Search

In regression analysis with heteroscedastic and/or correlated errors, the usual assumption is that the covariance matrix Sigma of the errors is completely specified, except perhaps for a scalar multiplier. This condition is relaxed in this paper by assuming only that Sigma has a certain pattern; for example, that Sigma is diagonal or partitionable into a diagonal matrix of sub-matrices. The method used for estimating Sigma is the standard procedure of equating certain quadratic forms of the observations (in this case, squares and products of residuals from regression) to their expectations, and solving for the unknown variances and covariances. A numerical example illustrates the method. [ABSTRACT FROM AUTHOR]

In the paper "Sources of Productivity Increases in the U.S. Passenger Airline Industry," a log-linear cross-section function was estimated with several independent variables of the "dummy" or "shift" form. In this type of estimating procedure, the dummy variable is interpreted as a shift in the intercept due to some discrete event that is thought to influence the dependent variable. However, the intercept of the estimated linear transformed equation is a multiplicative factor in the equation before transformation. The nontransformed equation, of course, has no intercept. For this reason, the impact of technical change on cost cannot be determined from the equation directly. Using some values for each of the continuous independent variables, use the equation to calculate cost. In this case the mean values of the logarithm were used. If the value of each of the dummy variables is zero, then this calculated cost is interpreted as the expected annual cost of running a propeller driven plane for the sample geometric seat miles, fuel-cost, etc. Replicate this procedure with the addition of a value of unity for the first dummy variable. This is interpreted as the expected annual cost of running a turboprop plane for the sample geometric mean values of the continuous independent variables.