Empirical studies of gender segregation by occupation must be founded on ri
gorous measurement procedures. There appears to be a consensus that any ind
ex used in the analysis of time-series or international cross-section emplo
yment data must be either margin-free or decomposable to yield a margin-fre
e component. On the other hand, Charles and Grusky (1995) advocate the use
of multiplicative log models from which a margin-free odds ratio can be der
ived. In this paper I contrast the construction and interpretation of the i
ndex of dissimilarity and the Karmel-MacLachlan index with the multiplicati
ve modeling of gender segregation and the associated log index.