NETCOM =f3(WORKPL, POPDEN, TAXBAS, UNEMP1),
with positive impacts expected for all four variables.
Estimating these three specifications using the well known Ordinary Least Squares linear
regression method and omitting variables which did not meet the significance level of 10 percent,
the results in Table 1 were derived.
(Table 1)
For the OUTCOM specification, the coefficients for PSH1766, IPHOUS and WORKPL behaves
as expected. The coefficient for UNEMP is found to be negative, indicating that municipalities
with high unemployment have structural problems related to geographical and professional
immobility of the workforce. The LM-L test for an omitted spatial autoregressive structure as
well as the LM-E test for a spatial Durbin proces and the LM-EL test for a combined process
strongly indicates the presence of local spatial patterns. Especially, the high LM-E value indicates
a strong spatial clustering pattern in the fit of modelled behaviour to actual behaviour. Combined
with the high LM-L value, the presence of local industrial clustering is an evident cause to
commuting.
For the INCOM specification, the coefficients for WORKPL, POPDEN and TAXBAS behave
as expected. However, it is not possible to trace a significant impact from unemployment in
neighbourhood municipalities, UNEMP1. The high values of all three LM tests for spatial
patterns strongly indicates the importance of local industrial clustering as a factor attracting
commuters.
For the NETCOM specification, the coefficients for UNEMP1 and WORKPL were as expected,
whereas no significant impact were found from POPDEN and TAXBAS. A straightforward
interpretation of these findings is that in-commuting and the (surplus) net-commuting are
conceptually different: It is clear that people commute to urban areas where high incomes may
be earned. But this in-commuting is to an important degree outweighted by out-commuting of
people who are not able to fit into these jobs. Consequently, the impact of urbanisation degree
and income levels are washed out when fitted to net-commuting. The LM tests for the NETCOM
specification are very high, indicating a strong presence of local industrial patterns attracting
commuters.
Combining the spatial autoregressive process with the causal model, the results in Table 2 were
estimated.
(Table 2)
A combination of the spatial Durbin process and the causal gave the estimates in Table 3.
(Table 3)
Finally, a causal model combined with a spatial Durbin and an autoregressive specification were
estimated. However, as the coefficient for the autoregressive terms were found to be strongly
insignificant in all three models, these results are not found interesting and therefore not reported.