Skip to main content
U.S. Department of Energy
Office of Scientific and Technical Information

Tests of a simple multinomial hypothesis when the sample is not large

Conference ·
OSTI ID:5498943

Outlying cells can be identified, based on the binomial distribution; the Bonferroni inequality then gives a conservative overall test. If instead, the Pearson chi-squared statistic is used, its distribution can be approximated as follows: condition on the counts in the first several cells, and approximate the conditional distribution as a shifted gamma distribution, matching the first two moments. Combine these approximate conditional distributions to obtain an approximate unconditional distribution. Generalized Chebyshev inequalities give bounds on the accuracy of the approximation; the fineness of the decomposition can be adjusted during the calculations to guarantee any desired accuracy. In examples, a good approximation is achieved by conditioning on those cells with the smallest probabilities.

Research Organization:
EG and G Idaho, Inc., Idaho Falls (USA)
DOE Contract Number:
AC07-76ID01570
OSTI ID:
5498943
Report Number(s):
CONF-810842-3; ON: DE82005955
Country of Publication:
United States
Language:
English