Krichevets, Anatoliy N.

Relevance.

Nowadays the researchers commonly use a limited set of standard procedures and statistical coefficients when they develop psychometric instruments and investigate their structure. The routine using of such procedures without taking into account the specific features of psychometric scales can lead to incomplete or even inadequate results. In this context detailed consideration of the structure of psychometric instruments seems to be important and it may demand various non-standard ways of statistical analysis. 

Objectives. 

To conduct detailed analysis of the results of two intelligent subtests at the item level and to assess the sufficiency and adequacy of using standard methods for estimation of reliability and structural validity for these subtests.

Methods.

We analyze the data collected in intelligence testing of a large sample of respondents (11335 young adults). The respondents passed the KR-3 battery. In this study we examine in detail the structure of the subtests "Syllogisms" and "Analogies". Specifically, we estimated the reliability of the scales by the Cronbach's alpha coefficient, and the structure at the item level using the confirmatory factor analysis.

Results and conclusions.

Estimation of the reliability of the scales by Cronbach's alpha coefficient showed the importance of taking into account the time limitation, which is commonly used in intelligence tests. On the other hand, a detailed analysis of each subtest items made it possible to find out an additional factor which was not originally proposed in the factor structure. It is the factor of higher-order abilities of abstract analysis, whilst the subtest originally aimed at esteeming the special abilities. Confirmatory factor analysis showed improvement of fit when this factor was added. The results allow to conclude that the researcher may miss important properties of scales if one does not perform a detailed analysis of testing procedures and the structure of subtest at the item level, and so one may draw incomplete or inadequate conclusions about their psychometric properties.