Measurement equivalence of neuropsychological tests across education levels in older adults

Abstract

Objective. The objective was to determine whether neuropsychological tests provide an equivalent measure of the same psychological constructs in older adults with low versus higher levels of education. Method. Confirmatory factor analysis was used to evaluate the fit of a three-factor model (Verbal Ability, Visuospatial Ability, Long-Term Retention) to scores from the neuropsychological battery of the Canadian Study of Health and Aging (CSHA). Measurement equivalence of the model across lower educated (LE; ≤8 years) and higher educated (HE; ≥9 years) participants was evaluated using invariance testing. Results. The measurement model demonstrated adequate fit across LE and HE samples but the loadings of the 11 tests onto the three factors could not be constrained equally across groups. Animal Fluency and the Token Test were identified as noninvariant tests of Verbal Ability that, when freed from constraints, produced a partial metric invariance model. Scalar invariance testing identified the Buschke Cued Recall Test and Block Design as measures with invariant factor loadings but noninvariant intercepts. Analyses were replicated in age- and sex-matched subsamples. Conclusions. Metric and scalar invariance across HE and LE samples was achieved for seven of the 11 tests in the CSHA battery. Animal Fluency and the Token Test were noninvariant measures of Verbal Ability, suggesting that cognitive processes underlying performance on these tests may vary as a function of education. In addition, scores from Block Design and the Buschke Cued Recall Test were observed to differ in their scale of measurement between HE and LE examinees.

Keywords:

• Cognitive reserve
• Psychometrics
• Older adults

Data reported in this article were collected as part of the second wave of the Canadian Study of Health and Aging (CSHA). The CSHA was coordinated through the University of Ottawa and the Federal Government’s Laboratory Centre for Disease Control. Waves 1 and 2 of the CSHA core study were funded by the Seniors’ Independence Research Program, through Health Canada’s National Health Research and Development Program (NHRDP) [project number 6606-3954-MC (S)]. Additional funding was provided by Pfizer Canada Incorporated, through the Medical Research Council/Pharmaceutical Manufacturers Association of Canada Health Activity Program, NHRDP [project number 6603-1417-302R]; by Bayer Incorporated; and by the British Columbia Health Research Foundation [project number 38 (93–2)], [project number 34 (96–1)]. The CSHA was coordinated through the University of Ottawa and the Federal Government’s Laboratory Centre for Disease Control. P.W.H.B. was supported in part by a Canada Graduate Scholarship Master’s Award (2009–2010) and a Doctoral award (2011–2013) from the Canadian Institutes for Health Research. The authors report no financial or other relationships that could be interpreted as a conflict of interest affecting this manuscript.

Notes

¹ Kurtosis values for the Token Test were similarly elevated in the age-matched and sex-matched HE and LE subsamples. Missing data never exceeded 5% in any of the subsamples.

² Single-sample CFA suggested adequate model fit in the age-matched [HE subsample, χ²(41, N = 115) = 63.17, p = .01, χ²/df = 1.54, CFI = .93, NNFI = .90, SRMR = .08, RMSEA = .07; LE subsample, χ²(41, N = 115) = 45.28, p = .30, χ²/df = 1.10, CFI = .99, NNFI = .99, SRMR = .03, RMSEA = .03] and the sex-matched samples [HE subsample, χ²(41, N = 120) = 80.26, p < .001, χ²/df = 1.96, CFI = .90, NNFI = .87, SRMR = .08, RMSEA = .09; LE subsample, χ²(41, N = 120) = 43.98, p = .35, χ²/df = 1.07, CFI = .99, NNFI = .99, SRMR = .02, RMSEA = .03]. This was further supported by a MGCFA across education levels in the age-matched subsample [χ²(82, N = 230) = 108.45, p = .03, χ²/df = 1.32; CFI = .96, NNFI = .92, SRMR = .04, RMSEA = .04] and the sex-matched subsample [χ²(82, N = 240) = 124.25, p = .002, χ²/df = 1.52, CFI = .94, NNFI = .90, SRMR = .06, RMSEA = .05].

³ Full metric invariance similarly could not be established in the age-matched, Δχ²(8, N = 230) = 20.67, p = .01, or the sex-matched subsample, Δχ²(8, N = 240) = 19.29, p = .01. Partial metric invariance was obtained in both subsamples by relaxing constraints on the Verbal Ability factor loadings: age-matched subsample, Δχ²(4, N = 230) = 8.75, p = .07; sex-matched subsample, Δχ²(4, N = 240) = 6.54, p = .16. As with the full sample, factor loadings for Animal Fluency and the Token Test were noninvariant in the sex-matched sample, and relaxing constraints on these factor loadings resulted in partial metric invariance, Δχ²(6, N = 240) = 9.63, p = .14. In the age-matched sample, only Animal Fluency was noninvariant, and relaxing the constraint on the factor loading for this test resulted in partial metric invariance, Δχ²(7, N = 230) = 12.82, p = .08.

⁴ Constraint of model intercepts significantly decreased model fit in the sex-matched subsample, Δχ²(12, N = 240) = 28.59, p < .01. As with the full sample, release of constraints on intercepts associated with the Buschke Cued Recall Test and the Block Design test resulted in partial scalar invariance, Δχ²(10, N = 240) = 16.50, p = .09. In the age-matched subsample, constraint of model intercepts associated with invariant factor loadings did not significantly influence model fit, Δχ²(14, N = 230) = 19.86, p = .09.

⁵ Constraint of error terms associated with invariant model intercepts resulted in significant loss of fit in both the sex-matched, Δχ²(14, N = 240) = 30.36, p < .001, and the age-matched, Δχ²(21, N = 240) = 37.63, p < .05, subsamples.