reported. For example, Landon and Smith (1997)
examine five different functions choosing the reciprocal
square root form, Oczkowski (1994) reports
a log-linear form, and Nerlove (1995) compares
log-linear, log-log and Box–Cox transformations.
The log-linear form has been applied in a number
of published studies, including Oczkowski (1994,
2001), Nerlove (1995), and Combris
et al
. (1997).
In our case, following Oczkowski, the results of
applying a RESET-test to the linear, log-linear, and
log-log functional forms lead us to prefer the loglinear
specification, with log(Price) as the dependent
variable (see Table 3). An examination of the
correlation matrices for the coefficient estimates
suggests that no serious degree of multicollinearity
is present in the data. Moreover, we take note of
the point stressed by Oczkowski (2001) that serious
correlation between a single measure quality
regressor and the error term would point to
measurement errors and lead to inconsistent OLS
(ordinary least squares) estimates. To test for that,
we conducted a standard Hausman test using the
average of the quality ratings for each producer
label as an instrumental variables.
6
As is clear from