Substance Use 8~ Misuse, 34(2), 269-295, 1999 Measurements, Instruments, Scales, and Tests Characteristics of Inconsistent Respondents Who Have “Ever Used” Drugs in a School-Based Sample Ohidul Siddiqui, Ph.D.,* Joshua A. Mott, Ph.D., Tammy L. Anderson, Ph.D., and Brian R. Flay, Ph.D. Prevention Research Center, University of Illinois at Chicago, Chicago, Illinois, USA ABSTRACTS This study examines the predictors of inconsistent responses from adolescents to questions about whether they ever used alcohol, cigarettes, and marijuana. Male adolescents had significantly higher rates of inconsistent responses than female adolescents. Black and Hispanic adolescents had significantly higher rates of inconsistent responses regarding ever using alcohol and cigarettes (only for Black) than White adolescents. The subjects’ living status and academic achievements were significant predictors of inconsistent responses regarding ever using marijuana. Thus, these results are consistent with the notion that inconsistent responses may bias the estimation of the prevalence of ever using drugs in multivariate analyses. Key words. Random-effects models; Inconsistent responses; Longitudinal study; Smoking prevention *To whom correspondence should be addressed at Georgetown Medical Center, 2233 Wisconsin Ave. NW, Suite 440, Washington DC 20007, USA. 269 Copyright 0 1999 by Marcel Dekker, Inc. www.dekker.com 270 SIDDIQUI ET AL. INTRODUCTION . Longitudinal surveys of adolescent drug use rely heavily on self-report measures. The reliabilities of such measures are therefore of concern to researchers (Martin and Newman, 1988; Needle et al., 1989; Single et al., 1975). As a result, a considerable amount of empirical research has been devoted to identifying sources of adolescents’ inaccurate responses to questions about personal drug use. One way to do this is to examine the consistency of adolescent drug use selfreports over time.* Inconsistencies in adolescents’ reporting of drug use over time may result from a variety of causes including lying, errors in recall, and random errors. When asked to report on socially disapproved behaviors such as drug use, adolescents’ responses may be influenced by factors such as the type of social image they desire to project about themselves. For example, adolescents may often conceal their true actions and underreport drug-use behaviors. Alternatively, some adolescents may “brag” or overreport about engaging in a disapproved behavior even when they have not. In this regard, the “truthfulness” of adolescent responses to drug-use questions could be expected to be influenced by their misunderstanding of the purpose of the survey, personal characteristics of the data collector (gender, race/ethnicity, age), the site of data collection, confidentiality issues, or the implications of the survey findings. There are also factors which might increase the chance that adolescents will unintentionally respond inconsistently over time. Recall bias makes it questionable that adolescents can always accurately remember their past drug-use behaviors. They may also be uncertain of definitions such as “ever using” drugs. For example, adolescents who have used substances once or twice may not consider themselves “users” at all. In this regard, the complexity of the responses requested, the time of day when questions are asked, or ambiguity in the wording of the questionnaire items may be important causes of inconsistent responses. Single et al. (1975) investigated reasons for the inconsistent reporting of drug use among adolescents within waves and between waves of a longitudinal survey. They concluded that most of the inconsistencies reflected uncertainty regarding the definition of drug use as well as recall error. The results of Bachman and O’Malley (198 1) suggested that the tendency to “telescope” historical eventsthat is, to accidentally recall events of the distant past into more recent time periods-may also produce inconsistencies in drug-use data. In particular, they reported that the frequency of drug use during the past month was roughly three times larger than would be estimated based on reports of use during the past year. These and other causes of the misreporting of drug use may have important con*However, while consistency of reporting is a necessary component of accuracy, it does not guarantee it because a subject may report inaccurately all of the time. INCONSISTENT sequences for cused more cl dividuals, and responses to s Luetgert ; and frequency did not affect it did affect tl the subjects M and more like: that among hi port “ever” 01 consistent wil quality of dru nal Survey of than heavy UE Users of i of licit drugs. likely to unde results of Ped among Norwc use. In this sl written surve sponses show amphetamine of marijuana While Si of experimen claimed that able response likely to con1 responses ova social accept; In this regard experimental 14% of male males being 1 to marijuana responses va ing more inc by Fendrich underreporti] SIDDIQUI ET AL. ,N rely heavily on self-report mearefore of concern to researchers Single et al., 1975). As a result, (as been devoted to identifying lestions about personal drug use. .cy of adolescent drug use self? drug use over time may result recall, and random errors. When s such as drug use, adolescents’ s the type of social image they , adolescents may often conceal .viors. Alternatively, some ado: in a disapproved behavior even less” of adolescent responses to .tenced by their misunderstandacteristics of the data collector ection, confidentiality issues, or the chance that adolescents will . Recall bias makes it questionmber their past drug-use behav;uch as “ever using” drugs. For once or twice may not consider jmplexity of the responses re:d, or ambiguity in the wording ses of inconsistent responses. ie inconsistent reporting of drug :n waves of a longitudinal ‘suries reflected uncertainty regardor. The results of Bachman and “telescope” historical events: past into more recent time peIg-use data. In particular, they : past month was roughly three sorts of use during the past year. lg use may have important connponent of accuracy, it does not guar.ime. . .- INCONSISTENT RESPONSE 271 sequences for the validity of self-report data. As a result, other research has focused more closely on the identification of the behavioral characteristics of individuals, and aspects of their environment, that are associated with inconsistent responses to survey questions about drug use. Luetgert and Armstrong (1973) examined how anonymity, recency of use, and frequency of use affected subjects’ self-reported marijuana use. Anonymity did not affect whether the subjects identified themselves as users or nonusers, but it did affect the type of use that the subjects reported. As anonymity decreased, the subjects were less likely to report occasional present use (experimental use) and more likely to report frequent past use. Bachman and O’Malley (1981) found that among high school seniors, experimental users were much less likely to report “ever” or “current” use of drugs than extensive users. Their findings were consistent with the findings of Mensch and Kandel (1988), who examined the quality of drug data over two waves (1980 and 1984) of the National Longitudinal Survey of Youth (NLSY) and found that light users of drugs were less likely than heavy users to acknowledge their use. Users of illicit drugs may also be less likely to report their behavior than users of licit drugs. According to Mensch and Kandel(l988), adolescents are especially likely to underreport illicit drug use (other than marijuana). This is similar to the results of Pedersen (1990) who used a longitudinal study of lifestyle and drug use among Norwegian adolescents to investigate the reliability of self-reports of drug use. In this study, 1,936 high school students (aged 12-19 years) completed a written survey and completed it again 7-8 months later. The legal drug-use responses showed the highest consistency while the reports of hard drug use (e.g., amphetamine, heroin) showed the least consistency. The consistency of reports of marijuana and inhalant use were somewhere in the middle. While Single et al. (1975) argued that such inconsistencies in the responses of experimental users were the result of recall error, Mensch and Kandel (1988) claimed that this was more due to the tendency of some to give socially desirable responses with maturation. If this were the case, then subjects who are more likely to conform to societal norms with age would be more inconsistent in their responses over time. This would be due to the fact that their views regarding the social acceptability of their earlier responses had changed at a later interview date. In this regard, Mensch and Kandel(1988) found that of the subjects who reported experimental marijuana use (1-5 times) in 1980, 19% of females compared to 14% of males denied ever using marijuana in 1984. They interpreted this as females being more likely to be concerned with the social desirability of a response to marijuana use than males. The same study also concluded that inconsistent responses varied significantly by race/ethnicity, with Blacks and Hispanics being more inconsistent than Whites. Further support for this notion was provided by Fendrich and Vaughn (1994) who used the same data set to investigate the underreporting of lifetime marijuana and cocaine use. The results of this study 272 . SIDDIQLJI ET AL. also suggested that a high rate of underreporting was evident for both marijuana and cocaine use, with Blacks and Hispanics displaying higher underreporting than Whites. While the level of current use, the type of substance used, anonymity, gender, and race have all been found to be significantly associated with the underreporting of drug use among adolescents, other studies have not found significant relationships in this regard. For example, Needle et al. (1983) compared drug-use self-reports taken from an in-school survey of 155 7th11 th grade adolescents, with self-reports taken from the same students while they were in their homes in the company of their families. Measures of within-wave and acrosswave inconsistencies were used. They found that adolescent self-reports were, in most cases, reliable and valid, and that the setting in which the respondents completed questionnaires did not, in general, result in any systematic reporting bias. O’Malley et al. (1983) examined data from a nationwide survey of 2,589 high school seniors (19761978) and a follow-up survey 1 or 2 years after graduation. Personal characteristics (gender, race/ethnicity) were not associated with inconsistencies in self-reports of substance use. Bailey et al. (1992) explored the consistencies of self-reports of the frequency of use and age of first use of alcohol and marijuana in a sample of 5,770 secondary school students. Two waves of data were collected between 1985 and 1988 using self-administered instruments. In contrast to the results of Pedersen (1990), results showed that self-reports were more consistent for lifetime marijuana use than for alcohol use. However, the reliability of reported age of first use was very low for both substances. Finally, Simon et al. (1996) examined reports of the onset of smoking taken from 2,597 junior high school students in 7th and 8th grades. They found that students’ intrapersonal characteristics including risk-taking, self-esteem, perceived stress, perceived susceptibility to social influence, and friends’ smoking were not associated with their misreporting on a smoking onset question. A summary table for the above literature review is given in the Appendix. It conveys that the validity of self-reports of drug use remain ambiguous. A number of studies concluded that self-reports of drug use were reliable and valid (Needle et al., 1983, 1989; O’Malley et al., 1983; Pedersen, 1990; Single et al., 1975). However, Bachman et al. (1981), Bailey et al. (1992), Fendrich and Vaughn (1994), Luetgert and Armstrong (1973), and Mensch and Kandel(1988) concluded that self-reports of drug use were not highly reliable. In these latter studies the level and type of current drug use (Bachman et al., 1981; Mensch and Kandel, 1988; Pedersen, 1990), anonymity (Luetgert and Armstrong, 1973), gender (Mensch and Kandel, 1988), and raceiethnicity (Fendrich and Vaughn, 1994; Mensch and Kandel, 1988) were significant predictors of inconsistent responses. It can also be seen that existing multivariate research in this area (Fendrich and Vaughn, 1994; Mensch and Kandel, 1988) is based on the analysis of a single INCONSISTENI data set (the b has used univ consistent resI of the variabk (in a longitud a multivariate prehensively c cents (i.e., ger religiosity, ant over time. While the ethnicity may no multivariatl and risk-takin; For example, attendance mz haviors such a use at a baseli in a socially d lead to a reca associated wit at Time 1 ma) partly reflect : b) they may bc when, in fact, criminality to tions (Harrisc extent that de 1985), those v to inconsisten could include ful of researcl terview, or to 4 For these rest among adole: sponses to qu In these s as predictors 1 hol, cigarette! longitudinal c tors of denyin earlier wave. SIDDIQUI ET AL. vas evident for both marijuana ring higher underreporting than >stance used, anonymity, genficantly associated with the ter studies have not found sigNeedle et al. (1983) compared ey of 155 7611 th grade adoidents while they were in their :s of within-wave and acrossdolescent self-reports were, in .n which the respondents comany systematic reporting bias. ionwide survey of 2,589 high ,’ 1 or 2 years after graduation. ere not associated with incon:t al. (1992) explored the conmd age of first use of alcohol 31 students. Two waves of data ‘-administered instruments. In showed that self-reports were or alcohol use. However, the 1 for both substances. Finally, of smoking taken from 2,597 s. They found that students’ self-esteem, perceived stress, ends’ smoking were not assoquestion. ew is given in the Appendix. se remain ambiguous. A num: use were reliable and valid Pedersen, 1990; Single et al., et al. (1992), Fendrich and rd Mensch and Kandel(1988) iighly reliable. In these latter man et al., 198 1; Mensch and tgert and Armstrong, 1973), ticity (Fendrich and Vaughn, predictors of inconsistent reesearch in this area (Fendrich used on the analysis of a single 1 . INCONSISTENT RESPONSE 213 data set (the National Longitudinal Survey of Youth). Research on other samples has used univariate analyses on two waves of data to find the predictors of inconsistent responses. Univariate analysis can not consider the independent effects of the variables under study or the correlation structure of a variable over time (in a longitudinal design). As a result, the goal of this current research is to use a multivariate methodology and a data set from Southern California to more comprehensively examine the associations between several characteristics of adolescents (i.e., gender, race/ethnicity, family structure, risk-taking, perceived stress, religiosity, and academic achievement) and the inconsistent reporting of drug use over time. While the literature review indicates that factors such as gender and race/ ethnicity may be predictors of inconsistent responses to questions about drug use, no multivariate research has focused on the roles that adolescent religiosity, stress, and risk-taking tendencies may play as determinants of the same (see Appendix). For example, adolescents who have strong ties to religion and church/temple attendance may be less willing to consistently report engaging in “deviant” behaviors such as drug use over time. While they may have indicated personal drug use at a baseline measurement, perceived pressures to answer drug-use questions in a socially desirable manner that are associated with increased religiosity could lead to a recanting of earlier reports of drug use over time. Stress may also be associated with inconsistency similarly. Adolescent smokers who are under stress at Time 1 may be more likely to misreport at Time 2 because, a) their stress may partly reflect perceived negative social responses to their current smoking, and b) they may be able to reduce some of this stress by saying that they don’t smoke when, in fact, they do. Finally, some recent literature which links deviance and criminality to inconsistent reporting on drug use may also inform our expectations (Harrison and Hughes, 1997) concerning adolescent risk-taking. To the extent that deviant adolescents are more likely to be risk-takers (Elliot et al., 1985), those who score higher on a risk-taking measure may also be more likely to inconsistently report their drug use over time. Reasons for this inconsistency could include the increased tendency of deviant youth to be suspicious or fearful of research interviewers, to perceive a stigmatization of drug use by the interview, or to define current use as high frequency use (Magma and Kang, 1997). For these reasons we suspect that increased religiosity, stress, and risk-taking among adolescents may be significant determinants of inconsistencies in responses to questions about drug use. In these analyses we considered these and other sociodemographic indices. as predictors of the consistency of adolescents’ self-reports of ever using alcohol, cigarettes, and marijuana. Specifically, we used a multilevel analysis in a longitudinal cohort study with four waves of data to identify significant predictors of denying ever having used a substance after having admitted using it at an earlier wave. 274 SlDDIQLJl ET AL. INCONSISTENT METHODOLOGY Subjects The data set for this study was collected as part of the Television, School and Family Project (TVSFP), a longitudinal study of adolescent smoking prevention in Los Angeles and San Diego (Flay et al., 1988, 1995). This project used a longitudinal cohort design with randomization of schools to various treatment conditions. Interventions were delivered to students within classrooms. A cohort of 6,695 students from 287 classrooms within 47 public schools was surveyed with self-administered questionnaires at four time points. In the cohort, 757 students were present at only one time point. As a result, they were dropped from the analysis. The remaining 5,938 children were present at a minimum of two successive time points, and their information is analyzed in this paper. The survey questionnaire covered demographics, tobacco use, other drug use, and psychosocial correlates of substance use. A detailed description of informed consent as well as of procedures for assuring the subjects confidentiality can be found in Flay et al. (1988). The questionnaire was administered by trained data collectors and took 45 minutes to complete. These data collectors were project staff whom the students had come to know quite well in group discussions over the course of the study. The project staff had an average of 10.5 years of classroom teaching experience (including an average of 3 years of junior high school teaching) prior to the implementation of this curriculum, and the majority had previously taught a drug education program (Sob01 et al., 1989). Within schools, classrooms were randomly selected and letters describing the study were sent to the parents of the children. Although they were made aware of the optional nature of the data collection, less than 8% of the subjects did not participate due to parental or personal reasons. The students were pretested in January 1986 when they were in the seventh grade (wave A), and completed an immediate postintervention questionnaire in April 1986 (wave B). A l-year follow-up was given in April 1987 in the eighth grade (wave C), and a second year follow-up in April 1988 in the ninth grade (wave D). The pretested students were 49.3% male and 50.7% female. They were 35.2% Hispanic, 32.8%White, 15.2% Black, and 16.8% Asian. By the 2-year follow-up, 3,155 (47%) of the original sample was present. At that .time, 47.6% were male, 36.1% were Hispanic, 34.8% were White, 10.3% were Black, and 18.6% were Asian. A breakdown of the overall, consistent, and inconsistent samples by the independent variables is presented in Table 1. Our precise operational definitions of consistency and inconsistency are described in more detail in the Univariate Analysis section. An analysis of the factors affecting attrition in this study (those lost from the baseline to the 2-year follow-up) suggested that those who dropped out tended .Th Sample characteris Gender: Male Female Race: African-America White Hispanicnatino Asian Risktakers: Not risk takers Risk takers Stress during last 3 Did not feel Felt Religiosity: Never go to chur SometimesIalwa: Livingstatus: Living with both Not living with b Academic performa Usual grade A Usual grade B Usual grade C Usual grade D or ‘Subjects who had 1 waves. bSubjects who had a across all waves. to have lower ac and be more lik was not a signit likely than otha 1996). Thus, p underrepresentel prevalence of in As a result, we c 1NCONSlSTENT RESPONSE SIDDlQUl ET AL. 275 Table 1. .The Distribution of the Independent Variables in the Sample: Overall and by Reporting Consistency elevision, School and : smoking prevention is project used a lonuious treatment conrsrooms. A cohort of Is was surveyed with cohort, 757 students re dropped from the minimum of two sucis paper. The survey ug use, and psychoinformed consent as can be found in Flay 1 data collectors and eject staff whom the ver the course of the ssroom teaching exhool teaching) prior Id previously taught ~ls, classrooms were to the parents of the ture of the data colparental or personal were in the seventh .on questionnaire in I 1987 in the eighth ! in the ninth grade ; female. They were sian. By the 2-year 4t that time, 47.6% % were Black, and t, and inconsistent Our precise operaibed in more detail (those lost from the lropped out tended Sample characteristics Gender: Male Female Race: African-American White Hispanic/Latin0 Asian Risk takers: Not risk takers Risk takers Stress during last 30 days: Did not feel Felt Religiosity: Never goto church Sometimes/always go to church Living status: Living with both parents Not living with both parents Academic performance in school: Usual grade A Usual grade B Usual grade C or F Usual grade D . Consistent reporters’ (N = 4,597) Inconsistent reporters (N= I.3411 Total sample (N = 5.9381 46.8 53.2 57.5 42.5 49.3 50.7 14.0 33.8 35.2 17.1 19.4 29.6 35.1 15.9 15.2 32.8 35.2 16.8 80.1 19.9 74.5 25.5 79.1 20.9 75.4 24.6 78.4 21.6 75.9 24.1 26.6 73.4 30.4 69.6 27.3 72.7 70.2 29.8 69.0 31.0 69.9 30.1 6.7 42.4 39.1 11.7 4.3 39.1 43.4 13.2 6.2 41.7 40. I 12.1 *Subjects who had consistent responses to ever using alcohol, cigarettes, and marijuana across all waves. bsubjects who had at least one inconsistent response to ever using alcohol, cigarettes, and marijuana across all waves. to have lower academic achievement, have lower tobacco and health knowledge, and be more likely to be cigarette smokers and marijuana users. While gender was not a significant predictor of dropout, African-Americans were also more likely than other raceiethnicities to have been’lost to follow-up (Siddiqui et al., 1996). Thus, problem-prone subjects (substance users) may be somewhat underrepresented in these analyses, and it could be expected that estimates of the prevalence of inconsistent reporting among these subjects may be conservative. As a result, we considered attrition as a predictor of inconsistent responses in the 276 SIDDIQUI ET AL. model to test whether the attriters were more likely to provide inconsistent reporting during the time they were in the study as compared to subjects continually present at all four waves. Variables of Interest and Their Measures Measurement of Individual Inconsistency Scores. Individual inconsistency scores for ever using alcohol, cigarettes, and marijuana were calculated separately at waves B, C, and D. Answers given at one wave were compared to those given at the preceding wave to measure individual inconsistency scores. Thus, transitions between any two waves are essentially a 2x2 contingency table (Table 2). Specifically, the numerator of the inconsistency measure are those subjects who claimed to have used a given substance at time X, and then claimed “never” to have used that substance at a time X+ 1 (cell “yes, no” in Table 1). The denominator of the inconsistency measure is all subjects in each of the four cells in Table 2. We calculated inconsistency scores for ever using cigarettes, alcohol, and marijuana in this manner. Race/Ethnic@ and Gender. Dummy variables representing Black, Hispanic, and Asian racial/ethnic categories (White is the reference group) were created from the race/ethnicity variable. Gender was coded as 1 = male or 0 = female. Risk-Taking. Risk-taking preference was assessed by three items: a) It is worth getting into trouble to have fun, b) I like to take risks, and c) I enjoy doing things people say shouldn’t be done. The possible responses of each of the items were 1 = this describes me very well, 2 = fairly well, . . . , 5 = not at all. Since the three items form a reliable scale (Cronbach alpha = .78), we summed them together and averaged them to form the risk-taking scale score. For the univariate analyses, we dichotomized the scale as 0 = I do not like to take risks (responses: 3,4, and 5) or 1 = I like to take risks (responses: 1 and 2). This measure is described further by Collins et al. (1987) and Sussman et al. (1990). Table 2. Possible Responses to “Ever Using” a Drug in between Two Adjacent Waves Ever use Wave I + I No Wave I No Yes (no, no) (yes, no) Yes (no, yes) (yes, yes) INCONSISTEN-I Perceive1 four items: a) thing that hap unable to cant have you felt r angered becat possible respo 4 = often. The adapted from and averaged (only in the u or 1 = stress ( Religiosil asked. The re: With WI as 0 = not wit Academi assessed by “I were 1 = A, 2 sured at wave Dropout tionnaires we] to follow-up a jects who wer ered as stayer Statistical P We cannc independent, 1 that assume in surveyed on n highly likely ments nested anced design t dents within 1 multilevel dat effects modelr models (Jenm SIDDIQUI ET AL. Ire likely to provide inconsistent retdy as compared to subjects continuwes ency Scores. Individual inconsistes, and marijuana were calculated iven at one wave were compared to re individual inconsistency scores. essentially a 2x2 contingency table e inconsistency measure are those bstance at time X, and then claimed : X + 1 (cell “yes, no” in Table 1). .e is all subjects in each of the four y scores for ever using cigarettes, variables representing Black, Hishite is the reference group) were ier was coded as 1 = male or 0 = LS assessed by three items: a) It is : to take risks, and c) I enjoy doJossible responses of each of the = fairly well,. . . , 5 = not at all. mbach alpha = .78), we summed risk-taking scale score. For the as 0 = I do not like to take risks ; (responses: 1 and 2). This mea) and Sussman et al. (1990). . ‘sing” a Drug Waves eI+l Yes (no, yes) (yes, yes) INCONSISTENT RESPONSE 277 Perceived Stress. The perceived stress of each subject was assessed by four items: a) In the last month, how often have you been upset because of something that happened unexpectedly? b) In the last month, how often have you felt unable to control the important things in your life? c) In the last month, how often have you felt nervous and stressed? d) In the last month, how often have you been angered because of things that happened that were outside of your control? The possible responses to each of the items were 1 = never, 2 = rarely, 3 = sometimes, 4 = often. The four items form a reliable scale (Cronbach alpha = .83) which is adapted from the work of Cohen et al. (1983). We summed the items together and averaged them to form the perceived stress scale score. We dichotomized (only in the univariate analyses) the scale as 0 = no stress (responses: 1 and 2) or 1 = stress (responses: 3 and 4). Religiosity. A question “How often do you attend church or temple” was asked. The responses were 0 = never or 1 = sometimes or often. With Whom Subject Lives. The living status of the subjects was coded as 0 = not with both parents or I = lives with both parents. Academic Achievement. The academic achievement of each subject was assessed by “What grades do you usually get in school.” The possible responses were 1 = A, 2 = B, 3 = C, and 4 = D, F. All of the above variables were measured at waves B, C, and D. Dropout Status. Subjects who completed each of the four waves of questionnaires were considered as stayers (coded as I), and subjects who were lost to follow-up at successive waves were considered as dropouts (coded as 0). Subjects who were absent in one wave but present at later waves were also considered as stayers. Statistical Analysis We cannot assume that the behaviors of students within each classroom are independent, hence traditional statistical models, including logistic regression, that assume independence are not appropriate for this analysis. Each student was surveyed on multiple occasions, thus the multiple responses of each student were highly likely to be correlated. The data are multilevel data (repeated measurements nested within students nested within classrooms) and represent an unbalanced design (the repeated measurements within students and the number ofstudents within classrooms are not constant). Statistical analysis of unbalanced multilevel data has been developed under a variety of names, including randomeffects models (Gibbons et al., 1988; Laird and Ware, 1982), random-coefficients models (Jennrich and Schlucher, 1986), variance component models (Dempster 278 SIDDlQUl ET AL. et al., 1981), hierarchical linear models (Bryk and Raudenbush, 1982), multilevel models (Jennrisch and Schlucher, 1986), two-stage models (Bock, 1989), and mixed models (Longford, 1987). Advantages of random-effects regression models (RRM) over the traditional statistical models for the analysis of longitudinal data have been described (Gibbons et al., 1988), and include an ability to allow for missing observations, subjects measured at different time points, and estimation of random person-specific effects. Covariates included in the model can be either time-variant or time-invariant. We used a random-effects ordinal regression model described by Hedeker and Gibbons ( 1994) and implemented in the MIXOR computer program (Hedeker, 1993) for the data analysis. Since at present this procedure has only been developed for two-level data (Hedeker, 1993), our strategy was to tit the two-level model of students within classrooms (treating classrooms as a random effect) and repeated observations within students (treating students as a random effect). It was always the case that the program was not able to estimate a nonzero student variance term for the repeated observations within the student model. We then based our results on the students within classrooms model. The dependent variables in the models were the inconsistency scores (1 = yes or 0 = no) at the three time points (waves B, C, and D). Among the independent variables, gender and race were time-invariant, and the remaining variables were time-variant. RESULTS Univariate Analysis . The percentages of students by racelethnicity and gender who had inconsistent responses at each wave on each of the three questions assessing whether they ever used cigarettes, alcohol, and marijuana are shown in Table 3. The percentages of inconsistent responses of ever using alcohol at times Tl-T2, T2-T3, and T3-T4 were 6.82, 5.05, and 5.32, respectively. The corresponding percentages for cigarettes were 4.64,4.29, and 5.61, respectively. The percentages of inconsistent responses of ever using marijuana at times Tl-T2, T2-T3, and T3-T4 were 2.1, 2.5, and 2.8. The percentages of inconsistent responses of ever using alcohol decreased over time, but for cigarettes and marijuana they increased over time. Between 8.6 and 10.4% of the respondents gave inconsistent response at least once to each of the three drug questions between any two time points. About 17% of the subjects who gave inconsistent responses to ever using cigarettes also gave inconsistent responses to questions concerning ever using alcohol and marijuana. Of the subjects who were present at each of the four time points, 24.3% had at least one inconsistent response across time. Male subjects had higher percentages of inconsistent responses than female subjects for each of the three drug questions at each time point. Blacks and His- INCONSISTENT SIDDIQUI ET AL. Raudenbush, 1982), multilevel 3ge models (Bock, 1989), and rdom-effects regression models le analysis of longitudinal data include an ability to allow for :nt time points, and estimation luded in the model can be eijam-effects ordinal regression rd implemented in the MIXOR analysis. Since at present this ata (Hedeker, 1993), our strattin classrooms (treating classions within students (treating that the program was not able repeated observations within le students within classrooms ‘e the inconsistency scores (1 C, and D). Among the indeiant, and the remaining vari- Id gender who had inconsisitions assessing whether they wn in Table 3. The percentat times Tl-T2, T2-T3, and : corresponding percentages r. The percentages of inconTl-T2, T2-T3, and T3-Ti ent responses of ever using arijuana they increased over ve inconsistent response at any two time points. About to ever using cigarettes also ver using alcohol and marihe four time points, 24.3% itent responses than female ime point. Blacks and His- INCONSISTENT RESPONSE 280 INCONSI: SIDDIQUI ET AL. panics had higher percentages of inconsistent responses than Whites and Asians (the percentages of inconsistencies for Whites and Asians were very close to each other). Similar patterns of inconsistent responses by gender and race/ethnicity were also found for each of the drugs and at each time point. The subjects who liked to take risks had higher percentages of inconsistent responses on each of the three drugs between time 1 and time 2. Differences in the rates between high and low risk-takers decreased across waves however. The subjects who score high on the stress scale had lower inconsistent response rates to each of the three questions than the subjects who scored higher, though the differences were very small. Subjects who never attended church had higher inconsistency rates of ever using alcohol, cigarettes, and marijuana compared to subjects who attended church. Subjects who lived with both parents had lower rates of inconsistent responses for each of the three drugs compared with subjects who did not live with both parents (these differences also decreased over time). Subjects with higher academic achievement had lower inconsistent response rates than subjects with lower academic achievement. The prevalence of ever using alcohol, cigarettes, and marijuana with respect to the total sample and by race/ethnicity and gender is shown in Table 4. Prevalence rates for alcohol use were the highest (58.7-78.7%), while rates for marijuana were the lowest (I 2.0-25.4%). Prevalence rates for cigarette use were intermediate (42.4-59.1%). The percentages for each of the drugs increased over time. The prevalence rates for each of the drugs was lowest for Blacks, highest for Whites, and intermediate for Hispanics and Asians. For each of the drugs, males had higher prevalence rates than females. Table 5 shows the prevalence of ever using alcohol, cigarettes, and marijuana with respect to the subjects who had consistent responses (defined as the consistent group) across waves and those who had at least one inconsistent response (defined as inconsistent group) in ever using alcohol, cigarettes, or marijuana across waves. The inconsistent group had higher prevalence rates of ever using alcohol than the consistent group at wave A, but at waves B, C, and D the prevalence was lower for the inconsistent group than for the consistent group. For ever using cigarettes and marijuana, the prevalence rates were higher for the inconsistent group at the first three waves, and lower at the fourth wave. These patterns held across race/ethnicity and gender. Tt R. G ‘b Multivariate Analysis ., Table 6 shows the parameter estimates and the P-values in parentheses from the random-effects regression models for each of.the three drugs. We modeled the inconsistent responses (yes/no) of ever using a drug (e.g., alcohol) across time, and as a function of time, demographic variables, intrapersonal variables, and dropout observec not signi juana. T response porting 1 at early 1 marijuar Blat cigarette of incon: SIDDIQUI ET AL. ponses than Whites and Asians I Asians were very close to each s by gender and racelethnicity h time point. The subjects who onsistent responses on each of ences in the rates between high ‘er. The subjects who score high : rates to each of the three questhe differences were very small. inconsistency rates of ever us) subjects who attended church. rates of inconsistent responses cts who did not live with both me). Subjects with higher aca;e rates than subjects with lower tes, and marijuana with respect ler is shown in Table 4. Preva-78.7%), while rates for marirates for cigarette use were inch of the drugs increased over was lowest for Blacks, highest Asians. For each of the drugs, :ohol, cigarettes, and marijuana sponses (defined as the consiseast one inconsistent response :ohol, cigarettes, or marijuana prevalence rates of ever using .t waves B, C, and D the pr&ar the consistent group. For ever ltes were higher for the inconat the fourth wave. These pat- e P-values in parentheses from f the three drugs. We modeled drug (e.g., alcohol) across time, s, intrapersonal variables, and . 281 INCONSISTENT RESPONSE Table 4. Percentages of Respondents Ever Using Alcohol, Cigarettes, and Marijuana at Different Waves: For the Total Sample and by Race/Ethnicity and Gender= Wave A Wave B Wave C Wave D A C M 58.7 42.4 12.0 62.1 46.8 14.2 71.7 55.1 20.2 78.7 59.1 25.4 A C M A C M A C M A C M 43.7 33.1 9.0 70.8 46.1 14.3 57.0 45.1 12.0 54.4 38.3 10.3 47.8 41.6 12.3 74.3 49.9 16.6 61.1 48.8 13.8 58.4 41.9 11.0 61.0 45.9 19.0 82.8 58.1 23.3 72.2 58.3 19.3 70.0 51.6 15.0 66.8 52.7 27.2 86.4 60.9 29.5 77.7 61.7 24.1 76.6 54.6 20.2 A C M A C M 55.8 39.8 8.3 61.6 45.0 15.6 61.3 43.8 9.9 64.9 50.2 18.4 74.0 54.9 16.7 73.6 55.9 23.3 81.2 60.5 23.5 77.4 57.7 27.7 Total sample: Race: Blacks Whites Hispanics Asians Gender: Females Males -~ ‘A: alcohol, C: cigarettes, M: marijuana. - \ dropout status. The intrapersonal variables were time-varying, that is, they were observed at the time we measured the inconsistent responses. The time effect was not significant for inconsistent responses regarding ever using alcohol and marijuana. The time effect was significant and had a positive effect on inconsistent responses regarding ever using cigarettes. This means that the subjects were reporting more inconsistent responses to ever using cigarettes at later waves than at early waves. This was not the case with respect to the reporting of alcohol and .. marijuana use. Blacks provided inconsistent responses regarding ever using alcohol and cigarettes significantly more than Whites. Hispanics had significantly higher rates of inconsistent responses regarding ever using alcohol than Whites, but their rates 282 INCON: SIDDIQUI ET AL. Table 5. Percentages of Respondents Ever Using Alcohol, Cigarettes, and Marduana at Different Waves: By Reporting Consistency, Race/Ethnic& and Gender’ Group1 (subjects who had consistent responses in ever using alcohol, cigarette, and marijuana across the waves: N = 4,597) Group 2 (subjects who had at least one inconsistent response in ever using alcohol, cigarette, and marijuana across the waves; N = I ,34 1) Race: Blacks: Group I Group 2 Wave A Wave B Wave C Wave D A C M 56.8 36.1 6.6 60.0 42.7 9.0 66.4 52.1 16.0 73.7 57.5 23.4 A C M 76.0 45.8 18.8 52.0 47.9 16.8 63.0 53.2 19.2 58.5 51.1 16.6 Male: ‘A: alcc A C M A C M 38.6 25.9 2.5 69.9 35.3 17.5 45.0 36.1 5.4 45.6 39.4 16.0 60.4 43.3 13.9 54.7 41.5 12.6 72.3 49.0 24.8 48.5 45.8 22.2 A C M A C M 68.8 37.8 7.5 81.9 49.7 21.5 78.1 43.4 10.0 53.6 52.0 18.7 86.5 54.2 18.3 67.6 58.5 24.6 90.7 61.4 30.6 67.5 51.8 14.9 A C M A C M 52.9 36.8 6.9 79.2 50.4 19.9 62.9 45.0 9.5 49.4 47.6 14.8 72.7 55. I 15.9 67.1 57.7 19.4 82.2 60.7 23.7 60.3 56.3 19.6 A C M A C M 50.1 37.0 6.6 68.6 36.6 12.5 59.2 40.9 7.6 52.7 41.2 11.5 71.9 50.4 13.0 53.9 44.7 15.0 79.5 57.0 20.8 57.0 43.7 11.8 for ci! White the ral sisten cantly sisten not lil using better to eve nifica cigarc of the F ceive, spans mode nal ri (attrit not a the ti Whites: Group 1 Group 2 Hispanics: Group I Group 2 Asians: Group 1 Group 2 ., Gender: Females: Group 1 A C M 54.3 35.6 5.5 64.9 40.9 7.1 77.5 52.6 13.9 85.0 60.9 23.6 (conrimed) 1 perce from SIDDIQUI ET AL. INCONSISTENT RESPONSE 283 Table 5. Continued t, Cigarettes, and Manjuana at Different ?ace/Ethnicity, and Gender a ave A Wave B Wave C Wave D 6.1 6.6 60.0 42.7 9.0 66.4 52.7 16.0 73.7 57.5 23.4 5.0 i.8 .8 52.0 47.9 16.8 63.0 53.2 19.2 58.5 51.1 16.6 6 9 5 3 I 45.0 36.1 5.4 45.6 39.4 16.0 60.4 43.3 13.9 54.7 41.5 12.6 72.3 49.0 24.8 48.5 45.8 22.2 78.1 43.4 10.0 53.6 52.0 18.7 86.5 54.2 18.3 67.6 58.5 24.6 90.7 61.4 30.6 67.5 51.8 14.9 62.9 45.0 9.5 49.4 47.6 14.8 72.7 55.1 15.9 67.1 57.7 19.4 82.2 60.7 23.7 60.3 56.3 19.6 . 59.2 40.9 7.6 52.7 41.2 11.5 71.9 50.4 13.0 53.9 44.7 15.0 79.5 57.0 20.8 57.0 43.7 11.8 6.8 Wave A Wave B Wave C Wave D Group 2 A C M 74.1 42.1 12.9 46.7 43.3 10.1 59.4 52.4 15.5 61.1 52.2 13.5 Group 1 A C M A C M 59.8 36.8 8.0 77.4 48.3 23.1 67.3 45.1 11.3 53.1 50.7 19.9 75.0 53.0 18.6 65. I 54.5 21.4 82.3 56.7 28.4 59.5 52.0 20.0 Males: Group 2 ‘A: alcohol, C: cigarettes, M: marijuana. for cigarettes and marijuana were not significantly different from the rates for Whites. The inconsistency rates for Asians were not significantly different from the rates for Whites with regard to each of the three drug questions. The inconsistent rates of ever using marijuana for Blacks and Hispanics were not significantly different from the rates for Whites. Males gave significantly more inconsistent responses than females to each of the three questions. Subjects who did not live with both parents gave significantly more inconsistent responses to ever using marijuana than the subjects who lived with both parents. Subjects with better academic achievement also had significantly fewer inconsistent responses to ever using marijuana. Living status and academic achievement were not significant predictors of inconsistent responses of ever using alcohol and smoking cigarettes. No three-way or two-way interaction effects were significant in any of the three models. We also considered several intrapersonal variables such as risk-taking, perceived stress, and church attendance as possible predictors of inconsistent responses of ever using alcohol, cigarettes, and marijuana in the random-effects models. None of the variables was significant, so we dropped them from our final random-effects regression models. We also considered the dropout status (attrition) of the subjects as a predictor of inconsistency responses. Attrition was not a significant predictor in any of the three models, so it was also dropped from the final models. DISCUSSION 64.9 40.9 7.1 77.5 52.6 13.9 85.0 60.9 23.6 (continued) These analyses suggest several interesting findings. In the overall sample the percentage of inconsistent responses in the reporting of ever using alcohol ranged from 5.05 to 6.82% between any two time points (Table 3). For cigarettes this Table 6. Significant Predictors of Inconsistent Responses to Ever Using Alcohol, Cigarettes, and Marijuana: Final Multivariate Model Dependent measures Predictors Time Race: Black Hispanic Asian Gender Living status Academic achievement In alcohol parameter (P-value) .679(001) .208(.028) .178(.112) .289(<.001) - In cigarette parameter (P-value) .010(.022) .435(<.001) .112(.310) .089(.454) .382(<.001) - In marijuana parameter (P-value) .041 (.884) .029(.890) -.340(.191) .736(<.001) ,598 (.009) .612 (C.001) SIDDIQUI ET AL. INCONSISTENT RESPONSE 285 percentage varied from 4.29 to 5.6 1, and for marijuana use it varied from 2.1 to 2.8. We calculated these percentages based on the subjects who responded at time “X” that they had ever used drugs but responded at time “X + 1” that they had never used drugs. We found that the percentages of inconsistent responses for ever using cigarettes were less than the percentages for ever using alcohol at early time points. At later time points these differences were diminished, however. The percentage of inconsistent responses for marijuana use was the lowest of all the drugs. This is consistent with the findings of Bailey et al. (1992) who found greater inconsistencies in self-reports of alcohol as compared to marijuana use. It is possible that subjects may be more likely to change their reports over time, or be more susceptible to recall error, when asked to report on past licit drug use (which occurs with relatively high frequency) as compared to past marijuana use. However, the notion that marijuana use is a “rarer” statistical event relative to cigarette or alcohol use may also, in part, explain the reduced number of inconsistencies associated with its reporting. That is, from a statistical perspective one might anticipate that the rarer an event, the less the likelihood of inconsistencies appears. Everything else being equal, the greater the percent never using a substance at time X, the less the possibility of an inconsistent response (defined as it is in this and most other studies) from time X to time X + 1. When inconsistency rates were considered by gender and race, several trends were notable. The Black and Hispanic adolescents were more inconsistent in responding to ever using drugs than the White adolescents. Mensch and Kandel (1988) and Fendrich and Vaughn (1994) also found this to be the case. Such differences by race/ethnicity might be the result of culturally-based confusion of definitions of ever using drugs, intentional erroneous reporting, or some other sources. The effects of raceiethnicity might also be confounded with other factors. Socioeconomic status (SES), for instance, might be linked to inconsistencies in self-reports of drug use and to the race/ethnicity of a subject. In this regard, the definition of ever use might differ across ethnic or SES groups such that experimental or lower-level users might classify themselves as nonusers, and this classification criterion might vary systematically by ethnic and socioeconomic standing. Male adolescents also gave more inconsistent responses than the female adolescents in all cases. In the multivariate analysis, gender remained a significant predictor of inconsistency for each of the three drug use questions. Therefore, we can conclude that the female adolescents reported their drug use more reliably than the male adolescents. This stands in contrast to Mensch and KandEl’s (1988) findings that female adolescents were more inconsistent than male adolescents in answering questions about ever using marijuana. These differences might be due to the different ages of the two study cohorts. Our cohort was SIDDIQUI ET AL. 286 younger (12-15 years old) than the cohort ( 1927 years old) used by Mensch and Kandel (1988). In addition, Mensch and Kandel (1988) relied on a univariate analysis of the gender difference while our findings are based on both univariate and multivariate analyses. Finally, coming from a “broken” family structure and getting poorer grades in school were significant predictors of inconsistent responses to ever using marijuana. However, these factors were not significant predictors of inconsistent responses to the questions regarding alcohol and cigarette use. The affirmative reporting of marijuana use reflects the subjects’ participation in an activity which is illegal for adults, whereas the reporting of cigarette and alcohol use does not. In addition, living status and academic achievement may serve as proxies for a subjects’ social class. Thus, these findings are consistent with the notion that youth in lower socioeconomic circumstances may feel a greater need to “lie” in response to questioning about illegal activities as they grow older. This would reflect an increasing lack of willingness to acknowledge personal participation in illegal activities to adults, or those in “authority,” over time. LIMITATIONS . While these analyses extend existing research in this area in several important ways, they also have some limitations of their own. First, the rate of attrition in this longitudinal sample is not insignificant. Largely because of the transfer from junior to senior high-school, half of the original subjects were lost from the sample by the final wave of the study. In many important respects (i.e., gender, perceived stress, risk-taking, and religiosity) the attriters were not different from those subjects who remained in the study at the final wave. However, the subjects who dropped out did report greater lifetime use of cigarettes and marijuana. Thus, as in other longitudinal studies, it remains a possibility that drug-use behaviors among inner-city high-school students are somewhat underrepresented in this sample. However, we also expect that this underrepresentation of problem-prone youth would conservatively bias our prevalence estimates of inconsistent reporting among high school adolescents (Magura and Kang, 1997). Second, these data are drawn from an urban sample of children from Southern California. As many of the existing health and social problems confronting adolescents are often concentrated among inner-city youth, this is a very relevant sample for these analyses. At the same time, these results require replication in a sample which reflects the geographic and socioeconomic characteristics of the United States as a whole (and, indeed, other countries as well) before wider generalizations can be made. Third, these analyses included no control for the socioeconomic situation of the adolescents. As has been suggested above, many of the associations between raceiethnicity, family structure, academic achievement, and inconsistent reporting may reflect underlying linkages between socioeco- INCONSISTEN nomic status be structured proxy for SE: analyses, 0th For example, at time “r tl had never us “X” might he consistency, baseline, ma: jects use dru definition the which this is between the responses mi complexity 1 (Lorenz, 199 ing; howevet be addressed These li useful pract3 of substance using multiv to study incc appropriate : are indepenc correct for tl adolescents drug-use quc move beyor sponses. For sistent respo levels of risk not signitica Confoundin. meaningful In long: sponses are indicate tha SIDDIQUI ET AL. s old) used by Mensch and 88) relied on a univariate e based on both univariate nomic status and the consistency of drug-use self-reports. Further analyses should be structured to test our assertions about the extent to which these factors may proxy for SES effects. Fourth, in addition to the inconsistencies measured in these analyses, other types of inconsistent responses are possible but not measurable. For example, we defined inconsistent reporters as those subjects who responded at time “X’ that they had ever used drugs but responded at time ‘X+ 1” that they had never used drugs. However, some of those who had, in reality, used at time “X’ might have indicated that they had never used at that time. This type of inconsistency, where subjects hide the fact that they have ever used drugs at the baseline, may in all likelihood occur as frequently as inconsistencies where subjects use drugs and then “change their minds” at later time points. However, by definition the censoring of the data does not allow us to determine the extent to which this is the case. Finally, all of our analyses presume linear relationships between the dependent and independent variables. The debate on inconsistent responses might benefit from newer artificial paradigms (chaos, uncertainty, or complexity theories) that do not presume linear systems of cause and effect (Lorenz, 1993; Casti, 1994). Analysis strategies of these types are very promising; however, they are beyond the scope of the present investigation and should be addressed in future research. and getting poorer grades : responses to ever using predictors of inconsistent rette use. The affirmative bation in an activity which and alcohol use does not. ray serve as proxies for a tent with the notion that a greater need to “lie” in grow older. This would ge personal participation ler time. is area in several imporI. First, the rate of attrily because of the transfer ejects were lost from the nt respects (i.e., gender, were not different from ave. However, the subigarettes and marijuana. ibility that drug-use be:what underrepresented representation of probnce estimates of incon1 and Kang, 1997). Sec‘lildren from Southern blems confronting adothis is a very relevant s require replication in ic characteristics of the Yell) before wider genno control for the sogested above, many of academic achievement, :es between socioeco- 281 INCONSISTENT RESPONSE CONCLUSION . These limitations not withstanding, the results of this study offer several useful practical and theoretical implications for empirical researchers in the field of substance use. Methodologically, these findings suggest the importance of using multivariate methodologies which can account for multiple levels of data to study inconsistent responses. To this point, prior analyses have made the inappropriate assumption that the behaviors of students nested within classrooms are independent of each other. In these analyses we used multilevel modeling to correct for this assumption and found that several demographic characteristics of adolescents remained significant predictors of their inconsistent responses to drug-use questions over time. These findings also suggest that it is important to move beyond the univariate analyses of the determinants of inconsistent responses. For example, although our descriptive analyses revealed that the inconsistent response rates for ever using alcohol and cigarettes were different by the levels of risk-taking, perceived stress, and church attendance, these variables were not significant predictors of inconsistent responses in tbe multivariate analyses. Confounding effects such as these can be misleading to those looking to draw meaningful theoretical conclusions from uncontrolled analyses. In longitudinal smoking prevention studies, subjects with inconsistent responses are often overlooked, deleted, or ignored. The results from this study indicate that this may lead to important misrepresentations of substance use 288 . SIDDIQUI ET AL. among high school adolescents. About 24.3% of the subjects gave at least one inconsistent response to questions about ever using drugs across time. In addition, there was a tendency for subjects who reported inconsistently with regard to one type of drug to misreport on other types of substance use as well. These inconsistent responses did not occur randomly. Significant predictors of inconsistent responses to ever using alcohol and cigarettes were gender and race. Gender, living status, and academic achievement were significant predictors of inconsistent responses to ever using marijuana. Perhaps more importantly, the inconsistent group had differential prevalence rates of each of the three drugs over time compared to the rates for the consistent group. Thus, this study reveals that reporting a) the percentages of inconsistencies in self-reports of substance use and b) the methods used to handle such inconsistencies are essential for any longitudinal analyses of lifetime substance use among adolescents. Simon et al. (in press) made a recommendation that future longitudinal researchers identify misreporters and omit them from their analyses. However, the results of this study suggest that such a deletion of inconsistent responders might result in biased prevalence estimates of ever using drugs. For this reason, researchers need to be cautions in calculating the prevalence rates of lifetime substance use based on adolescent selfreports in longitudinal studies. APPENDIX: ANALYSES OF INCONSISTENT RESPONSES TO ADOLESCENT SUBSTANCE USE A summary table of the existing literature is presented on the next two pages. SIDDIQUI ET AL. of the subjects gave at least one tsing drugs across time. In addiborted inconsistently tvith regard of substance use as well. These Significant predictors of inconettes were gender and race. Genrere significant predictors of inPerhaps more importantly, the es of each of the three drugs over ntp. Thus, this study reveals that self-reports of substance use and ies are essential for any longitulolescents. Simon et al. (in press) researchers identify misreporters results of this study suggest that t result in biased prevalence essearchers need to be cautions in nce use based on adolescent self- YSTENT RESPONSES TANCE USE presented on the next two pages. . .. ^ ___- c Authors and year Data source Number of subjects Needle et al., 1983 Survey conducted by the authors. spring 1980, USA I.55 juniorlsenior high school students and their families O’Malley et al.. 1983. Monitoring the Future Project, a study of high school seniors conducted by the Institute for Social Research. 1976 through 1978. USA Two wwcs of data were collected by the author. 1985 and 1988. USA 2.589 high school seniors Bailey et al.. 1992 Aim of study Method of data collection Type of analysis RCSUitS To test for differences in adolescents responses on a substance we questionnaire as a function of social setting To measure the frequency of drug use and related attitudes Self-reported in classroom and self-reported at home with their families participating Univariate analysis Adolescent self-reports are, in most cases. reliable and valid Self-reported in classroom Correlation analysis between time points of responses 5.770 secondary school sludcnts To explore the consistency of self-reports of the frequency of “se and age of tint use of alcohol and marijuana Self-reported in classroom Univariate analysis Gender and race were not associated with inconsistencies in self-reports of substance use. Discrepant reports of marijuana and alcohol use WcrC COrrelatcd within time Self-reports were nwre consistent for lifetime and marijuana UK thsn alc0h01 use. Reliability of reported age of first use was w-y low for both substances The consistency dropped over time. The legal drug-use r~spotts~ showed the highest WnSiStCnCy: hsrd d”I8S showed the lowest consistency Students intrapersonal characteristics were not associated with their misreporting of the smoking onset question Pedersen. 1990 Two WBWS of data were collected by the author. 1989. Norway 1.936 high school students (age 12-l 9 years) To measure the reliability of self-reports of drug use Self-reported in classroom Univariate analysis Simon et al.. 1996 Cigarette smoking assessment project. 1989 and 1990. USA 2.597 junior high school students To measure the prevalence of SttIOking Self-reported in classrwm Univariate analysis m P I- INCONSISTENT RESPONSE 291 ACKNOWLEDGMENTS Collection of data for this research was supported by Grant ROl-DA0348. The analyses reported here were completed with support from Grants ROlDA06307 and ROl -DA 10306 from the National Institute on Drug Abuse. REFERENCES BACHMAN, J. G., and O’MALLEY, P. M. (1981). When four months equal a year: lnconsistenties in student reports of drug use. Public Opin. Q. 45: 536-548. BAILEY, S. L., FLEWELLING, R. L., and RACHAL, J. V. (1992). The characteristics of inconsistencies in self-reports of alcohol and marijuana use in a longitudinal study of adolescents. J. Stud. Alcohol 53: 636-647. BENSON, P., and DONAHUE, M. (1989). Ten-year trends in at-risk behavior: A national study of African-Americans adolescents. J. Adolesc. Res. 4: 125-139. BOCK, R. D. (1989). Measurement of human variation: A two stage model. In R. D. Bock (Ed.), Multilevel Analysis o/Educational Data. New York, NY: Academic Press. BROOK, J. S. (1993). lnteractional theory: Its utility in explaining drug use behavior among African-American and Puerto Rican youth. Drug abuse among minority youth. In M. R. De La Rosa and J. L. Recio Adrados (Eds.), Methodological Issues andRecent Research Advances (NIDA Research Monograph 130). Rockville, MD: National Institute of Drug Abuse. BRYK, A. S., and RAUDENBUSH (I 982). Hierarchical Linear Models: Applications and Data Analysis Models. Newbury Park, CA: Sage. CASTI, J. L. (1994). Complexijication: Explaining a Paradoxical World through the Science of Surprise. New York, NY: Harper Collins. COHEN, S., KARARCK, T., and MERMELSTEM, R. (1983). A global measure of perceived stress, J. Health Sot. Behav. 24: 38%396. COLLINS, L. M., SUSSMAN, S., RAUCH, J. M., DENT, C. W., JOHNSON, C. A., HANSEN, W. B., and FLAY, B. R. (1987). Psychosocial predictors of young adolescent cigarette smoking: A sixteen month, three wave longitudinal study. J. Appl. Sot. Psychol. 17: 554-573. DEMPSTER, A. P., RUBIN, D. B., and TSUTAKAWA, R. K. (1981). Estimation in covariance components models. J. Am. Sfat. Assoc. 76: 341-353. ELLIOT, D. S., HUIZMGA, D., and AGETON, S. S. (1985). Explaining Delinquency and Drag Use. Beverly Hills, CA: Sage. FENDRICH, M., and VAUGHN, C. M. (1994). Diminished lifetime substance use over time. Pubtic Opin. Q. 58: 96-123. FLAY, B. R., BRANNON, B. R., JOHNSON, C. A.,HANSEN, W. B., ULENE, A. L., WHITNEYSALTIEL, D. A., GLEASON, L. R., SUSSMAN, S., GAVIN, M. D., GLOWACZ, K. M., SOBOL, D. F., and SPIEGAL, D. C. (1988). The television, school and family smoking cessation and prevention project: 1. Theoretical basis and program development. Prev. Med. 17: 585-607. FLAY, B. R., MILLER, T. Q., HEDEKER, D., SIDDIQUI, O., BRITTON, C. F., BRANNON, B. R., JOHNSON, C. A., HANSEN, W. B., SUSSMAN, S., and DENT, C. (I 995). The television, school and family smoking prevention and cessation project: VIII. Student outcomes &id mediating variables. Prev. Med. 24: 29-40. FIORE, M., NOVOTNY, T., PIERCE, J., HATZIANREU, E., PATEL, K., and DAVIS, R. (1990). Trends in cigarette smoking in the United States: The changing influence of gender and race. JAMA 261: 4!%55. 292 SIDDIQUI ET AL. GIBBONS, R. D., EATERNAUX, C., HEDEKER, D., and DAVIS, J. M. (1988). Random regression models: A comprehensive approach to the analysis of longitudinal psychiatric data. Psychopharmacol. Bull. 27: 73-77. HARRISON, L., and HUGHES, A. (Eds.) (1997). The Validity ofSe[f-ReportedDrug Use: Improving the Accuracy o/Survey Esrimafes (NIDA Research Monograph 167). Rockville, MD: National Institute of Drug Abuse. HEDEKER, D. (1993). MIXOR: A Fortran Program for Mixed-Effects OrdinalProbir and Logistic Regression. Technical Report. Chicago, IL: Prevention Research Center, School of Public Health, University of Illinois at Chicago. HEDEKER, D., and GIBBONS, R. D. (I 994). A random-effects ordinal regression model for multilevel analysis. Biometrics 50: 933-944. JENNRICH, R. I., and SCHLUCHER, M. D. (1986). Unbalanced repeated-measures models with structured covariance matrices. Biometrics 42: 805-820. LAIRD, N. M., and WARE, J. H. (1982). Random effects models for longitudinal data. Biometrics 38: 963-974. LONGFORD, N. T. (1987). A fast scoring algorithm for maximum likelihood estimation in unbalanced mixed models with nested random effects. Biometriha 74: 817-827. LORENZ, E. N. (1993). The Essence of Chaos. Seattle, WA: University of Washington Press. LUETGERT, M. L., and ARMSTRONG, A. H. (1973). Methodological issues in drug usage surveys: Anonymity, recency, and frequency. Inr. J. Addict. 8: 683-689. MAGURA, S., and KANG, S. (1997). The validity of self-reported cocaine use in two high risk populations. In L. Harrison and A. Hughes (Eds.), The Validiv of Self-Reported Drug Use: Improving the Accuracy of Survey Estimates (NIDA Research Monograph 167, pp. 227-246). Rockville, MD: National Institute of Drug Abuse. MARTIN, G. L., and NEWMAN, I. M. (I 988). Assessing the validity of self-reported adolescent cigarette smoking. J. Drug Educ. 18: 275-284. MENSCH, B. S., and KANDEL, D. B. (1988). Underreporting of substance use in a national longitudinal youth cohort: Individual and interviewer effects. Public Opin. Q. 52: 100-124. NEEDLE, R. H., JOU, S., and SU, S. S. (I 989). The impact of changing methods of data collection on the reliability of self-reported drug use of adolescents. Am. .I. Drug Alcohol Abuse 15: 275289. NEEDLE, R. H., McCUBBIN, H., LORENCE, J., and HOCHHAUSER, M. (1983). Reliability and validity of adolescent self-reported drug use in a family-based study: A methodological report. Int. J. Addict. 18: 901-912. O’MALLEY, P. M., BACHMAN, J. G., and JOHNSTON, L. D. (1983). Reliability and consistency in self-reports of drug use. Inr. J. Addict. 18: 805-824. PEDERSEN, W. (1990). Reliability of drug use responses in a longitudinal study. &and. J. Psycho/. 31: 28-33. PULKKINEN, L. (1993). Youthful smoking and drinking in a longitudinal perspective. J. Youth Adolesc. 12: 253-283. SIDDIQUI, O., FLAY, B. R., and HU, F. B. (1996). Factors affecting attrition in a longitudinal smoking prevention study. Prev. Med. 25: 554-560. SIMON, T. R., SUSSMAN, S., DENT, C. W., and STACY, A. W. (1996). Predictors of misreporting cigarette smoking initiation among adolescents. Eval. Rev. 20(S): 552-568. SINGLE, E., KANDEL, D., and JOHNSON, B. (1975). The reliability and validity of drug use responses in a large scale longitudinal survey. J. Drug Issues 5: 426-443. SOBOL, D. F., ROHRBACH, L. A., DENT, C. W., GLEASON, L., BRANNON, B. R., JOHNSON, C. A., and FLAY, B. R. (1989). The integrity of a smoking prevention curriculum delivery. Health Educ. Res. 4: 5967. . INCON: sUSSM CE UNITE To mc Prl Sn WINDL tot 99 Este 1 adoles y mari de inc lo laq signifi alcohb vivieo incon: coinci de la : Cette tions adole sexe cigan des I concc les nl const l’utili rep01 l’util SIDDIQUI ET AL. D., and DAVIS, J. M. (1988). Random regresthe analysis of longitudinal psychiatric data. le Validity of Self-Reported Drug Use: Improv- esearch Monograph 167). Rockville, MD: Nafor Mixed-Effects Ordinal Probit and Logistic revention Research Center, School of Public lom-effects ordinal regression model for mulUnbalanced repeated-measures models with .05-820. ‘ects models for longitudinal data. Biometrics for maximum likelihood estimation in unbalBiomelrika 74: 8 1 l-827. e, WA: University of Washington Press. Methodological issues in drug usage surveys: :I. 8: 683-689. f self-reported cocaine use in two high risk .), The Validity of Self-Reported Drug Use: )A Research Monograph 167, pp. 227-246). sing the validity of self-reported adolescent porting of substance use in a national longiffects. Public Opin. Q. 52: 1 O&1 24. pact of changing methods of data collection scents. Am. J. Drug Alcohol Abuse IS: 275IOCHHAUSER, M. (1983). Reliability and mily-based study: A methodological report. ‘I, L. D. (I 983). Reliability and consistency 324. s in a longitudinal study. &and. J. Psycho/. . ng in a longitudinal perspective. J. Youth actors affecting attrition in a longitudinal (, A. W. (1996). Predictors of misreporting rl. Rev. 20(5): 552-568. The reliability and validity of drug use re: Issues 5: 426-443. SON, I-., BRANNON, B. R., JOHNSON, smoking prevention curriculum delivery. INCONSISTENT RESPONSE 293 SUSSMAN, S., DENT, C. W., STACY, A. W., BUCIAGA, C., RAYNOR, A., TURNER, G. E., CHARLIN, V., CRAIG, S., HANSEN, W. B., BURTON, D., and FLAY, B. R. (1990). Peer group association and adolescent tobacco use. J. Abnorm. Psychol. 99: 349-352. UNITED STATES DEPARTMENT OF HEALTH AND HUMAN SERVICES (I 994). Preventing Tobacco Use among Young People. A Report ofthe Surgeon General. Atlanta, GA: US Department of Health and Human Services, Public Health Service, Centers for Disease Control and Prevention, National Center for Chronic Disease Prevention and Health Promotion, Office on Smoking and Health. WINDLE, M. A. (I 990). Longitudinal study of antisocial behaviors in early adolescence as predictors of late adolescent substance use: Gender and ethnic group differences. J. Abnorm. Psychol. 99: S&91. RESUMEN Este estudio examina 10s “predictores” de respuestas inconsistentes de adolescentes a preguntas acerca de si algun vez han probado alcohol, cigarrillos, y marijuana. Adolescentes masculiones a tuvierren significamente altos niveles de inconsistencia en respuestas a acerca de1 alcohol, cigarillos, y marijuana. A lo largo que las adolescentes femeninas. Negros y Hispanos (adolesc.) tuveron significamente altos niveles de respuestas inconsistentes acerca de haber usado alcohol y cigarillos (solo negros) que adolescentes blancos. El “estatus de vivienda” y hazarias academicas jueron predictores signiticantes de respuestas inconsistentes acerca de haber usado marijuana. En si, (asique) 10s resultados coincidan con la notion que respuestas inconsistentes puedan “bias” la estimation de la prevalencia de haber usado drogas en un analyses multivariados. RliSUMl! Cette etude examine les “predicteurs” des responses des adolescents aux questions au sujet de l’utilisation de l’alcool des cigarettes, et de la marijuana. Les adolescents du sexe masculin avaient des taux pies eleves que les adolescents du sexe feminin concernant l’utilisation pendant un certain temps de l’alcool, eds cigaretes et de la marijuana. Les adolescents noirs et hispaniques avaient des taux des reponses inconsistantes considerablement plus eleves que les blancs concernant si ils avaient jamais utilises de l’alcool et des cigarettes (seulment pour les noirs). Les conditions de vie de ces sujets et leur reussities academiques constituaient des signes “predicteurs” des reponses inconsistant au sujet de l’utilisation de la marijuana. Ainsi ces resultats sont d’accord avec l’idee que les reponses inconsistantes peuvent influencer l’opinion de la frequence de l’utilisation constant de la drogue dans les analyses “multivariates.” P
© Copyright 2026 Paperzz