Parental Influences on Adolescent Marijuana Use and the Baby Boom Generation

CHAPTER 3: NHSDA METHODOLOGY

3.1 Sample Design

The NHSDA is a survey of drug use in the general population that was conducted every two or three years from 1971 to 1988, and has been conducted annually as of 1990. The target population is the civilian, non-institutionalized population of the United States 12 years of age and older.¹ Persons with no fixed address, residents of institutional quarters (such as jails and hospitals) and active military personnel are excluded. Oversamples have been consistently implemented for individuals in the ages of highest risk for drug use (ages 12-34), African-Americans and Hispanics since 1985, and six metropolitan statistical areas in 1991-1993. Current tobacco users were oversampled in the 1993-1995 surveys.

All descriptive analyses, including prevalence rates, cross-tabulations and univariate and multivariate logistic regression analyses were estimated by SUDAAN (Shah et al., 1992), which uses a Taylor series linearization technique to adjust for design effects. All analyses were based on weighted data. The significance of the differences between odds ratios was evaluated through the Wald test.

3.2 Sampling of Dyads

In a number of households, two respondents were selected for participation in each survey conducted since 1974. Before 1991, one of these respondents had to be a youth 12 to 17; in more recent surveys the age restriction was eliminated. This report is based on data beginning with the 1979 survey, since dyadic data from prior surveys are not available. In 1985 only one respondent was interviewed in each househould. Some information on the sampling of dyads is provided in the NHSDA public release codebooks (SAMHSA, 1992; 1994; 1995; 1996a; 1996b; 1997) and in the 1991-1996 sampling reports prepared by Research Triangle Institute (Allred et al., 1997; Jones et al., 1992; 1993; Jones and Folsom, 1993; 1994; 1995; 1996; Research Triangle Institute, 1997).

Through the 1996 survey, no consistent sampling design was applied to the selection of multiple respondents within households. The dyads were not a representative sample of parent-child pairs living in households in the U.S. and their selection was affected by the within-dwelling unit sampling procedures. These took into account the ethnic classification of the household head, the age group composition of the household and, in the 1993-1995 surveys, the cigarette smoking within the last month of 18-34 year old residents.

Prior to 1993, the population was divided into four age groups: 12-17, 18-25, 26-34 and 35+ years old. In 1992, the 35+ age group was partitioned into those 35-49 and 50+ years old. From 1993 to 1995, those aged 18-25 and 26-34 were classified as 18-34 smokers and nonsmokers, and sampled accordingly. The selection probabilities of particular age groups were based on the desired sample sizes for each age group by race/ethnicity. If an age group was selected for a particular dwelling and the dwelling contained more than one person in that age category, the interviewer would consult a table to determine the specific person to be interviewed. The probabilities of selecting a person within an age group were based on the sample size desired for the age group by race/ethnicity and the number of households expected to contain this age group by race/ethnicity. No more than two persons in one age group could be interviewed in a household. Before 1991, one of the two respondents had to be a youth aged 12-17. In more recent surveys, the age restriction was removed. In 1993-1996, one 12-17 year old and one adult were selected in the majority of cases. When two adult selections were made, most included one smoker and one nonsmoker in 1993-1995, and one 18-25 year old and one 26-34 year old in 1996.

3.3 NHSDA Parent-Child Dyads: Dyad Identification and Overview

Although NHSDA surveys were initiated in 1971, dyadic data were first collected in 1974 but data from the 1974, 1976 and 1977 surveys are not available for analysis. As noted above, parent-child dyads were not ascertained in 1985.

Two methods were used to identify parent-child dyads in the data sets. (1) For survey years 1979, 1982, 1988, 1990, 1991 and 1992, dyads were identified using two relational variables provided to us by Research Triangle Institute. One variable, Relationship of the DU Pair (DURELAT), identified the type of dwelling unit-pair (dyad). Four types were coded: parent-child, husband-wife, other/other, and not in a unit pair. After the parent-child dyads were identified, a second variable, Parent-Child Code (P-C), differentiated parent from child and was used to create the parent-child dyad file. (2) For survey years 1993, 1994, 1995 and 1996, parents and children were linked by us using information from the public use data files. Two individuals were classified as a parent-child pair if they lived in the same household and identified themselves as being a parent or a child of the other respondent. For all survey years (1979-1996), we imposed the criterion that the presumed parent was at least 12 years older than the child. In the 1991-1996 surveys, parent-child pairs with a child aged 12-25 were selected for analysis. We excluded those where the child was 26 years or older, since this represented an atypical living arrangement for this older age group. According to census data, from 1980 to 1996, only 10.4% to 12.3% of young adults aged 25-34 lived at home, and 52.6% to 54.4% of those aged 18-24 did so (U.S. Census Bureau, 1999). From 1989 to 1996, 94.9% to 96.4% of those aged 12 to 17 lived at home (U.S. Census Bureau, 1990-1991; 1992a, b; 1994; 1996a, b; 1998c)². On average 97.0% of 12-17 year olds lived with their mothers; 77.0% lived with their fathers (U.S. Census Bureau, 1998a; b). Excluded dyads with a child 26 years old and older represented 2.4% of all the identified parent-child dyads in the surveys. Since parental self-identification of biological status in relation to the child was only ascertained beginning in 1994, and biological status could not be determined in prior years, both biological and non-biological parent-child pairs were included in the 1994 to 1996 surveys. In these three years, biological offspring accounted for 97.2% of the children in the dyads.

The analyses in this report are based on all ten available waves of parent-child data (1979, 1982, 1988, 1990-1996), including those conducted prior to 1991 where children were interviewed as part of a dyad unit pair only if they were aged 12-17. The 1994 survey included two components, 1994A and 1994B (see Section 3.5). A total of 9,463 parent-child dyads were identified in the aggregated 1979, 1982, 1988 and 1990-1996 surveys. Of the children, 88.6% were 12 to 17 years old, 11.4% were 18 to 25 years old. Close to 21% of all 12-14 year old NHSDA respondents were included in a parent-child dyad, 15.6% of all 15-17 year olds, and 3.1% of all 18-25 year olds (Table A.3.1). The aggregate and year-specific sample sizes of parent-child dyads with children aged 12-25 are reported in Table 3.1 by sex of parent and child. The number of parent-child pairs is much larger in recent than in earlier waves, reflecting increases in the total sample size of the surveys.

3.4 Dyad Level Weights

All dyad level weights were constructed and provided by Research Triangle Institute. The year-specific dyad level weights were derived from the person-level sampling weights. Each dyad level weight was the sum of the parent and child year-specific person-level sampling weights, divided by the total number of individuals in the household aged 12 and older. This constructed weight was applied to both members of each dyad.

For the 1979, 1982, 1988, 1990-1996 aggregate sample, new dyad level weights were constructed by multiplying the year-specific dyad weight for each case by the year-specific total number of dyads, and dividing by a constant³. The constant was the sum of the year-specific weights weighted by the number of cases per year, divided by the total cases in the aggregate sample (Personal communications, Michael Witt, 1/30/98, 5/26/98)

3.5 Changes in Interview Format in 1994

The NHSDA schedule is a structured one-hour personal interview, in which the drug questions are answered by respondents in self-administered modules. Until 1994, questions pertaining to tobacco use were answered orally to the interviewer.

In 1994, the interview schedule was substantially revised and selected data processing procedures were modified as well. Two versions of the interview schedule (Forms A and B) were administered in a split sample design to assess the consequences of the changes. Form A retained all the core questions from prior surveys and was administered to one-fifth of the respondents.

Form B constituted the revised version and was administered to four-fifths of the sample. The revisions included changes in the wording of selected drug-related questions; the presentation of the tobacco module in a self-administered format rather than being administered by the interviewer as in prior surveys; the addition of questions on mental health and access to mental health care. [See SAMHSA (1996c) for a detailed presentation of the major changes in the interview schedule]. In addition, different imputation procedures were implemented with data from Form B to define an individual as a drug user in the presence of missing data on drug use reports. In particular, while respondents who answered positively to any question dealing with use in the last twelve months were classified as marijuana users through 1994A, as of 1994B, only answers to the recency of use question were used.

The prevalence estimates of substance use differ between Forms A and B. Estimates of illicit drug use are higher in Form A than Form B, whereas estimates of licit drug use are higher in Form B than Form A, especially for last month tobacco use reported by adolescents. The rates of lifetime and last year marijuana use among age groups are slightly higher in Form A than Form B, with selected exceptions for adults aged 26 and older. For example, the rates of lifetime marijuana use are 16.0% in Form A compared with 13.6% in Form B among respondents aged 12-17; 43.4% compared with 41.9% among those aged 18-25; 56.9% compared with 52.7% among those aged 26-34; and 28.4% compared with 25.4% among respondents aged 35 and older (Table A.3.2). A similar pattern was observed in self-reported marijuana use in the last year. Given these differences in self-reported use, survey year specific analyses for 1994 were conducted for Forms A and B separately.

3.6 Advantages and Disadvantages of the NHSDA

The NHSDA has both advantages and disadvantages for the study of parent-child association in marijuana use.

The advantages include: (1) the large number of parent-child dyads, in which independent data on drug behavior were obtained from each person; (2) the inclusion of mothers and fathers, albeit only one parent per family; (3) the availability of detailed drug use information on both members of the dyads; (4) the overrepresentation of minorities; and (5) the multiple waves of data available since the late 1970's. The NHSDA surveys include families in which parents were members of the baby boom generation (i.e., the 1946-1964 birth cohorts), families in which the parents were born before 1946, and a very small number who were born after 1964.

The NHSDA also has limitations: (1) The dyads are not a representative sample of parent-child pairs living in households in the United States. Reflecting the gender distributions of parents who participated in the NHSDA as a whole, there is an under-representation of father-child dyads compared with the distribution of fathers residing in households in the U.S. population (see Chapter 4, Section 4.2). Rates of marijuana use are lower among parents and children interviewed as members of dyads than as household members (see Chapter 5, Section 5.1.c). (2) The sampling selection procedures for the overall sample and for multiple respondents in a household varied across survey years, so that the resulting dyadic samples are not necessarily similar over time. (3) The data set includes very little information about factors other than drug use, especially individual characteristics of young people, which are important predictors of drug use. As a result, few variables could be used as controls to provide more accurate estimates of the influence of parental drug consumption on offspring use or as predictors of child marijuana use. (4) Finally, the data are cross-sectional. While we interpret the observed association between parent and child as reflecting the influence of the parent on the child, it is possible, although not likely, that the association reflects in part the influence of the child on the parent.

These limitations notwithstanding, the NHSDA provides an unusual opportunity to address important questions related to the extent of parental influence on children's marijuana use in different parental birth cohorts.

3.7 Measurement of Variables

The definitions of variables was complicated by the fact that the format of the questions, and the coding and imputation of the variables underlying the constructed variables changed over the course of the 10 surveys included in the analysis. Most of the changes took place in the 1988 and 1994B surveys.

The variable labels specified below are from the public use codebooks. Variable labels denoted with an "X" are those that we constructed from available public use variables. Unless otherwise indicated, variables were available in all the surveys included in this report. All imputed variables were constructed by SAMSHA. Details about the construction of the drug use variables and other selected variables appear in the Technical Appendix: Construction of Drug Use and Other Variables.

The differential availability of variables across the surveys determined the selection and grouping of different surveys for analyses based on these variables.

3.7.a Marijuana Use Variables

Two dimensions of the child's marijuana use were examined to assess variations in patterns of parental influence on the child lifetime marijuana experimentation and current use: (1) ever used marijuana; and (2) used in the past year. Six marijuana variables were examined for parents: (1) ever used marijuana; (2) used in the past year; (3) a combination of use lifetime and past year to identify former users; (4) frequency of use in the lifetime; (5) frequency of use in the past year; and (6) frequency of use in the past 30 days.

Parent and child lifetime marijuana use (MRJFLAG): 0=never used marijuana; 1=ever used.

Parent and child past year marijuana use (MRJYR and MRJYRX). For parents and children, a binary variable (MRJYR) was created: 0=no marijuana use in the past year; 1=used in the past year. For parents, a trichotomous variable (MRJYRX) was also created: 0=never used marijuana; 1=former use, not in the past year; 2=used in the past year.

Parent frequency of lifetime marijuana use (MJTOTX): 0=never used marijuana; 1=used 1-10 times/days lifetime; 2=used 11-99 times/days lifetime; 3=used 100 or more times lifetime.

Parent number of days used marijuana in the past 12 months (MJYRFRQX): 0=never used marijuana; 1=former use, not in the past 12 months; 2=used 1-200 days in the past 12 months; 3=used 201-300 days in the past 12 months. Available in 1988 and 1990-1996. Text refers to the variable as "in the past year."

Parent number of days used marijuana in the past 30 days (MJDAY3OX): 0=never used marijuana; 1=former use, not in the past 30 days; 2=used 1-10 days in the past 30 days; 3=used 11-30 days in the past 30 days. Text refers to the variable as "in the past month."

3.7.b Other Drug Use Variables: Cigarettes, Alcohol, Cocaine

Parent lifetime cigarette use (CIGFLAG): 0=never used cigarettes; 1=ever used.

Parent past year cigarette use (CIGYRX): 0=never used cigarettes; 1=former use, not in the past year; 2=used in the past year.

Parent smoked 100 or more cigarettes in lifetime (CIG5PKX): 0=never used cigarettes or did not use 100 or more cigarettes in lifetime; 1=used 100 or more cigarettes in lifetime.

Parent number of cigarettes per day in the past 30 days (CIGMFRQX): 0=never used cigarettes; 1=former use, not in the past 30 days; 2=1-15 cigarettes per day in the past 30 days; 3=16-35 cigarettes per day in the past 30 days; 4=36 or more cigarettes per day in the past 30 days. Text refers to the variable as "in past month."

Parent number of cigarettes smoked daily (PACKSX): 0=never used cigarettes; 1=smoked 1-5 cigarettes per day; 2=smoked 6-15 cigarettes per day; 3=smoked 16-25 cigarettes per day; 4=smoked 26-35 cigarettes per day; 5=smoked 36 or more cigarettes per day.

Parent lifetime alcohol use (ALCFLAG): 0=never used alcohol; 1=ever used.

Parent past year alcohol use (ALCYRX): 0=never used alcohol; 1=former use, not in the past year; 2=used in the past year.

Parent frequency of past year alcohol use (IRALCFQX). Eight categories. See Technical Appendix, Section A.1.b.

Parent number of times very drunk or high in the past 12 months (DRUNKYRX). Ten categories. See Technical Appendix, Section A.1.b.

Parent quantity/frequency of alcohol use in the past 30 days (ALCMFRQX): 0=never used alcohol; 1=former use, not in the past 30 days; 2=used less than 2 drinks per day in the past 30 days; 3=used 2 or more drinks per day in the past 30 days. Available in 1988 and 1990-1996.

Parent lifetime cocaine use (COCFLAG): 0=never used cocaine; 1=ever used.

Parent past year cocaine use (COCYRX): 0=never used cocaine; 1=former use, not in the past year; 2=used in the past year.

Parent frequency of lifetime cocaine use (COCTOTX1 and COCTOTX2). Two versions were constructed. A four category version (COCTOTX1): 0=never used cocaine; 1=used 1-10 times/days lifetime; 2=used 11-99 times/days lifetime; 3=used 100 or more times/days lifetime. An eight category version is described in the Technical Appendix, Section A.1.b.

Parent frequency of past year cocaine use (IRCOCFQX). Nine categories. See Technical Appendix, Section A.1.b.

3.7.c Sociodemographic Variables

Ten sociodemographic variables were examined. See Technical Appendix , Section A.2.a. for further detail.

Parent and child age (IRAGE). Imputed variable, coded in years. Children were classified into three age groups: 12-14, 15-17 and 18-25 years old. Parents were classified according to their children's ages. Age ranges of parents for each of the three children's age categories were 24-73 years; 27-80 years; and 31-74 years, respectively.

Parent and child sex (IRSEX). Imputed variable: 1=male; 2=female.

Parent race/ethnicity (IRACE). Imputed variable, four categories: 1=white; 2=African-American; 3=Hispanic; 4=other. Across surveys, ethnicity was based on self-reports for 95.0% of individuals and on interviewers' observations in the remaining cases. To simplify the analysis of dyadic similarity, a single variable based on parental ethnicity was used to characterize the ethnicity of parent and child. In 97.1% of dyads, both respondents were in the same category.

Parent and child birth cohort (PCOHRTX and CCOHRTX). Constructed variable. Nine groups of parental birth cohorts (PCOHRTX) were identified based on exposure to historical periods of the marijuana epidemic during ages 15-18, as described in Chapter 2: 1=born 1945 and earlier; 2=born 1946-1948; 3=born 1949-1953; 4=born 1954-1956; 5=born 1957-1959; 6=born 1960-1962; 7=born 1963-1964; 8=born 1965 and 1967; 9=born 1968 and after.

Children were classified into five groups of birth cohorts, each aggregating five birth years. (CCOHRTX). Constructed variable, five categories: 1=born before 1964; 2=born 1965-1969; 3=born 1970-1974; 4=born 1975-1979; 5=born 1980-1984.

Parent and child marital status (IRMARIT). Imputed variable, five categories: 1=married; 2=divorced/separated; 3=widowed; 4=never married; 5=legitimate skip for respondents aged 12-17 in 1979 and 1982, and aged 12-14 in 1988, 1990-1996.

Parent and child education (EDUCCT2X). Recoded variable, five categories: 1=less than high school/drop out; 2=high school graduate; 3=some college; 4=college graduate; 5=in secondary school.

Child school dropout (CDRPOUTX). Constructed variable. Based on variable EDUCCT2X: 0=not a dropout; 1=dropout.

Household income (FAMINC). Five categories of total family income: 1=$0-$8,999; 2=$9,000-$19,999; 3=$20,000-$39,999; 4=$40,000-$74,999; 5=$75,000⁺. Available in 1990-1996.

Region of the country (REGION). Four categories: 1=Northeast; 2=North Central; 3=South; 4=West.

Population density (PDEN). Three categories based on the 1990 Census: 1=segment in an MSA with 1 million or more persons; 2=segment in an MSA with fewer than 1 million persons; 3=segment not in an MSA. Available in 1990-1996.

3.7.d Personal Characteristic Variables

See Technical Appendix, Section A.2.b, for further detail.

Parent depression in past 12 months (MDE1): 0=probable non-case; 1=probable case.

Parent anxiety in past 12 months (GAD1): 0=probable non-case; 1=probable case.

Parent and child delinquency in past 12 months (DELQX). Scores ranged from 0-12. Additive index based on 12 delinquency items. Available in 1991-1994A and 1995.

Child behavioral problem in past six months (PR_BEHAV). Administered only to 12-17 year olds in 1994B-1996: 0=no behavioral problem; 1=behavioral problem. Recoded binary variable derived from 100 items from the Achenbach Youth Checklist (YSR). The Behavioral Problem Scale was the summation of the Delinquent and Aggressive Behavior Syndrome raw scores. For boys a score of 19 and for girls a score of 17 indicated a behavioral problem.

Child emotional problem in past six months (PR_EMOT). Administered only to 12-17 year olds in 1994B-1996: 0=no emotional problem; 1=emotional problem. Recoded binary variable derived from 100 items from the Achenbach Youth Checklist. The Emotional Problem Scale was the summation of the Withdrawn, Anxious/Depressed and Somatic Complaints Syndrome raw scores (minus overlap for the unhappy/sad/depressed item). For boys a score of 17 and for girls a score of 22 indicated an emotional problem.

Parent and child perceived risk of occasional and regular marijuana use (RSKMJOCX and RSKMJRGX). Recoded variables, three categories: 1=great risk; 2=moderate risk; 3=slight/no risk. Two questions: "How much do you think people risk harming themselves physically and in other ways when (1) they smoke marijuana occasionally (RSKMJOCC); and (2) smoke marijuana regularly (RSKMJREG)?" The original four categories included: 1=no risk; 2=slight risk; 3=moderate risk; 4=great risk. Categories 3 and 4 were combined. Three-category reverse coded versions were created. Available in 1988, 1991-1994 (A and B) and 1995.

			Parents		Children2			Dyad Types
								Father-		Father-		Mother-		Mother-
Survey Year		N	Father	Mother	Son	Daughter		Son	%	Daughter	%	Son	%	Daughter	%
1979		693	267	426	353	340		164	27.7	103	17.2	189	24.6	237	30.5
1982		371	141	230	197	174		76	23.0	65	19.4	121	29.3	109	28.3
1988		289	70	219	165	124		38	15.8	32	12.5	127	41.9	92	29.8
1990		185	55	130	82	103		24	20.2	31	14.5	58	28.4	72	36.9
Subtotal		1,538	533	1,005	797	741		302	23.4	231	16.5	495	29.4	510	30.6
1991		1,646	541	1,105	840	806		289	18.3	252	21.1	551	31.1	554	29.6
1992		1,869	575	1,294	914	955		293	20.9	282	17.2	621	29.2	673	32.8
1993		1,213	359	854	615	598		190	20.5	169	20.7	425	27.5	429	31.3
1994A		229	69	160	143	86		39	23.5	30	16.3	104	34.8	56	25.4
1994B		880	270	610	440	440		151	22.8	119	16.0	289	30.5	321	30.8
1995		949	270	679	489	460		153	25.3	117	15.6	336	27.3	343	31.8
1996		1,139	305	834	569	570		151	19.4	154	21.3	418	30.7	416	28.6
Subtotal		7,925	2,389	5,536	4,010	3,915		1,266	20.9	1,123	18.6	2,744	29.5	2,792	30.9
Grand Total		9,463	2,389	6,541	4,807	4,656		1,568	22.1	1,354	18.3	3,239	28.8	3,302	30.7

¹Weighted estimates, unweighted N's.
²In 1979, 1982, 1988 and 1990, children aged 12-17 were selected. In all other years, children aged 12-25 were selected.
Source: SAMHSA, Office of Applied Studies, National Household Survey on Drug Abuse.

Footnotes for Chapter 3

² Data for 1980 to 1988 are available for 10-17 year olds as a group. The range is 96.0% to 96.8% (U.S. Census Bureau, 1981; 1982; 1985; 1986; 1987; 1988; 1989).

³ For selected analyses, aggregate samples consisting of the 1991-1994A and 1994B-1996 surveys were used. Dyad level weights were also constructed for these aggregate samples.

This page was last updated on June 03, 2008.

SAMHSA, an agency in the Department of Health and Human Services, is the Federal Government's lead agency for improving the quality and availability of substance abuse prevention, addiction treatment, and mental health services in the United States.