a variable. This may have been a problem,as this coefficient measures the total number of the economically active population in agriculture;here,we see a limitation of our data,which has affected our probit estimation.
R&D is statistically significant in all columns;this variable is positive in columns 3 and 9 as well,which agrees with our intuition—higher levels of expenditure on R&D leads towards the adoption of biofuel policies (NewGDP was used in this regression, due to the high correlation between GDP and R&D).The elasticity of R&D in column 3 is larger than those of Tarrifs and FeedstockP (Table A6, Appendix 6). This agrees with Berg (2004), who opines that having appropriate technologies is a fundamental factor contributing to successful biofuel production.The negative sign in column 6 is puzzling; R&D becomes positive in the probit model after PopInAgric in column7 is taken out.This may be another indication of the difficulty of utilising PopInAgric in the probit estimation.The significance of PopInAgric actually increases in column 3,compared to column 2.All other significance levels decrease in theOLS model.This is likely due to the lack of sufficient and consistent R&D data,which considerably minimised our sample size,which must be considered.Observing that the sample size decreases significantly,and that our pseudo R2 equals one in columns 7,8, and 10,indicates that this probit
regression may not be completely reliable.
Contrasting our intuition,Oil is negative and significant,while we expected this variable to be positive.In fact,Oil is only positive when GDP and R&D arere moved,which is demonstrated in columns 5 and 11.By dropping these variables ,though,only Oil and Tariffs remain statistically significant.We will see how our oil imports variable reacts when split into OECD and Non-OECD groups.
Though Almirall et al.(2010) find that a decrease in land leads to increase crop prices,the correlation of feedstock prices and biofuel policies in our sample is 0.01659. Thus,there is not a
correlation problem between these two variables.Furthermore,
regressing BioPol on FeedstockP leaves BioPol statistically in significant, as is demonstrated in Table 2.