The error in combining frequency groups for the x2 test is probably small if the groups contain only a small part of the total frequency. For small samples the test is also reasonably satisfactory. The difference between minimum x2 and the x2 of maximum likelihood is small.