Download presentation
Presentation is loading. Please wait.
Published byPaul Dorsey Modified over 6 years ago
1
Representativeness challenges for web surveys collecting information on drug users: a way for improving the cannabis market estimation in France. Stéphane Legleye1, François Beck1, Stanislas Spilka1,2 1 CESP, Inserm, Villejuif, France 2 Observatoire français des drogues et toxicomanies, France
2
First French web survey among cannabis users
Methods and some descriptive results Cannabis market estimation / limits Comparison with other studies Recommandations
3
The EMCDDA web survey The EMCDDA developes estimates of the size of the drug market in the EU based on data provided routinely to the EMCDDA by National Focal Points Problems: The quantities of substances and expenditures of different groups of drug users may be known in detail through small scale surveys without representativeness On the other hand, General Population Surveys generally do not have sufficiently large samples of drug users to provide robust information on amounts and expenditures In 2016, a pilot study, in a first phase, was conducted by EMCDDA in different countries across Europe to overcome these difficulties: Croatia, the Czech Republic, France, the Netherlands, Switzerland and the United Kingdom.
4
Recruitment: 4849 clics have been logged during the whole survey period : from 23 may to 1 september 2016 (101 days, i.e. 50 clics per day); 4849 clics on the web page 3738 accepted to participate (they had to give their agreement to participate) 2615 declared to live in France 2295 started at least one of the four modules relating their use drug habits 3000 used at least once one time last year cannabis or cocaïne or mdma or amphetamine (this was a necessary condition for participation) 1932 adults finally can be analysed : aged from 18 to 64 and with cannabis use in the past year
5
Strategy recruitment A dual strategy Through specialised websites
Links (banner) on 10 of the most important websites regarding drugs (self-support associations, prevention and information websites, etc.) flyer ; dedicated page on our web site; twitter posts during the whole survey period; The survey was relayed by several of our partners (university, research centres) a banner on the site of a student health insurance (only 30 clics); An advertisement posted on about pages Facebook pages: 3765 clics paid
6
A dual strategy Après 5 semaine de la première phase de recrutement, on n’observe plus que 25 connexions. Dès le début de la campagne de recrutement via facebook le nombre de connexion remonte rapidement pour retomber dès l’arrêt de la campagne!
7
Web survey vs GPS: vs All 62 67 38 33 42 29 17 9 3 40 27 30 49 22 23
% WEB SURVEY 2016 (n=1932) LAST GPS 2014 (n=1287) Facebook (n=804) Other (n=1128) All MALES 57 65 62 67 FEMALES 43 35 38 33 18-25 yo 78 60 42 26-34 yo 16 20 29 35-44 yo 4 13 17 45-54 yo 1 5 9 55-65 yo 2 3 SECONDARY SCHOOL 12 COLLEGE 48 40 27 UNIVERSITY 45 30 ALONE 52 47 49 22 COUPLE 25 23 15 COUPLE + children 6 ALONE + children 7 OTHERS 34
8
Web survey vs GPS: vs WEB SURVEY 2016 (n=1932) LAST GPS 2014 (n=1287)
% WEB SURVEY 2016 (n=1932) LAST GPS 2014 (n=1287) Facebook (n=804) Others (n=1128) All PROFESSIONNAL STATUS Missing 8 6 7 - Employed 32 39 36 55 Student 51 40 45 20 Unemployed 11 9 17 Others 2 4 3
9
Frequency of use? Two sets of questions on web:
number of RESIN smoking days number of WEED smoking days Frequency based on the sum of the two Frequency based on the maximum of the two CANNABIS SUM MAX GPS Year 14% 41% Month 1-9 23% 25% 30% Month 10-19 11% 10% Month 20-29 16% 4% Month daily 37% WEED WEB GPS Year 18% ? Month 1-9 36% Month 10-19 15% Month 20-29 16% Month daily GPS: only one question: « how many times did you smoke cannabis? »
10
Polydrug users and cannabis homegrowers in web respondents
Drug (past year) Web GPS Cocaine 40% 8% Amphetamins 33% 2% MDMA 49% 6% More cocaine and MDMA in non-facebook respondents: 43% vs 35% and 52% vs 46% respectively Selection of cannabis homegrowers in web respondents Weed frequency Missing Yes Total year 15% 4% month1-9 6% month10-19 3% 12% month20-29 2% 17% daily 1% 28% Total (n) 97 224 1932
11
Quality of cannabis expenditure reports in web (resin and weed)
(Based on SUM) N Obs % Miss Min Mean 95th Pctl Max year 260 70% 30 300 920 month1-9 442 57% 26 100 600 month10-19 215 29% 42 110 360 month20-29 308 13% 86 250 460 daily 707 14% 189 550 4000 High level of expenditure missing, especially for low frequencies of use Apparently abnormally high expenditure in low and high frequencies of use
12
Estimation procedure 2 types of data :
full data: observations WITH missing expenditures Completed data: respondents WITHOUT missing expenditure 2 frequencies of use: based on SUM or MAX Extreme values above p95 in each frequency of use recoded as non-response 2 estimation methods: IPW or matching IPW: missing expenditures are imputed in web respondents with all variables available (homegrowing) Matching: missing expenditures imputed after matching in GPS with GPS variables only (e.g. no homegrowing)
13
Some details (1) IPW: propensity score: (variables present in GPS and web) gender, age group, diploma, occupational situation, family situation, cocaine, amphetamines, MDMA (All IPW-weighted web samples were well balanced compared to the GPS: standard differences less than 0,1) IPW: imputation of missing expenditures in web: Predictive mean matching stratified by cannabis frequency of use Gender, age, age-squared, diploma, family situation, occupational situation, homegrowing, cocaine, amphetamines, MDMA
14
Some details (2) Matching based on propensity score:
Propensity score used for IPW, stratified by frequency of use Nearest neighbour with replacement: all GPS observations are matched Matching: Imputation in GPS (after matching) Predictive mean matching, stratified by frequency of use Gender, age, age-squared, diploma, family situation, occupational situation, homegrowing, cocaine, amphetamines, MDMA
15
Some details (3) Finally we get 8 estimations of total expenditure
2 data x 2 frequencies x 2 methods = 8
16
Results (P95 troncated expenditures)
Frequency based on SUM Frequency based on MAX IPW MATCHING FULL 1,779 M€ 2,216 M€ 2,032 M€ 2 399 M€ Complete 1 787 M€ 1,904 M€ 2,183 M€ 2 212 M€ Matching yields higher estimates than IPW (due to frequent users) No reliable mean to identify the best method Average annual expenditure=2,064 M€, min=1,779 M€, max=2,399 M€
17
Summary of estimations
EXPERIMENTAL ! Annual expenditure for all cannabis products: 2,060 M€ Annual expenditure for weed only: 1,020 M€ Annual expenditure for resin only: 1,040 M€ Weed contributes to the half of the estimation
18
Validity Internal validity: some evidences
Small differences between methods Matching and IPW led to balanced samples BUT no frequency of weed smoking and homegrowing in GPS External validity: uncertain ! Self-selection of frequent cannabis users that are polydrug users in the web survey High expenditures in the web survey: drug dealers? Collective purchases? lack of precision in the web survey Measurement mode effect (frequency of use, expenditures)?
19
Comparisons to previous estimations (1)
2005: M€: (Legleye et al. 2008) 2005 GPS, 2003 school survey (Espad) and adolescent survey (Escapad) Expenditures in Escapad number of joints on last occasion + method of supply price*frequency in GPS Partial disconnection between data sources 2016: 2,200 M€ [1,900; 3,100] (EMCDDA, 2016) 2014 GPS (same as used in this presentation) frequency*price arbitrary level of weed smoking frequency Price estimated by retail and seizure data (external sources) Disconnection between data sources
20
Comparison to previous estimations (2)
2010: 1,199 M€ (computations by the authors) 2010 GPS (n=1625 past-year cannabis users) Report of frequency, expenditures, weed smoking and homegrowing Single source for all data Very strong internal and external validity BUT: underreporting of expenditures? Due to data collection mode effect (social desirability)? Due to some lack of details: no distinction between weed and resin? Recent changes between 2010 and 2014: Number of past-year users raised by 29% Rise of weed homegrowing: 6% (reported) to 15% (IPW) evidence of a switch in the type of cannabis? Weed is more expensive than resin (1.5 times) Inflation of 5%
21
Recommandations for a replication
Try to recruit more diverse profiles Ask comparable web and GPS questionnaires On cannabis Frequency of use assessed the same way (more simple on Web) Frequency of herb smoking and homegrowing in GPS? Personal or collective purchases on Web On other drugs including alcohol and tobacco On sociodemographics Income Rationale: adding control variables “reduces” self selection and measurement biases ! Ask expenditures in a GPS subsample to control for mode effect
22
References Legleye S, Ben Lakhdar, Spilka S. (2008), Two ways of estimating the euro value of the illicit market for cannabis in France, Drug and alcohol Rev, 27(5), Ben Lakhdar C, Lalam N, Weinberger D (2016), L’argent de la drogue en France, Estimation des marches illicites en France, Rapport à la Mildeca, 2016 EMCDDA (2016), Technical report, Estimating the size of the main illicit retail drug markets in Europe
23
ANNEX: expenditures for weed only
24
ESTIMATION for Weed ONLY
Problem: weed smoking days is not known in GPS Wee keep the global frequency of cannabis use as stratification variable Weed expenditures in the web survey Weed frequency N Obs % Miss Min Mean 95th Pctl Max year 344 72% 9 50 300 month1-9 687 53% 21 70 400 month10-19 286 27% 45 150 month20-29 317 24% 87 210 500 daily 298 26% 161 1000 Some abnormally high expenditures All expenditures above P95 will be recoded as non-response in each frequency category
25
Results (P95 troncated expenditures) for HERB only
Frequency based on SUM Frequency based on MAX IPW MATCHING FULL 939 M€ 934 M€ 1,074 M€ 1,054 M€ Complete 929 M€ 1,066 M€ 1,103 M€ 1,031 M€ Matching yields higher estimates than IPW No real mean to identify the best method Average annual total expenditure=1,016 M€, min=929 M€, max=1,103 M€
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.