The exclusion restriction for this IV model requires that engine displacement does not affect CO2 except through
engine displacement. This appears to be satisfied as long as CO2 emissions depend only on fuel consumption. We
estimated the IV model for our four specifications. Reported standard errors are again clustered at the car producer
level and were computed according to the results in Shore-Sheppard (1996).
The results are reported in columns through of Table . Compared with the OLS estimates, the IV coefficients
on fuel consumption are about one log point higher and never below 0.99. This is the expected result; the
difference between IV and OLS estimates is consistent with attenuation bias in OLS due to measurement error. Tests
whether = 0.99 do not reject in specifications and but do in and .
A more dramatic change occurs in the case of the coefficients on the interaction between fuel consumption and
the dummy for diesel engines. The estimates of fall by approximately two log points and only in two case are
statistically significant at the 5 percent level (although they are all significant at the 10 percent level), which is a
substantial change in comparison with the OLS estimates. This, again, is an expected result if the attenuation bias
was more severe in the case of diesel engine cars. Note also that the coefficient on the dummy for diesel engine
suggests that diesel engines produce between 13.5 to 14.6 log points more CO2 per unit of fuel compared to fuel
engines, the proportional difference in CO2 emissions per unit off between diesel and petrol is 0.144 (in logs we
have log.
To summarize, it is reasonable to believe the elasticity of CO2 emissions with respect to fuel consumption is between
0.99 and 1, and this holds for petrol as well as, although less precisely, for diesel engines. This implies that
increasing fuel consumption by 1 percent result in close to one 1 increase of CO2 emissions.
The exclusion restriction for this IV model requires that engine displacement does not affect CO2 except throughengine displacement. This appears to be satisfied as long as CO2 emissions depend only on fuel consumption. Weestimated the IV model for our four specifications. Reported standard errors are again clustered at the car producerlevel and were computed according to the results in Shore-Sheppard (1996).The results are reported in columns through of Table . Compared with the OLS estimates, the IV coefficientson fuel consumption are about one log point higher and never below 0.99. This is the expected result; thedifference between IV and OLS estimates is consistent with attenuation bias in OLS due to measurement error. Testswhether = 0.99 do not reject in specifications and but do in and .A more dramatic change occurs in the case of the coefficients on the interaction between fuel consumption andthe dummy for diesel engines. The estimates of fall by approximately two log points and only in two case arestatistically significant at the 5 percent level (although they are all significant at the 10 percent level), which is asubstantial change in comparison with the OLS estimates. This, again, is an expected result if the attenuation biaswas more severe in the case of diesel engine cars. Note also that the coefficient on the dummy for diesel enginesuggests that diesel engines produce between 13.5 to 14.6 log points more CO2 per unit of fuel compared to fuelengines, the proportional difference in CO2 emissions per unit off between diesel and petrol is 0.144 (in logs we
have log.
To summarize, it is reasonable to believe the elasticity of CO2 emissions with respect to fuel consumption is between
0.99 and 1, and this holds for petrol as well as, although less precisely, for diesel engines. This implies that
increasing fuel consumption by 1 percent result in close to one 1 increase of CO2 emissions.
การแปล กรุณารอสักครู่..