Common-Coefficient, Panel-Data, or Pooled RegressionsΒΆ

It fairly frequently happens that we have similar data sets collected at different times or in different places. We then may wish to estimate regressions that combine the data sets in various ways. The comcoef command in G7 is designed for this purpose. It allows one to specify a number “common coeficients” in the regressions that follow. The method may be explained best with an example. In the following example, “cps78” and “cps85” are data banks of from the Current Population Surveys of 1978 and 1985, respectively. The first has data on 550 individuals, the second on 534 individuals. We want to regress the logarithm of an individual’s wage, lnwage, on the individuals education, ed. We want the coefficient of lnwage to be the same in both samples, but the interecepts may be different in the two years. Here is what we do.

f one = 1
comcoef 1 4
hbk cps78
lim 1 550
r lnwage = ! ed, one
hbk cps85
lim 1 534
r lnwage = ! ed, one
do

In the comcoef command, the “1” tells the program to use the same coefficient for the first 1 variable(s) in all the following regressions. The “4” is the total number of variables, counting the dependent variable, in the combined or “pooled” regression. Once the comcoef command has been received, G7 accumulates regressions until the do command is reached.

Only a limited list of commands is permitted between the comcoef command and the do command. They are the bank, hbk, lim, f, add, pause, quit, and r commands. The quit command only quits comcoef, not G7.

The comcoef command can be used under the chow command and under the catch command. It also puts the estimated coefficients into the rcoef series in the workspace bank, as usual. It does not enter a series of predicted values into the workspace bank, so gr * will not work following a comcoef regression. Nor are the variable names shown on the display of the results, for in general the variables with a common coefficient may have different names while variables with different coefficients may have the same name, such as “one” in the above example.

Previous topic

Seemingly Unrelated Regressions

Next topic

Non-linear Regression

This Page