Dear Statalist community,
I have learned of my mistakes and learned to interprete interaction effects.
I performed a simple OLS regression with a two-way interaction effect, x1 and x2 both are dummy variables.
The interpretation of x1 and x2 estimates is the differences to firms who have the other attribute to zero:
Firms who have x1 produce 1.3 less Y1 than firms who have x1=0.
Firms who have x2 produce more Y1 than firms who have x2=0.
Firms whi have x1 and x2 set to one, produce more Y than firms with x1=0 and x2=1, or vice versa.
My question is now, does this suffice to reliably state that firms who have x2=1 provide more Y than firms with x2=0?
If I drop variables with x2=1 (0) and then look at means, I get the following result. Is it realiable to just calcualte differences between these two means, i.e. (-2.45- (2.51)?
How could I then show what influence the interaction with x1 has?
I have learned of my mistakes and learned to interprete interaction effects.
I performed a simple OLS regression with a two-way interaction effect, x1 and x2 both are dummy variables.
Code:
. reghdfe Y x1##x2 c1 c2 c3 c4 c5 c6 c7 c8 c9, absorb(FIRM Year) cluster(Industry)
(dropped 27 singleton observations)
(MWFE estimator converged in 7 iterations)
HDFE Linear regression Number of obs = 487
Absorbing 2 HDFE groups F( 12, 21) = 63.45
Statistics robust to heteroskedasticity Prob > F = 0.0000
R-squared = 0.4268
Adj R-squared = 0.1758
Within R-sq. = 0.0595
Number of clusters (Industry) = 22 Root MSE = 0.7048
(Std. err. adjusted for 22 clusters in Industry)
------------------------------------------------------------------------------
| Robust
Y | Coefficient std. err. t P>|t| [95% conf. interval]
-------------+----------------------------------------------------------------
1.x1 | -1.305025 .0995242 -13.11 0.000 -1.511997 -1.098053
1.x2 | .3349848 .1583003 2.12 0.046 .0057813 .6641883
|
x1#x2 |
1 1 | 1.501429 .3299879 4.55 0.000 .8151814 2.187676
|
c1 | -.0236591 .0586029 -0.40 0.691 -.1455305 .0982123
c2 | -.0256179 .0314577 -0.81 0.425 -.0910377 .0398018
c3 | .1965687 .1940449 1.01 0.323 -.2069697 .6001071
c4 | .0366049 .0571484 0.64 0.529 -.0822417 .1554514
c5 | .0089514 .0059729 1.50 0.149 -.0034699 .0213726
c6 | -.5765208 .5296322 -1.09 0.289 -1.677951 .5249097
c7 | .0045407 .0007158 6.34 0.000 .0030521 .0060293
c8 | .0104786 .0261328 0.40 0.692 -.0438676 .0648248
c9 | -.3780107 .2715036 -1.39 0.178 -.9426333 .186612
_cons | -5.923346 2.962192 -2.00 0.059 -12.08356 .2368706
Firms who have x1 produce 1.3 less Y1 than firms who have x1=0.
Firms who have x2 produce more Y1 than firms who have x2=0.
Firms whi have x1 and x2 set to one, produce more Y than firms with x1=0 and x2=1, or vice versa.
My question is now, does this suffice to reliably state that firms who have x2=1 provide more Y than firms with x2=0?
If I drop variables with x2=1 (0) and then look at means, I get the following result. Is it realiable to just calcualte differences between these two means, i.e. (-2.45- (2.51)?
How could I then show what influence the interaction with x1 has?
Code:
. drop if x2==1
(381 observations deleted)
. univar Y
-------------- Quantiles --------------
Variable n Mean S.D. Min .25 Mdn .75 Max
-------------------------------------------------------------------------------
Y 133 -2.45 0.81 -4.41 -2.91 -2.34 -1.92 -0.59
. drop if x2==0
(133 observations deleted)
.
. univar Y
-------------- Quantiles --------------
Variable n Mean S.D. Min .25 Mdn .75 Max
-------------------------------------------------------------------------------
Y 381 -2.51 0.77 -4.53 -2.98 -2.43 -2.00 -0.56
Comment