To The Editor:

As a faculty member who teaches in the area of testing and assessment, I applaud and support the efforts of John Hunter and his colleagues regarding their criticism of employing the .05 level of statistical significance in social science research ("Psychologists Debate Accuracy of 'Significance Test,'" August 16). The truth of the matter is that use of the .05 level of significance does tend to inappropriately embellish the utility of research results that are statistically significant in a technical sense but of marginal relevance within a more pragmatic context.

There are at least two plausible ways of dealing realistically with this problem. First, we could utilize a smaller 'cutoff' point when interpreting the statistical significance of research results. Although acknowledging that there is only a 5 percent probability that the results obtained could have been produced by chance alone seems suitable, in reality a 5 percent margin of error is much too high to establish substantive confidence in the inherent meaningfulness of those results. Ideally, it might be beneficial to employ a much smaller 'cutoff' point. I have always felt that the .001 level of statistical significance, which is intrinsically much more rigorous, would be more appropriate as the 'standard' for social science investigations.

Second, we could eliminate the use of 'cutoff' points altogether. Researchers could simply state the exact level of statistical significance when reporting results. For example, if the results obtained are significant at the .037 level, then that is what should be reported when summarizing research findings. By handling issues of statistical significance in this manner, the reader would be left to decide whether or not the findings are truly 'significant,' and, accordingly, how much importance to assign the conclusions and implications that are drawn from those results.

John Hunter has a valid observation that researchers need to consider very seriously. Consequently, I anxiously await the recommendations of Robert Rosenthal's panel regarding this issue. Those recommendations could profoundly shape the way research results are interpreted in the social sciences for years to come.