Expands the Capabilities of SPSS and Excel 

Program Dynamic

Researchers often want to average a set of correlations.  The most common scenario for this is in the context of meta-analysis.  ZumaStat offers a program for averaging correlations.

The program asks the user to input the correlations and the sample sizes, up to 20 correlations.  The program can also input data from an ascii file, up to 1,000 correlations.  After entering the correlations, the desired confidence interval, and the number of correlations, pressing Calculate opens an ascii file with the results of the analysis. 

The program is unique in that it applies both fixed effect and random effect models in the calculation of the average correlation.

How it Appears on Your Screen

The Output

The output appears as follows:

ZumaStat first prints out the input data so that if you print the output, you have a record of the data used in the analysis.

The first average is based on a fixed effects model.  This model is used if one can assume that each of the sample correlations is selected from populations with a common population correlation value (e.g., each population has a true correlation value of 0.40).  The average correlation in the output is a point estimate of the population correlation.  It was 0.423.  ZumaStat provides a confidence interval for the estimate and a z test that tests the null hypothesis that the true population correlation is zero.

ZumaStat also reports a Q test which often is called a test of homogeneity.  It tests the null hypothesis that all of the samples come from a population with a common population correlation.  If the Q test is statistically significant, then one rejects the null hypothesis of a common population correlation.

The second average is based on a random effects model.  A random effects model is used if one assumes that the samples come from populations that can have different population correlation values.  One does not assume a common population correlation for all samples.  ZumaStat provides the point estimate of the average correlation across the populations, a confidence interval for the estimate, and a z test of the null hypothesis that the true average correlation across populations is zero.

Two different statistical approaches have been suggested for applying a random effects model, a theta method and a Q method.  ZumaStat calculates both.  The Help menu of ZumaStat provides references to the relevant literature.