抄録
The purpose of the present study was to examine the quality of statistical testing in
foreign language teaching research in Japan. We reviewed t-tests, X2-tests and
ANOVAs reported in the articles published in Language Education & Technology
(LET) from vol. 38 to vol. 49, and calculated the post-hoc statistical power of each test.
The findings of the present study were summarized as follows: (a) the sample size of
most of the studies in LET ranged from 20 to 60, (b) the median of the effect sizes (in
the case of t-tests) showed middle to large levels (d = 0.40-0.80), but (c) the statistical
powers of many studies signified severely low levels (almost the 80% of the two-
sample t-tests failed to show the statistical power greater than .80). The tendencies
were quite likely to have originated chiefly from the inappropriate designs of the
experiments or surveys, especially, mismatches between the targeted effect size and
the actual sample size. We assert the importance of setting proper sample sizes based
on a priori power analysis and precision analysis.
