A Stata package to "wreck" any regression with a variable being transformed according to log(+1) or IHS (when there are zeros or negative values in the data) by applying a scaling factor on the transformed variable to achieve an arbitrary coefficient or semi-elasticity estimate.
- Stata
From Stata:
net install wreckitreg, from("https://raw.githubusercontent.com/tt-econ/wreckitreg/main/pkg")
In Stata, after installation:
. sysuse nlsw88, clear
(NLSW, 1988 extract)
. replace wage = 0 if wage < 2
(26 real changes made)
. wreckitreg wage tenure grade, coefvar(tenure) value(.03) absorb(age)
(setting technique to nr)
Iteration 0: f(p) = .00011345
Iteration 1: f(p) = .00011345
Iteration 2: f(p) = .00011345 (not concave)
Iteration 3: f(p) = .00011345 (not concave)
Iteration 4: f(p) = .00011345 (not concave)
(switching technique to dfp)
Iteration 5: f(p) = .00011345
Iteration 6: f(p) = .00011345
Iteration 7: f(p) = .00011345
Iteration 8: f(p) = .00011345
Iteration 9: f(p) = 6.240e-07
(switching technique to bfgs)
Iteration 10: f(p) = 1.379e-11
Iteration 11: f(p) = 9.497e-15
Iteration 12: f(p) = 1.384e-16
Iteration 13: f(p) = 2.626e-18
Note: Results may not be obtainable outside the range between the two semi-elasticity values be
> low due to machine precision:
Semi-Elasticity Value with a scale of 1.00000000e-10: .0193485116
Semi-Elasticity Value with a scale of 1.00000000e+44: .155239028
Scaling wage by a factor of .16653186 will yield a semi-elasticity with respect to tenure (eval
> uated at the means of covariates) of .03 for variable transformation: ln(1 + .1665319 * wage)
> . Resulting regression:
---------------------------------------------------------------------------------------------
(1) (2)
Coefficient Semi-Elasticity
---------------------------------------------------------------------------------------------
Job tenure (years) 0.0131*** 0.0300***
(11.57) (11.57)
Current grade completed 0.0473***
(19.35)
Constant 0.0728**
(2.24)
---------------------------------------------------------------------------------------------
t statistics in parentheses
* p<0.10, ** p<0.05, *** p<0.01
Theoretical limit cases:
As scale -> 0, semi-elasticity -> .0193485117215812
As scale -> infty, abs(semi-elasticity) -> infty
Note that the semi-elasticity estimate may not change monotonically with the scale.
See the help file for more: In Stata, type help wreckitreg after installation.
"When Are Estimates Independent of Measurement Units?" (Thakral and Tô 2023)
Ⓒ 2022–2023 Neil Thakral and Linh T. Tô