1) Time Series Processor (TSP) from TSP Intl. (California);
2) STATA from STATA Corp. (Texas);
3) LIMDEP from Econometrics Ltd. (NY University).
From the usability standpoint, it is important to stress that these packages have been developed in
the academic environment to meet special research needs, giving less weight to the user interface, which so
far is essentially a simple character interface. These packages are adopted in advanced university lectures
and learning to use them represents an important sunk cost for researchers; this gives the usability an
idiosyncratic element overriding the weight of objective differences.
As far as the availability of panel estimation commands is concerned, it is possible to verify an
overlapping of the three packages examined.
The algorithms required in this paper for the panel estimation, i.e. the fixed and random effects
estimators, can be easily interpreted as Generalized Least Squares (GLS) estimates; for this purpose it is
sufficient to evaluate the error variance-covariance matrix.
Even though it is possible to perform an interactive estimation, in order to preserve the ability to
reproduce the exercises, it has been decided to use programs written in the languages native to the
packages. The code relative to these three programs is shown at the end of Appendix 1. A simple visual
inspection of the code shows the similarity of the packages considered, as witnessed by the number of lines
of code and the output produced. In the use of the STATA package some statistical tests must be explicitly
requested.
The main goal of this comparison was to validate the numerical results. As can be checked from the
tables shown in Appendix 1, this goal has been fully reached. The numerical value of the estimated
coefficients for the fixed and random effects models have been checked by writing the estimation algorithms
using the Speakeasy-Modeleasy Plus environment, a package featuring a compact expression for matrix
processing.20 It is important to mention the fact that the output of STATA also presents a value for the
constant. It is, of course, irrelevant for the fixed effects estimator. In fact, after the slope estimation, this
constant has been computed with the following fοrmula a within = -ɪ- ∙ (yit - β within ■ Xit ) .
∑ Ti
i=1
20 Modeleasy Plus, EMCC (Palm Harbor, Florida)
10