NYC-WTC_000187725

NYC 9/11 Public Portal Document ^^sa<ei3s»-

•l

63

The econometric technique of multivariate regression analysis was applied to the sample of sales parcels stratified by building class, utilizing a stepwise least squares method to systematically estimate the best fitting parameters of each equation.Because the method incorporates statistical controls within the model, through use of cross-sectional data and a wide array of site, neighborhood, access and transit related independent variables, the need for a control area is negated. The multivariate regression approach allows us to determine the positive or negative effect of each independent variable, and the significance of these effects on the dependent variable, the unit price of land. NYSTA is stratified by separate equations specified for each building class and, separately, for vacant land by each transit policy variable. The equations and graphic representations are presented in the next section of this chapter. Statistical tests of modeling results permit an overall assessment of each equation. Because NYSTA is based upon cross-sectional rather than time series data, goodness of fit results as reflected in a high value are not to be expected. Although this suggests that the model does not fully predict the value of land as a function of the specified independent variables, it does alter its ability to isolate the effect of transit accessibility variables on land value. f

Specification of the Equations In developing separate equations by building class for NYSTA, the specification was guided by various forms of hedonic price models but based, after an iterative process of stepwise regression analysis, upon statistically considerations. The functional form of the basic regression model is:

y = Po-^o * Pi^i * • ■ * PfAp + £

where Y is the vector of dependent variable values, X is the matrix containing the values of the independent variables, 6 is the vector of parameters, and e is the vector of error components. The specification may be linear, logarithmic, log-linear, or quadratic. Although a wide range of explanatory variables were searched for values that minimized the sum of squared residuals, only some twenty of the sixty available variables were utilized in the model. For those calibrated in the model. Table 3.2 presents the name of each variable, its type, and a brief definition.

stratification by borough or use of borough dummies might improve explanatory powers since relationships probably differ by borough;

CITY_21MC100_006662294

NYC-WTC_000187794