I am trying to run a gravity model with the inclusion of an interaction term composed of three variables. My dependent variable is gvc_total, which is the bilateral penetration in the global value chain of an exporter in the domestic economy of an importer. The interaction term is c.lag1_lnRD_imputed##i.exp_developing##i.income_cl assification_imp. lag1_lnRD_imputed corresponds to the bilateral difference in year t-1 in domestic regulation between two countries (exp and imp), exp_developing is a dummy variable equals to 1 if the exporter is a developing country and income_classification_imp is a categorical variable with three classes, in which importers are classified as 2 (low-income), 3 (middle-income) and 4 (high-income). I would like to see whether higher levels of bilateral regulatory differences have a higher impact on developing countries and how this is differentiated according to the income level of the importer.
This is a sample of my data:
Code:
* Example generated by -dataex-. For more info, type help dataex clear input str5(exp imp) double year float(gvc_total lag1_lnRD_imputed exp_developing income_classification_imp) "ARG" "AUS" 2005 .4810562 . 1 4 "ARG" "AUS" 2006 .7095908 3.080612 1 4 "ARG" "AUS" 2007 3.686711 3.0960314 1 4 "ARG" "AUS" 2008 2.6161735 3.040033 1 4 "ARG" "AUS" 2009 2.3592985 2.9926226 1 4 "ARG" "AUS" 2010 2.80107 2.946606 1 4 "ARG" "AUS" 2011 4.7547293 2.920012 1 4 "ARG" "AUS" 2012 4.7634377 2.899041 1 4 "ARG" "AUS" 2013 4.5947137 2.8946085 0 4 "ARG" "AUS" 2014 5.387508 2.8502626 1 4 "ARG" "AUS" 2015 4.823783 2.8395286 1 4 "ARG" "AUS" 2005 .0025410315 . 1 4 "ARG" "AUS" 2006 .003603552 3.080612 1 4 "ARG" "AUS" 2007 .005169975 3.0960314 1 4 "ARG" "AUS" 2008 .005921311 3.040033 1 4 "ARG" "AUS" 2009 .6192643 2.9926226 1 4 "ARG" "AUS" 2010 .007792954 2.946606 1 4 "ARG" "AUS" 2011 .010513017 2.920012 1 4 "ARG" "AUS" 2012 .009914982 2.899041 1 4 "ARG" "AUS" 2013 .4977893 2.8946085 0 4 "ARG" "AUS" 2014 .008768263 2.8502626 1 4 "ARG" "AUS" 2015 .007165529 2.8395286 1 4 "ARG" "AUS" 2005 .7004743 . 1 4 "ARG" "AUS" 2006 1.2754846 3.080612 1 4 "ARG" "AUS" 2007 .14972857 3.0960314 1 4 "ARG" "AUS" 2008 .596966 3.040033 1 4 "ARG" "AUS" 2009 .45553005 2.9926226 1 4 "ARG" "AUS" 2010 .5267684 2.946606 1 4 "ARG" "AUS" 2011 .01068677 2.920012 1 4 "ARG" "AUS" 2012 13.515596 2.899041 1 4 "ARG" "AUS" 2013 .0933937 2.8946085 0 4 "ARG" "AUS" 2014 .06100044 2.8502626 1 4 "ARG" "AUS" 2015 2.423828 2.8395286 1 4 "ARG" "AUS" 2005 .03694565 . 1 4 "ARG" "AUS" 2006 .013627814 3.080612 1 4 "ARG" "AUS" 2007 .02500085 3.0960314 1 4 "ARG" "AUS" 2008 .036339752 3.040033 1 4 "ARG" "AUS" 2009 .014212865 2.9926226 1 4 "ARG" "AUS" 2010 .02837868 2.946606 1 4 end
Code:
. asdoc ppmlhdfe gvc_total c.lag1_lnRD_imputed##i.exp_developing##i.income_clas
> sification_imp lag1_lntariffs lag1_RTA ln_dist $dummy_list if exp!=imp, a(exp_
> sector_time imp_sector_time) vce(cluster pair_sector_id) replace nest cnames(o
> verall GVC participation) dec(3) add(RESET Test, 0.000, AVE, 0.000)
(dropped 5066 observations that are either singletons or separated by a fixed ef
> fect)
(legend: p: exact partial-out s: exact solver h: step-halving o: epsilon b
> elow tolerance)
Converged in 16 iterations and 53 HDFE sub-iterations (tol = 1.0e-08)
HDFE PPML regression No. of obs = 657,981
Absorbing 2 HDFE groups Residual df = 68,928
Statistics robust to heteroskedasticity Wald chi2(21) = 8540.75
Deviance = 15224547.31 Prob > chi2 = 0.0000
Log pseudolikelihood = -8430762.747 Pseudo R2 = 0.9148
Number of clusters (pair_sector_id)= 68,929
(Std. Err. adjusted for 68,929 clusters in pair_sector_id)
-------------------------------------------------------------------------------
| Robust
gvc_total | Coef. Std. Err. z P>|z| [95% Conf. Interval]
--------------+----------------------------------------------------------------
lag1_lnRD_i~d | -.1894304 .0315448 -6.01 0.000 -.2512571 -.1276038
1.exp_devel~g | 0 (omitted)
|
exp_develop~g#|
c. |
lag1_lnRD_i~d |
1 | .4557452 .1341563 3.40 0.001 .1928036 .7186868
|
income_cla~mp |
2 | 0 (omitted)
3 | 1.418432 .3873059 3.66 0.000 .6593261 2.177537
4 | 0 (omitted)
|
income_cla~mp#|
c. |
lag1_lnRD_i~d |
2 | -.014681 .0655308 -0.22 0.823 -.1431191 .113757
3 | -.4596913 .1467517 -3.13 0.002 -.7473193 -.1720632
4 | .1397922 .0333796 4.19 0.000 .0743694 .2052151
|
exp_develop~g#|
income_cla~mp |
0 3 | 0 (empty)
1 2 | .8923923 .4241467 2.10 0.035 .0610801 1.723704
1 3 | 0 (omitted)
1 4 | 1.512856 .3679944 4.11 0.000 .7916001 2.234111
|
exp_develop~g#|
income_cla~mp#|
c. |
lag1_lnRD_i~d |
0 3 | 0 (empty)
1 2 | -.3549914 .1515992 -2.34 0.019 -.6521204 -.0578624
1 3 | 0 (omitted)
1 4 | -.4380782 .1400441 -3.13 0.002 -.7125596 -.1635969
|
lag1_lntari~s | -.1761927 .0182922 -9.63 0.000 -.2120447 -.1403407
lag1_RTA | .136915 .0368821 3.71 0.000 .0646273 .2092026
ln_dist | -.8003638 .0218646 -36.61 0.000 -.8432176 -.75751
contig | .3244154 .0291999 11.11 0.000 .2671846 .3816462
comlang_ethno | .1493362 .0364861 4.09 0.000 .0778247 .2208477
comcol | -.2656031 .1177075 -2.26 0.024 -.4963055 -.0349006
comrelig | -.0009697 .0558059 -0.02 0.986 -.1103473 .1084079
col45 | .1237388 .0835462 1.48 0.139 -.0400087 .2874862
comleg_post~s | .2450281 .024703 9.92 0.000 .196611 .2934452
transition_~e | -.1048861 .039355 -2.67 0.008 -.1820204 -.0277517
sibling | -.1014199 .0794286 -1.28 0.202 -.2570972 .0542573
_cons | 12.48745 .2074874 60.18 0.000 12.08079 12.89412
-------------------------------------------------------------------------------
I would have two questions:
1) in this case, the red-output states that the impact of regulatory differences is higher for developing economies (than for developed economies which is the baseline) when the importer is high-income( 4) and low-income as well (2). Is that interpretation correct?
2) I saw that in this context the postestimation analysis usually includes a plot with margins, would you also recommend further post-estimation analysis?
I am sorry for the length of the question, I hope to see some interesting comments and ideas anyway.
Thank you in advance for your time.
0 Response to Analysis with interaction term with three variables
Post a Comment