>Any
>competent statistician will tell you that inferences based on sample N=27
>split four ways are not very robust - to say the least.
The 27 cases were not a "sample", it wasn't a sample of the 27 victims, but the entire number of victims:
"My listeners were surprised to learn that the majority of those we Californians have put to death have been white. But they were astonished when I added this statistic: Among the 27 victims of these 12 condemned prisoners, [since 1992] not a single one was African-American! (Five were Asian; three were Latinos; and 19 were white.)"
"Sampling error" is a risk when a small number of cases are being used to represent a larger population, but since it wasn't a sample, sampling error can't apply. Your objection on the basis of possible sampling error is misguided.
Bill Bartlett Bracknell Tas