A MONTE-CARLO INVESTIGATION OF THE ROBUSTNESS OF THE WALD, LIKELIHOODRATIO, AND MEDIAN TESTS WITH SPECIFIED SYMMETRICAL AND ASYMMETRIC MARGINAL DISTRIBUTIONS
Jf. Reed, A MONTE-CARLO INVESTIGATION OF THE ROBUSTNESS OF THE WALD, LIKELIHOODRATIO, AND MEDIAN TESTS WITH SPECIFIED SYMMETRICAL AND ASYMMETRIC MARGINAL DISTRIBUTIONS, Computer methods and programs in biomedicine, 39(3-4), 1993, pp. 131-136
The effect of nonnormality on the Type I (tau) error when comparing tw
o independent binomial proportions (P) or the nonparametric alternativ
es, the Median (Me), Wald (W), and Likelihood Ratio (LR), has not been
investigated. If these selected tests are overly conservative the imp
lied loss of power would moderate their practical use. The purpose of
the present study was to investigate the impact of nonnormality on sma
ll to moderate sample sizes on the estimated tau for alpha = 0.10, 0.0
5, and 0.01 for the P, Me, W, and LR tests. Samples were generated fro
m nine long-tailed symmetric and asymmetric distributions using a mult
iplicative congruential generator. For each marginal distribution and
for a variety of sample sizes, the proportion of samples for which the
test statistic exceeded the 10, 5, and 1 percentage points was tabula
ted. For data that mimic a symmetric distribution, the median test uni
formly yields an empirical alpha considerably less than tau, while the
likelihood ratio test consistently overestimates tau for small sample
s (n less-than-or-equal-to 15) over all symmetric distributions and em
pirical alpha levels. For asymmetric distributions, the median test ag
ain yields an empirical alpha significantly less than tau. Similar und
erestimates of tau were found for the chi-square (2 df), chi-square (4
df) log normal, and gamma (2, 1) distributions. The likelihood ratio
test consistently overestimates tau for small samples (n less-than-or-
equal-to 15) over all asymmetric distributions and empirical alpha lev
els. The independent proportions test produces an empirical alpha clos
est to tau for n = 10 for all asymmetric distributions. The most consi
stent test procedure, disregarding sample size or distribution charact
eristics, is the independent proportions test procedure.