TECHNICAL EXPLANATION OF ESTIMATION OF TOTAL, DIRECT, AND INDIRECT EFFECTS*
As we explained in chapter 1, a primary goal of this book is the estimation of the total, direct, and indirect influences of a broad range of explanatory factors on rates of entry to marriage and cohabitation. Estimation of such effects requires properly specified models that correctly indicate temporal and causal ordering. By “total effect” we mean the total amount of influence of a factor through all potential causal mechanisms. A total effect can be further decomposed into a “direct effect” component and an “indirect effect” component. An indirect effect operates through a particular intervening factor or set of factors. A direct effect is the residual influence unexplained by all the intervening factors. Because of the ambiguities associated with causal ordering, we estimate these effects in multiple models with different causal assumptions. In doing so we regularly estimate effects with minimum and maximum controls. This provides a method for bounding the sizes of our effect parameters. At the same time, we recognize that the omission of relevant factors and poor assumptions about exogeneity and endogeneity could result in all of our estimates being biased and different from the “true” effects in the “real” world. That is, without experimental data we cannot be sure that extraneous factors have not biased our estimates.
We utilize a thirty-one-year intergenerational panel study to help us in specifying appropriate causal models. The ordering of attributes by generation and time helps us to get the causal ordering correct. Although we recognize that causal or-
*Li-Shou Yang collaborated in the analysis and writing of this appendix.