I am struggling with the interpretation of Treated x Post x Log(1+Z) with Z being a continuous positive variable.
My variable Treated takes 3 values (0 for control, 1 and 2) and Post takes 2 values (0, 1). Z is a continuous variable like distance (in km).
1- I understand that 1.treated#1.post and 2.treated#1.post are the partial effect of the treatment relative to the control group, 0.treated#1.post, with ln(1+Z)=0, i.e. Z=0
However, Z is stricly positive. Why aren't the coefficients omitted ? What does they mean ?
2- My variable Z increases (exogenously) after the treatment. How do I interpret 1.treated#1.post#c.lnZ and 2.treated#1.post#c.lnZ ? I am not sure what question it answers. Is it :
- "Is the effect of Z different because of the treatment?", or
- "What is the effect of Z being greater after the treatment?"
3- How do I understand the magnitude of 1.treated#1.post#c.lnZ and 2.treated#1.post#c.lnZ ?
One standard deviation of log(1+Z) (.2662474) increases my outcome variable Y by (0.26*30035=) 7809 on my first treated group ?
Any help would be very appreciated.
Here's a (silly) example with the auto data:
HTML Code:
use "C:\Program Files (x86)\Stata15\ado\base\a\auto.dta", clear gen treated=0 if mpg<18 replace treated=1 if mpg>=18 & mpg<24 replace treated=2 if mpg>=24 gen post=0 if rep78<=3 replace post=1 if rep78>3 gen lnZ=log(1+weight) sum lnZ Variable | Obs Mean Std. Dev. Min Max -------------+--------------------------------------------------------- lnZ | 74 7.979106 .2662474 7.473637 8.484877 r; t=0.00 11:15:39 . . reghdfe price ib0.treated##ib0.post##c.lnZ, noabsorb (MWFE estimator converged in 1 iterations) HDFE Linear regression Number of obs = 74 Absorbing 1 HDFE group F( 11, 62) = 8.02 Prob > F = 0.0000 R-squared = 0.5873 Adj R-squared = 0.5141 Within R-sq. = 0.5873 Root MSE = 2055.9924 ------------------------------------------------------------------------------------ price | Coef. Std. Err. t P>|t| [95% Conf. Interval] -------------------+---------------------------------------------------------------- treated | 1 | 121104.6 53046.34 2.28 0.026 15066.45 227142.7 2 | 156977.3 58501.97 2.68 0.009 40033.52 273921 | 1.post | 286430.3 67555.75 4.24 0.000 151388.3 421472.3 | treated#post | 1 1 | -246480.9 75195.96 -3.28 0.002 -396795.5 -96166.37 2 1 | -339319.7 83140.1 -4.08 0.000 -505514.4 -173125 | lnZ | 20706.8 5809.84 3.56 0.001 9093.091 32320.5 | treated#c.lnZ | 1 | -15026.52 6432.404 -2.34 0.023 -27884.72 -2168.331 2 | -19577 7199.596 -2.72 0.008 -33968.79 -5185.217 | post#c.lnZ | 1 | -34890.05 8200.045 -4.25 0.000 -51281.71 -18498.4 | treated#post#c.lnZ | 1 1 | 30035.15 9177.67 3.27 0.002 11689.25 48381.05 2 1 | 41973.65 10325.43 4.07 0.000 21333.41 62613.89 | _cons | -161752.2 48150.94 -3.36 0.001 -258004.6 -65499.83 ------------------------------------------------------------------------------------
0 Response to Interpreting triple interaction term with continuous variable
Post a Comment