The Strengthening Washington D.C. Families Project: A Randomized Effectiveness Trial of Family-Based Prevention

Over time, this site began to recruit families from the surrounding neighborhood as well as the center residents.

Data were not collected on the families who did not enroll. It is therefore not possible to determine how the recruitment process influenced the characteristics of the sample.

The program was designed to be delivered once per week over a 14-week period. In SWFP, 16 of the 35 classes were run using twice per week sessions over a 7-week period. This local adaptation was approved by the program developer.

Each FT condition included a child and a parent training class as well as a family training class, so the total number of each parent and child training class delivered was 35 (17 plus 18).

Two follow-up surveys were also conducted at 6 and 18 months following the completion of the intervention. Data from these surveys were examined using latent growth curve analyses to assess the extent to which any of the four experimental conditions resulted in longer term changes in the same outcome measures examined in this report. The results of this exploration are reported in a separate report (Wilson, 2004). No long-term effects of any of the experimental conditions were found. Because Wilson's report uses a different method of analysis and a different sample (e.g., only those families include in the follow-up surveys), page limitations do not allow us to include the details of the analysis in this report. It will be submitted for publication separately.

Items measuring parental substance use, child lifetime use, and parental depression were also included in the surveys. The variability in parent and child self-reports of substance use was low, and these measures did not converge with other scales that should have been related. In addition, several youths reported a decrease in lifetime drug use from pre to post, possibly indicating low validity in these measures. These scales and items were not analyzed for this report.

Several of the original scales were not normally distributed and their response formats differed from other scales with which they were to be combined. To resolve both problems, square root transformations were taken for the non-normally distributed scales prior to combining them with the other scales. Specifically, to create the family supervision and bonding factor, the scale representing closeness to parents (which was skewed and had a range of 1–4) was transformed by taking the square root in order to average it with the parental supervision scale (which had a range of 1–2). To create the parenting skills factor, confidence in parenting and consistency in discipline (both of which were measured on 1–5 scales and were skewed) were also transformed by taking the square root before combining them with the other parenting skills scales that had 1–2 ranges. For these two scales, a small number of cases (2–4) were also trimmed to pull the small number of square root transformed scores that were greater than 2 into the required 1–2 range. After these original scales were combined into higher order scales, the distributions of the resulting scales were again checked. The distributions of these higher order scales were closer to normal than were the original scales. Nevertheless, for the three higher order scales whose distributions contained more than a minimal amount of skew (rebellious behavior, family supervision and bonding, and parenting skills), analyses were conducted using both untransformed and transformed variables. These analyses produced essentially the same results, so the untransformed results are presented.

A manual (Cowan & Gottfredson, 2000) containing the wording of all items, their response formats, their sources, and instructions for recoding items to obtain a common metric among items in each original scale is available from the first author.

Child and parent reports of problem behavior did not load on a single factor, as anticipated. The correlations between child and parent reports of similar behaviors were modest. The correlation between child reports of rebellious behavior and parent reports of antisocial behavior, for example, was .25. An extensive analysis of the validity of the child and parent reports utilizing measures available from school records for a subset of the sample indicated that neither could be rejected as obviously invalid. Correlations of child reports of child characteristics generally had slightly higher correlations with measures of school performance taken from school records than parent reports of child characteristics, but both were correlated in the expected direction. Therefore, the two sets of measures of problem behavior were retained for analysis but were kept as separate factors.

The numbers of observations differ by type of session because each observation of a family skills training condition generated three distinct class observations—one for each of the child, parent, and family training sessions. The observations of classes in the CT and PT skills training conditions generated only one observation each.

Each family was coded as having attended at least one session if any family member attended any of the 14 sessions. The total sessions attended for each family was the number of sessions attended by any family member.

Trainer turnover generally occurred between classes so that minimal disruption of ongoing classes took place.

That is, 70% of the parents who completed a pre-test also completed a post-test.

These parent baseline measures included all five of the parenting skills scales used in this study; parent reports of their target child's antisocial behavior and school progress; parent reports of their own alcohol problems and illicit drug use; race; age; income; and number of children.

Regression would be expected given the high-risk nature of the population and participant self-selection for child problem behavior. As a further check, we examined gain scores for participants at different ranges of the pre-test distributions. This analysis revealed that families who tested at the more negative end of the distribution at the time of the pre-test improved the most between pre- and post-test. The association of gain score with pre-test was highly significant and did not differ by experimental condition.

In analyses that examined each of the original scales separately, the F-test for parent reports of social skills reached statistical significance (F = 5.62; p = .05).

That the content of the program is less important than other aspects of the program, such as the opportunity it affords to make social connections, is underscored by the fact that the parents assigned to the MT and CT groups were able to answer 76% of the content items correctly without benefit of the parent sessions. Knowledge of the parenting skills covered in the training session is probably not the most important active ingredient. This finding corresponds to findings from drug prevention research indicating that increasing knowledge about substances and their effects does not reduce substance use (Botvin, 1990).