Inventory Development
 Split-Half Reliability


Whereas Cronbach's alpha represents the average of all possible item splits of an instrument, split-half analyses provide data-based estimates on only one split. This procedure requires that the original TSI be randomly split into two equivalent sets of items. Thus each item set would represent one of two item pools drawn from one content domain-in this case, the domain of "teacher stress." The correlation of the scores from the two halves is then inspected. If each of the subsets of items is truly representative of the content domain from which they were drawn, then each short form's alpha should be reasonably large, and the correlations between forms should prove at least moderate in magnitude (American Psychological Association, 1974). Limited split-half reliability data for the initial form TSI have been collected and are reported in Fimian (1986c).