The missing data effect

Michael J. Mueller inquired:

Q. It turns out after I ran a plot for Pct. women vs. vote for Reagan in 80 I got a plot that looks pretty normal with an r of .53 and an r-squared of .28. If you remember on my mult. regression the ending b value for women was 2.47 and the starting correlation was .692 with reagan80 (this is in the describtives part of the run). The one weird thing I guess is that the plot r between the two variables is different from the correlation given in the descriptives section of the multiple regression. How can all this happen?

A. Check the Ns underlying the two correlations. If you compute the correlation in a scattergram between two variables, all available cases on both variables will be involved. Regression computes intercorrelations only among the cases that have data on ALL variables. So I guess that your scattergram N is based on 51, and your regression is based on another N, perhaps 49. Let me know if this isn't so.