same across models and are of similar magnitude, depending on whether error
components are present. The key effects pertaining to investment cost and energy
savings have the expected positive and negative effect on utility, respectively, and
are highly significant. Because the standard deviations for the parameter distribu-
tions of these two coefficients are statistically insignificant (i.e. ui1 = ui2 =0∀i),
there is no empirical evidence for taste heterogeneity beyond the variation that
is captured by the interaction effects.
The results of these interaction terms must be interpreted with respect to
the coefficients of Cij and ∆Qij . For example, increasing energy consumption
is seen to attenuate both the negative effect of cost and the positive effect of
energy savings, which is consistent with the intuition that high energy-consuming
households are less responsive to changes in energy expenditures. Likewise, the
error components variants of the model indicate that access to information, as
measured by the relative availability of certified auditors within a 20 kilometer
radius of the household, has a dampening effect on the negative influence of
cost that is of roughly the same magnitude as energy consumption. Income,
on the other hand, exacerbates the negative influence of costs. An explanation
for this finding is not immediately forthcoming, other than to speculate that
wealthier households may be more aware of other, more profitable investment
opportunities than housing. Finally, households living in the eastern part of
Germany experience less disutility from the investment cost than their western
counterparts. This result is expected, since the East-German building sector
was in dire need of rehabilitation before reunification. This led in the 1990s
to an extensive wave of refurbishment on the territory of the former German
Democratic Republic.
Regarding the question of model fit, a comparison of the log-likelihoods sug-
gests that little is gained from the incorporation of heterogeneous preferences
using the random parameters logit, an unsurprising finding given the insignifi-
16