Conditional direction dependence analysis in linear models with SPSS macros and custom dialogue implementation

[ACCESS RESTRICTED TO THE UNIVERSITY OF MISSOURI AT AUTHOR'S REQUEST.] In nonexperimental data, the causal ordering of variables can be examined with Direction dependence analysis (DDA), a statistical method that utilizes various asymmetry properties of the linear regression model to validate a postulated explanatory model against plausible causally reversed alternative models. However, standard DDA assumes that the observed causal effect is constant for all subjects and does not consider the conditional effect of a third variable on direction dependence, which may lead to biases and/or compromised power. The present work relaxes this assumption by proposing conditional direction dependence analysis (CDDA). CDDA examines the direction of effect when a moderator is present and extends standard DDA by combining the pick-a-point approach and variable purification technique, which enables researchers to examine the direction of effect at a certain moderator value. The results of two Monte-Carlo simulation studies are reported which evaluate the performance of CDDA. The first simulation study shows that the observed power of DDA tests vary across moderator values when a t hird variable moderators the main effect. The second study shows that, under certain conditions, CDDA is able to identify the true data-generating mechanism when a third variable determines the direction of causal flow. SPSS macros and auxiliary custom dialogues are provided for easy implementation of CDDA procedures, which is illustrated with a worked example. A real-world example is given for illustrative purpose.