4.2 Results from the estimations
In order to complete the estimations two estimators were considered: (i) the 2SLS
estimator and (ii) the 3SLS estimator. To determine which to choose, the Hausman (1978)
specification test was performed16. In the context of comparing the 2SLS estimator with
the 3SLS estimator, the strategy is to see if 3SLS improves over 2SLS, (see e.g. Doan,
1996 or Greene, 1994). The 3SLS estimator is more efficient only in the presence of
correlation between the disturbances in the structural equations. Otherwise, 3SLS reduces
to 2SLS. Thus, the estimates of the 2SLS should be identical to the 3SLS if a hypothesis
that no correlation between the disturbances in the structural equations is true. This is H0
in the test and a rejection of it, hence, implies that we should use 3SLS. It will be evident
from the subsequent tables that H0 could be rejected in all cases.
Table 4.2. 3SLS estimations of Equations 13a and 13b, total employment in manufacturing and
producer services in 2000.
|
Variable |
Parameter |
Estimates |
Estimates (producer services)__________ |
|
Intercept_____________________________ |
a, δ________ |
-3.67 (-2.9)*___________ |
-4.86 (-4.32)*____________ |
|
Acc. producer services |
Φ1__________ |
0.06 (5.66)* |
- |
|
Acc. manufacturing |
Y1 |
- |
0.01 (0.66) |
|
Wage-sum |
Φ 2__________ |
0.00005 (3.45)* |
- |
|
Knowledge intensity |
Y 2 |
- |
52.98 (5.20)* |
|
Dummy urban regions |
Φ 3, Y3 |
0.85 (1.14) |
-2.14 (-1.55) |
|
Interaction variable (D*Pa ) |
Ф 4__________ |
0.12 (3.01)* |
- |
|
Interaction variable (D*Ma ) |
Y 4 |
- |
0.15 (2.8)* |
|
Adj. R2___________________________ |
- |
0.64______________ |
_________0.43_________________ |
|
No. of observations__________________ |
- |
81________________ |
_________81_________________ |
|
Hausman Specification test__________ |
- |
_________________51.00 (12.59)________________________ | |
*)denotes significance at the 0.05 level.
**)denotes significance at the 0.1 level.
***) t-values are presented within brackets. For the Hausman specification test, the figure within
brackets is the critical value at the 0.05 level.
Table 4.2 presents the results from a 3SLS estimation of the equation system in
Equation (13a) and (13b) over the total employment in manufacturing and the total
employment in producer services. It is evident that all the estimates have the expected
signs. Wage-sum has a positive impact on manufacturing employment and the average
knowledge intensity of the workforce has a positive impact on the producer service
employment. Since the interaction variables are significant and positive in both equations,
regional size-effects are present in both equations. More precisely, this implies that the
manufacturing sector respond more to the accessibility to producer services in urban
regions compared to non-urban regions. Likewise, the relationship between the
16 The test was performed in the RATS package and follows the standard procedure suggested in the
accompanying manual by Doan (1996).
14