Project Summary/Abstract
Most empirically supported treatments for conduct problems in children are implemented by parents, but a
recent meta-analysis found them to be half as effective now as 50 years ago. Basic parenting research has
failed to improve these treatments partly because of its inadequate causal validity. Of the two basic methods
for analyzing longitudinal change, ANCOVA-type residualized-score analyses have been preferred for causal
estimates, but they are biased by stable pre-existing differences on outcome scores, according to recent
critiques. But the alternative, difference-score analysis, has less statistical power and cannot easily test crucial
Pretest X Treatment interactions. This study intends to address those disadvantages by testing extensions of a
promising innovation called dual-centered ANCOVA to multi-occasion data. This is a modification of Huitema’s
quasi-ANCOVA that centers posttest scores (new) as well as pretest scores (quasi) around their pretest group
means. Centering pretests make the two change-score results consistent with each other, whereas centering
posttest scores around pretest group means then keeps all change scores unchanged. It avoids the bias in
standard ANCOVA by duplicating the results of difference-score analyses, but retains ANCOVA’s advantages:
more statistical power and ability to test Pretest X Treatment interactions.
For “true” causal effects to compare for testing bias, this research simulates two 3-occasion datasets to fit the
null hypotheses of each traditional change-score analysis. It also tests longitudinal data on corrective actions
by parents or professionals shown to be effective in randomized trials. Depression treatments (medications
and therapy), child hospital visits, and timeout will each be tested in two longitudinal datasets. Research
questions include: (1) Do analyses across three occasions still get contradictory results from the two standard
change-score analyses? (2) Does dual-centered ANCOVA duplicate the results of pure within-person
difference-score analyses, parallel to the robust consistency across both change-score analyses after pretest
matching? (3) If standard residualized analyses and pretest matching continue to contradict the known
effectiveness of these corrective actions, do Pretest X Treatment interactions reconcile the discrepancies by
showing effectiveness for the most at-risk cases? (4) Can the identification and control of Time-1 confounds be
improved by similar focusing on pure within-person changes and group-mean centering?
We expect this study to document biases that explain why all corrective actions appear to be harmful in
residualized analyses, whether implemented by parents (disciplinary responses, talking to youth about deviant
behavior and peers) or by professionals (therapy, prescription medications, child care subsidies, foster care,
job training programs). Removing these pervasive biases have the potential to improve the ability of basic
parenting research to approximate valid causal inferences more closely, which is necessary to improve the
next generation of parent-implemented psychotherapies for at-risk children.