outcome paths), estimated for each of the three outcome constructs, (c) a three-occasion simple
model with within-time mediation paths and auto-regression paths, conducted separately for the
three pain outcomes, and (d) a fully recursive three-occasion model with all possible cross-time
paths estimated separately for the three pain outcome constructs. This final model was followed
by tests of nested reduced models to determine the best fit to the data.
Measurement Models
Estimation of baseline (Time 1) measurement models. At the Time 1, three separate
measurement models were estimated to analyze the relationships among pain, pain medications,
and each pain outcome construct. Separate measurement models were conducted for each pain
outcome: physical functioning, emotional functioning, and social functioning. The measurement
phase was necessary to evaluate the strength and fit of the proposed models. Model fit was
determined by evaluating several fit indices. A good model fit has a Chi-square value < 112 times
the degree of freedom and a non-significantp value. The literature indicates that the Chi-square
value is often inflated in a large sample analysis, resulting in p-values < .05. Thus, in a study
such as this in which there was a large sample size, model fit was not determined by the Chi-
square value alone. Additional incremental and absolute fit indices evaluated included the root
mean square error of approximation (RMSEA), normed fit index (NFI), relative fit index (RFI),
incremental fit index (IFI), Turker-Lewis index (TLI), and comparative fit index (CFI).
Incremental index values (e.g. NFI, RFI, IFI, TLI, & CFI) for a good fit were expected to be
above .9 on a scale of 0-1.0 (1.0 being a perfect fit) and absolute fit index values were expected
to be close to 0.00 (e.g., RMSEA less than .05).
Each Time 1 (baseline) measurement model was specified to indicate the proposed
relationship paths between the endogenous latent constructs (e.g., physical functioning,
emotional functioning, and social functioning), the endogenous indicator variables (e.g., physical
64