Skip to main content

Table 4 Parsimony-adjusted goodness of fit for 4 SEM models of CFA and causal cascade pathways

From: Psychometric characteristics of the chronic Otitis media questionnaire 12 (COMQ – 12): stability of factor structure and replicability shown by the Serbian version

 

Chi-sq

df

RMSEA

AIC

N better models found in 300 k permutations

CFA1

59.934

47

0.068

145.934

88

CFA2

64.037

49

0.072

146.037

24

CFA3

78.008

51

0.095

156.008

11

SEM

182.394

87

0.136

278.394

2

  1. This table reflects the deletion of marginally loading items, and the addition of variables and links for one SEM (as in the Figure) to supplementing the CFA with construct validity links from disease activity, HL, and SF-36. Significance of chi-squared is uninformative (as data always differ highly significantly from a model’s predictions), but chi-squared values are the basis of calculating other indices of model fit. Lower df for CFA1 means fewer degrees of freedom in the residual error, as more df are being ‘spent’ on links for the sake of goodness of fit (GoF), making a more complicated, but less parsimonious model (CFA1&2 compared to CFA3). The RMSEA (root mean square error approximation) is the most generally used index [18] of GoF. The lowest (best) value here of RMSEA, 0.068, represents good fit, verging on excellent (criterion RMSEA <0.05) and all loading links are very highly significant except the weaker cross-loadings (candidates for elimination). This is for the first CFA with four items (Qs 5, 6, 7 and 9) cross-loading on a second factor, plus all three of the inter-factor correlation links realised. These links are next reduced to two (CFA2), and then to one (CFA3, this also omitting the non-significant link between the hearing and activities/uptake factors). As determined by the quality of the EFA solution, all these CFAs are highly parsimonious, with the parsimony-weighted Akaike Information Criterion undercutting the reference ‘saturated’ value of 180. The SEM (last row) adding the three presumably causal links from disease activity, HL and the SF-36 mental well-being factor re-derived for this sample comes close to absolute parsimony (Akaike AIC 278.394 only slightly higher than the target saturated value of 270) but with cross-loading of Question 5 on ear problems marginal at p = 0.085
Back to article page

Health and Quality of Life Outcomes

ISSN: 1477-7525

Contact us