Reliability and Item Analysis Introductory Overview - Final Remarks

The Reliability and Item Analysis features were designed to make the construction of sum scales as efficient as possible. STATISTICA handles a large number of items (e.g., 300 items in a single scale), allows the user to identify and delete "bad" items efficiently, provides various options for estimating the number of items that should be added to the scale, and allows the user to examine the correlation of the scale with other measures after correcting for attenuation (validity). The Examples provide additional details concerning the utility of the various options and statistics provided in the program.

Tetrachoric correlations

In educational and psychological testing, it is common to use yes/no type items, that is, to prompt the respondent to answer either yes or no to a question. An alternative to the regular correlation coefficient in that case is the so-called tetrachoric correlation coefficient. Usually, the tetrachoric correlation coefficient is larger than the standard correlation coefficient, therefore, Nunnally (1970, p. 102) discourages the use of this coefficient for estimating reliabilities. However, it is a widely used statistic (e.g., in mathematical modeling), and the Reliability features include different ways to compute this coefficient.