with certain signs at certain places, in certain situations) but real. Ady himself recognised that his study did not prove that billboards caused accidents but merely provided suggestive results for further investigation. Unfortunately follow up experiments were not carried out.

MADIGAN-HYLAND

7.16 The New York based agency Madigan-Hyland Inc. was commissioned to provide a report on billboards and their effects on advertising. This has proven to be one of the most controversial of all reports of its kind, and it is unfortunate that the report was never published, and was, therefore, unobtainable for this literature review. However, in 1963 the author of the report (Daniel Greenbaum) wrote a letter to the Chairman of the New York State Thruway Authority, which summarised its findings. This letter was placed in the Congressional Record. It is this letter that is used as the basis for the following description.

Method

7.17 Two years worth of data were analysed relating to the New York State Thruway (1961-1962). As with the other studies, observers were sent out to collate data relating to the exact locations of signs and other 'similar devices' that could, in theory, be seen by motorists. Accident records were obtained from the relevant police authority, and divided into two main groupings: accidents that occurred where the motorist could see these signs and accidents that occurred where the motorist could not see these signs.

7.18 It was felt simply to analyse all data would be too confusing and would create too large a database. Therefore all accidents were omitted except for accidents which were specifically classified as 'driver inattention' by the investigating state trooper. Rather remarkably, accidents at toll booths and interchanges were omitted from the study (despite the fact that they accounted for 25% of the accidents in the group), on the grounds that as other causal factors (such as 'the need to locate money for toll payment') were prevalent there and that this would therefore bias the study. Correlations were then calculated. It was discovered that 13.1% of the 1118 miles of highway had a high proportion of visible advertising, but 32.6% of the 1,550 accidents attributed to driver inattention occurred in these areas. Annually, 1.7 accidents occurred per mile at a 'high advertising' area, whereas only 0.5 accidents occurred per mile at a 'low advertising' area.

7.19 Traffic volume was controlled for by creating three separate sub-groupings of areas of high, medium and low-density traffic flow. Within these groupings, low and high advertising areas were correlated with accidents. In all three areas, high advertising areas had higher accident rates than low advertising areas (Neuberger, 1963).

Problems

7.20 The key problems with the Madigan-Hyland study are its concentration on 'driver inattention' reports only, and the degree to which it controlled for traffic volume. As has been stressed earlier, 'driver inattention' is merely a classificatory device, and its use will be context specific. Given that proponents for the existence of billboard distraction have argued that distraction may be unconscious (and that this argument has been backed up by Theeuwes, and Godjin, in press) it is not clear why drivers who were unconsciously distracted should state 'I was distracted' when asked why the accident happened. Moreover, despite the fact that traffic volume was controlled for, it was hardly controlled in a sophisticated (statistical) fashion. It was this methodology that led Blanche to describe their approach as 'erroneous' and 'immature' (!) (Champion, 1971: 134).

7.21 On the other hand, other studies (McMonagle 1952 a & b) have argued that intersections etc. should be correlated with specific distraction type events: specifically those involving search strategies. It could be argued that including these events could have increased the level of correlation.

7.22 Therefore, yet again, the Madigan-Hyland study provides a suggestive correlation, which should have been the basis for a long series of replicated scientific studies to investigate whether this was just a statistical 'fluke' or whether it indicated a genuine effect. However, to the best of our knowledge this was never carried out.

NEW JERSEY GARDEN STATE PARKWAY

7.23 This (1965) correlation study was carried out by Ernest Blanche on the New Jersey Garden State Parkway. Data related to accidents were collated for the years 1961, 1962 and 1963: producing 3902 accident reports in total. In order to carry out the research, official 'recorders' were sent out to list every off-road feature of the entire 173 mile parkway. This included 'bridges, overpasses…official signs of all kinds….advertising signs within 1300 feet of the parkway' (Blanche, 1965: 24). Therefore this was a study not just of billboards but of all roadside features.

7.24 To carry out the correlation study itself, a chart was drawn up dividing up the parkway into units of a tenth of a mile. On this chart every accident in the three-year timescale and every roadside feature was plotted. Traffic volume data was also provided. Then correlations between features and accidents were created, both individually and grouped together.

7.25 The study failed to find a correlation between advertising signs, safety signs, or any other roadside feature, and accident rates. It should be remembered that this was a correlation study and that even if a correlation effect had been found then this would not prove that billboards caused accidents, but would only suggest that more research was needed. However the fact that no such effect was discovered is suggestive. Needless to say, the negative results have provoked criticism.

7.26 Wachtel and Netherton argue that because Blanche counted roadside signs up to 1300 feet from the parkway, any effect caused by specific signs close to the parkway would be 'swamped' by the extraneous 'noise' produced from the further away signs (Wachtel and Netherton, 1980). This may be true for the 'grouped' data, but Blanche makes it clear that individual variables were cross-correlated with accidents as well, such that this effect should not appear.

7.27 The fact that Blanche used all the accident data available in the relevant database and not just fatalities (as discussed above) is another problematic element of the research: also questionable is the fact that (unlike Holohan, below), no inter-rater reliability data was calculated as to the identification of roadside features. However the key point is that even if Blanche's findings are entirely justified they do not necessarily contradict those of Madigan-Hyland above. It is entirely possible that individual signs on specific roads in specific relationships to other signs may well have a negative impact on road safety although other signs do not. The Garden State Parkway is a multi-lane motorway, with a low number of highway deaths. However, (judging by the photographs provided, and its geographical location) it carries a high volume of traffic, and has an extremely high number of external roadside features. Therefore it would not seem to be the ideal place to test the theory that the genuinely dangerous situations are of low arousal, monotonous situations in which the position of a single, highly conspicuous sign on a seemingly safe area of road (for example a concealed sharp bend) can distract drivers and cause accidents.

INFORMAL, SMALL SCALE STUDIES

7.28 Champion (1971) carried out a brief survey of traffic in Melbourne, based on accident records provided by the Victorian Traffic Commission. He discovered that there was a negative correlation between accidents and billboards: however he failed to use controls, and the study used extremely small numbers.

McMONAGLE (MICHIGAN) STUDY

7.29 J.C. McMonagle carried out two studies in the 1940s. The first was a lab study, which will be discussed in the 'experimental' section below. The second, however, was a correlational study (McMonagle, 1952 a and b).

7.30 A 70 mile stretch of rural road was selected in Michigan. All road and roadside features were catalogued, and the road itself divided into 1000-foot sections. Local police agreed to record the distance of the accident from one of the 1000 foot markers in all accident reports. The study itself was carried out between 1946 and 1949. Two kinds of statistical analysis were performed on the data. First, accidents were classified according to their distance from each individual feature. Second, standard correlations were produced between the number of accidents and the number of various roadside features (in terms of feature density).

7.31 The results of the initial analysis demonstrated that intersections seemed to be an important feature in terms of accidents, and so the road was further divided into 'intersection' and 'non-intersection' areas. These analyses demonstrated that intersections are hazardous, with the accident rate on the sections of road classed as 'intersections' being double the rate for non-intersection sections of road. However McMonagle notes that intersections became even more dangerous when there were a large number of roadside features around them.

7.32 In terms of the correlation studies, roadside features were divided into six categories: 'Taverns', 'Gas Stations/Garages', 'Stores', 'Other Establishments', 'Design Features', and 'Advertising signs'. Correlations between these features and accidents were then studied. It was discovered that on their own these features had little effect, but when the same analyses were run with the road divided into 'intersections' and 'non-intersections' a correlation was discovered with Other Establishments, Advertising Signs, Taverns and Gas Stations. It should be noted, however, that these figures reflect amalgamations of numbers. That is, it seems to be the presence of intersections and all four of the above categories that lead to an increased rate of accidents.

7.33 Therefore more research was done to discover if there was a relationship between individual variables and accident rates (still on the 'intersection' area of road). When this was done it was discovered that only Gas Stations and Taverns were still associated with accidents, and that with Gas Stations the correlation was weak. Looking at the whole road, it was then found that the Taverns category had a link with accidents even on the 'non-intersection' of road.

7.34 The conclusion of the McMonagle study is simple: accidents are correlated with road complexity. As a general rule straight roads where traffic can flow reasonably easily (again, in general, higher traffic density is associated with higher accident rates, although the relationship is not linear) will be safer than congested roads with intersections, taverns and gas stations. Of course this might alter if the road becomes too straight, featureless and monotonous, leading to low arousal and 'highway hypnosis', and, hence the possibility of distraction from 'singletons'. Nevertheless the statement that 'road complexity is usually bad for road safety' is probably a worthwhile general heuristic. According to McMonagle it is unlikely that billboards, signs, and other roadside 'distracters' help directly to increase accident rates. Instead, systems features of the road itself are correlated with accidents.

7.35 There are two basic criticisms possible of the McMonagle study. Firstly, (and this point must be repeated), this is only a correlational study: McMonagle did not prove that road complexity causes accidents. Secondly, McMonagle did not present data as to whether his correlations were statistically significant or not.

HOLOHAN FIELD BASED ANALYSIS

7.36 The Holohan study was interesting in that it took account of the fact that signs are different. This was a specific study of signs, and not all roadside features, as with the Michigan and Minnesota studies. It also took into account the situational aspect of sign placement and its relationship to accidents. It should be noted that the experiments that were performed by Holohan (described below) dealt with situations where important safety signs were difficult to recognise when surrounded by other signs. This should be borne in mind when discussing his results (Holohan, 1979).

Methodology

7.37 Sixty intersections that had had at least one accident in 1975 were selected in Austin, Texas. Junctions with very high or very low traffic flow were eliminated from the study, and steps were taken to ensure the intersections were all roughly of the same type (that is, they were all streets intersected by other streets at a roughly 90 degree angle). Intersections were also divided into 'traffic lights' and 'stop signs' junctions.

7.38 Signs were categorised into size, kind of sign (public or private) and colour. Then police reports of studies were taken (using all the available evidence (direction of vehicle, probably cause, and responsible party). Alcohol and speeding related accidents were eliminated (as were accidents occurring at night, it being assumed that signs could not distract if they could not be seen).

7.39 Observers were sent out 'into the field' and categorised every sign visible from these intersections on the three dimensions above. Inter-rater reliability data was produced to ensure that all the signs were correctly logged. Then correlations were calculated between intersections which had low or high accident rates, and the various sign variables discussed above.

7.40 It was discovered that whereas traffic signal approaches showed no correlations, at stop sign intersections there was a link between accidents and the presence of a number of signs, large signs, and colour (specifically, non-red signs). However, Holohan argued that this effect was possibly the result of rates of traffic flow, and that the data would have to be re-analysed to take this into account. This did reduce the level of correlation. However, there was still a link between number of signs, non red-signs, private signs and (particularly) large signs and the accident rate at stop sign junctions.

7.41 The conclusions Holohan drew were as follows: that in a situation of visual search, drivers could be distracted by extraneous phenomena which were of the same medium as the searched for object. That is, at traffic lights it was reasonably easy to spot the lights, but in a situation where there were a large number of signs, the driver would be slowed down in terms of making the visual search, and an accident could occur in this 'gap' period. This supports the research quoted in section 3.29 above: visual search was slowed down when there were irrelevant stimuli available when subjects did not know what they were looking for. This would be the case here: drivers would not necessarily know that a stop sign was at the intersection (that is, who had right of way) and would be searching to see if one was present. This search was slowed by a large number of commercial, large signs.

7.42 Again, we should be careful of reading too much into a correlational study. However, Holohan's research is backed up by current research, and bears out Blanche's point above. It begins to look as though the question 'do billboards cause accidents?' is meaningless. However it may be possible to ask and answer the question 'will this sign, in this situation, at this time, tend to lead to a higher accident rate?'

OREGON STATE HIGHWAY

7.43 Whilst not being strictly a distraction study, the Oregon State Highway Study nevertheless shows some details as to 'systems' features of traffic accidents that help to explain traffic accident rates (Head, 1959).

Methodology

7.44 Observers collated data on 426 'sections' of a highway in Oregon. Roadside features were noted and divided into seven sections of which two are of particular importance in terms of the current study: traffic signals and shops ('commercial units'). Accident data were obtained from the Oregon State Highway Department for the years 1954 and 1955. Traffic flow data were obtained, and the number of accidents per million vehicle miles was calculated to allow for traffic volume. Correlations were then obtained.

Results

7.45 Head discovered (as with Holohan above op cit.) that the number of intersections per mile correlated with accident rates (note: this relation is at its strongest when traffic volume is high and gradually fades away as traffic volumes drop). Also he discovered that intersections with traffic lights were correlated strongly with accident rates. Other correlations of interest were that between 'indicated speed' and accident rates (this time a negative correlation: the lower the speed the higher the accident rate). However the strongest and most important of all correlations was that between Commercial Units and accident rates. Average daily traffic was correlated with accident rates but the relationship was not linear (it tended to become stronger as volume increased).

7.46 The results of Head's study demonstrate some facts that are intuitively obvious but are sometimes forgotten in correlation studies. They support Holohan's study (above): complex driving situations tend to be more associated with accident rates than simple driving situations. That is, roads with high traffic volume, many intersections, many shops, and many traffic lights will tend to produce more accidents than roads where the opposite is the case. The key point is that billboards, advertising signs, adverts on taxis and so forth tend to be situated in these areas (in fact, most shops prominently display signs advertising themselves, thereby creating a confound: it would be difficult to tell whether any statistical effect was caused by the sign or by the shop). Yet again, even if a correlation between signs/adverts etc. and accidents was demonstrated, this would not prove causation.

MINNESOTA RURAL TRUNK HIGHWAY

7.47 Another correlation study was the Minnesota Rural Trunk Highway study, which was carried out in 1951. 510 miles of Minnesota (mainly rural) highway were selected and divided into 2,600 road sections. Each section was then studied by observers, and every relevant off-road feature was noted and entered into a database (the report does not go into detail as to how this was done, nor whether inter-rater reliability was calculated). Correlations were then calculated.

7.48 The major conclusion of the report as a whole supports the theories of Head. That is, high speed, low-to-medium traffic volume, low complexity (i.e. few intersections) roads are generally safer than the contrary.

7.49 There are two major findings of the Minnesota study which are of interest here. The first is (as mentioned earlier) accident rates for curves which were preceded by a long stretch of road are higher than those which were preceded by a short stretch of road. This lends support to the idea that low arousal leads to lowered attention, and that when a curve occurs, drivers are less prepared to cope with it. It should also bring to mind Ady's study of the advertising sign located on a curve which was strongly associated with accidents.

7.50 The second major finding relates to advertising signs and commercial units. In terms of commercial units, sections of road within 300 feet of a commercial unit were selected for analysis. Traffic volume was controlled for. It was discovered that accident rates for the road sections close to commercial units was considerably higher than that for the control section.

7.51 Advertising signs were observed and the data collated. It was found that 95.6% of the signs were within 60 feet of the roadway centre line, and 51.4% larger than 12 square feet. The signs were of various shapes and sizes (e.g. diamonds, horizontally rectangular, oval), and were mainly coloured red, yellow, black, white, and blue. However only 5.8% of them were neon or illuminated in some other way. The study noted that 14% of the signs were located on commercial units advertising the produce, but did not go on to state whether other signs were located on commercial units. Again, here is a potential confound: commercial units tend to imply intersections (where drivers can get back onto the freeway) but also advertising signs.

7.52 Traffic accident data were then calculated and compared with road sections with data pertaining to frequency of signs per mile. Traffic volume data was also calculated for the same sections of road. The tests were significant, indicating a positive link between number of signs and number of accidents.

7.53 More data were then calculated to see whether the presence of signs at intersections had a statistically significant relation to accidents. Again the results were positive (Staffeld, 1953).

THE RUSCH STUDY

7.54 In 1951 W.A. Rusch published the results of the Iowa study into the safety effects of billboards. Again it was a correlation study, so the 'correlation does not imply causation' rule applies. It should also be noted that there are issues of ecological validity: as Rush states, Iowa was (and is) more rural and less 'developed' in terms of business than New York or Los Angeles. However, to counteract this, unlike most of the studies above, this analysis concentrated on cities/towns with a population of over 5,000 (Rusch, 1951).

Methodology

7.55 The State of Iowa was divided into four geographical areas, and classed according to population. Within each area, subgroups were created called A-B, X and Y areas. A-B areas were areas of a mile or half mile in length where at least 90% of the advertising and roadside business being studied were situated. The X areas were a mile beyond the 'A-B' group, and the Y areas were one mile beyond the X area (these were therefore the same 'kind' of road, but had less advertising). There were few differences in traffic volume between the A-B, X, and Y areas, with rates averaging at between 2,500 and 4,000 vehicles per day. All roads (with a few minor exceptions) were two lane.

7.56 Accident records were obtained from the Iowa State Highway Commission for 1947 and 1948. Accidents were divided into three main groups on the basis of these reports: accidents attributed to roadside businesses, accidents attributed to attentional problems, and accidents attributed to 'other' causes. Obviously 'raw' figures of total accident rates were created as well.

Analysis and Results

7.57 It was discovered that the greatest number of accidents for each year occurred in the A-B area. Moreover, accidents associated with attentional problems were particularly associated with the A-B group. This was the case in both years in which the study took place. In terms of 'adjusted' accidents per 100 miles the same basic patterns occurred. Rusch concluded that inattention accidents were associated with areas where advertising and businesses were more prevalent.

Discussion

7.58 Again, Rusch's study was suggestive. However it has two major difficulties. Firstly it is not clear from the published data how 'areas rich in advertising and businesses' were identified. The second major problem is the 'correlation does not imply causation' rule. Of course, this applies to all the correlation studies. However in Rusch's the problem is particularly acute in that advertising and business location data was collated, not separated. It is clear from other studies that business location (and what comes with business location: heavier traffic, slower speeds, more intersections) is associated with higher accident rates. It is not clear in Rusch's paper how these confounding factors were controlled for, if at all. It may well be, therefore, that the effects noted were simply effects caused by road/systems factors.

VERSACE

7.59 The Versace study of 1960 was a study of two lane rural highways in Oregon. 1,400 miles of the Oregon highway system were selected, and fourteen categories created including average daily traffic, land-width, and number of structures. Versace does not make clear how these elements were selected or created. A factor analysis was then carried out to identify the relationship of these elements to accidents.

7.60 Versace's discoveries were similar to the other studies above, in that what he terms 'traffic conflict' (a mixture of traffic volume, number of driveways and number of intersections) was most strongly associated with accidents. He failed to find significant correlations with many other features and none of relevance to this study.

7.61 The problems with Versace's study are the usual ones. It is not clear where he obtained his data, how he calculated numbers of 'roadside structures', or whether advertising signs/billboards/other signs were included as a 'roadside structure'. Given these problems, the relevance of his study to the current research is limited (Versace, 1960).

EXPERIMENTAL STUDIES

7.63 As well as these mainly correlational studies (with the exception of Ady), there have also been two major experimental studies. Here the main problem is 'ecological validity': that is, how accurately the experimental set up reproduces 'real world' driving situations.

JOHNSTON AND COLE

7.64 It should be noted that both the Holohan and Johnston/Cole studies used young (early late 'teens/early 'twenties) psychology students as subjects. This is not representative of the average driving population. Moreover, whereas the field studies above tended to study rural areas, it must be assumed that the majority of these students were white, middle class, and had a shared cultural background. This, again, creates a problem with ecological validity.

7.65 Johnston and Cole carried out a study in the mid 1970s to see whether irrelevant information could distract subjects from driving-like tasks (Johnston and Cole, 1976).

Method

7.66 Subjects were placed before a curved white screen, on which photographs could be projected. Subjects were equipped with a joystick for task performance. Arrows pointing right or left appeared on the screen at various intervals (infrequently to measure 'leisurely driving'; more frequently to mimic 'demanding driving'). The task was for the subject to move the joystick in the direction indicated by the arrow. At the same time coloured adverts and signs (of the sort frequently found on billboards) were broadcast on the screen. Subjects attempted to carry out the task while 240 of these were projected randomly on the screen. In some experiments as well as these images, a small 'spot' target would appear at two random locations occasionally throughout the experiment. The subject was to press a button held in his/her left hand whenever they were seen. The tests were to correctly move the joystick in the direction indicated, and to spot the 'target' as quickly as possible, despite the possibly distracting effects of the adverts and images. Five experiments were conducted in total, investigating various effects of the distraction and task performance.

Results

7.67 In terms of the main task (the arrow task) it is clear that there were two parameters: task performance and task time. That is, how accurately subjects moved the joystick in the indicated direction, and how quickly they 'spotted' the arrows. The second, 'target' task, obviously had only the second parameter. In two of the experiments, there was a decrease in accuracy on the 'arrow' task, and in three others there was an increase in detection times (in both situations when distracters were present (only one of these experiments showed both effects at the same time)). Interestingly, in two experiments, accuracy increased under conditions of distraction: Johnston and Cole interpreted this as evidence for Hebb's 'arousal' theory: that is, in a boring experiment colour adverts helped maintain arousal and increase task performance. However, there seemed to be a concommitant increase in detection time of the target (when present) in this situation (that is, when searching the visual field for important information).

Conclusions

7.68 Johnston and Cole explicitly took Hebb's theory as the model for their experiment, and, as they stated themselves, failed to consider Broadbent's 'filter' theory or to test for it. Nevertheless they did discover an effect that bears out Hebb's arousal theory of distraction, in both senses: that information can increase or decrease task performance depending on circumstance. The main problem with their experiment (apart from ecological validity: this was not sufficiently close to a real-life car situation to prove that the same effect would occur on the road) was that the effects (though statistically significant) were small. On the other hand, this would explain why there seems to be a statistical link between billboards and accidents and yet why this contributory factor rarely turns up in accident databases. It should be noted that Johnston and Cole's experiment only took account of average scores: it is possible that some individual adverts may have had a far more distracting effect than others.

HOLOHAN

7.69 The other major study in this field is the Holohan experiments (a companion to Holohan's field studies, above) (Holohan et al, 1978).

Method

7.70 56 psychology students were asked to sit in front of a visual display on which slides were projected by a computer, which also measured reaction times of subjects. Each subject was presented with 106 slides, some of which (but not all) showed a normal traffic 'stop' sign (white lettering on red background). When this was seen, subjects were to press a 'target present' button wired up to the computer as quickly as possible. When the target was missing the subjects were to press a 'target missing' button. The experimental effect was to be provided by the distracters that were also shown on the same slide.

7.71 Distraction effects were grouped along three main categories: number, colour and location. The 'stop' sign would be surrounded by either 2,4,6, or 10 distracters (4.45 cm. square elements containing common words mimicking commercial signs) which were classified by similarity of background colour. That is, they were either identical to the 'stop' sign (red), similar (orange) or dissimilar (blue green or black). Distracters were also either close or far away from the 'stop' sign. Therefore, there was a total of 48 different combinations of these variables to test.

Results

7.72 The results were positive for the tested hypothesis. That is, search times were slowed by a larger number of distracters, which were similar to or identical to the target sign in terms of colour, and close to the target. The deciding factor seems to have been proximity: all distracters close to the target slowed reaction time. However, number of distracters and colour (that is, the larger the number of distracters and the closer in colour they were to the target the stronger the effect) also produced a strong effect, when the distracters were further away. That is, regardless of any other variable, proximal distracters always slowed reaction time. It was only when this was controlled for that other factors came into play.

Problems

7.73 The key point again is ecological validity. Without belabouring the point, it must be asked how similar searching for a target sign on a two dimensional screen is to actual driving tasks. Moreover, subjects were given a maximum time of 1.5 seconds per slide to identify the target: given that searching for a 'stop' sign would normally be at an intersection, it is possible that drivers in real world situations would have slightly more time (perhaps three or four seconds) to look for any relevant sign. Thus, the effect, though real, might not manifest itself in the 'real world'. On the other hand, this bears out McMonagle's finding about the effect of distracters on junctions and intersections.

CONCLUSIONS

7.75 It is likely that there is more than one kind of 'distraction'. We suggest there are two main features of external-to-vehicle driver distraction.

7.76 Firstly, 'search mode' distraction. The scientific evidence seems overwhelming that in a situation where the driver does not know what s/he is looking for (in other words, s/he is unsure whether a 'stop' sign, for example, exists or not) the existence of 'distracters' in the area in which s/he is looking, will slow up the search. This kind of distraction is particularly associated with junctions.

7.77 Secondly 'broad attention' distraction. This is in a situation where the driver does not realise there is a threat, and is not looking for anything in particular. Ironically enough s/he is now particularly at risk from distraction. This mode is particularly prevalent on motorways, and highways. Of course, mostly when the driver feels that s/he is not at risk, there is in fact no risk. The danger comes when the situation changes without the driver realising it. This would be the case on a sharp bend at the end of a long featureless section of road, and it is significant that the 'distracting' billboard in the Ady study was in just such a place.

7.78 A subdivision of this would be 'phototaxis' in which the driver is distracted from the road by the flashing lights of a stationary vehicle or billboard. This would be a case of 'absorption' and it is possible that accidents caused by drivers becoming 'engrossed' or 'absorbed' in a mobile telephone conversation are also of this sort. The 'Catch-22' of attention studies is that in some circumstances signs (or other sources of information) may help the driver maintain his/her alertness. It is vital to know when signs might help the driver and when they might distract him/her, but there has been little or no research on this issue.

7.79 The evidence suggests that 'abrupt onsets', primary colours, bright lights, and 'volume of information' are important in attracting attention. Therefore, flashing neon signs, information-rich signs (with moving images for example), sexually or otherwise explicit signs, would be particularly likely to attract attention from the road. It is likely that these strictures are particularly apposite regarding 'search' mode distraction at junctions. Here, high information and 'novel' signs would slow down search patterns even more by increasing visual 'clutter' and hence would lessen further the time available for the driver to make decisions.

7.80 Despite this, it is still not proven whether billboards attract attention from driving or not. Certainly there is a large amount of scientific evidence suggesting they might under certain circumstances, and a few suggestive correlation and laboratory studies suggesting they do. However all the studies are flawed: either because they are correlation studies, because they are too small scale to draw conclusions from, or because of issues of ecological validity.

7.81 With signs becoming increasingly prevalent and increasingly explicit, it is vital that research is done on this topic, and genuine scientific evidence produced to guide planners and local authorities in terms of policies and procedures for sign placement.

« Previous | Contents | Next »