Functional status, life-space mobility, and quality of life: a longitudinal mediation analysis

Separate models were fit for PCS and MCS primarily for the ease of presentation of results and because each variable was considered a separate end point. It is possible to estimate one complete model containing both PCS and MCS. Given that the PCS and MCS were originally constructed to be orthogonal, there should be little difference in the results based on separate models and results from a single model considering both PCS and MCS. Indeed, when considering the final models reported in the current analysis, there were no substantive differences in findings when comparing separate estimation models for PCS and MCS and a single, combined model.

The nature of the calculation of the LSC score limits the applicability of internal consistency reliability and thus Cronbach’s alpha is not reported for the LSC; as stated earlier, past research reports high levels of test–retest reliability for this measure.

As suggested by one anonymous reviewer, in addition to examining chi-square change and model fit statistics for assessing measurement invariance, one should consider whether any meaningful differences in the estimates of interest occur between models with constrained invariant parameters and those where parameters are freely estimated. In the present study, the mediated effects were highly similar between models with measurement invariance for functional status and those with non-invariance. For example, the total mediated effect estimate for the PCS Model 2 assuming invariant thresholds and loadings (see Table 2) was −0.88 (p < 0.0001); this same effect was −0.90 (p < 0.0001) in a model with non-invariance. An examination of the time-specific indirect effects revealed similar results; p values for these indirect effects differed in the third decimal place, if at all.