Skip to main content

LUP Student Papers

LUND UNIVERSITY LIBRARIES

Accounting for Partial Measurement Invariance in Score Computation: Does it Matter for Diagnostic Personality Assessment in Europe?

von Westerman, Marie LU (2024) PSYP01 20241
Department of Psychology
Abstract
A lack of full scalar measurement invariance (MI) across groups of test takers is widely assumed to render sum scores incomparable between groups. Computing factor scores from partial-scalar-MI models has been proposed as a remedy. To what extent this scoring method affects personality assessment in diagnostic contexts has yet not been investigated. The present master thesis explored this question re-analysing an example dataset with the responses of men and women from four northern European countries (n = 7504) to the Big-Five questionnaire IPIP-NEO-120. First, each of the Big-Five scales was tested for full and partial scalar MI across nationality-by-gender groups with multigroup confirmatory factor analysis. Second, the study compared... (More)
A lack of full scalar measurement invariance (MI) across groups of test takers is widely assumed to render sum scores incomparable between groups. Computing factor scores from partial-scalar-MI models has been proposed as a remedy. To what extent this scoring method affects personality assessment in diagnostic contexts has yet not been investigated. The present master thesis explored this question re-analysing an example dataset with the responses of men and women from four northern European countries (n = 7504) to the Big-Five questionnaire IPIP-NEO-120. First, each of the Big-Five scales was tested for full and partial scalar MI across nationality-by-gender groups with multigroup confirmatory factor analysis. Second, the study compared Big-Five scores computed from partial-scalar-MI models with sum scores. The analyses revealed that for each Big-Five personality trait (a) full scalar MI had to be rejected, and (b) the two scoring methods yielded highly similar trait scores on average. However, for up to 6% of the sample, trait scores computed from partial-scalar-MI models did not fall within the 95%-confidence interval of respective sum scores. Depending on the reason for the personality assessment, these differences might be relevant in high-stakes situations. Practitioners should be aware of the potential effect of accounting for partial scalar MI in score computation. Future research could (a) try to replicate the results in more diverse and representative samples, and (b) examine which scoring method yields in fact the least biased trait scores. (Less)
Please use this url to cite or link to this publication:
author
von Westerman, Marie LU
supervisor
organization
course
PSYP01 20241
year
type
H2 - Master's Degree (Two Years)
subject
keywords
Personality Measures, Five Factor Personality Model, Measurement Invariance, Test Scores
language
English
id
9174296
date added to LUP
2024-09-11 16:27:40
date last changed
2024-09-11 16:27:40
@misc{9174296,
  abstract     = {{A lack of full scalar measurement invariance (MI) across groups of test takers is widely assumed to render sum scores incomparable between groups. Computing factor scores from partial-scalar-MI models has been proposed as a remedy. To what extent this scoring method affects personality assessment in diagnostic contexts has yet not been investigated. The present master thesis explored this question re-analysing an example dataset with the responses of men and women from four northern European countries (n = 7504) to the Big-Five questionnaire IPIP-NEO-120. First, each of the Big-Five scales was tested for full and partial scalar MI across nationality-by-gender groups with multigroup confirmatory factor analysis. Second, the study compared Big-Five scores computed from partial-scalar-MI models with sum scores. The analyses revealed that for each Big-Five personality trait (a) full scalar MI had to be rejected, and (b) the two scoring methods yielded highly similar trait scores on average. However, for up to 6% of the sample, trait scores computed from partial-scalar-MI models did not fall within the 95%-confidence interval of respective sum scores. Depending on the reason for the personality assessment, these differences might be relevant in high-stakes situations. Practitioners should be aware of the potential effect of accounting for partial scalar MI in score computation. Future research could (a) try to replicate the results in more diverse and representative samples, and (b) examine which scoring method yields in fact the least biased trait scores.}},
  author       = {{von Westerman, Marie}},
  language     = {{eng}},
  note         = {{Student Paper}},
  title        = {{Accounting for Partial Measurement Invariance in Score Computation: Does it Matter for Diagnostic Personality Assessment in Europe?}},
  year         = {{2024}},
}