Eguchi and Matsuura (1990, Biometrics 46, 415-426) noted that the gene
ralized Stevens test statistic for the Hardy-Weinberg law for human le
ukocyte antigen (HLA) data yields an excessively large value when no d
ouble blanks are observed. In this paper, we investigate this aberrant
case. The inflated value of the test statistic is caused mainly by a
seriously biased Bernstein's estimator of the recessive. We propose th
e generalized Stevens test using bias-corrected Bernstein's estimators
for HLA data with no observed double blanks. An example illustrates t
he correction of the test statistic.