The Solomon 4 group experimental design has been famous for the true-experimental design which could maintain the internal validity and support the external validity more or less. Nevertheless, it has been insufficient on the discussion of appropriate...
The Solomon 4 group experimental design has been famous for the true-experimental design which could maintain the internal validity and support the external validity more or less. Nevertheless, it has been insufficient on the discussion of appropriate statistical treatment. The flowchart listing the statistical tests and their sequence in analyzing the Solomon four-group experimental design was presented in this study.
The frame of the Solomon four-group experimental design is as follows;
Group Sample pretest treatment posttest
Group Ⅰ R O_1 X O_2
Group Ⅱ R O_3 O_4
Group Ⅲ R X O_5
Group Ⅳ R O_6
The flowchart began with test A, a 2×2 analysis of variance(ANOVA) on the means of O_2, O_4, O_5, O_6, which examines the interaction of pretest×treatment. Pretest sensitization is indicated if this test is statistically significant. This flowchart noted that a statistically significant interaction should be followed up with a simple effects test on the means of the pretested groups(O_2, O_4) and a simple effects test on the means of the unpretested groups (O_5, O_6). If test above is not statistically significant, the analysis continues with the test of the main effects for the experimental versus control group.
Conditioned on these preliminary tests not being statistically significant, the process continues with an analysis of covariance(ANCOVA) on O_2, O_4, with O_1, O_3 as a covariate.(Alternatives to the ANCOVA are test, a gain score analysis, and the test a repeated measures ANOVA). If the ANCOVA is not statistically significant, a t test is conducted on O_5, O_6
On the other hand, Braver & Braver suggested using Stouffer's Z(Stouffer, Suchman, DeVinney, Star, & Williams, 1949) to combine the results of the last two traditional tests used ANCOVA and t test, which will include all available data in the experiment. But Sawilowsky, Kelley, Blair, Markman(1994) present a Monte Carlo demonstration that the experiment-wise error rate inflates nearly triple norminal alpha. Despite the ability of Stouffer's Z to combine all available data into a single statistic, Braver & Braver's method is needed more research. Consequently, there are still not solved the problem which could combine the data of 4 groups in Solomon 4 group experimental design synchronously.