Longitudinal models for two time points and different constructs?

[Evelyn Mare]

Dear all,

I have a general question I have been wondering about. I have a mini-panel with two waves. Disregarding covariates for a moment, I want to assess if job stress in wave 1 affects the satisfaction in wave 2.

The most intuitive approach for me would now be to regress the satisfaction at wave2 on the reported stress in wave 1, while controlling for the "baseline" satisfaction in wave1. However I was wondering if there is not a way to test that with a more longitudinal model.

Two issues worry me when doing so:

-1.) I have only two time-points (2 waves)
-2.) The two measures for satisfaction differ across those two waves (different range and slightly different question).

Is there still a way to estimate a "true" longitudinal model, or to assess the development in stress more directly?

Cheers,

Evelyn

[Phil Bromiley]

You'll increase your chances of a helpful answer by following the FAQ on asking questions - provide Stata code in code delimiters, readable Stata output, and sample data using dataex.
You present several problems. Normally, we would think of this as a dynamic panel model (i.e., a model where we include y(t-1) on the rhs) but that requires recognizing that the lagged satisfaction may be endogenous (see xtivregress, xtdpdml, and the xtabond related estimators). If you have some truly exogenous variables that should not influence satisfaction, you may be able to estimate this as a dynamic model.
Measuring things differently over time adds noise but I suspect the basic statistical problems remain.