Page 10 - IJPS-8-1

International Journal of
Population Studies Validity and reliability of Mini-Mental State Examination in older Chinese

analysis. The ultimate decision to determine the number of form, the factor loadings are also equivalent across the
factors was based on the outcomes of the above-mentioned two age groups. The next level of constraint that adds to
methods and the interpretability of the EFA results. the invariance is the intercepts. It was also referred to as
scalar equivalence (Mullen, 1995), indicating the existence
2.3.2. CFA of strong factorial invariance (Meredith, 1993). While the
After determining the number of factors, CFAs were mean is further constrained across the groups, it is referred
carried out using the package lavaan function cfa with the to as strict factorial invariance, indicating systematic group
specification of ordered = TRUE to indicate binary CFA differences in means matrices are due to group differences
which was performed. The weighted least square mean and in common factor score distributions (Yoon and Millsap,
variance adjusted estimator was used for the estimation. 2007). The factorial invariance testing was carried out
It used diagonally weighted least squares to estimate the using function cfa from the package lavaan (Rosseel,
model parameters. Model fit for CFAs was assessed using 2012) with the specification of group.equal to “loadings,”
multiple indices, consisting of the χ statistic (Bollen, 1989; “intercepts,” and “means” for metric, scalar, and strict
2
Jöreskog, 1993), comparative fit index (CFI; Bentler, 1990), invariance, respectively.
Tucker and Lewis index (TLI; Tucker and Lewis, 1973), 3. Results
(RMSEA; Browne and Cudeck, 1992), and standardized
root mean square residual (SRMR; Brown, 2006). The 3.1. Descriptive statistics
χ , RMSEA, and SRMR assess how well the covariances
2
predicted from the estimates reproduced the sample Descriptive statistics were generated using the function
covariances (Pomplun and Omar, 2001). The CFI and TLI dfSummary from the package summarytools (Comtois,
assess the degree of fit of the proposed model accounted 2021). Table 1 shows the mean and standard deviation of
for the sample covariances. RMSEA values approximating the 23 MMSE items for the study and the two age groups of
0.06 demonstrate a close fit of the model (Browne and young-old and old-old.
Cudeck, 1992; Hu and Bentler, 1999). A value <0.08 is 3.2. EFA
generally considered a good fit for SRMR (Hu and Bentler,
1999). CFI and TLI values of 0.90 (Bentler, 1990) and 0.95 3.2.1. Factorability
(Hu and Bentler, 1999), respectively, indicate an acceptable The KMO measure of sampling adequacy was 0.9,
and good fit for the model. indicating the adequacy of undertaking factor analysis.

2.3.3. Factorial invariance Similarly, the Bartlett test of sphericity also indicated the
sufficiency numbers of significant correlations that it was
After determining the CFA model that best described the unlikely the population correlation matrix was an identity
structure of the MMSE, factorial invariance testing was matrix (χ = 69,555; P < 0.001). The individual MSA that
2
carried out to examine the degree of invariance concerning ranged from 0.77 to 0.97 also gave the same conclusion.
age. Factorial invariance is a concept applied in the context
of psychometric analysis of an inventory (Revuelta, 3.2.2. Determining number of factors
Franco-Martinez, and Ximénez, 2021; Schürer, van The various methods of determining the number of factors
Ophuysen, and Behrmann, 2021; Putnick and Bornstein, gave diverse outcomes that ranged from 1 to 7. However,
2016) to measure the degree of invariance assurance for the majority of the fit indicators showed either a six-factor
a categorical variable about its applicability level. In the or a seven-factor solution. The heterogeneous correlation
current context, this concept postulates the psychometric correlogram with the correlation coefficients shown in
properties of MMSE, and its applicability for the two Figure A1 and without correlation coefficients shown in
age groups, young-old and old-old. The first invariance Figure A2 indicated either a six-factor or a seven-factor
testing is configural invariance, commonly referred to as solution. The scree plot and Horn’s parallel analysis showed
pattern invariance. This testing procedure aims to examine a six-factor solution (Figure A3). Out of the six fit indices
whether both the young-old and old-old have the same from Table A1, RMSEA, BIC, and SABIC suggested seven
MMSE factor structure (Cheung and Rensvold, 2002; factors. While the Velicer’s minimum average partial
Horn and McArdle, 1992; Vandenberg and Lance, 2000). procedure criteria (map) showed a five-factor model,
While the configural invariance is satisfied, the next step the two vss1 and vss2 indicated a one-factor and a two-
is to set the factor loadings to be equal across the two age factor solution, respectively. The Kaiser’s greater than 1
groups. This is generally referred to as metric invariance. rule indicated a six-factor solution with a moderately high
Metric invariance is built on configural invariance by 61.57% cumulative percentage explained (Table A2). In
requiring that in addition to the equality of the structural summary, both the six-factor and seven-factor models

Volume 8 Issue 1 (2022) 4 https://doi.org/10.36922/ijps.v8i1.1285