I run the following treatment effect regression:
where y1 is the outcome,
ctreat is the treatment variable that is zero for the control group but continuous for the treatment group,
y0 is the outcome variable at baseline
district is the district (location) indicator
cid is the cluster identifier
The output is as follows
I want to compare the treatment effect at specific thresholds of the treatment variable (a kind of dose response analysis) using the margins command. I did this at the 25th, 50th and 75th percentiles of the treatment variable as follows:
The results are as follows:
My question is why the t-statistics and the p-values are exactly the same for all Contrasts? Did I do something wrong?
Code:
qui regress y1 ctreat y0 i.district, cluster(cid)
where y1 is the outcome,
ctreat is the treatment variable that is zero for the control group but continuous for the treatment group,
y0 is the outcome variable at baseline
district is the district (location) indicator
cid is the cluster identifier
The output is as follows
Code:
Linear regression Number of obs = 2,663
F(14, 134) = 85.44
Prob > F = 0.0000
R-squared = 0.4692
Root MSE = 24.35
(Std. err. adjusted for 135 clusters in cid)
------------------------------------------------------------------------------
| Robust
y1 | Coefficient std. err. t P>|t| [95% conf. interval]
-------------+----------------------------------------------------------------
ctreat | .0027545 .0009769 2.82 0.006 .0008224 .0046865
y0 | .0183763 .058014 0.32 0.752 -.0963652 .1331179
|
district |
2 | 6.233628 4.117701 1.51 0.132 -1.910467 14.37772
3 | 27.25729 5.079016 5.37 0.000 17.21188 37.3027
4 | 13.79617 8.248232 1.67 0.097 -2.517393 30.10974
5 | 10.68128 5.32056 2.01 0.047 .1581417 21.20442
6 | 1.774933 5.046554 0.35 0.726 -8.206271 11.75614
7 | -12.30428 5.431784 -2.27 0.025 -23.0474 -1.561155
8 | -21.82611 4.653727 -4.69 0.000 -31.03037 -12.62185
9 | -41.36124 3.739212 -11.06 0.000 -48.75675 -33.96573
10 | -4.462939 7.175909 -0.62 0.535 -18.65564 9.729757
11 | -14.43221 3.751938 -3.85 0.000 -21.85288 -7.011527
12 | -55.43812 3.669914 -15.11 0.000 -62.69657 -48.17967
13 | -1.066209 4.157414 -0.26 0.798 -9.28885 7.156432
|
_cons | 95.93494 5.71724 16.78 0.000 84.62724 107.2426
------------------------------------------------------------------------------
Code:
g d=ctreat if ctreat>0 _pctile d, p(25 50 75) margins, at(ctreat = (0 `r(r1)' `r(r2)' `r(r3)')) pwcompare(pv)
The results are as follows:
Code:
Pairwise comparisons of predictive margins Number of obs = 2,663
Model VCE: Robust
Expression: Linear prediction, predict()
1._at: ctreat = 0
2._at: ctreat = 354.7357
3._at: ctreat = 603.0507
4._at: ctreat = 1064.207
-----------------------------------------------------
| Delta-method Unadjusted
| Contrast std. err. t P>|t|
-------------+---------------------------------------
_at |
2 vs 1 | .9771054 .346526 2.82 0.006
3 vs 1 | 1.661079 .5890943 2.82 0.006
4 vs 1 | 2.931316 1.039578 2.82 0.006
3 vs 2 | .6839738 .2425682 2.82 0.006
4 vs 2 | 1.954211 .6930521 2.82 0.006
4 vs 3 | 1.270237 .4504838 2.82 0.006
-----------------------------------------------------
My question is why the t-statistics and the p-values are exactly the same for all Contrasts? Did I do something wrong?
