Article 3: Estimating under- counting of Road Casualties in Scotland Article 3. Estimating under-counting of road casualties in Scotland
1.1. It has always been recognised that not all accidents are reported to the police and there is no legal requirement to do so.
1.2. This article summarises and extends previous analysis produced by Department for Transport (DfT) and Scottish Government / Transport Scotland. It considers alternative sources of data on road accidents and casualties, and provides some estimates of the overall number of road casualties in Scotland. An earlier version of this article was discussed with the Liaison Group on Road Accident Statistics (LGRAS) in June 2011.
1.3. This work also meets Requirement 2 of the UK Statistics Auth§ority Report 61, Statistics on Transport in Scotland. This states: "For Reported Road Casualties Scotland and Key Reported Road Casualties Statistics, develop a best approximation of the numbers of casualties based on research into the under-counting associated with the Stats 19 form."
1.4. It should be noted that more serious accidents are more likely to be reported to the police, although it is possible that many less serious accidents are reported for insurance or legal purposes. Reporting of road traffic accidents also varies by type of road user involved, so the estimates and percentages below should be applied to sub groups of the Stats 19 casualty data with caution.
1.5. There is no evidence to suggest that the trend in total road casualties is any different to the trend in levels of reported road casualties. Stats 19 remains the most reliable and complete data source for monitoring injury accidents on Scotland's roads and the statistics published in Key Reported Road Casualties and Reported Road Casualties Scotland remain the most complete and reliable source of this data.
2. What is a road casualty?
2.1. The data used in Key Reported Road Casualties and Reported Road Casualties Scotland (RRCS) are collected via the Stats 19 form from police forces. This is the same data source used by DfT to collect road accident circumstances, vehicle and casualty data in the rest of Great Britain. The following paragraphs define the terms "accident" and "casualty" in the context of Stats 19.
2.2. "The Statistical returns include only those accidents which result in personal injury, which occur on roads (including footways), in which a vehicle is concerned, and which become known to the police. The vehicle need not be moving and it need not be in collision. The statistics are therefore of injury road accidents only: damage only accidents are not included in the figures." 
2.3. A casualty is "A person killed or injured in an accident."
"A fatal injury is one which causes death less than 30 days after the accident
A serious injury is one which does not cause death less than 30 days after the accident, and which is one (or more) of the following categories:
(a) an injury for which a person is detained in hospital as an in-patient
Or (b) any of the following injuries (whether or not the person is detained in hospital): fractures, concussion, internal injuries, crushings, severe cuts and lacerations, severe general shock requiring treatment
Or (c) any injury causing death 30 or more days after the accident.
A slight injury is any injury which is neither fatal nor serious - for example a sprain bruise or cut which is not judged to be severe, or slight shock requiring roadside attention."
2.4. The definition of a road traffic accident or casualty will vary in other data collections, for example in hospital admissions or in death registrations.
3. Reporting of road accidents
3.1. There is no legal requirement to report road accidents to the police and it has long been recognised that this will lead to under-counting of accidents and casualties in the Stats 19 data.
3.2. The 1998 Road Traffic Act (section 170), and subsequent amendments, and Sections 261 and 262 of the Highway Code describe the duties of the driver to report an injury accident. They merely require drivers/riders to report to the police within 24 hours if they are unable to exchange documents, (these reports only relate to the need for information to support insurance claims and establish criminal liability, and not the detailed statistical coding). There is no requirement on the public to report an injury accident, if those involved are able to exchange documents, even though they may still attend hospital for treatment.
3.3. The Stats 19 requirement for statistical information relating to a police report on a road accident differs in some respects to the requirement of the road traffic act. These differences, which mainly relate to the definition of public highway/road, the allowable reporting period for an injury road accident to the police, and the treatment of injury accidents involving cyclists, are set out in the definitional guidance to the Stats 19 requirement in a document known as 'Stats 20', the DfTs Instructions for the Completion of Road Accident Reports.
3.4. The figures published in Reported Road Casualties Scotland and Key Reported Road Casualties therefore relate only to accidents and casualties reported to the police, which is a subset of all road accidents and casualties.
4. Under-reporting of road accidents and casualties
4.1. Whether an accident is reported to the police will depend on a number of factors and the likelihood of the police becoming aware will increase as the severity of the accident and injuries increases.
4.2. The Stats 19 collection covers accidents where the police attended the scene and also self reported accidents ie where a member of the public reports the accident at a police station. Stats 19 data relate to injury accidents only and do not include non-injury accidents, many of which may be reported to the police.
4.3. There are three aspects of under-counting:
a. Under-reporting, where accidents are not reported to the police
b. Under-recording, where reported accidents are not recorded on the Stats 19 form
c. Misclassification, where the police officer reports what they find at the scene of the accident, but where for example, some casualties may appear uninjured, or a serious injury may appear slight.
4.4. There will be variation in the levels of each of these by type of accident and type of road user as well. For example, where someone falls off a bicycle and injures themselves they would be unlikely to report to the police and are not obliged to, where as a collision involving a number of vehicles on a motorway is more likely to receive police attendance whether it is required by the parties or not.
4.5. It would be possible to produce estimates of under-counting by type of road user however the margins of error around the estimates for individual types of user would be very wide.
5. Other data sources
5.1. As stated above, there is no legal requirement for parties involved in a road accident to report it to the police, even if someone is injured. People injured in a road accident will not necessarily report to hospital either so there is no administrative data source which can provide a complete record of road accidents or casualties.
5.2. There are other sources of data which enable comparison with the Stats 19 return to estimate under-counting, though the definitions set out in Stats 20 differ from these alternative sources which makes direct comparisons with other sources difficult.
5.3. Death records: National Records of Scotland (NRS - Previously the General Register Office for Scotland) hold death records for Scotland which includes cause of death. There will be slight differences between the Stats 19 data and the NRS death registration data due to the different definitions used in the data collections. For example the NRS data will include deaths which occurred on roads other than public roads and deaths which occurred more than 30 days after the accident (which would count as a seriously injured casualty in Stats 19). The Stats 19 data will include people involved in a road accident in Scotland but whose death occurs outwith Scotland as the death registration will be made in the country of death. As reported in RRCS 2009 and RRCS 2010 Article 2, the death statistics and the Stats 19 trends and totals are very similar. As the deaths data and the numbers killed reported on Stats 19 track very closely and given the likelihood of police becoming aware of fatal accidents, it can be concluded that if there is under-counting of fatalities it will only be by a very small number (e.g. a delayed reaction to an accident) so the published figure is the best estimate.
5.4. Hospital Admissions data: The NHS record admissions to hospital and identify those that are a result of a road traffic accident, though the definition will differ from that used in Stats 19. The figures are based on periods of care (episodes) under a particular consultant so patients can be counted more than once (e.g. if they transfer to another consultant). Casualties admitted to hospital could be recorded as killed or seriously injured on Stats 19, and some casualties who appear to be slightly injured at the scene may turn out to be more serious. Also some injuries recorded on Stats 19 will be treated at A&E and through GPs without admission to hospital.
5.5. Analysis published in Reported Road Casualties 2009 and figures published by NHS Information Services Division (ISD) shows the Stats 19 numbers of killed and seriously injured casualties are about 70% of the hospital admissions numbers (after a step change in the early 1990s) suggesting some under-counting in Stats 19. This is supported by work for the "Safety Net" project linking Stats 19 and Scottish Hospital in-Patient data. There will be some injuries recorded by hospitals as road casualties that have not been reported to the police e.g. if someone is able to drive away from the scene after exchanging details. However, there will also be injuries recorded by hospitals as road casualties that may have been injured in accidents in private car parks, for example, which would not be collected under the Stats19 system.
5.6. Survey data: The UK wide National Travel Survey (NTS) and the Scottish Household Survey (SHS) contain questions about road traffic accidents and injuries. Published DfT analysis in Reported Road Casualties Great Britain 2009 uses the results of the NTS but the Scottish sample size is too small to look at the results for Scotland. The Scottish Household Survey includes a question asking respondents whether they have been injured in a road accident in the last 12 months and, if so, there is a second question asking whether the police became aware of it. Survey follow up work by DfT suggests some over-reporting in the survey data for example when people are asked to recall whether the accident occurred in the last 12 months. These reasons are explored in the 2009 DfT article.
6. Estimating under-counting of road casualties
6.1. A number of different approaches to estimating overall road casualties can be taken using the data sources above. As stated earlier, the number killed is very similar to the deaths data, however for simplicity, the initial analysis of data sources below uses figures relating to the numbers of casualties of all severities.
Scottish Household Survey
6.2. As a follow up question for those reporting being injured in a road accident in the last 12 months, respondents are asked whether the police became aware of the accident. Combining data from 2006 to 2009 enables analysis by sub group of road user. The police became aware of 68% of those injured who were drivers or passengers, 56% of those who were pedestrians and only 29% of cyclists. Cyclists make up a small proportion of casualties reported in the SHS, so the average for all road users is 63% (95% confidence interval +/- 3.8%). This figure will reflect casualties who would be recorded as serious or slight on Stats 19. As slight injuries are less likely to be reported and form a larger proportion of injury accidents, this will skew upwards the percentage not reported.
6.3. This 63% figure from the SHS is supported by Hospital admissions data for casualties in road traffic accidents (which will include some of those killed, most seriously injured and some slightly injured casualties) which suggests casualties reported to the police are around two thirds of those admitted to hospital. As stated above, there will be some seriously injured casualties (e.g. with a broken toe) where the injured party is able to exchange details at the scene and drive away, then report to hospital later.
6.4. Applying these percentages to the total numbers of reported road casualties provides an estimate of between 20,000 and 22,600 using SHS confidence intervals with a central estimate of 21,300 casualties.
6.5. It is also possible to look at the question about involvement in an accident in the last 12 months. Combining data from 2006 to 2009 gives an estimate of 1.2% (95% confidence intervals of +/- 0.1%) of the adult population having been injured in a road traffic accident in the last year.
6.6. Applying this percentage to the population of Scotland, 5,222,100 gives a number of casualties between 58,500 and 67,900 with a central estimate of 63,200. However, we know from Reported Road Casualties 2009, table 25, that levels of injuries to children are 12% of reported injuries to adults (The child population of Scotland is 22% of the adult population). Applying the Scottish Household Survey estimate of 1.2% to the adult population in Scotland and increasing this by 12% to include injuries to children gives an estimate of 55,300 casualties.
6.7. Estimates based on the 1.2% appear very high given levels of reporting to the police. Even using the lowest figure above (55,300) and applying the proportion of cases where the police were made aware (63%), this would suggest 35,000 casualties are reported to the police each year, and Stats 19 returns are completed for less than 40% of these. Discussion with the police forces suggests that this is not the case so, either the 1.2% or the 63% is incorrect. Evidence from other sources suggests that the 63% is the more accurate, particularly for seriously injured casualties.
6.8. There are a number of reasons why the 1.2% figure will be an over estimate. DfT research identified "recall" issues, e.g. people saying yes to being injured in an accident in the last 12 months even if the accident happened outwith this period. There are also differences in what respondents perceive as an injury road accident and the definitions used in the statistical return e.g. injuries sustained on private ground. In Reported Road Casualties Great Britain 2009, Article 5, DfT decided not to use the figure for the percentage of people injured in the last year derived from the NTS as they felt this was an overestimate of injury accidents on public roads.
DfT best estimates
6.9. For Reported Road Casualties Great Britain 2009, DfT published an estimate of the number of road casualties in Great Britain of between 610,000 and 780,000 casualties per year, with a central estimate of 700,000. As a percentage of the Great Britain population, this works out as between 1% and 1.28% injured in a road accident each year, similar to the SHS percentage. Comparing these with Stats 19 data suggests between 28% and 36% of casualties are reported to the police. Both of these approaches could be applied to Scotland figures if similar levels of reporting and severity are assumed.
6.10. Applying the estimates of 1% to 1.28% of the GB population to the Scotland population would give a similar figure to the SHS data. However, reported road casualties as a percentage of the population are lower in Scotland than in Great Britain (2.55 per thousand population compared to 3.64 per thousand for GB). This suggests that the figure is an over estimate, unless the proportion of accidents reported to the police in Scotland is much lower than that in England and Wales. There is no evidence to suggest this as, for example, the SHS suggests 63% which is similar to the 60% suggested by the National Travel Survey and British Crime Survey.
6.11. Comparing the DfT estimate of casualties with the levels of reported road casualties in the UK suggests between 28% and 36% of all road casualties are reported through Stats 19. Applying these proportions to the Scotland casualty figures (13,334 in 2010) gives a lower estimate of between 36,600 and 46,800, with a central estimate of 42,000.
6.12. These results still seem high when compared to the published data and the levels of reporting and police awareness suggested by the SHS and British Crime Survey data.
Combining data sources for serious injuries
6.13. The published DfT best estimates uses 'Comparison of Hospital and Police Casualty Data: A National Study' (Simpson, 2006). This work took hospital admissions and linked them to Stats 19 data, finding 56% of those with serious injuries from a road traffic accident could be matched to a Stats 19 record. This work also suggests that some casualties recorded as slightly injured on the Stats 19 form received treatment in hospital that would count as a serious injury using the Stats 20 definitions. It is estimated that the number of Stats 19 seriously injured casualties should be multiplied by 1.52 to include this misclassification of severity.
6.14. 'Linking Stats 19 and Scottish Hospital In-Patient data for the SafetyNet project' (Broughton, 2010) used Stats 19 data as the basis for matching and found 58% of seriously injured casualties recorded on Stats 19 could be matched with hospital records. This project used data from 1997 to 2005, almost 50,000 Stats 19 records and the level of matching dropped to 53% in later years. This suggests the multiplier to account for slight casualties that should be serious is 1.50 to include this misclassification of severity.
6.15. This research and the Scottish Household Survey data allow an estimate to be calculated in two different ways, one using levels of under-reporting and the other using matching rates with hospital data.
Published number of seriously injured casualties (2010):
Police aware but no Stats 19 match [i / 89%]:
Total seriously injured casualties not reported to the police [ii / 63%]:
6.16. The research also shows that some casualties recorded as slight injuries on Stats 19 should be categorised as serious injuries.
Proportion of Stats 19 seriously injured casualties matched to hospital data [i x 53%]
Apply percentage to calculate number of slightly injured casualties that should be recorded as seriously injured [iv x 50%]
Total serious casualties plus slightly injured casualties that should be recorded as seriously injured [iii + v]
6.17. It would also be possible to apply the proportions used in (ii) and (iii) above to the figure calculated in (v) which when added to (iii) gives an upper estimate of 4,430 serious casualties.
Published number of seriously injured casualties (2010):
- Of which not matched to hospital data [vii x 47%]
- Number matched to hospital data [vii x 53%]:
Of those in hospital, scale up to include slightly injured casualties that should be recorded as seriously injured [ix x 50%]
Total cases in hospital (using % of hospital data that can be matched to Stats 19) [(ix + x) / 56%]
Total seriously injured casualties (those in hospital + Stats 19 not in hospital) [viii + xi]
6.18. Of these, 67% are recorded on Stats 19 which is an over estimate when compared to the SHS data which suggested 63%. This would suggest a figure of nearer 4,000, the difference being seriously injured casualties reporting to neither the police nor to hospital.
6.19. These estimates are slightly below the figures that would be derived if the scaling factors from the research projects were applied, however the hospital definition of a road traffic accident differs from that used in the Stats 19 data collection system, and the severity of casualties (serious and slight) that report to hospital are likely to be different to from those who do not ie as severity increases the likelihood of going to hospital will increase. The more serious of the slightly injured casualties are the ones that are likely to seek medical attention.
Combining data sources for slight injuries
6.20. Slightly injured casualties are less likely to report to hospital so there are fewer sources to draw on to estimate levels of under-counting. The Simpson research mentioned above suggested there were just under a quarter of slightly injured casualties where the casualty said the police were aware of the accident which could not be matched to the Stats 19 data. Applying this and the 63% figure from the SHS suggests a total of 23,300 slightly injured casualties, though some slightly injured casualties should be recorded as serious, which could reduce this figure and others would not report to hospital or the police which could push the figure closer to 30,000. This would mean nearer to one in three slightly injured casualties recorded on a Stats 19 return, however there is little evidence to use to firm up this estimate above the 23,300 figure.
7. Best Estimates
7.1. A number of data sources have been considered to arrive at the estimates below
The published figure is the best estimate. The number of those killed in 2010 was 208.
SHS data suggests just under two thirds of serious and slight injuries are reported to the police and research projects that matched hospital data with police data suggest around half of seriously injured casualties could be matched to a Stats 19 return. Although most seriously injured casualties are likely to attend hospital, a casualty with a broken toe, which would count as a serious injury by virtue of Stats 20, may report to neither the police nor hospital.
The SHS proportion of accidents reported to the police and the proportion of health records matched to Stats 19 will include a number of cases that would not be counted as a serious injury Stats 19. Taking misclassification, under-reporting and under-recording into account, these data sources produce an estimate of around 4,000 seriously injured casualties in 2010, which compares with a published Stats 19 figure of 1,964.
An estimate for the numbers of slightly injured casualties not reported to the police, or reported to the police but not recorded on Stats 19, is harder to narrow down, however available data sources suggest an overall estimate of around 23,300 slightly injured casualties, compared to a published Stats19 figure for 2010 of 11,162. This is the area where there will be the largest difference between reported and actual injuries as it includes sprains, bruises and cuts which may just result in an exchange of details (assuming that a party other than the driver was injured or otherwise involved) and so need not be reported to any authority. Accidents involving slightly injured casualties make up the largest proportion of casualties, so applying a different rate to these has the largest effect on overall estimates and some other possible sources are discussed above.
7.2. Combining the estimates for killed and seriously injured casualties provides an overall estimate for the numbers who are killed or seriously injured of 4,200 casualties per year, compared to a published Stats 19 figure of 2,172. There is no reason to suppose that there has been any change in the likelihood of an accident being reported to the police, so this estimate does not affect the published trends.
7.3. Combining the estimated number of slightly injured casualties with the estimate for killed and seriously injured casualties is possible but not particularly helpful given that the level of confidence in the former estimate is much lower than that for killed and seriously injured casualties. The resulting total would be very sensitive to any changes in the proportion of slightly injured casualties in non-reported accidents. Also the focus tends to be on particular severity of casualty eg KSI (Killed and Seriously Injured) or pedestrian casualties, so applying a single estimate of under-counting to a subset of casualty figures would provide a distorted picture.
8.1. There are several ways to approach the estimation of under-counting of road casualties which provide a wide range of possible values.
8.2. For the reasons outlined and particularly the similarity with the National Records of Scotland data, the number of casualties reported as killed through the Stats 19 system is the best estimate.
8.3. The definition of serious injuries used in Stats 20 covers a range of injuries some of which will result in hospital admissions and some of which will be treated in other ways. Given the different definitions used in Stats 19 and the hospital admission data and the percentage responding to the SHS to say an accident had been reported to the police, it is estimated around half of seriously injured casualties are recorded as seriously injured on the Stats 19 form.
8.4. For slight injuries it is much harder to narrow down the estimates to a single number. Casualties whose injuries are very slight may have no reason to report to the authorities if details were exchanged at the scene and will possibly require no medical treatment. The figures above produce a very broad range, however using the SHS data and research suggests an estimate of just over two slightly injured casualties for each one recorded on Stats 19.
8.5. As slightly injured casualties make up the large proportion of total casualties, the uncertainty around this estimate would make a single figure for all accidents very difficult to interpret and for most purposes interest is in fatal and serious accidents or particular groups of casualties, where applying a single percentage under-counting of road casualties based on all severities would not be helpful.
8.6. These estimates are lower than the DfT estimates, which is what we would expect given rates of casualties per head of population in the two countries.
Edinburgh EH6 6QQ
Telephone: 0131 244 7254