Representation of women in CS: how do we measure a program's success?

This paper presents an analysis of the data on representation of women in US liberal arts computer science departments, using 10 years' worth of IPEDS data. What began as a search for departments with exemplary representation ended with the conclusion that the representation data is too unstable to be a useful measure of success: The correlation between average representation values in consecutive five-year periods is small, only r=0.156, and not significantly different from zero (p=0.143, t=1.477) for this set of institutions. Other metrics and sample populations are considered with similar results. This result has important implications for studies assessing the impact of departmental interventions on the representation of women, as well as research on the factors that influence representation.