Meta Analysis Page Logo Meta - Analysis:
Methods of Accumulating Results Across Research Domains
Page 6
Home Contents Abstract Introduction Links Contact Info

Introduction and History

There are two types of Quantitative Review procedures. One type involves the combination of probability values or Z scores. The procedures for this method was developed in parallel during the 30's by Cochran (1937), Fisher (1932), Pearson (1933) and Tippett (1931). These procedures were developed to address the need in agricultural research to combine the results of a number of independent tests, all of which were planned to test a common hypothesis. An alternative approach was also developed by Fisher in 1932, the r to Z transformation.

The demands of World War II served to assist in the development of combinatorial procedures. In their landmark study on the American soldier, Stouffer and colleagues during the 1940's developed a probability combination method. A more recent version of the combinatorial procedure is Winer's (1971) method of combining independent t tests.The other type of meta-analysis is the accumulation of effect sizes, correlation coefficients or to Cohen's d statistic. Thorndike (1933) was among the earlier researchers to accumulate results across studies using an average correlation. He also corrected the observed variance of results across studies for sampling error (unreliability). The intent of this procedure was to integrate differing research on intelligence.

While procedures for averaging correlations were available since the 1930's, as noted above, and were discussed in various behavioral statistics texts (cf. McNemar, 1969), these procedures generally involved the use of Fisher's r to Z transformation, or were generally not used. Unfortunately no guidelines existed that allowed for a "dimensionless" statistic which could be used as a rubric or common statistic which would be independent of any specific measurement unit. Cohen (1977) developed one such statistic now in common use, the effect size statistic, or d. it was originally developed for use in statistical power analysis and to estimate the optimal sample size for a study.

Previous Home Contents Intro Top Next