Technical Report #1

Creation of the Data Set for Analysis

The basic data set used for analysis in this report has been the product of a series of critical decisions made and processes completed over the course of four years. Each step and decision was based on careful review of the data themselves as well as other available information, consideration of alternatives and implications, and often consultation with key advisors within ACYF and the National Transition Demonstration Study’s Technical Work Group.

Because the National Transition Demonstration Study incorporated data obtained from a variety of informants (children, families, teachers, principals, school records, national databases), the construction of the final data set is based on the concept of family units. Unique seven-digit family/child identification numbers were assigned to each child participating in the study, indicating: (10 the cohort (one of 2); (2) the site (one of 31); (3) Head Start affiliation (former Head Start or non-Head Start); (4) unique family affiliation (a 3-digit identifier); and (5) whether or not the child had a sibling in the study (a single digit identifier). All data that were associated with a given child (i.e., from the family interview, child assessments, child-specific teacher questionnaire, and information from school records) was linked to the child, on all relevant forms, via this family/child identification number. A family unit, then, was defined as a child/family unit (defined by the unique identification number) that had, in kindergarten (at baseline), some associated data -- either a family interview, a child assessment, a teacher questionnaire, or information from the school records. Any combination of the four data sources could exist for the family unit at any given time point throughout the four-year longitudinal follow-up period. In April 1995, at the midpoint of the study, a total of 12, 067 family units were involved in the National Transition Demonstration Study.

A series of data clarification processes were completed in the ensuing four years, each resulting in some modification to the number of observations within the data set. The steps of clarification and the consequences in terms of lost cases (totaling 1,238 cases) are outlined in Table 1 below. The deletion of one site from the analysis sample (because of failure to submit adequate documentation of random assignment of schools) led to the largest reduction in sample size (a total of 785 cases removed from the analysis sample). The additional steps together removed total of a 435 cases from the analysis sample. The final analysis set, then, includes 7,515 cases -- 90 percent of the original sample.

Of these 10,829 family units included in the final data analysis set, 7,515 were former Head Start participants and the remainder (3,314) did not participate in Head Start prior to enrolling in kindergarten. Thus, the final analysis set used as a basis for child and family analyses reported in this final report includes the 7,515 former Head Start children and families.

There was some variability in the completeness of data within and across rounds for a given family unit. For example, a family interview could be present but the child assessment and teacher questionnaire missing, or the family unit could have data available in kindergarten but not in the following year. Table 2 summarizes the availability of data forms of certain types across rounds of data collection, singly and in key combinations.