Regression Estimates of UC Benefit Generosity
Table 3.3 of Chapter 3 displays two sets of estimates of the index of UC benefit generosity, G. The estimates in column (5) are simply means based on data from the 1990s (or for as many years as available). The estimates in column (6) were derived from regression equations. Table C.1 provides details of these regressions. It displays estimated slopes, t-ratios, goodness-of-fit measures (adjusted R2s and standard errors), average cost rates, average unemployment rates, and the years of the estimation periods and the number of observations.
In developing regression-based estimates of G, it was decided to fit homogeneous regressions where the unemployment rate (TUR) explained the average cost rate (B) with no other explanatory variables and no intercept. Thus, the estimated coefficients show overall estimates of G with no controls for possible changes in statutory UC provisions or other factors that may have occurred during the estimation periods. Use of this specification means that the adjusted R2 can be substantially negative, indicating that the regression performs worse than estimating simply the mean of the dependent variable (B). In fact, four fits are so poor that the adjusted R2 falls below — 0.25.
Note also that several regressions are based on very few data points. Eight regressions had fewer than 10 observations, 4 from CEE- FSU countries.