efficiency measures. Thus the Debreu (1951)-Farrell
(1957) efficiency measures which we calculate may
overstate the inclusive Koopmans (1951) definition
of efficiency. We address the empirical significance
of slacks in Section 5.
3.2. The second step regression analysis: rationale
and method
After calculating efficiencies, it is natural to seek
to explain their variation. Earlier studies have sought
to explain variation in calculated efficiencies by
means of a second step regression, in which the
calculated efficiencies are regressed on a set of
exogenous variables using OLS or, because the efficiencies are censored variables, tobit methods. However a potentially serious shortcoming of this approach is that a part of the variation in the calculated
efficiencies can remain unaccounted for, ending up
mixing with the white noise error term and contaminating the estimated regression coefficients. Here we
adopt a different approach, in which the unexplained
part of efficiency variation is separated from the
white noise error term.
We specify the second step explanatory regression
as an SFA model, rather than as an OLS or tobit
model. In an SFA regression model the error term
contains two components, a normally distributed
white noise component, and a one-sided component,
which in this case captures that part of efficiency
variation which is not associated with the explanatory variables included in the model. This specification allows us to decompose variation in calculated
efficiencies into systematic and random parts. The
systematic part captures the effect of the exogenous
variables on calculated efficiency variation, and the
random part is captured by the one-sided error component. The entire variation in calculated efficiencies
is thereby assigned to systematic and random sources.
The second step SFA regression model is specified as
~St = O ( z f , ; / 3 ) exp( os ' - z/I), (3)
where ~/t is the vector of effficiencies calculated in
the first step using DEA, Z ft is a vector of exogenous variables, /3 is a vector of parameters to be
estimated, Vft is a symmetric white noise error component, and r f t > 0 is a random error component.
The systematic part of variation in calculated efficiencies is captured by
efficiency measures. Thus the Debreu (1951)-Farrell
(1957) efficiency measures which we calculate may
overstate the inclusive Koopmans (1951) definition
of efficiency. We address the empirical significance
of slacks in Section 5.
3.2. The second step regression analysis: rationale
and method
After calculating efficiencies, it is natural to seek
to explain their variation. Earlier studies have sought
to explain variation in calculated efficiencies by
means of a second step regression, in which the
calculated efficiencies are regressed on a set of
exogenous variables using OLS or, because the efficiencies are censored variables, tobit methods. However a potentially serious shortcoming of this approach is that a part of the variation in the calculated
efficiencies can remain unaccounted for, ending up
mixing with the white noise error term and contaminating the estimated regression coefficients. Here we
adopt a different approach, in which the unexplained
part of efficiency variation is separated from the
white noise error term.
We specify the second step explanatory regression
as an SFA model, rather than as an OLS or tobit
model. In an SFA regression model the error term
contains two components, a normally distributed
white noise component, and a one-sided component,
which in this case captures that part of efficiency
variation which is not associated with the explanatory variables included in the model. This specification allows us to decompose variation in calculated
efficiencies into systematic and random parts. The
systematic part captures the effect of the exogenous
variables on calculated efficiency variation, and the
random part is captured by the one-sided error component. The entire variation in calculated efficiencies
is thereby assigned to systematic and random sources.
The second step SFA regression model is specified as
~St = O ( z f , ; / 3 ) exp( os ' - z/I), (3)
where ~/t is the vector of effficiencies calculated in
the first step using DEA, Z ft is a vector of exogenous variables, /3 is a vector of parameters to be
estimated, Vft is a symmetric white noise error component, and r f t > 0 is a random error component.
The systematic part of variation in calculated efficiencies is captured by
การแปล กรุณารอสักครู่..