Examination of Response Formats without Anchoring Items

Abstract

Bartlett's chi squared, Cattell's scree test, and the Kaiser criterion for the number of factors were examined for susceptibility to error variance using non-anchored responses ( N = 292) to various sets of fixed-response formats (4-, 5-, and 7-point Likert and numeric, lines with end points anchored, and several forms of human faces). Results demonstrated that Cattell's scree test is the most conservative under these conditions followed by Bartlett's chi squared with alpha set at .001. The Kaiser criterion was generally the most liberal. The factor pattern of the first principal component for each set of response formats was then examined for consistency using congruence coefficients. High congruences were found among most formats with the exception of 4-point formats. Of formats tested 4-position response formats seem least susceptible to systematic error based both on eigenvalue criteria and congruence with other formats.