Problem with margins in a survey.
Dear Statalists,
I am working with survey database. I have run the next logit regression for a subsample of my data, where the dependent variable (fin11) is a binary one. Furthermore, i am including fixed effects at industry level (i.isic).
This is the ouput of the regression:
It can be see that some of the levels of the industry fixed effects are not estimated, because for those omitted industries there is only one observation for which the covariates and the dependent variable do not take missing value.
I am interested on obtaining average marginal effects of all covariates, so I run the next code:
And this is the output:
It can be seen that I do not get the margins of the industry fixed effects, even for those industries that did not be omitted in the logit regression. However, I got it for the covariates. Is this issue a major problem? I mean, it is legitimate to use the margins that I have got for the covariates?
Finally, I have a doubt regarding subpop command. The number of observations in th elogit regression is 122,771, meanwhile in the margins estimation 34,083, why such a big difference?
Thank you,
Ibai
I am working with survey database. I have run the next logit regression for a subsample of my data, where the dependent variable (fin11) is a binary one. Furthermore, i am including fixed effects at industry level (i.isic).
Code:
svyset, clear svyset idstd [pweight=wt], strata(strata) singleunit(scaled) svy, subpop(if loan_duration<=4 & k8==100): logit fin11 i.n_outcome lcar1 lnemployees i.ownership i.k9 ln_dlabproduc_2 k7 k21 k3a i.isic
Code:
note: 11b.isic != 0 predicts success perfectly
11b.isic dropped and 1 obs not used
note: 12.isic != 0 predicts failure perfectly
12.isic dropped and 1 obs not used
note: 65.isic != 0 predicts failure perfectly
65.isic dropped and 2 obs not used
note: 71.isic != 0 predicts success perfectly
71.isic dropped and 1 obs not used
note: 93.isic != 0 predicts success perfectly
93.isic dropped and 1 obs not used
note: 74.isic omitted because of collinearity
Survey: Logistic regression
Number of strata = 1,023 Number of obs = 122,771
Number of PSUs = 122,771 Population size = 7,992,600
Subpop. no. obs = 27,735
Subpop. size = 1,453,296
Design df = 121,748
F( 46, 121703) = 3.59
Prob > F = 0.0000
---------------------------------------------------------------------------------------------------------
| Linearized
fin11 | Coef. Std. Err. t P>|t| [95% Conf. Interval]
----------------------------------------+----------------------------------------------------------------
1.n_outcome | -.3410065 .1234849 -2.76 0.006 -.5830349 -.0989782
lcar1 | .097119 .1033377 0.94 0.347 -.1054212 .2996592
lnemployees | .2238808 .0475924 4.70 0.000 .1306004 .3171612
|
ownership |
Foreign | .3300358 .1857271 1.78 0.076 -.0339862 .6940577
|
k9 |
State-owned banks or government agency | .4543845 .174829 2.60 0.009 .1117225 .7970466
Non-bank financial institutions | -.1643362 .2303354 -0.71 0.476 -.6157898 .2871173
Other | -1.502558 .30923 -4.86 0.000 -2.108644 -.8964726
|
ln_dlabproduc_2 | .0548861 .0094919 5.78 0.000 .0362822 .0734899
k7 | -.2306878 .1387156 -1.66 0.096 -.502568 .0411924
k21 | .2545136 .117209 2.17 0.030 .024786 .4842413
k3a | -.5454588 .1690183 -3.23 0.001 -.876732 -.2141857
|
isic |
11 | 0 (empty)
12 | 0 (empty)
15 | .5761103 .8422641 0.68 0.494 -1.074713 2.226934
16 | .5525625 .9970459 0.55 0.579 -1.401631 2.506756
17 | .2459134 .8569859 0.29 0.774 -1.433765 1.925592
18 | .4972503 .8647682 0.58 0.565 -1.197681 2.192182
19 | .6810023 .9549176 0.71 0.476 -1.19062 2.552625
20 | .5277384 .9299466 0.57 0.570 -1.294942 2.350418
21 | -.2068407 .9720362 -0.21 0.831 -2.112016 1.698334
22 | .1737072 .925783 0.19 0.851 -1.640812 1.988227
23 | 1.525126 1.239943 1.23 0.219 -.9051413 3.955392
24 | .3956102 .8714038 0.45 0.650 -1.312327 2.103547
25 | .8913126 .8912435 1.00 0.317 -.85551 2.638135
26 | .949387 .8757977 1.08 0.278 -.767162 2.665936
27 | .2927997 .9028069 0.32 0.746 -1.476687 2.062286
28 | .7221179 .8472834 0.85 0.394 -.9385435 2.382779
29 | .4974671 .8955593 0.56 0.579 -1.257814 2.252749
30 | 3.646133 1.43535 2.54 0.011 .8328701 6.459396
31 | .4336457 .8994272 0.48 0.630 -1.329217 2.196508
32 | 1.546787 1.064917 1.45 0.146 -.5404316 3.634006
33 | .4358964 .9968934 0.44 0.662 -1.517998 2.389791
34 | .3470953 .9126551 0.38 0.704 -1.441694 2.135884
35 | 1.621231 1.055895 1.54 0.125 -.4483058 3.690768
36 | .7194533 .9259729 0.78 0.437 -1.095438 2.534345
37 | 1.731182 1.203947 1.44 0.150 -.6285341 4.090899
45 | .6625817 .8719648 0.76 0.447 -1.046455 2.371618
50 | .9029901 .8975743 1.01 0.314 -.8562407 2.662221
51 | .5655833 .8539491 0.66 0.508 -1.108143 2.239309
52 | .6118768 .8502834 0.72 0.472 -1.054665 2.278418
55 | .6197807 .8848717 0.70 0.484 -1.114553 2.354115
60 | .8436712 .8972863 0.94 0.347 -.9149952 2.602338
61 | 2.158662 1.097073 1.97 0.049 .0084165 4.308908
62 | 1.178337 1.22349 0.96 0.336 -1.219683 3.576358
63 | .5864199 .8711842 0.67 0.501 -1.121087 2.293927
64 | .6458872 1.009906 0.64 0.522 -1.333513 2.625287
65 | 0 (empty)
70 | 2.118628 1.332885 1.59 0.112 -.4938037 4.73106
71 | 0 (empty)
72 | .5602221 .9209123 0.61 0.543 -1.244751 2.365195
74 | 0 (omitted)
93 | 0 (empty)
|
_cons | -1.411011 .8950048 -1.58 0.115 -3.165206 .3431832
---------------------------------------------------------------------------------------------------------
Note: 222 strata omitted because they contain no subpopulation members.
Note: Variance scaled to handle strata with a single sampling unit.
I am interested on obtaining average marginal effects of all covariates, so I run the next code:
Code:
margins, dydx(*)
Code:
Average marginal effects Number of obs = 34,083
Model VCE : Linearized
Expression : Pr(fin11), predict()
dy/dx w.r.t. : 1.n_outcome lcar1 lnemployees 1.ownership 2.k9 3.k9 4.k9 ln_dlabproduc_2 k7 k21 k3a 12.isic
15.isic 16.isic 17.isic 18.isic 19.isic 20.isic 21.isic 22.isic 23.isic 24.isic 25.isic 26.isic
27.isic 28.isic 29.isic 30.isic 31.isic 32.isic 33.isic 34.isic 35.isic 36.isic 37.isic 45.isic
50.isic 51.isic 52.isic 55.isic 60.isic 61.isic 62.isic 63.isic 64.isic 65.isic 70.isic 71.isic
72.isic 74.isic 93.isic
---------------------------------------------------------------------------------------------------------
| Delta-method
| dy/dx Std. Err. t P>|t| [95% Conf. Interval]
----------------------------------------+----------------------------------------------------------------
1.n_outcome | -.0723683 .0266629 -2.71 0.007 -.1246271 -.0201094
lcar1 | .0203329 .0216112 0.94 0.347 -.0220248 .0626906
lnemployees | .0468718 .009898 4.74 0.000 .0274719 .0662717
|
ownership |
Foreign | .0665421 .0357958 1.86 0.063 -.003617 .1367013
|
k9 |
State-owned banks or government agency | .0935447 .034765 2.69 0.007 .0254059 .1616834
Non-bank financial institutions | -.0362868 .051395 -0.71 0.480 -.1370201 .0644465
Other | -.3274578 .0584824 -5.60 0.000 -.4420823 -.2128332
|
ln_dlabproduc_2 | .011491 .0019731 5.82 0.000 .0076238 .0153581
k7 | -.0482969 .028923 -1.67 0.095 -.1049856 .0083917
k21 | .0532851 .0244308 2.18 0.029 .0054011 .1011692
k3a | -.1141976 .0351164 -3.25 0.001 -.1830252 -.0453699
|
isic |
11 | 0 (empty)
12 | . (not estimable)
15 | . (not estimable)
16 | . (not estimable)
17 | . (not estimable)
18 | . (not estimable)
19 | . (not estimable)
20 | . (not estimable)
21 | . (not estimable)
22 | . (not estimable)
23 | . (not estimable)
24 | . (not estimable)
25 | . (not estimable)
26 | . (not estimable)
27 | . (not estimable)
28 | . (not estimable)
29 | . (not estimable)
30 | . (not estimable)
31 | . (not estimable)
32 | . (not estimable)
33 | . (not estimable)
34 | . (not estimable)
35 | . (not estimable)
36 | . (not estimable)
37 | . (not estimable)
45 | . (not estimable)
50 | . (not estimable)
51 | . (not estimable)
52 | . (not estimable)
55 | . (not estimable)
60 | . (not estimable)
61 | . (not estimable)
62 | . (not estimable)
63 | . (not estimable)
64 | . (not estimable)
65 | . (not estimable)
70 | . (not estimable)
71 | . (not estimable)
72 | . (not estimable)
74 | . (not estimable)
93 | . (not estimable)
---------------------------------------------------------------------------------------------------------
Note: dy/dx for factor levels is the discrete change from the base level.
It can be seen that I do not get the margins of the industry fixed effects, even for those industries that did not be omitted in the logit regression. However, I got it for the covariates. Is this issue a major problem? I mean, it is legitimate to use the margins that I have got for the covariates?
Finally, I have a doubt regarding subpop command. The number of observations in th elogit regression is 122,771, meanwhile in the margins estimation 34,083, why such a big difference?
Thank you,
Ibai
