Sixth,consider the relationship between the alternative distributions, Gamma(4,0.5), Chi-Square(1) and the simulated power. Figure 6summarizes the power analysis for the Gamma(4,0.5) and Chi-Square
(1) alternative distributions. For a fixed sample size and a significance level, powers for these
two alternative distributions were exactly the same. As in the previous alternative distributions, the PML-test outperformed all other four exponentiality tests across all sample sizes and significance levels. The LF-test was in the last place on the power curve. The powers for the VS-test and S-test were identical for a fixed sample size and a significance level. The D-test demonstrated the superior power than the VS-test and the S-test for small sample sizes across all significance levels but this relationship was reversed for medium to large sample sizes. For sample
sizes at least 200, the powers for all five tests were equivalents which were close to 1. As compare with the previous alternative distribution (Gamma(0.55,0.412)), powers for these two alternative distributions decrease across all sample sizes and significance levels. It is relevant to note that the shape parameter (k) was changed from 0.412 to 0.50 which caused the decrease in power. It appears that as the value of the shape parameter (k) approaches that of the null distribution (k=1), the simulated powers decreases
Before considering the power for next two alternative distributions, it is indispensable to revisit that the
Chi-Square distribution is a special case of Gamma distribution (equation 8). This study previously showed that the power for the Gamma distribution depends only on the shape parameter (k). Null distributions were generated using the exponential (θ=5) for power simulation.