John R. Schuerman

Factors associated with an increased impact of child sexual abuse

Data is presented identifying factors associated with the impact of sexual abuse on children. A group of 369 sexually abused children and a comparison group of 318 children recruited from the community were compared on a parent-completed behavior rating scale. Data describing the abused children were also available from a 38-item symptom checklist completed by the childs social worker. Using a score based on the symptom checklist as the measure of the impact of sexual abuse, 15 variables were in the final regression equation explaining 42% of the variance in impact. Using a score based on parent-generated data, 5 variables were in the final equation explaining 20% of the variance. The significance of the variables in identifying factors associated with an increased impact of abuse is discussed.

The Effects of Sexual Abuse on Children A Multidimensional View

Data were collected at the Sexual Assault Center in Seattle on 369 sexually abused children and a community comparison sample of 318 not-abused children. Data describing the behavior of these children were collected from the childs parent and for the abused children from the social worker. Samples differed on a number of variables and these variables were used as control variables in the analysis. Results indicate that abused and not-abused children appear behaviorally different on a set of factors and clinical dimensions constructed from the parent-completed measure. Suggestions for future research are provided.

Putting families first: An experiment in family preservation

This volume is a comprehensive evaluation of the largest randomized experiment to date on placement prevention programs: the Family First program in Illinois. It offers insights into the tensions between policies advocating family preservation and those favoring out-of-home placement. The authors conclude by suggesting that placement prevention programs are but one component in a comprehensive effort to reform the child welfare system, and that those efforts should include both improvements in the foster care system and more refined decision-making in individual cases.

Multiple Regression II

In the social sciences we often have a large number of variables that we think might have some importance in explaining a dependent variable. We could compute a regression equation involving all of the variables (assuming that we have more cases than variables, which is not always the case). However that course of action is problematic for at least two reasons: 1. The use of a large number of variables will spuriously inflate the R2 as described in chapter 3. 2. Using large numbers of variables violates the principal of parsimony, which says that we should keep our explanations as simple as possible.

Understanding Decisions about Child Maltreatment

To understand how decisions are made in abuse/neglect cases by the child welfare system, the authors asked child welfare experts and protective service line workers to make decisions about actual child abuse and neglect cases on the basis of written summaries of the cases. Respondents included 27 experts and 103 line workers. Regression analyses found that workers and experts emphasized the same case characteristics in making their decisions, but the decisions were not well structured in the sense that they were not well predicted by case characteristics. Individual experts and workers varied widely in the decisions they made on identical cases. The authors conclude that decision making in the child protective system is inconsistent, with errors of two kinds: failing to remove children from their families when that is called for and removing children when it is unnecessary. Progress must be made in developing decision-making criteria that are consistent, preserve family integrity, and promote the well-being of children.

Decisions on placement and family preservation. Agreement and targeting.

Vignettes presented to experts and workers in the child welfare field were used to explore the degree of agreement on decisions to place children in substitute care or to refer them to family preservation services. The design allowed for investigation of the problem of targeting in family preservation programs. Findings indicate considerable inconsistency in decisions among experts and workers, particularly in decisions to refer to family preservation and other in-home services. Contrary to the stated intentions of family preservation services, a majority of the referrals to these programs do not involve children who would have been placed in the absence of these programs.

Principal Components Analysis

We study principal components analysis for two reasons, first because the ideas in it are central to much of multivariate analysis and secondly as a data reduction technique. By the latter we mean that we start with a relatively large number of variables and hope to wind up with a smaller number of variables that include most of what was in the larger set of variables. It is sometimes said that principal components analysis and factor analysis reveal the structure of a set of observations.

Repeated Measures Analysis

Repeated measures analysis is concerned with a common problem in social science research and evaluation. Often we have a situation in which we have measured the same thing at two or more points in time (called occasions), all of the measures having been taken on the same sample. Usually we assume that there are equal time periods between the occasions. For example, in the study of research teaching discussed earlier in this book we measured several outcome variables (knowledge of research, attitudes toward empirically based practice, etc.) at three points in time, the beginning of the school year, the end of the year, and again a year later. In chapter 8 we discussed the simplest kind of repeated measures analysis, one involving only two points in time and we used t tests and multivariate extensions of the t test to test for the differences in the two points in time. Now we will extend the analysis to more points in time.

Multivariate Tests of Means

Investigators in the human services almost always have more than one outcome variable. Frequently in such studies the researcher simply performs separate statistical tests on all of the dependent variables. There are, however, serious drawbacks to such a procedure. The first is that the overall probability of Type I error will be larger than the decision level set for each separate test. To put this simply, if we test lots of hypotheses separately at the .05 level, there is a high probability that at least one of them will be rejected “just by chance.” Thus, it is desirable that some way be found to test all of the hypotheses simultaneously. We thus come to one of the core motivations for multivariate analysis: the problem of multiple dependent variables. It turns out that the procedure for the multivariate test of means closely parallels that for single variables. So we begin with a review of the test for single variables.

Other Multivariate Techniques

In chapters 3 and 4 we considered multiple regression, a common technique for exploring the relationship between a number of independent variables and a single dependent variable. In this chapter we consider two procedures to use when we have several dependent variables and several independent variables (all of which are equal interval): multivariate multiple regression and canonical correlation. The latter will be dealt with only briefly.