The l998 Maryland Adolescent Survey was administered to samples of sixth, eighth, tenth, and twelfth graders in public middle and high schools in every school system in Maryland. Certain special schools, such as home and hospital schools, non-home vocational schools, and special education centers were not included in the study. The schools excluded from this sampling frame are listed in Appendix A.
To ensure a statistically generalizable result for each grade at the school system level, the study used a multistage stratified cluster sample. First, the required sample size for each local school system was determined based on the system enrollments in each grade and the desired level of measurement precision (p<.05). Second, the number of classes to be surveyed within each school system was calculated using class sizes from the previous survey as a surrogate for the class sizes for the survey administration date in December, l998. Finally, the number of schools required in the sample was dictated by the number of classes to be studied. Where possible, two classes per grade were selected in each school.
The sample was designed to ensure an equal probability of selection for every student at each grade level in each school system.
Selection of Schools
In large school systems, the schools were stratified using participation in the school lunch program and ethnic composition. The number of students to be sampled for each grade level were allocated to each stratum proportional to the number of students enrolled at schools in the stratum. Schools were selected with probabilities proportional to the enrollment at the sixth eighth and twelfth grades respectively. In schools where the twelfth grade was selected, the tenth grade was also selected. After a school was selected, two classes per grade were randomly drawn from each school.
In medium size school systems, stratification was unnecessary because the majority of schools were included in the sample. In these school systems, the opportunity for each schools selection was again proportional to its enrollment for each grade, exactly as was done for each stratum in the large systems. In small school systems, all schools were sampled, but the number of classes was adjusted to assign each student in the county the same probability of selection.
Prince Georges County was oversampled, as a large proportion of non-responding schools was expected based on prior survey administrations.
Selection of Classes
In most cases, the desired two classes per school were selected for participation. However, in order to meet sample size requirements, there were instances where more than two classes were sampled within a school as well as instances where either only one class was available to be surveyed or where only one class from a school was needed.
Classes were designated as eligible for selection according to the criterion that all students in the school within the survey grades were enrolled in them. In grades six and eight, these were English classes. In grades ten and twelve, some were English classes and the remaining classes sampled were drawn from within specified time blocks. Each school provided a list of classes within the specific time block (e.g., between 1:15 p.m. and 2:00 p.m.) during which all survey eligible students were enrolled in one or another class. The classes sampled were drawn from the lists of classes within the designated time block.
Within each selected class, every survey-eligible sixth, eighth, tenth, and twelfth grade student was asked to complete a questionnaire. Other enrollees were excused from participation. Table 2.1 indicates that between 12% and 21% of enrolled students in each of the grades were in the initial sample statewide and between 9% and 11% were in the final sample.
Weighting of Responses
In accordance with the sampling procedures, each school system was selected as a stratum. Within each school system, schools were selected for participation; within each school, classes were designated; within each class, all eligible students were requested to participate. This approach led to equal probabilities of selection for each student at each grade level. However, in order to control for differential participation rates, responses were weighted to account for the race/ethnicity and gender totals in each school system. Thus the weights were adjusted so as to add up to each total.
For students with missing race/ethnicity or gender, the missing category was imputed using a hotdeck approach. A student from the same school and grade was randomly selected and the ethnicity or gender category of the randomly selected student was assigned to the student with a missing value, for weighting purposes only.
As part of this process it was noticed that American Indians were over-represented. In some instances their number exceeded the number of American Indians enrolled at that grade in the given school system. For this reason the students who replied American Indian were treated as if they had a missing response, and the state enrollment figures were adjusted proportionately to distribute American Indian counts among the other ethnic groupings. Again this was done only for purposes of weight adjustments.
Survey Return Rates
A total of 36,847 questionnaires were sent to schools for completion by the designated sample of their students. Of these, 22,140 were returned and analyzed. Table 2.2 shows the number of survey forms sent out and returned for each participating school system. Return rates varied from a high of 84% to a low of 15%. With the exceptions of Prince Georges County (overall) and Kent County (twelfth grade), participation rates were sufficient to permit generalization of responses at the school system level for each grade.
As indicated in Table 2.3 below, the proportions of males and females who participated in the study from each of the four grades surveyed reflect those enrolled in these grades in the state as a whole. The proportions of respondents from each of the categories of race/ethnicity on which data were collected (Table 2.4) also reflect the proportion of the students enrolled in each of the grades studied. Table 2.4, however, suggests a small degree of over sampling of white students and a corresponding under representation of African American students. This slight imbalance is rectified in the data analysis when responses are weighted. The American Indian population is also over represented in the sample. It was hypothesized that some non-Indian students may have incorrectly read the option "American Indian" as "American". Therefore, data on American Indians are not used in the weight adjustments.
The survey consisted of three questionnaire forms. Form One was designed for administration to sixth graders, Form Two for eighth and tenth graders, and Form Three for twelfth graders. All three forms included sections on students background characteristics, drug knowledge, attitudes, and use patterns; family relationships; and drug availability. In addition, students completing Forms Two and Three were asked about any negative effects they had experienced from substance use; parental and peer approval of substance use; and estimates of degrees of risk associated with substance use. In 1998, students completing Forms Two and Three answered questions about safetyat school, going to and from school, and in their neighborhood. Twelfth graders completing Form Three were asked additional questions about alcohol, drugs, and driving. Form Three is included in Appendix B.
The content of the alcohol, tobacco, and other drug use items in the 1998 MAS questionnaire was similar to the 1996 content. However, there were two additions/revisions to the 1998 questionnaire.
As in previous administrations, the 1998 questionnaire was pretested in four classrooms and reviewed by staff in the Department of Transportation; the Department of Health and Mental Hygiene; the Division of Planning, Results, and Information Management; the Division of Student and School Services; and the Maryland Adolescent Survey contacts in each school system. Although the questionnaire changed little from 1996, Forms One and Two were pretested in two sixth and two eighth grade classes in two middle schools in Baltimore County. As in the 1996 pretest, particular attention was paid to the readability of items in order to determine if younger children could understand the terminology and language of substance use. Of the four classes used for the pretest, one in each grade was characterized by school administrators as reading below grade level. The data derived from the pretest indicated that sixth and eighth graders had no difficulty understanding the questions. Accordingly, no changes were made in the questionnaire as a result of the pretest.
In each participating school, questionnaires were administered in the classes that were identified by sampling procedures. One county (Prince Georges) required students to receive parental consent to complete the questionnaire. The 1998 survey was conducted in all school systems on December 8, 1998.
Questionnaire packets were distributed to each participating school principal with instructions as to which classes were selected for the survey. Principals distributed the envelopes, which contained forms and administration instructions, to the designated survey administrator (teachers or others) for each class (Appendix C).
In most cases, teachers administered the questionnaire. In a few instances, questionnaires were administered by other school personnel. Administration procedures were pilot tested with one classroom teacher in order to assure that the administration instructions were clear and requirements were feasible. Survey administrators were responsible for requesting student participation, distributing forms, delivering instructions, and returning the collected questionnaires. In addition, they were instructed to assure students of the voluntary nature of their participation and the confidentiality of their responses.
In each classroom, the questionnaires were collected from the students by one of the participants, placed in a large envelope, and sealed in order to assure respondents confidentiality and protect their privacy. School principals were instructed to return all survey forms (completed and blank) to a designated survey repository site.
GENERALIZING THE SURVEY RESULTS
As described earlier in this chapter, the survey sample allows generalization of responses at the school system level for each surveyed grade (with the exception of Prince Georges and Kent County). Johnston, OMalley, and Bachman (1999), in their report of the national survey results on drug use from the Monitoring the Future Study, found that survey results, such as those from the MAS, represent an accurate estimate of drug use, despite the fact that the estimates rely on self-reported measures of drug use. They believe there is a "high level of validity" in the measures obtained.
Johnston, OMalley, and Bachman also discuss whether the twelfth grade findings can be generalized to "dropouts" (students who do not finish high school). While many have hypothesized that dropouts use drugs more than students who stay in school, these researchers for the Monitoring the Future study found that the increased use by dropouts theory does not always hold true. They conclude however, that until such time as good trend data are gathered directly from dropouts, estimates on incidence and prevalence of drug use among the school aged population are limited to students who are in school and participate in the survey. The MAS results, therefore, are only generalizable to those students who are in school.