1 – The surveys

6French surveys devoted, at least partly, to illicit drug consumption among young people generally use one of three data collection methods: telephone interviews (Baudier et al., 1998), face-to-face interviews (Lagrange et al., 1997) or self-administered questionnaires (Choquet and Ledoux, 1994; de Peretti and Lesellbaum, 1995; de Peretti, 1996; Baudier et al., 1997; de Peretti and Lesellbaum, 1999; Ballion, 1999; Navarro et al., 1999). The accuracy of reporting on illicit behaviour is difficult to measure, particularly among adolescents, even if the interviewers have been specially trained and the respondents have been told that their answers are confidential and anonymous. When the questions touch upon “sensitive” topics, adults appear to answer more easily by telephone than in face-to-face interviews. This has been shown in France with respect to surveys on lifetime drug use among persons aged 18-75 (Guilbert et al., 1999). For other sensitive subjects, however, such as sexual behaviour, the comparison of these two methods did not show such a clear-cut difference (ACSF, 1992; Riandey and Firdion, 1996; Le Vaillant, 1993) [3]. In fact, other factors such as the personal characteristics of the respondents and of the interviewers also play a role: the interviewer effect has often been demonstrated (Deville, 1994; Berthier et al., 1996), particularly in surveys on sensitive topics (Firdion and Laurent, 1998). As concerns adolescents, they seem to find it easier to talk face-to-face to young interviewers (Lagrange et al., 1997). The surveys used here are based on the other two methods of data collection.

7The Baromètre Santé jeunes 97/98 is a telephone survey (Baudier et al., 1998). The data were collected through a computer-assisted telephone interview system (CATI) that optimizes the quality of the survey by allowing data entry checks and saves interviewer’s time through computerized management of telephone calls, appointments, filtering, etc. The fieldwork took place in November and December 1997, with a sample of 4,115 young people representative of the population aged 12-19 living in metropolitan France. The sample was established on the basis of 39,300 phone numbers randomly selected from the national telephone directory. Among these, 1,598 did not correspond to the target (wrong or professional numbers), 27,983 did not correspond to the survey universe (71 non French-speaking households and 27,891 households with no persons aged 12-19), 4,101 were impossible to reach despite ten attempts on different days and at different times. If the contacted household had more than one person belonging to the targeted age group, the respondent was selected according to the next birthday method. Among the 5,639 eligible households, 983 refused to participate, 325 persons were impossible to reach or unable to respond, 144 persons personally refused to respond and 37 gave up in mid-interview. The survey had been announced by mail, before the first telephone call. The representative quality of the sample was ensured through statistical adjustment taking into account age, sex, region of residence and type of housing. The Baromètre Santé jeunes surveys collect information concerning behaviour, knowledge and attitudes of adolescents in matters of health. Only adolescents aged 15-19, that is 2,675 persons, were asked about illicit drug consumption. This precautionary measure was taken after a pilot survey conducted in October 1997 showed that younger adolescents were frequently embarrassed to talk about drugs. The research team also believed that approaching such a topic outside the framework of a prevention campaign and without offering any possibility for fuller discussion could easily be troubling for an ill-informed person, as is usually the case of young adolescents.

8In the CADIS survey on deviant behaviour among secondary school students, the data were collected using the self-administered questionnaire method. The survey was conducted in schools in two stages (April-May 1997 and November-December 1997) on 10,004 public secondary school students (Ballion, 1999). Six school districts (Amiens, Bordeaux, Créteil, Grenoble, Poitiers, Rennes) were chosen on the basis of their size and geographical location, and 85 schools were randomly selected according to size, type (general and technical on one hand, vocational on the other) and success rate in the baccalauréat (the secondary school graduation diploma) in 1996 within their respective school districts. In each selected school, four to six classes were chosen on the basis of grade. These classes were sampled exhaustively (all the students of the class were interviewed). The sampling was thus a multi-stage cluster sampling [4]. The students completed the questionnaires themselves, under the supervision of a member of the school’s medical staff, or if not of a main school counsellor or a social worker. Five schools refused to take part in the survey, and in each case the reason given by the principal was that the school had a “drug problem”. These five schools were not replaced and no adjustment was made to the sample. The truancy rate noted by the person responsible for supervising the filling-in of the questionnaire averaged 3% in general and technical schools and 5% in vocational schools. In addition, 85 questionnaires were considered unusable, which reduced the sample to 9,919 students. This survey covers deviant behaviour among secondary school students (risky behaviour, transgressions, victimization, consumption of psychoactive substances, etc.). It was presented to the students as a “survey on the schooling conditions of secondary school students”.

9This article is devoted mainly to the comparison between these two surveys. However, we will also from time to time refer to a third source: a survey that the OFDT commissioned the National Institute of Pedagogical research (INRP) and the University of Paris X to take on Parisian secondary school students and psychoactive drugs (de Peretti and Lesellbaum, 1999) [5].

2 – Systematic differences

10On the issue of illicit psychoactive substances, these three studies often ask rather similar questions, even if differences in formulation can sometimes hamper comparison. Nonetheless, the three surveys raise the question of the use of illicit substances during the last twelve months in very similar terms [6] and at the same stage in the questionnaire (towards the end, after a question concerning alcohol and tobacco consumption). It is thus possible to compare the results obtained for these declared uses. The question about past-year use is more “sensitive” than that pertaining to lifetime use, since the latter can sometimes be perceived as an initiation entitling the respondent to answer questions of opinion on drugs. This hypothesis is supported by American research which compares estimates on drug use obtained through self-administered questionnaires and in face-to-face interviews. The estimates are always higher with the self-administered questionnaire, and the differences are largest when the question touches upon a recent period. This is particularly true for questions concerning the use of cocaine during the month preceding the survey, since reported use is 2.4 times higher when the information is collected through a self-administered questionnaire. In comparison, the lifetime consumption of cannabis is only 1.6 times higher and that of alcohol during the last month is only 1.06 times higher; in the latter case, the difference between the two collection methods is not significant (Turner et al., 1992).

11Table 1 summarizes the prevalence of the use of different substances during the year obtained in the three surveys (i.e. the proportion of respondents having used a given substance at least once during the last twelve months). The differences between the Baromètre Santé jeunes survey and the CADIS-OFDT self-administered survey are relatively large, and what is more, systematic. Regardless of the substance under consideration, the observed prevalence is higher for the in-school self-administered survey. Even if one sums the items “hallucinogens” and “ecstasy” of the Baromètre survey for comparison with the items “ecstasy + LSD” of the CADIS-OFDT survey, the declared prevalence is lower in the Baromètre Santé jeunes survey. Similarly, the prevalence is higher in the INRP-Paris X-OFDT survey than in the Baromètre. These results must nevertheless be read with caution, since they concern only the Paris school district. The differences between the two in-school surveys (higher prevalence in Paris for cannabis and lower for other illicit substances) are probably linked to regional differences.

Table 1

Prevalence of use of different illicit psychoactive drugs during the last year, by source

Substance Baromètre Santé jeunes n =2,675 CADIS-OFDT survey n =9,919 INRP-Paris X-OFDT survey n =875 Cannabis 22.8% 29.8% 34.4% Cocaine 0.1% 1.9% 0.8% Heroin 0.0% 1.7% 0.2% Ecstasy(1) 0.5% 3.4% 1.5% Hallucinogens 0.7% item absent item absent Inhalants(2) 0.3% 5.7% 1.0% Other drugs 0.1% 4.1% item absent All substances(3) 22.8% 33.5% 35.3% (1) Ecstasy +LSD for the CADIS-OFDT survey. (2) Glue, solvents for the INRP-Paris X-OFDT survey. (3) The prevalence for all substances is lower or equal to the sum of prevalences in the column, since the dif ferent substances are not mutually exclusive. Sampled population: young people aged 15-19 (whether or not they attend school) for the Baromètre Santé, public secondary school students for the CADIS-OFDT survey, Parisian secondary school students (from both public and private schools) for the INRP-Paris X-OFDT survey. Sources: CFES, Baromètre Santé jeunes, 1997; CADIS-OFDT survey on deviant behaviour among secon dary school students, 1997; INRP-Paris X-OFDT survey on Parisian secondary school students and psy choactive drugs, 1998.

Prevalence of use of different illicit psychoactive drugs during the last year, by source

12As concerns reported use of illicit drugs during the year, we observe a systematic difference between the telephone survey and the self-administered questionnaire survey. In order to understand the reasons behind this difference, it is important to note that the two surveys do not target the same population: each collection method creates its own specific selection biases.

1 – Comparability of the two samples

13The populations surveyed by the Baromètre Santé jeunes and the CADIS survey match only partially. The Baromètre survey interviewed young people aged 15-19, who were either in secondary school (lycée), middle school (collège), higher education or not in school, while the CADIS-OFDT survey interviewed only secondary school students, among whom a small number were aged under 15 or over 19. To be valid, the comparison must be limited to the persons covered by both surveys, that is secondary school students aged 15-19. The comparison will thus be limited to this sub-group and to cannabis use during the previous year, because it is the only substance whose prevalence is sufficiently high to allow detailed statistical analysis. Once these adjustments are performed on the two samples, we are left with 1,681 observations for the Baromètre and 9,280 for the CADIS-OFDT survey; the respective prevalence of cannabis use is 21.3% and 29.7%. Contrary to what might have been expected, the exclusion of non-students from the Baromètre Santé jeunes survey leads to only a slight decrease of prevalence (which reached 22.8% before the samples were made comparable). If young people not attending school are more likely to use cannabis than those who are, this is because their average age is higher (Langemeijer, 1997); the age effect is limited by the sample being restricted to ages 15-19.

14For an even stricter comparison, we should have also excluded from the Baromètre sample the students of agricultural and private secondary schools, but this information was not available. However, a separate study conducted by CADIS in agricultural secondary schools shows that the prevalence of illicit drug use in these schools is close to that observed in other schools (Ballion, 1998). In addition, we would have had to exclude from the school survey the young people who were not reachable by phone—and thus not covered by the Baromètre survey—either because their household does not have one (this is true of about 4% of French households) (Beck et al., 2000), or because their phone numbers are unlisted (one in five households) (Ambroise and Mauris, 1999).

15It must be noted that refusals are much more frequent in phone surveys. A few absent students and a few unanswered questions (never in excess of 4%) affect the in-school survey, whereas in telephone interviews, several obstacles have to be overcome before the interviewer can start collecting information. Thus, even though the respondents themselves rarely refuse to answer (3.3% of individual refusals and 0.9% dropping out in mid-survey), refusals on the part of the household (i.e. by the parents) are much more frequent (17.4%), though not excessive for a survey of this type. For comparative purposes, the corresponding rates were respectively 14%, 1% and 8% in the survey on knowledge, attitude, beliefs and practices in matters pertaining to AIDS in France (Grémy et al., 1999) and 8%, 5% and 15% in the 95/96 Baromètre Santé survey for adults (Baudier and Arènes, 1997). Nevertheless, the major problem is that we know nothing of the young people who could not be reached by phone.

16We must also say a word about those who remain excluded from both types of surveys (because they are not in school and are difficult to reach by phone) and who represent a rather specific group in terms of psychoactive substance use. These can be young people who are hospitalized for relatively long periods, homeless youth or students who skip classes and spend most of their time outside their home. The survey conducted in 1997 and 1998 by Unit 472 of the French National Institute of Health and Medical Research (INSERM) among young people who are in the care of the Protection judiciaire de la jeunesse (PJJ – Court protection of youth) as a result of either their own behaviour or that of their kin, provides some clues despite the very low response rate (17%). Young people interviewed in this context began to use illicit drugs, and particularly cannabis, at an earlier age than average (Choquet et al., 1998).

2 – The differences remain after controlling for age and sex

17All the available data confirm that consumption of illicit drugs in general and of cannabis in particular is strongly linked to sex and age. Boys use illicit drugs more than girls, and use tends to increase with age, at least until 18 (Beck et al., 2000). If the CADIS sample were older and/ or more female than that of the Baromètre, this could explain the differences observed between the two surveys. In fact, the teenagers interviewed in school are slightly older than those interviewed by phone: 16.8 years against 16.6, this difference being significant at the usual level of 5%. On the other hand, there are slightly fewer boys in the CADIS sample (45.5% against 47.4%), but in this case the difference is not significant. In order to take into account the specific structure of each sample, it is better to compare the surveys at the same age and for the same sex. Figures 1 and 2 show the evolution by sex of cannabis use during the year between ages 15 and 19. The random sample of the Baromètre Santé jeunes was designed to allow the computation of a confidence interval (of 95%) for this prevalence, at each age and for each sex. For the in-school survey, we were unable to calculate the confidence interval due to lack of information on the clusters. Figure 1 shows that according to the two surveys, the prevalence among boys undergoes similar variations between ages 15 and 19. The curve corresponding to the CADIS data is always above the other, and is usually outside of the confidence interval of the Baromètre curve, whose limits are marked by the two thin lines.

Figure 1

Past-year use of cannabis among male secondary school students aged 15-19, by data source (%)

Past-year use of cannabis among male secondary school students aged 15-19, by data source (%)

18The same result can be observed among girls (Figure 2). The two prevalence curves have the same general shape, those of the CADIS data being higher and outside the confidence interval of the Baromètre curve (except at age 19, when this interval becomes wider, reflecting the small number of female secondary school students aged 19 in the Baromètre Santé jeunes survey).

Figure 2

Past-year use of cannabis among female secondary school students aged 15-19, by data source (%)

Past-year use of cannabis among female secondary school students aged 15-19, by data source (%)

19Thus, for the same age and the same sex, there remains a difference between the two surveys. Moreover, other factors linked to the consumption of illicit drugs must be controlled as well. Previous analyses have shown that this consumption is more frequent among teenagers who have repeated a school year, among those attending vocational schools, those who often miss school for no valid reason, and also among those living in a household whose head is a manager or an executive (Choquet and Ledoux, 1994, 1999; de Peretti and Lesellbaum, 1999).

20The higher prevalence of cannabis use observed in the school survey could be due to the overrepresentation of these particular characteristics. In fact, the students of vocational secondary schools are a little more numerous in the CADIS-OFDT survey (21.7% as against 19.8%), and those who have repeated a year at least once are a little more numerous in the Baromètre survey (50.7% as against 49.0%), but in neither case are these differences significant. On the other hand, truancy is more frequently reported in the CADIS-OFDT survey, even if the questions are not asked in the same way. In the latter survey, 40.8% of those aged 15-19 reported having already missed school “with no valid excuse”, whereas only 21.6% of the Baromètre survey declare having already “skipped class”; the difference in this case is highly significant.

21As concerns family background, the social characteristics of the parents are very different in the two surveys. For example, the proportion of blue-collar workers is much higher in the Baromètre survey (34.8% as against 21.1% in the CADIS-OFDT survey), and the opposite is true for lower-level white-collar workers (14.6% against 22.9%). This difference is probably due to the data collection method. Over the phone, the respondent reports the exact occupation of the household’s person of reference, and the interviewer then attributes a code number to the occupation using a detailed coding scheme, asking the respondent for additional information if necessary. In contrast, on paper, respondents mark what they think is the correct socio-occupational category, with a much greater risk of error. The differences in the opposite direction observed for blue-collar and lower level white-collar workers may reflect the difficulty that adolescents have in distinguishing between these two categories [7]. On the other hand, as concerns the category “managerial and higher intellectual occupations”, the proportions are almost identical in both samples (18.5% for the CADIS-OFDT survey, and 17.8% for the Baromètre survey). If we suppose that adolescents have less trouble identifying the socio-occupational category of the head of household if the latter is a senior executive or a university professor, this category then has the same weight in both samples. Under this hypothesis, the difference observed in the prevalence of cannabis use during the last year is not related to differences in the occupational status of the parents [8].

22How then are we to account for this difference? Within the limits of what is comparable, the samples are quite similar, and for this reason the difference cannot be attributed to a selection bias or to chance [9]. The only thing we can pinpoint is the greater frequency of truancy reported in the in-school self-administered survey. This finding is rather surprising, inasmuch as there is a strong correlation between truancy and sex or the fact of having repeated a year, and these two factors are distributed similarly in both samples. Thus, instead of explaining the difference in cannabis use by the difference in truancy rates, it might be more appropriate to look for a common explanation for both phenomena, by analysing the impact of the data collection method.

1 – Survey context or data collection method?

23Before investigating the impact of the data collection method, we must look at the context of each survey. The CADIS and CFES surveys do not use the same questionnaires and do not investigate exactly the same issues, even though most are quite similar. The topics dealt with just prior to the questions concerning illicit drug consumption may well influence reported use. Thus, in the CADIS-OFDT survey, the presence of questions about deviant behaviour in the wide sense of the term, including actions which are particularly hard to disclose such as theft, violence, racket, as either victim or perpetrator, may make the questions pertaining to drug use appear less “sensitive” to the respondents. This seems all the more probable since these questions appear towards the end of the questionnaire. The survey conducted by the INRP in Parisian secondary schools does not deal specifically with deviant behaviour, but in that survey, the prevalence of cannabis use during the year is even higher than that of the CADIS survey. Although the specificity of the Parisian population should make us cautious when drawing conclusions, the INRP-Paris X-OFDT survey suggests that the impact of context is not enough to explain the disparity.

24Another aspect linked to methodology must be considered to explain the differences in observed prevalence: the CADIS school survey asks about cannabis use during the past twelve months, but not about lifetime use. Some persons who have tried drugs may wish to have the fact “acknowledged” even if the event took place more than a year before, or may consider that since it could have happened more recently, it should be recorded as such, a reaction they would not have had if, as in the Baromètre survey, an earlier question had asked about lifetime drug use. The tendency of respondents to “collapse dates” and include earlier events among recent ones (that the questionnaire is asking about) (Auriat, 1996) may also decrease if it is possible to report the same event as having happened within a longer span of time. However, this hypothesis also is seriously undermined by the INRP–Paris X–OFDT survey, which included both questions and still yielded the highest prevalence of the three surveys.

25That leaves the possible impact of the data collection method. The consumption of cannabis may be easier to admit by ticking a box in an anonymous questionnaire than by saying the answer over the telephone, especially if the respondent is a teenager and the interviewer an adult. In the Baromètre Santé survey, to ensure that adolescents who are not often at home are not underrepresented, the respondents were contacted every day except Sunday at times when they were most likely to be at home (from 5pm to 9pm on weekdays, from 2pm to 9pm on Wednesdays and from 2pm to 6pm on Saturdays). At those times, the respondents could very well not have been alone in the house, and this might explain their reticence to confess an illicit activity their parents, more often than not, knew nothing about (Riandey and Firdion, 1996). Even if the questions of the phone interview are formulated very carefully, so that they can be answered discreetly by saying yes or no and without having to speak aloud the words that might give away the subject (Giami, Spira et al., 1996), an adolescent may nonetheless hesitate to reveal illicit behaviour in the presence of a family member. In fact, in cases where the respondent reported not being alone during the conversation, or where the interviewer indicated that the respondent seemed not to be alone or was not freely answering the questions, the prevalence of cannabis use was much lower (17% during lifetime and 14% during the past year, compared to 25.7% and 21.3% respectively, for the entire group of secondary school students aged 15-19 interviewed in the Baromètre survey). In addition, Aquilino et al. (2000) examined the influence of the presence of bystanders on the answers of teenagers in self-administered questionnaires on drug and alcohol use, and showed that the respondents were less likely to report cannabis and alcohol consumption when a parent was present [10]. These elements support the hypothesis that cannabis use is easier to acknowledge in a self-administered questionnaire completed in school.

26Similar results were obtained among adults, in particular concerning the significant influence of partner presence on answers to questions concerning marriage and marital life (Aquilino, 1991, 1993) or sexual behaviour (Firdion, 1996). On the basis of the ACSF survey, Firdion shows in particular that among persons aged 18-69, lifetime illicit use of drugs is underreported if a third party is present during a telephone interview. In the case of the CADIS-OFDT survey, the presence of a person supervising the survey may have introduced such a bias. Although it is not possible to study the impact of context on this bias since we have no precise information concerning the function of such persons, we do know that they usually belonged to the school medical staff, and that the school counsellors were rarely involved (Ballion, 1999).

2 – Data collection method and sensitive questions

27Our hypothesis is as follows. Teenagers interviewed by phone at home, in the presence of, or not far from, family members, hesitate to acknowledge illicit drug use. More generally, if this is the case, one will observe a systematic underreporting on “sensitive” issues—issues pertaining to behaviour which the respondent’s parents do not know about— in the Baromètre survey, whereas on other issues no difference will be observed between the Baromètre and the CADIS-OFDT survey. For instance, students skip classes a priori without their parents’ knowledge, but parents do know whether their child has repeated a year or not. Effectively, truancy is much more often reported in the self-administered survey (40.8% as against 21.6%), although the proportion of students who have repeated a year at least once is almost identical. The observed difference for truancy tends to increase with age (Figure 3).

28Our hypothesis is supported by findings on two other sensitive issues, one concerning tobacco, the other suicide. For daily smokers, the survey results are very similar: 20% of the teenagers surveyed by CFES smoke between 1 and 9 cigarettes per day, and 21% in the CADIS-OFDT survey, while for those smoking at least 10 cigarettes per day, the percentages are respectively 12% and 13%. In contrast, the results for occasional smokers differ significantly: 7% in the Baromètre as against 17% in the CADIS-OFDT survey, and this difference remains relatively stable between ages 15 and 19 (Figure 3). One may suppose that, in general, the parents of a regular smoker are aware that their child smokes. If smoking is occasional, however, they tend to be unaware, since this type of smoker is more likely to hide the fact from his or her parents. This type of consumption is thus more “sensitive” and less often reported in a phone interview.

29As concerns suicidal thoughts or suicide attempts, one can also compare the results of the Baromètre survey with those of the 1993 INSERM survey on adolescents (Choquet and Ledoux, 1994), an in-school self-administered questionnaire survey among students aged 11-19 (the CADIS-OFDT survey did not ask any questions on this topic). In the Baromètre Santé jeunes survey, 10.5% of the students aged 15-19 reported having had suicidal thoughts or attempted suicide, as against 23.4% of those aged 11-19 surveyed in 1993. However, if the question concerns suicide attempts followed by hospitalization, which the parents therefore know about, the results are identical: 1.3% for the 15-19 year-olds surveyed in 1997 and for the 11-19 year-olds surveyed in 1993 (Baudier et al., 1998). It thus appears that when interviewed at home by phone, adolescents tend to underreport behaviour that is unknown to their parents.

Figure 3

Truancy and occasional use of tobacco among secondary school students aged 15-19, by data source (%)

Truancy and occasional use of tobacco among secondary school students aged 15-19, by data source (%)

30In order to quantify the probable effect of the data collection method on the results, we performed a logistic regression by merging the samples of the two surveys (keeping the weighting of the Baromètre survey). The dependent variable is the prevalence of cannabis use during the past year. There are three types of independent variables.

31— Socio-demographic indicators: sex, age (in completed years beyond age 15, this variable being an integer between 0 and 4) and whether the head of the household occupies a managerial or higher intellectual occupation.

32— Educational indicators: attends a vocational school, has repeated a year (at least once) and truancy (has skipped class at least once in the past 12 months).

33— Impact of the data collection method: the sample to which each observation belongs.

34In addition, since a socio-demographic or educational indicator may not have the same incidence on the dependent variable in both samples, we will also compute interaction terms between these indicators and the data collection method [11].

35Table 2 makes it possible to sort out the numerous potential independent variables. For the model that includes all the variables, the table shows the level of significance corresponding to each [12]. All the socio-demographic and educational indicators have a significant effect on the modelled prevalence, except for vocational school attendance, whose effect becomes negligible when the other indicators are taken into account. The data collection method also has a significant effect. For the interactions between the data collection method and the previous indicators, the only ones that are not negligible are those pertaining to the socio-occupational category of the head of household and to truancy.

Table 2

Selection of independent variables for modelling the prevalence of cannabis use during the previous year

Combined CADIS-OFDT and Baromètre samples (n = 10,961) Potential independent variables Level of significance Socio-demographic indicators Sex <0.001 Age <0.001 Head of household in managerial or higher intellectual occupation <0.001 Educational indicators Vocational school 0.200 Repeated a year <0.001 Truancy <0.001 Data collection method <0.001 Interactions between data collection method and other indicators Data collection method sex 0.907 Data collection method head in managerial or higher intell. occupation 0.036 Data collection method vocational school 0.156 Data collection method repeated a year 0.287 Data collection method truancy <0.001 Sources: CFES, Baromètre Santé jeunes 1997; CADIS-OFDT survey on deviant behaviour among secondary school students, 1997.

Selection of independent variables for modelling the prevalence of cannabis use during the previous year

36For the final model, we retained the variables that have a significant effect in Table 2 [13]. Table 3 shows the results in detail. For sex, age, and socio-occupational category (SOC) of the head of household, repeating a year and truancy, the results confirm relations that were already well established by previous research. The data collection method also has a specific effect on reporting. All things being equal, a teenager interviewed by self-administered questionnaire is 1.69 times more likely to report having used cannabis during the year than a teenager interviewed by phone at home [14].

37The interactions considered in the model also require some comment, even though the interpretation of interaction terms is always a difficult exercise. In the first place, the link between truancy and cannabis use during the year is stronger for the adolescents interviewed in the Baromètre survey. For them, truancy multiplies by 5.20 the likelihood of cannabis use, taking into account the interaction with the data collection method [15]. These two types of behaviour are precisely those we believe are underreported by phone: the “real” link between both forms of behaviour may be reinforced by a common reporting bias. From this angle, we may interpret the odds ratio by considering that a teenager who declares during a phone interview never having “skipped class” is five times more likely to also report never having “smoked pot” during the year. The second, less significant interaction (see Table 2), shows that the link between the socio-occupational category of the head of household and cannabis use (the children of executives are more likely to have used cannabis during the past year) is also a little stronger for the teenagers interviewed in the Baromètre survey. In order to interpret this interaction, we may consider that the bias introduced by underreporting in phone interviews is also socially determined, and that this bias is less powerful in families of executives. As it happens, the odds ratio measuring the data collection method falls to 1.20 for the child of an executive [16].

Table 3

Results of logistic regression on the prevalence of cannabis use during the previous year

Combined CADIS-OFDT + Baromètre samples (n =10,961) Independent variables Odds ratio CI 95%(1) Sex Boy 1.63 [1.49; 1.78] Reference: girl -1- Age 1.11 [1.06; 1.16] Reference: 15 -1- SOC of head of household Managerial, higher intellectual occupation 1.49 [1.32; 1.68] Reference: other -1- Repeated a year At least once since attending school 1.25 [1.13; 1.39] Reference: never -1- Truancy At least once during past year 2.84 [2.58; 3.13] Reference: never -1- Effect of data collection method Self-administered (CADIS-OFDT) 1.69 [1.39; 2.05] Reference: phone interview (Baromètre) -1- Interaction data collection method and truancy Baromètre absent at least once 1.83 [1.38; 2.43] Reference: other -1- Interaction data collection method and SOC Baromètre head of family managerial or higher intellectual 1.41 [1.01; 1.96] Reference: other -1- (1) Confidence interval 95%. How to read the table: For each variable, the reference category is indicated in italics. By construction, the odds ratio is 1 for this reference category. For age, the odds ratio measures the impact of each additional year beyond age 15. Sources: CFES, Baromètre Santé jeunes 1997; CADIS-OFDT survey on deviant behaviour among secondary school students, 1997.

Results of logistic regression on the prevalence of cannabis use during the previous year

38These results suggest that phone interviews not only lead to underestimation of cannabis use but also distort the links observed between cannabis use and other related factors, to the extent that the latter are also underreported (truancy, for example) or interact with the reporting bias just mentioned (as with the SOC of the head of household).