Table I.
Ordinal Probit MIMIC Model Moderated by Age and Disability.a
| Indirect Effects on Latent Construct: | Direct Effects to Latent Y Indictors:b | |||
|---|---|---|---|---|
| Subjective Patient Financial Strain | Insurance/Finances for Future Health Needs | Difficulty Paying Bills | Worry about Healthcare Expenses | |
| X Variables | b (S.E.) | b (S.E.) | b (S.E.) | b (S.E.) |
| Sex | −.020 (.149) | |||
| Education | .0035 (.047) | |||
| Marital | .289 (.322) | −.190 (.332) | .115 (.313) | |
| On-leave/Unemployed | .380 (.238) | −.653 (.277) | ||
| Employed | .071 (.192) | −.495 (.269) | ||
| Age | −.001 (.009) | .020 (.011) | −.045 ***** c (.012) | |
| Missing 1st Subcomposite | −.049 (.146) | |||
| OFFS | .304 ** (.134) | .141 ***** c,d (.148) | .186 ***** c (.137) | |
| Disability Days | .022 (.139) | |||
| Age2 | −.0047 (.00042) | −.00054 ** c (.00061) | ||
| OFFS2 | −.037 (.061) | −.124 **** (.071) | .041 *** c,d (.088) | |
| Age*OFFS | .002 (.007) | .016 * c (.009) | −.013 (.007) | |
| Age*Disability Days | b = 0 in final model | .022 (.019) | ||
| Disability Days*OFFS | b = 0 in final model | −.274 (.170) | ||
| Age*Disability Days* OFFS | b = 0 in final modele | −.035 **** (.012) | ||
NOTE: MIMIC = Multiple Indicators-Multiple Causes Regression; OFFS = Objective Family Financial Stress Index
n = 267. Fit of the data to the MIMIC model is good based on various indices: Chi-Square (23 df) = 16.074 at p = .8521; Descriptive Fit Value = .0030 < 1.5; and Goodness of Fit = .9958. Although the Root Mean Square Residual = .129, which exceeds the value of .08 recommended by Hu and Bentler [23] for adequate model fit, the MIMIC model moderated only by age (and not disability) has a Root Mean Square Residual = .079. Thus, the deterioration in residuals fit can be attributed to a) the information loss when software memory limitations necessitated the collapse at the mean of the continuous disability days variable into a dummy variable; and b) the specification loss of the curvilinear term.
The respective lambda loadings, representing acceptable fit of the latent y-indicators to the unidimensional latent construct, are 1.000, .6011, and .8915.
The total effect (i.e., direct plus indirect effects) is statistically significant at p < .05 when this unique, direct effect is added to the shared, indirect effect of the latent construct.
Not significant in the MIMIC model based only on observations with non-missing data within the first subcomposite of OFFS (n = 200). Missing data occurs within two of the variables (family income and out-of-pocket costs) used to calculate the first subcomposite. Interpretations of the follow-up simple slopes plots (Figures 2–4) are unaffected.
This final model is more valid than the one that also estimates the indirect effects since disability days co-moderates only one of the shared and unique effect relationships. Statistical inferences are unaffected.
p < .05
p < .02
p < .01
p < .005.