The analytical approach used here is designed to address several methodological concerns. • In the present state of knowledge, supply-side estimates are unlikely to be reliable.
This is because the trade in illicit drugs and the organisation of supply within the UK are very difficult to observe in a systematic way.
• No single source of survey data will be adequate for measuring total demand, because of incomplete coverage and non-response.
• Survey response is potentially an important issue, since the characteristics of survey non-respondents may differ systematically from those of respondents.
• The available survey data may be subject to measurement error. In particular, there is reason to believe that some survey respondents may under-report their use of drugs. • No estimate of market size will be completely reliable, so indicators of reliability
should be offered to users of the estimates.
Unlike much of the earlier work in this area (for example, Rhodes et al., 1997, and Bramley- Harker, 2001), the authors do not make a formal distinction between “regular” and “occasional” users, since these terms are ambiguous and they define consumer groups which cannot be surveyed directly. The analysis is based on a partition of the population into three groups, which are more directly accessible by survey.
Juveniles, aged 10 to 16, observed through a school-based survey.
Non-arrestee adults, defined as people aged 17 to 65 who do not experience arrest within the reference year; they are observed through a survey of the household-resident population.
Arrestee adults, defined as those aged over 16 who experience arrest within the reference year; they are observed through a survey of arrestees in police custody.
Children aged under ten and non-arrestees aged over 65 are assumed to have zero drug consumption. Each of the three groups is partitioned further, according to their response behaviour in the relevant surveys; thus, in principle, the authors allow non-respondents to be distinguished from respondents. The explicit treatment of non-response means that it is possible to explore the consequences for measurement of making different assumptions about the behaviour of non-respondents.
A major distinction between the authors’ approach and the earlier NERA analysis is that this study does not assume that all regular drug users are necessarily subject to arrest and thus representable solely by evidence from the arrestee population. Moreover, unlike NERA, this study does not “scale up” evidence from the arrestee population to represent the hypothetical wider set of potentially arrestable drug users, since the source of evidence on the general population appears to represent adequately those who have experienced arrest but are not in the population of current arrestees.
The full approach to measurement is set out schematically in Figure 4.1. The process has several stages.
Stage 1Imputation of frequencies of drug-use for survey non-respondents. This is done by fitting a statistical ordered-response model to data on respondents and then using it to predict
the probabilities of each category of consumption-frequency for all non-respondents. The explanatory variables in these models are variables which can be observed for both respondents and non-respondents. For the household-population survey, these variables are mainly descriptors of the local area; for the arrestee survey, they include demographic characteristics and circumstances of arrest. Non-response is a relatively minor problem for the survey of juveniles and the authors make no adjustment in that case.
Stage 2Adjustment of the estimated distribution of frequency-of-use to deal with under- reporting by survey respondents. This is difficult, owing to the lack of external checks on the validity of responses. The authors did not use adjustments for under-reporting in the main estimates but did report the results of an experimental adjustment procedure, which makes use of the biological tests made on saliva samples from arrestee survey respondents.
Stage 3Use of assumptions about average drug prices, quantity consumed per episode of use and purity to convert frequency-of-use to predicted annual consumption and expenditure rates. This process is a simple matter of multiplying by assumed average price, purity, and rate of use per day to convert predicted frequency-of-use probabilities into expected consumption. However, these conversion factors are based on the rather thin evidence currently available and they introduce the largest element of uncertainty into the estimation process.
Stage 4Construction of population aggregates for England and Wales using external information on the sizes of the adult non-arrestee, juvenile and arrestee sub-populations. The authors first estimated the size of the arrestee population, allowing for the fact that many individuals appear multiple times in official arrest statistics. They then combined estimated consumption figures for the juvenile, adult non-arrestee and arrestee populations, using their relative sizes as weights.
Stage 5Extension of the estimates from an England and Wales to a UK basis. For this, official estimates of the size of the Scotland, Northern Ireland and England and Wales populations, together with information on the (proportionately smaller) arrestee population in Northern Ireland were used. The assumption underlying the procedure is that juveniles, arrestees and non-arrestees have essentially the same behaviour patterns in all nations of the UK and that rates of drug use by drug users are also uniform.
Stage 6Comparison with external indicators of the structure of the drugs market. The authors compared the estimated market structure with the composition of drug seizures, to give a necessarily rough assessment of the concordance of the demand-side estimates with (arguably) the most reliable supply-side indicator.
Stage 7Calculation of indicators of the reliability of measured aggregates. The authors used an ad hoc method of combining two sources of uncertainty into a single rough indicator of the likely scale of error in their estimates. For each of the parameters fixed by external assumption (average street prices, purities and consumption amounts per use-day), the estimation process was simulated by drawing a large number of alternative values from an a priori range of uncertainty (assuming a uniform distribution). For each of these replications, the authors estimated total market size, computing confidence intervals (by means of a bootstrap method) for the intermediate quantities which are constructed from survey data. The total variation produced by this simulation process then gives an indication of the range of uncertainty in the final results.
Statistical analysis for AS respondents
Statistical analysis for adult non- arrestee OCJS respondents.
Predict frequency-of- use probabilities for AS non-
respondents, using predictor variables observable for both respondents and non-respondents Estimate frequency- of-use rates for AS respondents
Predict frequency-of- use probabilities for OCJS non-
respondents Estimate frequency- of-use rates for OCJS respondents
Optionally, adjust AS frequency-of-use probabilities for under- reporting Construct population aggregates using information and assumption about relative sizes of the arrestee and non-arrestee populations, split into respondent & non-respondent groups England & Wales Estimates Check against external data Explore sampling error & other uncertainties Comparative analysis of
drug test results and self report in AS and NEW- ADAM to use as an approximation to misreporting probabilities Convert to pure form using FSS purity estimates Final UK estimates Statistical analysis for SS respondents Estimate prevalence rates for SS respondents Predict frequency-of- use rates for SS drug users, from OCJS frequency-of use rates
Convert frequency-of-use probabilities to annual quantities and expenditures, using assumptions and/or data on average prices and average quantity per episode of use