1. See the articles in Engle, and Granger (1991) for a discussion of these issues in the single time series context.
2. Comparative Democracy: The Economic Development Thesis
3. The choice of divisor here is irrelevant, since PWLS depends only on relative weights. These relative weights are completely determined by the numerator. In particular, none of the disadvantages shown by PWLS in the Monte Carlo experiments is in any manner a consequence of our choice of divisor in equation 21.
4. While BLB indicate that they corrected their estimates for panel heteroskedasticity, our reanalysis indicates that they did not do so. Burkhart, and Lewis-Beck (1994, 905) state that “heteroskedasticity was corrected … with the ‘force homoskedastic model’ option in Microcrunch … .” This confusingly named option does not produce weighted least squares. The Microcrunch User's Guide states: “The normal specification for the GLS-ARMA model is heteroskedastic error (i.e., the estimator includes a weighted least squares analogue)… . Users may override that default by specifying ‘Y’ to a Homoskedastic Error prompt… .” (Atunes and Stimson 1988, 47).