where U, denotes a disturbance term having standard properties and where the f, (j" = 0, Y,P) are constant across i. While there have been discussions in the literature regarding such matters as: (1) methods of measuring Hi, Y, and Pi [5; 10; 11]; (2) problems of aggregation and the relative merits of micro and metro data [12; 17; 19; 20] (3) simultaneous equations bias and the neglect of the supply side [18; 23] and (4) the specification errors created by omitting the price term and other variables [7; 15; 16; 24], virtually all estimates have been constructed from a variant of this model. The log-linear specification is convenient because it allows the income and price elasticities of housing expenditure (Pr, and P,) to be estimated by a direct application of ordinary least squares (OLS). However, as will be argued momentarily, equation (1) may have a questionable basis in theory for two reasons. First, measures of transport costs are frequently excluded as regressors in housing expenditure functions even though such measures are theoretically relevant variables.' Second, and this point will be stressed in the discussion to follow, income and price elasticities are likely to vary across observation units. As a result, existing estimates of these elasticities must be interpreted as complicated weighted averages of their underlying and observation specific counterparts. Both of these problems are of potential importance because accurate estimates of these income and price elasticities of expenditure are useful not only in developing housing policy but also in studying issues involving suburbanization and property taxation.