23.3 Appropriateness of the Methods

SAS code for conducting the analyses in the previous section is given in Appendix G. Table 23.1 summarizes the p-values for the treatment group effect for each of these analyses. Although the exercises are assigned for practice in using SAS to perform commonly used statistical analyses, some of these methods would clearly be inappropriate to use in practice. Let's take a look at the results.

Table 23.1. Summary of Significance of Treatment Group Effects from Exercises

Analysis	Method	p-Value for Treatment Group effect
	Analyses Using the SCORE Variable
1	Two-sample t-test	0.0422
2	Wilcoxon rank-sum test	0.0464
3	One-way ANOVA	0.0422
4	Two-way ANOVA with interaction	0.0702
5	Two-way ANOVA without interaction	0.0378
6	Three-way ANOVA with all interactions	0.1036
7	Three-way ANOVA main effects model	0.0493
8	Linear regression	--
9	ANCOVA / unequal slopes	0.5663
10	ANCOVA / equal slopes	0.0257
11	Stratified ANCOVA / equal slopes	0.0252
	Analyses Using the RESP Variable
12	Chi-square test	0.4510
13	Fisher's exact test	0.4794
14	Cochran-Mantel-Haenszel test	0.4113
15	Logistic regression, trt effect only	0.4523
16	Logistic regression, trt and age effects	0.3576
17	Logistic regression, trt, age, sex, center, trtxcenter effects	0.6984
	Analyses Using the SEV Variable
18	Chi-square test	0.1216
19	Generalized Fisher's exact test	0.1277
20	Mantel-Haenszel, row mean scores differ / table scores	0.0514
21	Mantel-Haenszel, row mean scores differ / mod. ridit scores	0.0468
22	Wilcoxon rank-sum test	0.0472
23	CMH, row mean scores differ / table scores	0.0445
24	CMH, row mean scores differ / mod. ridit scores	0.0471
25	Proportional odds model, trt effect only	0.0447
26	Proportional odds model, trt and age effects	0.0244
27	Proportional odds model, trt, age, sex, center	0.0313