Split-half coefficients

1. To compute the split-half coefficients, recall the Reliability Analysis dialog box.

   

2. Select Split-half as the model.
3. Click Statistics.

   

4. Select Scale in the Descriptives for group and deselect Item and Correlations.
5. Click Continue.
6. Click OK in the Reliability Analysis dialog box.

Ideally, in order to obtain a good estimate of the reliability of a survey, we would like to administer the survey twice to the same group of people and then correlate the two sets of results. However, this is often impractical because bias may be introduced in the second set of answers or because respondents may be unwilling or unable to take the survey a second time. One solution is to compute Cronbach's alpha. Another is to split the items into two groups and then to compare these groups as if they were two separate administrations of the same survey.

If the items are entered in order, the procedure splits them so that the first four are in one group and the last three are in the other. The Correlation between forms is simply the correlation between the sums of the items in each group. The Equal-length Spearman-Brown coefficient is then computed using the formula for the standardized item alpha for two items, inserting the Correlation between forms as the correlation. The Guttman Split-half coefficient is computed using the formula for Cronbach's alpha for two items, inserting the covariance between the item sums of two groups and the average of the variances of the group sums.

Notice that different splits of the items will produce different estimates of the reliability coefficient. When they are split so that each group contains items that are highly correlated within the group but not between groups (as has been done here), the split-half coefficients will be close to their lowest values. When highly correlated items are paired off and placed into separate groups, then the split-half coefficients will reach their highest values.

Next