Laserfiche WebLink
<br />White's general test for heteroskedasticity was performed on the final regression <br />model. A positive results indicates large variances. Weighted least squares could be used to <br />correct for this, but there were several possible variables that might be appropriate as the <br />weight. However, only one variable can be used as the weight. Thus to avoid this problem <br />we corrected for heteroskedasticity using White's Heterosk'ed~,ticity-Consistent Standard <br />Errors and Covariance (Ulien et al. 1994), which reduces the standard errors and makes the <br />t-statistics more accurate. Table n reports the t-statistics after the standard errors are <br />corrected. R2's for each of the regressions were calculated by an ordinary least squares <br />estimation procedure using the transformation of the independent variable by lambda. <br /> <br />STATISTICAL AND PROPERTY VALUE RESULTS <br /> <br />Statistical Results <br />The final regression equation contains three property variables that have minimal <br />correlations among each other, three neighborhood or locational variables, three demographic <br />variables and two stream improvement variables, one from each restoration "package". This <br />can be found in the first column of Table n. Stream restoration measures in Tables I and n <br />are simply dummy variables for whether or not that measure was performed in the project. <br />For all the regression results in Table n, improvement size, lot size, garage, per capita <br />income, travel time to work, mean age, unemployment rate and lambda are significant at <br />least at the five percent level. Coefficients on the non-stream variables remain stable even <br />when different stream variables are used in the regression. Creek distance is significant at <br />least at the ten percent level. The stability of restoration measures is indicated by the <br /> <br />12 <br /> <br />. <br /> <br />. <br /> <br />. <br />