Risk and Protective Factors for Adolescent Drug Use:
Findings from the 1999 National Household Survey on Drug Abuse

Appendix B: Discussion of Missing Values for School Domain Factors

As mentioned in Chapter 2, the 1999 National Household Survey on Drug Abuse (NHSDA) computer-assisted interviewing (CAI) data contained a large number of missing values for questions related to youths' experiences and beliefs about school. The 1999 NHSDA CAI questionnaire was completed by 25,357 youths aged 12 to 17, and the questions regarding school had between 6,000 and 7,000 missing values, representing approximately 25 percent of the youth sample (see Table 2.4 for sample sizes for the school factors). The principal reason for these missing data are that only 19,306 of the 25,357 youths—or 76.1 percent—answered "yes" to the following survey question: "Have you been enrolled in any type of school at any time during the past 12 months." Although some percentage of these youths who did not answer "yes" did so because they truly were not enrolled in school during the past 12 months, a comparison with other national estimates suggests that the 1999 NHSDA CAI estimate of the percentage of youths enrolled in school is too small. The Current Population Survey (CPS) report published by the U.S. Bureau of the Census estimated that 96.9 percent of youths aged 12 to 17 were enrolled in public or private schools in October 1999 (U.S. Department of Commerce, 1999). It is possible that some of the missing values could be due to youths who attended home schools. However, the 1999 National Household Education Survey estimated that only 1.7 percent of youths under the age of 17 attended home schools (Bielick, Chandler, & Broughman, 2001).

Given these other national estimates, it is likely that many youths who did not answer "yes" to the school enrollment questions in the 1999 NHSDA CAI questionnaire had in fact been enrolled in school during the past 12 months. One possible cause for these errors is that some youths (especially younger ones) might not have understood what it meant to be "enrolled in school." A second possibility is that some youths who were interviewed during summer vacation may have answered "no" to the enrollment question because they were not currently attending school. These errors, in combination with "skip patterns" that were programmed into the computer- assisted interview, are the principal reasons for the large number of missing values for the school domain factors. CAI techniques allow for the programming of skip patterns, in which questions judged to be irrelevant for a respondent based on his or her answers to previous questions are not presented to the respondent. For example, if a respondent has previously identified himself as a male, the computer-assisted interview could be programmed to skip subsequent questions related to experiences with pregnancy for that respondent. For the 1999 NHSDA CAI data, the computer-assisted interview was programmed so that only youths who answered "yes" to the question about school enrollment were presented with the other questions related to school. Because of this, only the 76.1 percent of youths who answered "yes" to the school enrollment question provided data for school domain questions.

Due to these issues, any analyses that include items from the school domain are only possible for the subset of youths who correctly answered the question about school enrollment. The effect of this can be seen in the smaller sample sizes for the school domain factors listed in Table 2.4 compared with the sample sizes listed for the other domains in Tables 2.1 through 2.3 (see Chapter 2).³² This also affected the sample size of the final prediction models presented in Chapter 4. In those models, an observation that had a missing value for any of the covariates was dropped from the analyses. The inclusion of items from the school domain resulted in a reduction of the sample size for these models. For example, the final model for past year marijuana use included 16,411 of the 25,357 youths who completed the 1999 NHSDA CAI questionnaire (see Table 4.6 in Chapter 4).

If the responses of the subset of youths who answered "yes" to the school enrollment question differed from the responses for the full sample of youths, any results relating to the school variables could be biased. To determine whether the estimates of school-related questions for the responding subsample of youths were biased, some of the analyses involving the school domain factors were repeated using an adjusted set of sample weights. These adjusted weights were based on the population demographic characteristics of youths by age, race/ethnicity, and gender.

Table B.1 presents the descriptive statistics for each risk and protective factor in the school domain, using the adjusted sample weights. These results were very consistent with the descriptive statistics computed using the original sample weights (see Table 2.4 in Chapter 2). None of the mean values for the continuous predictors differed by more than 0.04, and none of the percentages for the categorical factors differed by more than 0.04 percent. Table B.2 presents the means for continuous factors or percentages for categorical factors, using the adjusted sample weights, by race/ethnicity, gender, and age group. Again, the results were very similar to the means or percentages computed using the original sample weights (see Table 2.8 in Chapter 2).

Table B.3 presents the simple odds ratios (ORs) with past year marijuana use for each of the school domain factors. These ORs are very similar to those that were computed using the original sample weights (see Table 3.5 in Chapter 3). None of the ORs differed by more than 0.09. Table B.4 presents these simple ORs with past year marijuana use using the adjusted sample weights, by race/ethnicity, gender, and age group. These ORs were also very similar to those computed using the original sample weights (see Table 3.10 in Chapter 3). Finally, Table B.5 presents the ORs for past year marijuana based on the adjusted sample weights, after adjusting for the set of demographic variables. Again, the results were very similar to the ORs computed using the unadjusted sample weights (see Table 3.14 in Chapter 3).

Because the comparison of estimates using the school domain factors indicate that using the adjusted sample weights resulted in minimal differences in the estimated results of school domain risk and protective factors, the analyses presented in this report using the original sample weights are considered to be unbiased.

Table B.1 Means or Percentages, Standard Deviations, and Quartiles of Youths Aged 12 to 17 Reporting School Domain Risk and Protective Factors, Using Adjusted Sample Weights: 1999

¹ Specific questions and distributions for school domain constructs are presented in Tables A.4 and A.8 (see Appendix A).
² Means are given for continuous variables, and percentages are given for categorical variables (marked with a percent sign).
³ Cronbach's alpha is a measure of the internal consistency of the individual items used to create multiple-item scales and is a function of the average intercorrelation between the items as well as the number of items. Cronbach's alpha values range from 0 (no correlation between items) and 1 (perfect correlation between items).
⁴ Response options for overall feelings toward school were 1 = Hated going to school, 2 = Didn't like going to school, 3 = Kind of liked going to school, and 4 = Liked going to school. Response options for belief about the meaningfulness and importance of school work were 1 = Never, 2 = Seldom, 3 = Sometimes, and 4 = Always. Response options for importance of school work to later life were 1 = Very unimportant, 2 = Somewhat unimportant, 3 = Somewhat important, and 4 = Very important. Response options for interest in courses at school were 1 = Very boring, 2 = Somewhat boring, 3 = Somewhat interesting, and 4 = Very interesting.
⁵ Multiple substance scales take the mean of responses for marijuana, cigarette, and alcohol use.
⁶ Response options were 1 = Strongly disapprove, 2 = Somewhat disapprove, and 3 = Neither approve nor disapprove.
⁷ Response options were 1 = None, 2 = Some, 3 = Most, and 4 = All.
Source: SAMHSA, Office of Applied Studies, National Household Survey on Drug Abuse, 1999.

Table B.2 Means or Percentages and Standard Deviations of School Domain Risk and Protective Factors among Youths Aged 12 to 17, by Race/Ethnicity, Gender, and Age, with Adjusted Weights: 1999

Note: Care should be taken in interpreting statistically significant differences in this table. With a large sample sizes, very small differences between groups can reach statistical significance.

¹ Means are given for continuous variables, and percentages are given for categorical variables.
² Includes those other than whites, blacks, and Hispanics (i.e., Asians, American Indians or Alaska Natives, Native Hawaiians or other Pacific Islanders).
³ p-value derived from F-tests for continuous variables and chi-square tests (df = 3) for dichotomous variables.
⁴ p-values derived from t-tests for continuous variables and chi-square tests (df = 1) for dichotomous variables.
⁵ Multiple substance scales take the mean of responses for marijuana, cigarette, and alcohol use.

Table B.3 Unadjusted Odds Ratios and Confidence Intervals (95 Percent) of School Domain Risk and Protective Factors and Past Year Use of Marijuana among Youths Aged 12 to 17, Using Adjusted Sample Weights: 1999

OR = odds ratio; CI = confidence interval.

¹ The questions used to measure each of the factors are provided in Appendix A (Table A.4). The coding and distribution of the responses for each factor are provided in Table 2.4.
² ORs are based on separate logistic regression models of past year marijuana use for each of the factors. ORs have not been adjusted for demographic differences. ORs > 1.0 indicate that the odds of past year marijuana use increased with each unit increase in the predictor. ORs < 1.0 indicate that the odds of past year marijuana use decreased with each unit increase in the predictor. For risk factors, each unit increase in the predictor generally indicates an increased risk of marijuana use. For protective factors, each unit increase in the predictor generally indicates a higher level of protection against marijuana use. An OR of 4.78 for the perceived prevalence of marijuana risk factor indicates that the odds of past year marijuana use increased 4.78 times with each unit increase in the perceived prevalence of marijuana question.
³ Multiple substance scales take the mean of responses for marijuana, cigarettes, and alcohol.

Table B.4 Unadjusted Odds Ratios and Confidence Intervals (95 Percent) of School Domain Risk and Protective Factors and Past Year Use of Marijuana among Youths Aged 12 to 17, Using Adjusted Sample Weights, by Race/Ethnicity and Gender: 1999

OR = odds ratio; CI = confidence interval.

¹ Includes those other than whites, blacks, and Hispanics (i.e., Asians, American Indians or Alaska Natives, Native Hawaiians or other Pacific Islanders).
² The questions used to measure each of the factors are provided in Appendix A (Table A.4). The coding and distribution of the responses for each factor are provided in Table 2.4.
³ ORs are based on separate logistic regression models of past year marijuana use for each of the factors, run separately for each of the categories of race/ethnicity and gender. ORs have not been adjusted for demographic differences. ORs > 1.0 indicate that the odds of past year marijuana use increased with each unit increase in the predictor. For risk factors, each unit increase in the predictor generally indicates an increased risk of marijuana use. For protective factors, each unit increase in the predictor generally indicates a higher level of protection against marijuana use.
⁴ Multiple substance scales take the mean of responses for marijuana, cigarette, and alcohol use.

Table B.5 Adjusted Odds Ratios (Controlling for Demographics) and Confidence Intervals (95 Percent) of School Domain Risk and Protective Factors and Past Year Use of Marijuana among Youths Aged 12 to 17, Using Adjusted Sample Weights: 1999

OR = odds ratio; CI = confidence interval.

¹ The questions used to measure each of the factors are provided in Appendix A (Table A.4). The coding and distribution of the responses for each factor are provided in Table 2.4.
² ORs are derived from multiple logistic regression models adjusted for age, gender, race/ethnicity, number of parents in home, household income, county type, and geographic region. ORs are based on separate logistic regression models of past year marijuana use for each of the factors. ORs > 1.0 indicate that the odds of past year marijuana use increased with each unit increase in the predictor. For risk factors, each unit increase in the predictor generally indicates an increased risk of marijuana use. For protective factors, each unit increase in the predictor generally indicates a higher level of protection against marijuana use.
³ Multiple substance scales take the mean of responses for marijuana, cigarettes, and alcohol.

This page was last updated on July 17, 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.