On tests for sex preferences

This paper presents a critical evaluation of three widely used tests for sex preferences: sex ratio, parity progression ratio and ordinary least squares regression of birth interval. We show that under some appropriate conditions, the sex ratio is a valid test for sex preferences. The methods of parity progression ratio and OLS regression of birth interval fail to deal with right censoring and time varying covariates, which reduce the power of the tests. We suggest the use of hazard estimation to test for sex preferences. We demonstrate the differences among the tests by analyzing the retrospective fertility histories of the Chinese and the Malays in Malaysia. We find that unlike the two conventional methods, the hazard estimation gives clear and strong evidence of sex preferences among the Chinese in Malaysia.I am indebted to Gary Becker, James Heckman and Thomas Mroz for their many valuable comments and encouragement. I have also benefited from the discussions with V. Joseph Hotz, Mark Rosenzweig, Lester Telser and Robert Willis. Research support from the Bradley Foundation and the Hewlett Foundation is gratefully acknowledged. Computational facilities are supported by NICHD Grant HD-19226 and NSF Grant SES-84-11242 to the Economics Research Center/NORC.