used for all but the motel-hotel model in this study.3 It was felt the specified models did not severely violate any assumption of OLS. Multiple correlation coefficients were generally low, indi- cating the "independent variables" (in the variable specification section of this report) were nearly independent of each other. Area of store and area of restaurant may be correlated in some department stores. However, this correlation was offset by stores in the sample that did not have a restaurant. Size of restaurants and bars in motels and hotels may be related to motel size; but, again motels in the sample without these facilities lowered cor- relation. The general OLS model can be specified as follows: Wi= f(sZ, z21, z,, ..., z i) (A.4) where (zi, zi, ..., zmi) are m independent variables and Wi is the dependent variable. In the context of this study, Wi is the average amount of water purchased per month by the ith firm. Equation (A.4) could be expressed in the following linear form (for i=1, 2, ... n observations) : Wi = f o + + 822+ A 3 32z +P z3i + pfmz, + e6 (A.5) where 3= parameters designating the constant term (3o) and the slope coefficients (pS, k =1, 2, 3, .., m); E = the disturbance term. The assumption regarding this linear model include: 1. Ei is distributed normally, 2. E (eL) = 0, 3. E (e2,) = a2, and 4. E [iE] = 0, i = j. It is also assumed there is no multicollinearity and that the z, are fixed and can be measured without error.4 Models can also be used to depict curvilinear relations. Cur- vilinear equations were needed in this study because economies of scale and curvilinear price response were expected to exist. Several alternative equation forms can be used to represent cur- vilinear relations. 3Heteroskedasticity was present in the OLS models for the motel-hotel group. A generalized least square (GLS) procedure used to remedy the problem is discussed at the end of this Appendix. 4See Kmenta [14, pp. 202-203] or Johnston [13, pp. 121-123] for defini- tions and a review of these concepts.