To estimate (2) for the different levels of industry aggregation we use the random-
effects estimator (RE) since we do not assume that the exogenous variables are strate-
gically correlated at the industry level. The Breusch and Pagan test for unobserved
heterogeneity as well as the Hausman test statistically confirm the use of a random ef-
fects model. Additional tests for consistency of the estimated error matrix, the modified
Wald test for groupwise heteroscedasticity as well as the Wooldridge test for autocorre-
lation show that the traditional error terms are indeed driven by a heteroscedastic error
structure as well as autocorrelation. Thus, we use the robust Huber / White / Sandwich
estimator for all regressions. With this procedure, also taking possible outliers into
account, we are able to assure the consistency and the comparability of the estimation
results. The results are presented in Table 2.
Table 2: Effects on the Indistries’ High Skill Labor Ratio
|
Whole |
Manufacturing |
Disaggregated | ||
|
Economy |
Sector |
Industry Levels | ||
|
ω |
-0.0263 (-0.34) |
-0.0424 (-0.54) |
-0.0623 (-0.78) |
-0.0498 (-0.89) |
|
Q |
1.26e-06 |
1.43e-06 |
-2.87e-06 |
-1.62e-06 |
|
(0.51) |
(0.57) |
(-1.02) |
(-0.81) | |
|
VS |
1.8137 |
2.1545 |
- |
- |
|
i VS (Y) |
- |
- |
-1.1867 (-0.44) |
- |
|
i VS (X) |
- |
- |
9.1045*** (2.62) |
- |
|
i VS (Y → L) |
- |
- |
- |
-2.9060 (-1.16) |
|
i VS (Y → H) |
- |
- |
- |
-9.3228** |
|
i VS (X → L) |
- |
- |
- |
28.6351** (2.21) |
|
i VS (X → H) |
- |
- |
- |
7.1781*** (2.54) |
|
t |
0.0988*** (7.00) |
0.1093*** (7.05) |
0.1072*** (7.03) |
0.1183*** (11.49) |
|
cons. |
0.4782 (1.39) |
0.3406 |
0.4760 (1.13) |
0.5988** (2.23) |
|
Obs. |
190 |
165 |
165 |
330 |
|
Groups |
25 |
20 |
20 |
40 |
|
Prob > chi2 |
0.0000 |
0.0000 |
0.0000 |
0.0000 |
Endogenous variable: within industries’ high skill labor ratio (H/L); (t-Statistics in parantheses);
* / ** / *** indicates significance at 10 / 5 / 1 percent
12