Dear STATA gurus!
I have been reading this forum for days, and have yet to fully understand how psmatch2 can be interpreted. I will provide everything I have here. Thank y'all in advance.
I run the following command:
I get the following results:
Next, I look at ATT. Treated .0349 and controls .03070. There is a POSITIVE difference and it is significant.
"ffs" is my binary 0/1 variable. I then say that when the value is 1, the effect is stronger; otherwise, weaker.
So my first question: Is the above statement correct?
Next, "export_ipri" variable is my interaction variable (IV multiplied by moderator). So what I want to find is whether the effect of my moderator differs on the outcome (lateROA) depending on the value of "ffs."
So the second question: In psmatch2 do I need to include the IV as well? (the IV is "export" the moderator is "ipri").
The third question: Would be it right to argue that the effect of "export_ipri" on my outcome (lateROA) is stronger when "ffs" = 1 (since the ATT is significant and the difference is positive). If so, then would it be correct to interpret it in this way: the effect of "export_ipri" is negative (and significant) on lateROA and it is MORE negative (stronger effect) when "ffs" = 1?
The fourth question: Do I even care if anything is significant in that probit model? (in this case specifically "export_ipri")
Thank you so much for your help. I have been reading everything for days and have yet to find the answers to the above questions.
Happy Holidays!
I have been reading this forum for days, and have yet to fully understand how psmatch2 can be interpreted. I will provide everything I have here. Thank y'all in advance.
I run the following command:
Code:
psmatch2 ffs export_ipri year_control emplognew capitinv i.fyear i.gsector, outcome(lateROA) noreplacement
Code:
Probit regression Number of obs = 9,182
LR chi2(26) = 459.97
Prob > chi2 = 0.0000
Log likelihood = -2837.1257 Pseudo R2 = 0.0750
------------------------------------------------------------------------------
ffs | Coefficient Std. err. z P>|z| [95% conf. interval]
-------------+----------------------------------------------------------------
export_ipri | -.0168623 .0085528 -1.97 0.049 -.0336256 -.0000991
year_control | -.0049055 .0005325 -9.21 0.000 -.0059492 -.0038619
emplognew | -.0358342 .0120888 -2.96 0.003 -.0595279 -.0121405
capitinv | .1667146 .1296518 1.29 0.198 -.0873982 .4208273
|
fyear |
2008 | .027891 .1502858 0.19 0.853 -.2666638 .3224458
2009 | .0705865 .149435 0.47 0.637 -.2223007 .3634737
2010 | .0708849 .1466351 0.48 0.629 -.2165146 .3582844
2011 | .0984372 .126006 0.78 0.435 -.14853 .3454043
2012 | -.0208677 .1272739 -0.16 0.870 -.27032 .2285846
2013 | .039797 .1265287 0.31 0.753 -.2081947 .2877886
2014 | -.0275057 .1279527 -0.21 0.830 -.2782884 .2232769
2015 | -.0097454 .1280022 -0.08 0.939 -.260625 .2411343
2016 | -.10004 .1290941 -0.77 0.438 -.3530598 .1529798
2017 | -.1036191 .1292263 -0.80 0.423 -.3568979 .1496597
2018 | .1134629 .1204432 0.94 0.346 -.1226015 .3495273
2019 | .053735 .1229188 0.44 0.662 -.1871815 .2946515
|
gsector |
15 | .2415728 .1198503 2.02 0.044 .0066706 .4764751
20 | .205864 .1093709 1.88 0.060 -.0084989 .420227
25 | .7184278 .1097162 6.55 0.000 .503388 .9334677
30 | .9132355 .1197168 7.63 0.000 .678595 1.147876
35 | -.0202379 .116347 -0.17 0.862 -.2482738 .207798
40 | .1536092 .1311221 1.17 0.241 -.1033855 .4106038
45 | .1611879 .1079818 1.49 0.136 -.0504525 .3728283
50 | .9800344 .1233752 7.94 0.000 .7382234 1.221845
55 | -.2942186 .3381075 -0.87 0.384 -.956897 .3684599
60 | .2228457 .2027682 1.10 0.272 -.1745727 .6202641
|
_cons | -1.242982 .157199 -7.91 0.000 -1.551086 -.9348774
------------------------------------------------------------------------------
----------------------------------------------------------------------------------------
Variable Sample | Treated Controls Difference S.E. T-stat
----------------------------+-----------------------------------------------------------
lateROA Unmatched | .034948008 .035744586 -.000796577 .004373 -0.18
ATT | .034948008 .030707496 .004240513 .006855752 0.62
----------------------------+-----------------------------------------------------------
Note: S.E. does not take into account that the propensity score is estimated.
| psmatch2:
psmatch2: | Common
Treatment | support
assignment | On suppor | Total
-----------+-----------+----------
Untreated | 8,226 | 8,226
Treated | 956 | 956
-----------+-----------+----------
Total | 9,182 | 9,182
"ffs" is my binary 0/1 variable. I then say that when the value is 1, the effect is stronger; otherwise, weaker.
So my first question: Is the above statement correct?
Next, "export_ipri" variable is my interaction variable (IV multiplied by moderator). So what I want to find is whether the effect of my moderator differs on the outcome (lateROA) depending on the value of "ffs."
So the second question: In psmatch2 do I need to include the IV as well? (the IV is "export" the moderator is "ipri").
The third question: Would be it right to argue that the effect of "export_ipri" on my outcome (lateROA) is stronger when "ffs" = 1 (since the ATT is significant and the difference is positive). If so, then would it be correct to interpret it in this way: the effect of "export_ipri" is negative (and significant) on lateROA and it is MORE negative (stronger effect) when "ffs" = 1?
The fourth question: Do I even care if anything is significant in that probit model? (in this case specifically "export_ipri")
Thank you so much for your help. I have been reading everything for days and have yet to find the answers to the above questions.
Happy Holidays!
