Advanced issues in correlation and causality
To go beyond the text,
discuss two sophisticated strategies designed to allow causal inferences from correlational data (cross-lagged panel analysis and path analysis). Discuss how these strategies deal with temporal precedence and third variables. Point out that because cross-lagged panel analysis makes some questionable assumptions about temporal precedence and third variables, its popularity seems to be waning. Conclude by discussing partial correlations, stressing that partial correlations allow us to determine whether a relationship between two variables can easily be explained by a certain third variable. For example, if, in a certain sample, we find a correlation between gender and intelligence, we may wonder if the relationship may simply reflect that the women in the sample are older than the men. By
computing a partial correlation, we can determine whether age could account for the relationship. Note that, conceptually, we could simply look for the correlation between gender and intelligence within each age group and see if the relationship held when "controlled" for age. Students can easily compute partial correlations by using this Excel spreadsheet.
You may also want to talk about the pitfalls of using multiple regression to try to make cause-effect statements. Ted Goertzel`s (five page) article "Myths of murder and multiple regression" in the January/February 2002 issue of the Skeptical Inquirer illustrates the probelms of trying to use multiple regression to address to make cause-effect statements by showing that multiple regression has been used to "prove" a variety of conflicting "truths," such as that legalizing abortion (a) decreased crime and (b) increased crime. The article also does a good job of emphasizing the problems of controlling for third variables as well as the problem that a regression equation that has been constructed to fit an existing data set may not be useful for predicting future data. To prepare students for a discussion of multiple regression, you might assign pages 530-535 in Appendix D.
Back to Chapter 7 Main Menu