Manually created interaction term and ## command

[Mark Amador]

Hi,
this is my first time using Stata.
I would like to include an interaction term in my model:

egen voc = group(vis)

xtreg foreign_revenues state_dom##voc state_min##voc log_age log_asset log_gdp_capita i.sector i.year, robust

where state_dom and state_min are dummy variables and voc is 4 different groups.

Can anyone tell me how to implement the same xtreg function manually creating interaction term instead of ##?

Thanks very much

[Carlo Lazzaro]

Mark:
welcom to this forum.
Strange question indeed (usually, details about the opposite approach are requested).
Anyway, you may want to replace your code with the following one:

xtreg voc foreign_revenues state_dom foreign_revenues state_dom*voc state_min state_min*voc log_age log_asset log_gdp_capita i.sector i.year, robust

[Mark Amador]

I'm sorry, I'd have posted the whole code:

 
 gen state_dom = 0  replace state_dom=1 if state_ownership>=25  gen state_min = 0  replace state_min=1 if state_ownership<25 & state_ownership>0  egen voc = group(vis) //CME=1, Emergent LME=2, Hierarchically Coordinated=3, State Led=4.  gen statedomCME = 0 gen statedomELME = 0 gen statedomHC = 0 gen statedomSL = 0 replace statedomCME = state_dom*voc if voc==1 replace statedomELME = state_dom*voc if voc==2 replace statedomHC = state_dom*voc if voc==3 replace statedomSL = state_dom*voc if voc==4  xtset id_company year //definisco un binomial panel NxT  xtreg foreign_revenues state_dom##voc log_age log_asset log_gdp_capita i.sector i.year, robust  xtreg foreign_revenues state_dom i.voc statedomCME statedomELME statedomHC statedomSL log_age log_asset log_gdp_capita i.sector i.year, robust

Why do I get different results for the two xtreg commands?

[Mark Amador]

What I want to know is how I can implement the first xtreg instruction using manually created variable instead of using ##

[Carlo Lazzaro]

Mark:
from

definisco un binomial panel

I would say that your probably not English mother-tongue.
Please note that Stata community appreciates real given and family names (for reasons that are well explained in FAQ).
That said, you would increase your chances of getting helpful replies if you posted a readable code (copy and paste from a do-file is a good approach) and share not only what you typed, but what Stata gave you back (ie, outcome tables), too.

[Mark Amador]

I'm from Italy and that's my real name.

gen state_dom = 0
replace state_dom=1 if state_ownership>=25
gen state_min = 0
replace state_min=1 if state_ownership<25 & state_ownership>0

egen voc = group(vis) //CME=1, Emergent LME=2, Hierarchically Coordinated=3, State Led=4

gen statedomCME = 0
gen statedomELME = 0
gen statedomHC = 0
gen statedomSL = 0
replace statedomCME = state_dom*voc if voc==1
replace statedomELME = state_dom*voc if voc==2
replace statedomHC = state_dom*voc if voc==3
replace statedomSL = state_dom*voc if voc==4

xtset id_company year

xtreg foreign_revenues state_dom##voc log_age log_asset log_gdp_capita i.sector i.year, robust
xtreg foreign_revenues state_dom i.voc statedomELME statedomHC statedomSL log_age log_asset log_gdp_capita i.sector i.year, robust


What I'd like to know is how I can implement the first xtreg instruction using manually created variable instead of using ##

Random-effects GLS regression Number of obs = 1450
Group variable: id_company Number of groups = 145

R-sq: within = 0.0845 Obs per group: min = 10
between = 0.3117 avg = 10.0
overall = 0.2956 max = 10

Wald chi2(28) = 234.04
corr(u_i, X) = 0 (assumed) Prob > chi2 = 0.0000

(Std. Err. adjusted for 145 clusters in id_company)

Robust
foreign_reve~s Coef. Std. Err. z P>z [95% Conf. Interval]

1.state_dom -.0171317 .0199518 -0.86 0.391 -.0562366 .0219731

voc
2 .1701311 .100678 1.69 0.091 -.0271941 .3674564
3 .1191134 .0910838 1.31 0.191 -.0594076 .2976343
4 -.0505043 .1084388 -0.47 0.641 -.2630405 .1620319

state_dom#voc
1 2 .1275343 .0299164 4.26 0.000 .0688992 .1861694
1 3 .0250271 .0373816 0.67 0.503 -.0482395 .0982936
1 4 -.0776619 .0588236 -1.32 0.187 -.1929541 .0376304

log_age .0283696 .0381954 0.74 0.458 -.046492 .1032311
log_asset .102811 .0618559 1.66 0.096 -.0184244 .2240464
log_gdp_capita -.065378 .039678 -1.65 0.099 -.1431453 .0123894

sector
2 .3483784 .1328379 2.62 0.009 .0880209 .6087359
3 -.1213994 .0706451 -1.72 0.086 -.2598613 .0170626
4 .0783069 .1183086 0.66 0.508 -.1535738 .3101875
5 .4735978 .0767551 6.17 0.000 .3231607 .624035
6 .1200187 .0683085 1.76 0.079 -.0138634 .2539008
7 -.0488898 .0809016 -0.60 0.546 -.2074539 .1096744
8 -.0078801 .1304892 -0.06 0.952 -.2636343 .247874
9 -.1194326 .057179 -2.09 0.037 -.2315014 -.0073638
10 .2788296 .094348 2.96 0.003 .0939108 .4637483

year
2009 -.0099008 .0050901 -1.95 0.052 -.0198771 .0000756
2010 -.0028758 .0113268 -0.25 0.800 -.025076 .0193244
2011 .0017951 .0135426 0.13 0.895 -.0247478 .028338
2012 .0060339 .0148849 0.41 0.685 -.02314 .0352078
2013 .0165934 .0167894 0.99 0.323 -.0163131 .0495
2014 .0274765 .0176467 1.56 0.119 -.0071105 .0620634
2015 .0293138 .016608 1.77 0.078 -.0032373 .0618649
2016 .0265384 .0174468 1.52 0.128 -.0076567 .0607335
2017 .0246151 .0190346 1.29 0.196 -.0126919 .0619222

_cons -.306811 .5414033 -0.57 0.571 -1.367942 .7543199

sigma_u .23393536
sigma_e .08043675
rho .89427267 (fraction of variance due to u_i)





Random-effects GLS regression Number of obs = 1450
Group variable: id_company Number of groups = 145

R-sq: within = 0.0845 Obs per group: min = 10
between = 0.3117 avg = 10.0
overall = 0.2956 max = 10

Wald chi2(28) = 234.04
corr(u_i, X) = 0 (assumed) Prob > chi2 = 0.0000

(Std. Err. adjusted for 145 clusters in id_company)

Robust
foreign_reve~s Coef. Std. Err. z P>z [95% Conf. Interval]

state_dom -.0171317 .0199518 -0.86 0.391 -.0562366 .0219731

voc
2 .1701311 .100678 1.69 0.091 -.0271941 .3674564
3 .1191134 .0910838 1.31 0.191 -.0594076 .2976343
4 -.0505043 .1084388 -0.47 0.641 -.2630405 .1620319

statedomELME .0637672 .0149582 4.26 0.000 .0344496 .0930847
statedomHC .0083424 .0124605 0.67 0.503 -.0160798 .0327645
statedomSL -.0194155 .0147059 -1.32 0.187 -.0482385 .0094076
log_age .0283696 .0381954 0.74 0.458 -.046492 .1032311
log_asset .102811 .0618559 1.66 0.096 -.0184244 .2240464
log_gdp_capita -.065378 .039678 -1.65 0.099 -.1431453 .0123894

sector
2 .3483784 .1328379 2.62 0.009 .0880209 .6087359
3 -.1213994 .0706451 -1.72 0.086 -.2598613 .0170626
4 .0783069 .1183086 0.66 0.508 -.1535738 .3101875
5 .4735978 .0767551 6.17 0.000 .3231607 .624035
6 .1200187 .0683085 1.76 0.079 -.0138634 .2539008
7 -.0488898 .0809016 -0.60 0.546 -.2074539 .1096744
8 -.0078801 .1304892 -0.06 0.952 -.2636343 .247874
9 -.1194326 .057179 -2.09 0.037 -.2315014 -.0073638
10 .2788296 .094348 2.96 0.003 .0939108 .4637483

year
2009 -.0099008 .0050901 -1.95 0.052 -.0198771 .0000756
2010 -.0028758 .0113268 -0.25 0.800 -.025076 .0193244
2011 .0017951 .0135426 0.13 0.895 -.0247478 .028338
2012 .0060339 .0148849 0.41 0.685 -.02314 .0352078
2013 .0165934 .0167894 0.99 0.323 -.0163131 .0495
2014 .0274765 .0176467 1.56 0.119 -.0071105 .0620634
2015 .0293138 .016608 1.77 0.078 -.0032373 .0618649
2016 .0265384 .0174468 1.52 0.128 -.0076567 .0607335
2017 .0246151 .0190346 1.29 0.196 -.0126919 .0619222

_cons -.306811 .5414033 -0.57 0.571 -1.367942 .7543199

sigma_u .23393536
sigma_e .08043675
rho .89427267 (fraction of variance due to u_i)

[Carlo Lazzaro]

xtreg foreign_revenues state_dom voc state_dom*voc log_age log_asset log_gdp_capita i.sector i.year, robust

[daniel klein]

Carlo's codes in #2 and #7 will not work. Even if we add the (old) xi prefix, the code will not produce the intended results.

Here are two ways of doing what Mark wants to (but should probably not) do:

xi : xtreg ... i.state_dom*i.voc i.state_min*i.voc

egen state_dom_voc = group(state_dom voc) , label
egen state_min_voc = group(state_min voc) , label
xtreg ... i.state_dom_voc i.state_min_voc

Either way, Mark will not obtain correct results from margins and possibly other post-estimation commands and there really is no reason to want to create the interactions by hand.


Added in Edit:


Reading

gen state_dom = 0
replace state_dom=1 if state_ownership>=25
gen state_min = 0
replace state_min=1 if state_ownership<25 & state_ownership>0

from #6 is puzzling. If state_ownership has missing values, then those will be coded 1 in state_dom which is probably not what Mark wants. Also, even if the two variables are correctly created, they seem to be collinear; including both in one model does not make sense.

There might be more problems, I did not read through the whole code.

Best
Daniel

[Nick Cox]

Cross-posted at https://stackoverflow.com/questions/...eraction-terms

Please note our cross-posting policy, explicit in the FAQ Advice you were asked to read before posting: You are asked to tell us about it.