Applying the standard bootstrapping technique with corrections
for heteroskedasticity for a sample of the 1997 Urban Household
Survey in China, the present paper attempts to test (1) whether the
commonly used decomposition methods for gender earnings
differentials give significantly different results, and (2) whether the
explained component is significantly different from the unexplained
component (which is commonly referred to as discrimination) within
each decomposition method. Based on a national data set, the
empirical results indicated some significant differences in both tests.
The implication of the results is that the proposed bootstrapping
technique can be regarded as a guideline on applying which
approach to decompose gender earnings differentials among
different methods without losing important information, and on
evaluating the relative importance of the decomposition components
for any chosen method.