2 Previous United Kingdom Studies
In this section we consider previous recent studies carried out in the United Kingdom that aimed to estimate the size of the illicit drugs market or to estimate the social and economic costs of drug use.
We begin by discussing the markets studies.
2.1 Sizing the market for illicit drugs
Bramley-Harker (2001), Sizing the UK market for illicit drugs
In a Home Office funded study, Bramley-Harker attempted to estimate the size of the drugs market in the UK, using data primarily for England and Wales. The study attempted to estimate the size of the market for six drugs; amphetamines, cannabis, cocaine (powder), crack cocaine, ecstasy and heroin. The study is described as an illustration of the methodology rather than the application of a firm tool to assess the size of markets. The report clearly set out the assumptions and used data available for 1998. Although there had been some attempts in the past to estimate the prevalence of problem drug use (such as the use of opiate and / or crack cocaine) by Frischer et al (2004), Bramley-Harker chose instead to derive information on the number of drug users from arrest data. He used the NEW- ADAM study (Bennett, 2000) to estimate levels of drug use by those who had been arrested and combined that information with information on the number of arrests. This programme was a survey of arrestees where interview data was augmented with urine samples to identify drugs in the respondents system.
The NEW- ADAM study at that time only sampled arrestees at four sites; Liverpool, Nottingham, Sunderland and South Norwood (an area of South London near Croydon). Those four sites are unlikely to be representative of England and Wales in terms of levels of drug use (as Liverpool and Nottingham both come in the highest ten out of 149 English Drug Action Team areas for the prevalence of problem drug use; Hay et al, 2006). It is, however, less clear as to whether they would be representative of levels of drug use amongst those arrested for crime, which is the more important issue.
The study used a "demand" rather than a "supply" approach, in that it considered the amounts of drugs used by drug users and built on that information to get a national figure for the amount of drugs used, rather than trying to estimate the amount of drugs entering the UK based on production, trafficking and successful seizures data. Bramley-Harker divided drug users (and hence the market) into three groups; regular users, occasional users and users in prison. He also limited his analyses to amphetamines, cannabis, crack cocaine, ecstasy, heroin and powder cocaine.
Bramley-Harker suggested that surveys are not appropriate for quantifying many patterns of illicit drug use, particularly heroin or crack cocaine use. He noted that the capture-recapture method can be used to estimate the size of heroin using populations, but suggested that it was less useful for estimating drug use prevalence nationally. This was, however, prior to large scale Home Office funded studies that provided local and national estimates of the prevalence of problem drug use for England, derived in part using the capture-recapture method (Hay et al, 2006), and national estimates of the prevalence of problem drug use in Scotland have been available since 2000 (Hay et al, 2001). He also considered multiplier methods, such as the mortality multiplier, for use in estimating prevalence and discussed what they saw as the benefits of the multiple indicator method (Bramley-Harker, 2001, p.6).
The commissioning body (the Home Office and H.M. Customs and Excise) were keen to receive a methodology based on data that are routinely available and updated on an annual basis. That is perhaps why the study focused on the NEW- ADAM data (which they considered would be a continuing data source) rather than more one-off studies such as capture-recapture or multiple indicator method studies.
To derive estimates of the number of drug users from NEW- ADAM, Bramley-Harker followed a process that first involved estimating the number of people in England arrested for all offences. This is then combined with information derived from the NEW- ADAM sample on the proportion of those arrested who were regular drug users.
Next they attempted to estimate the probability that a regular user is arrested within a year. This information can be combined with the total number of people arrested and the proportion of those arrested who were regular drug users to derive estimates of the total number of people using individual drugs. They then estimated the number of regular users who are in the general population ( i.e. not in prison), again using information from NEW- ADAM (this time the respondent's prison history).
They used that approach to estimate the number of amphetamines, crack cocaine, heroin and powder cocaine users, but suggested that they prefer to use household survey data for quantifying cannabis and ecstasy use. They therefore limited estimates of cannabis and ecstasy use to 'occasional' use.
The Bramley-Harker study estimated the total value of the UK drug market (inclusive of amphetamine, cannabis, cocaine, crack cocaine, ecstasy and heroin) in 1998 to be £6.6bn. Heroin took the largest share of the market, at £2.3bn, followed by crack cocaine at £1.8bn. The size of the crack cocaine market was much bigger than the powder cocaine market which was just over £350m. In terms of quantities, this would equate to just over 31 tonnes of heroin, around 18 tonnes of crack cocaine and about 4.5 tonnes of cocaine powder. The amount of cannabis in the UK market was estimated to be just over 480 tonnes a year.
In his referee comments, Pudney suggested that Bramley-Harker was erroneously using NEW- ADAM to calculate the probability that a regular drug user is arrested. He suggested that NEW- ADAM could actually only be used to calculate the probability that, once arrested, a regular drug user goes on to be re-arrested. More broadly, he noted that different drugs may lead to different arrest rates, something not accounted for in the Bramley-Harker analyses. This was particularly relevant for comparing crack cocaine use with powder cocaine use, where one could speculate that a heavily addicted crack user may have more chance of arrest than someone only using powder cocaine.
Pudney noted some of the issues related to Bramley-Harker's approach of using a regression model to calculate the rates at which regular users use different drugs. After suggesting that this was flawed, he went on to note that since it is likely that Bramley-Harker under-estimated the number of regular users and then over-estimated their drug consumption, the final estimates may not have been too badly biased. He did, however, suggest that this could be a reason for an apparent disparity between the size of the powder cocaine market and the crack cocaine market.
In his comments, Pudney rejected Bramley-Harker's approach of splitting the drug using population into three groups (regular, occasional and prison users) and suggested that a more appropriate split would be between users who had been arrested (within the past year) and users that had not. The basis for this suggestion was rooted in the availability of data, in that an offender based survey may be the best route to quantifying drug use amongst arrestees and another survey, such as the British Crime Survey ( BCS) would be the best route of quantifying drug use in the non-arrestee population. It should be remembered that arrestees (in this instance) are people arrested for any relevant crime, not just drugs possession or supply offences. He was, however, sceptical of the ability of NEW- ADAM to accurately portray levels of drug use across the total population in England and Wales that have been arrested. He also made a brief comment about the inappropriateness of extrapolating the BCS to provide estimates of the size of the Scottish and Northern Ireland drug markets.
The rest of Pudney's referee comments were concerned with making suggestions for improving the methodology, and indeed when Pudney was later commissioned to provide estimates for 2003/04 he described his report as making such methodological improvements. As previously mentioned, he suggested that it was more appropriate to split the population into arrestees and non-arrestees. He suggested that one primary reason for rejecting the regular / occasional drug use split used by Bramley-Harker was that levels of drug use form a continuum that cannot be easily split into two groups.
In summary, Pudney concluded that while the Bramley-Harker study was a worthy first attempt at estimating the size of the UK drugs market, it should not be regarded as a reliable benchmark. He suggested that the lack of measures of inherent uncertainty in the estimates was a serious failing. Finally, he suggested that established surveys, such as the BCS, should consider redesigning their drug use components to make them more appropriate for estimating levels of drug consumption.
Pudney et al (2006), Estimating the size of the UK illicit drug market
When Pudney and colleagues estimated the size of the UK illicit drug market in 2003/04, they again chose to include only amphetamines, cannabis, crack cocaine, ecstasy, heroin and powder cocaine. They also primarily focused on England and Wales, but suggested that their approach to extrapolating to the UK level was better than Bramley-Harker's, since they took into account the difference in age structure between Scotland and England and Wales. However, they still ignored differences in some of the more fundamental inputs into their estimates, such as arrest rates (which are used to estimate the number of drug users) and consumption rates.
While noting that their estimates were improvements on previous ones partly due to the availability of better data, they also suggested that their estimates were better as they also included drug use by juveniles (aged 10 to 16) and because of the various statistical innovations they employed, such as correcting for non-response and under-reporting in surveys.
As with the previous study, the size of the drug using population was derived from surveys, rather than more sophisticated methods for estimating prevalence which were in the public domain at that time, in particular the work of Frischer et al who used the multiple indicator method alongside Drug Action Team ( DAT) area level capture-recapture estimates for Brighton, Greater Manchester (10 DAT areas), Liverpool and areas within Greater London (12 DAT areas) to provide a national estimate for England. This may, however, be because of the perceived need to use data from sources that would be systematically updated every year (as would the surveys Pudney and colleagues settled on).
As suggested in the comments on Bramley-Harker's earlier work, they split the adult population into two groups - those that had been arrested (within the reference year) and those that had not. Arrests were not restricted to obvious drug-related arrests such as possession or supply offences. They used the Offending, Crime and Justice Survey ( OCJS) carried out in 2003 to examine drug consumption by non-offending adults and used the Arrestee Survey ( AS) carried out in 2003/04 to quantify drug use by arrestees. The AS can be seen as a more comprehensive and updated version of the NEW- ADAM survey used by Bramley-Harker, however the Arrestee Survey takes saliva samples instead of urine samples to corroborate recent drug use. Although still beset by the issue of low response rate (often more to do with gaining access to an arrestee rather than the arrestee refusing to participate) the AS is far more representative and larger, with a sample size of just over 7,500 in 2003/04.
While initially it may appear strange to use an offending survey ( OCJS) to look at non-arrestees, it was possible from that data source to isolate respondents who had not been arrested in the last year. Self report of arrest was used to identify those who had been arrested. Respondents could, however, be reporting that they were arrested for minor, non-notifiable offences such as drunk and disorderly. These offences do not appear in official statistics therefore there could be an issue of comparing offending behaviour in OCJS with the published offence statistics. Information about drug use and drug consumption by juveniles came from the Schools Survey ( SS) carried out in 2003. Where possible, Pudney and colleagues compared and contrasted the results from these surveys with other surveys (such as the BCS, the Youth Lifestyles Survey ( YLS) and the previous NEW- ADAM study). Thus Pudney and colleagues quite rigorously and systematically reviewed all available drug prevalence / drug consumption information and, in all likelihood, used the best available data. They went further and considered and updated these data in light of issues such as the wording of questionnaires, the validity of self reports against biological measures (such as urine or saliva tests) and non-response in surveys. They also considered under-reporting in surveys, but concluded that they could not effectively account for this issue.
After concluding that the responses in the AS were not reliable enough to quantify amounts spent on drugs (or indeed amounts used) Pudney and colleagues also systematically compared the price and consumption data from a wide range of sources, including the National Criminal Intelligence Service ( NCIS), the Forensic Science Service ( FSS) and the Independent Drug Monitoring Unit ( IDMU). The FSS data was particularly useful for examining purity, whereas the IDMU data was used mainly in comparison with more official data, as it is, in part, derived from unrepresentative surveys from night clubs or music festivals. Pudney and colleagues also cast their net wider than the UK, considering Australian data.
Pudney and colleagues did not feel they had relevant information on the number of drug users in England from sources such as Frischer's (Frischer, Heatlie and Hickman, 2004) (for more problematic patterns of drug use such as heroin use) or more directly from the BCS (in terms of cannabis use). They therefore estimated the size of the drug using population by combining the probability of an individual being arrested and the probability that, given arrest, they use drugs (and similarly the probability that, given they had not been arrested, they use drugs). Once these probabilities were combined, they also required information on the number of arrests to extract the number of drug users out of this equation. Obtaining information on the number of arrests appeared to be less than straightforward, eventually involving the authors estimating the size of the England and Wales population that are not living in households (split by age group) in order to obtain what they consider to be more reliable estimates of the number of arrests in England and Wales in 2003/04.
As they did not explicitly estimate the number of drug users within their analyses, instead arriving at it by combining probabilities of arrests, probabilities of drug use if arrested (or if not arrested) and estimated numbers of arrests, they did not present information on the number of people they estimated to be using drugs (hence not allowing the reader to make direct comparisons with Bramley-Harker's prevalence estimates or the other estimates available at that time, such as Frischer's). They only presented the estimates of the size of the markets (in aggregate street quantities, aggregate pure quantities and aggregate expenditure) for both England and Wales and extrapolated to the UK level. They also apportioned the market by the three population groups they have created; juveniles, adult non-arrestees and adult arrestees. In total they estimated the size of the market in England and Wales to be £4.645 billion, with an estimate for the UK of £5.271 billion. Pudney and colleagues quite rightly attempted to examine the statistical variation in these estimates (in a similar manner to which a 95% confidence interval reflects the statistical variation in a point estimate) and presented the error bounds as plus or minus £1.2 billion for the English estimate and £1.3 billion for the UK estimate. They noted that their estimate of £4.6 billion is less than the £5.5 billion estimate of Bramley-Harker and noted that part of the difference could be due to falling drug prices; it could also be due to the methodological differences and differences in the data sources used. They did, however, note that in terms of individual drugs, the value of the powder cocaine market in the 2003/04 report was over 300% larger than in the original report.
2.2 Estimating the economic and social costs of illicit drug use
There have been two studies commissioned by the Home Office to estimate the social and economic costs of illicit drug use in England and Wales. The first, carried out by Christine Godfrey and colleagues (2002), estimated the costs in 2000. The second updated these costs in terms of methodology and the data used in 2003/04 (Gordon et al, 2006). Both studies estimated the cost to society as a result of class A drug use only, while this current study aims to estimate the cost to society as a result of a wide range of illicit drug use including class A drugs as well as drugs such as cannabis. The rationale for the Home Office funded studies also differed slightly to the current study in that their aim was to produce estimates that "represent the baseline against which the effects of any changes in policy could be assessed" (Godfrey et al, 2002, pg 4). Therefore the decision to include costs was based partly on whether the cost would change as a result of changes to government policy. This current study on the other hand aims to provide a comprehensive measure of the costs to society as a result of illicit drug use, whether or not they are likely to be influenced by policy changes.
A study with a similar remit to this current one is the 2002 study that estimated the cost of illicit drug use in Canada as well as the cost of alcohol and tobacco use (Rehm et al, 2006). Although we are focusing on estimating the cost of illicit drugs on society, part of the remit for this current study is to consider extending the model to include alcohol and tobacco. Therefore, while the Canadian study is not as relevant as the Home Office commissioned studies in terms of its geographical and social profile, it is a useful study to examine given its similar remit.
Methods for constructing the cost model
Both Home Office studies estimated the cost to society by considering the consequences of class A drug use by three different drug using populations, in terms of levels of use as well as problems associated with these different levels of use. These groups were labelled young recreational, older regular and problematic drug users. Problematic users were defined as those whose use of drugs was an essential part of their life and this intense use was impacting negatively on their life. Where this was not the case, then the person was defined as a recreational user; and was defined as a young recreational or older regular user depending on whether the person was under or over 25 years.
This split seems sensible when considering the possible consequences of problematic drug use, which would by definition, be more serious than those experienced by recreational drug users. Thus, by attaching costs to the consequences and combining with prevalence figures for recreational and problematic drug users a more realistic and precise model of the social and economic costs of class A drug use can be constructed. However, while it was appropriate to split the recreational group by age in order to model the impact of specific policies on young recreational drug users, there is not such a need to do so in this current study.
Costs (or consequences) were considered for five different domains: health, work, driving, crime and other social consequences for each of the three drug using populations. They were then categorised into one of six different groups depending on who bore the cost ( e.g. the user themselves, families, the public sector etc). Although the five domains seem quite comprehensive, the rationale for using the estimates to model policy changes meant that some costs were not included in the model since they would not be changed by policy changes. Therefore costs such as treatment and law enforcement costs were not included in the model. Other costs, for example drug driving costs, were included in the model but it was not possible to estimate a cost due to the limitations on available data relating to drug driving.
In contrast, the Canadian study in line with the 'International Guidelines for Estimating the Costs of Substance Abuse' (Single et al, 2001) did not split the drug using population when considering what consequences, or costs should be included in their model. The categories of costs included in the Canadian model were similar to Godfrey et al's cost domains, namely: direct health care costs, direct costs of law enforcement attributable to illegal drugs and other substance-attributable matters and indirect costs such as lost productivity. However, their list was more comprehensive and included costs such as inpatient specialised treatment and costs for prevention and research.
Estimating the cost consequences
The authors of both Home Office reports on the social and economic costs of class A drug use used two methods to estimate the costs included in their model. Where possible they used risk probabilities (the probability that a person will experience a certain consequence as a result of their drug use), calculated using sources such as the NTORS study (National Treatment Outcomes Research Study) which gathered a wide variety of information on the health, criminal and social status of problematic drug users. The risk probability was then applied to the prevalence estimates for each of the three drug using populations, thus producing an aggregated estimate of the cost of class A drug use that modelled the different consequences in terms of severity and level experienced by the three different drug using populations.
Where it was not possible to use risk probabilities, particularly pertinent to the young recreational and older regular users, the authors used available data on the current level of consequences. For example, it was not possible to estimate the probability of a drug related death by a young recreational user using available data sources on young people's drug use. Instead the estimate was based on the number of class A drug related deaths for the year 2000 who were defined as young recreational drug users in this study.
The first Home Office study (Godfrey et al, 2002) estimated the social and economic costs of class A drug use in 2000 to be just under £12bn while the updated study (Gordon et al, 2006) estimated the total cost for 2003/04 to be around £15bn.