Luis A. Afonso
Posts:
4,725
From:
LIsbon (Portugal)
Registered:
2/16/05


Chisquared gof tests Distributions DO depend on sample size
Posted:
Sep 11, 2012 9:12 AM


Chisquared gof tests Distributions DO depend on sample size ________________Part ONE_____________________
Program ChiK10 shows how this ?blasphemy? is really true. People learn that Test Distribution varies only with the number of classes k. We were able to state that varying the sample size, n, we get differences in what concerns the fractiles 0.95 and 0.99. Samples,1 million for each n, as they were normal N(0,1) and k=10. In rigor, they were not simulated: the classes were filed by random (0, 1) values because we chose to let them limited by the CDF quantities Fi(j/n), j=1, 2, . . . , 10, resulting in a exact procedure because we are only concerned with the number of items contained in each class, not the real normal item values. To follow up . . .
Luis A. Afonso
[Economists (statisticians) were invented to confer some credibility to meteorologists. . .]

