4Elements (4El) Personality Inventory: Computerized Adaptive Personality Assessment in the Work Environment
Abstrakt
Description The 4Elements personality questionnaire is based on the metaphor of the four elements (Fire, Water, Earth, Air). It consists of 100 items divided evenly into four factors, and it uses a short scale ("yes", "no", "don't know").
Firstly, it was standardized in 2008. The analysis performed so far shows superb psychometric properties of this questionnaire regarding factor structure, reliability, and validity.
To improve the assessment's quality and efficiency, we decided to explore the possibilities of adaptive administration of the questionnaire. Data from 2011-2018 were used for this purpose (N = 13,298, 58.5\% females, average age 36.7, SD = 9.7).
Due to the nature of the scale, a polytomous model was considered, but since the results showed that the middle "don't know" responses behaved more as missing, a unidimensional 2PL model was eventually used. We performed the Post Hoc analysis using the catR package with the item parameters and the response patterns of the respondents.
To ensure content validity, we used content balancing. We compared the item selection methods MFI and bOpt (Urry's criterion) and used a random selection of items to assess the effectiveness of these methods.
While the MFI method worked better for Air and Earth factors, the bOpt method worked better for the factors Water and Fire. We compared the ML and EAP methods to evaluate the accuracy of the level of ability estimation.
In most cases, the EAP method proved to be better. The level of SE(θ)max 0.80, which is the value considered sufficient for personality inventories.
The adaptively administered test achieved the same accuracy as the full-length test using an average of half the items (Pearson's ṙ = 0.93). Due to the promising results of the performed Post Hoc simulations, we took steps to create a CAT version 4El.
We revised and expanded the item pool and started data collection to calibrate new items. As part of further simulations on a new sample, we will choose a suitable item exposure method.
The administration of the fixed test currently runs in a web-based application developed specifically for this purpose. We plan to implement the new adaptive features later this year.