I face the following issue. I am running a logit regression on survey data (individual women from two periods (pre and post 2009) and in two treatment areas (non-BH and BH). This is done in a difference-in-difference setting so at the end I am explicitly interested in the interaction effect.
My function is specified in three ways (basic model, basic+state dummies, basic+state dummies+control variables). By including a factor variable i.surveystate I want to account for state fixed effects. When running my logit regression everything works out fine. As soon as I try to make use of any margins command (e.g. margins, dydx(*); margins, at()) Stata states the marginal effect for the indicator BH (i.BH (dummy)) and for all the different values of the state indicator (i.surveyystate (factor)) are "not estimable". This only happens in model 2 and 3 while Stata supplies me with the marginal effect of BH in model 1. While Stata says the marginal effects for BH and surveystates are not estimable in model 2 and 3 it does give me the marginal effect of other dummy/factor variables though (e.g. muslim, literacy, wealthindex). Below you can find the three model codes I have used as well as the output from model 1 and model 3. Here my issues are shown. Based on this problem I am also not able to use many other commands like mcp, marginsplot etc. as the result of margins for BH is always non estimable. Later I also want to run a model with a binary#continous interaction and face the same issue. This happens no matter if I calculate the interaction effect by hand post2009TPXBH=(post2009TP*BH) or use the factor and # notation (Model 1 as an example of pre-calculated interaction effect, model 3 example of factor notation to crate interaction effect).
Any held would be greatly appreciated!
Kind regards Caspar
Model 1:
Code:
svy:logit teenpreg i.BH i.post2009TP i.post2009TPXBH margins, dydx(*) vce(unconditional)
Code:
svy:logit teenpreg i.BH i.post2009TP i.post2009TP#i.BH i.surveystate margins, dydx(*) vce(unconditional)
Code:
svy: logit teenpreg i.BH i.post2009TP i.post2009TP#i.BH i.muslim i.urban i.kanuri i.hhheadmale i.literacy i.wealthindex i.edulevel c.eduyears i.edulevelpartner c.eduyearspartner i.polygamoushh i.surveystate margins, dydx(*) vce(unconditional)
Code:
. svy:logit teenpreg i.BH i.post2009TP i.post2009TPXBH
(running logit on estimation sample)
Survey: Logistic regression
Number of strata = 148 Number of obs = 22,035
Number of PSUs = 2,269 Population size = 22,012,938,401
Design df = 2,121
F( 3, 2119) = 124.16
Prob > F = 0.0000
---------------------------------------------------------------------------------
| Linearized
teenpreg | Coef. Std. Err. t P>|t| [95% Conf. Interval]
----------------+----------------------------------------------------------------
1.BH | 1.28489 .0830362 15.47 0.000 1.12205 1.447731
1.post2009TP | -.2222882 .0485641 -4.58 0.000 -.3175264 -.1270499
1.post2009TPXBH | -.444609 .1087976 -4.09 0.000 -.6579702 -.2312479
_cons | -.341325 .0321343 -10.62 0.000 -.4043431 -.2783069
---------------------------------------------------------------------------------Code:
. margins, dydx(*) vce(unconditional)
Average marginal effects Number of obs = 22,035
Expression : Pr(teenpreg), predict()
dy/dx w.r.t. : 1.BH 1.post2009TP 1.post2009TPXBH
---------------------------------------------------------------------------------
| Linearized
| dy/dx Std. Err. t P>|t| [95% Conf. Interval]
----------------+----------------------------------------------------------------
1.BH | .3059899 .0175181 17.47 0.000 .2716355 .3403444
1.post2009TP | -.0521801 .0113664 -4.59 0.000 -.0744706 -.0298896
1.post2009TPXBH | -.0996184 .0229866 -4.33 0.000 -.144697 -.0545398
---------------------------------------------------------------------------------
Note: dy/dx for factor levels is the discrete change from the base level.Code:
. svy: logit teenpreg i.BH i.post2009TP i.post2009TP#i.BH i.muslim i.urban i.kanuri i.hhheadmale i.literacy i.wealthindex i.edulevel c.eduyears i.edulevelpartner c.eduyearspartner i.polygamoushh i.surveystate
(running logit on estimation sample)
note: 1011.surveystate omitted because of collinearity
Survey: Logistic regression
Number of strata = 148 Number of obs = 14,387
Number of PSUs = 2,179 Population size = 14,146,244,185
Design df = 2,031
F( 52, 1980) = 31.06
Prob > F = 0.0000
-----------------------------------------------------------------------------------
| Linearized
teenpreg | Coef. Std. Err. t P>|t| [95% Conf. Interval]
------------------+----------------------------------------------------------------
1.BH | .2549566 .1734003 1.47 0.142 -.0851043 .5950176
1.post2009TP | -.3188818 .0525346 -6.07 0.000 -.4219091 -.2158545
|
post2009TP#BH |
1 1 | -.0673073 .1045352 -0.64 0.520 -.2723146 .1377
|
1.muslim | .3358439 .0803647 4.18 0.000 .1782381 .4934497
1.urban | .0831391 .0599677 1.39 0.166 -.0344655 .2007437
1.kanuri | -.2364794 .158307 -1.49 0.135 -.5469404 .0739816
1.hhheadmale | .1217891 .0831064 1.47 0.143 -.0411937 .2847718
1.literacy | .0651338 .0817359 0.80 0.426 -.0951611 .2254287
|
wealthindex |
poorer | .0918043 .0657692 1.40 0.163 -.0371779 .2207864
middle | -.0885208 .0767987 -1.15 0.249 -.2391332 .0620915
richer | -.2907834 .0869054 -3.35 0.001 -.4612165 -.1203504
richest | -.8785281 .1157358 -7.59 0.000 -1.105501 -.6515549
|
1.edulevel | -.1782213 .1062559 -1.68 0.094 -.3866034 .0301607
eduyears | -.0665808 .0120163 -5.54 0.000 -.0901464 -.0430153
1.edulevelpartner | .1496471 .1040683 1.44 0.151 -.0544445 .3537388
eduyearspartner | -.0227653 .0111104 -2.05 0.041 -.0445543 -.0009763
1.polygamoushh | -.0330811 .0575886 -0.57 0.566 -.14602 .0798577
|
surveystate |
zamfara | -.1928214 .151785 -1.27 0.204 -.490492 .1048492
katsina | .2620451 .1615701 1.62 0.105 -.0548152 .5789055
jigawa | .3076793 .1573823 1.95 0.051 -.0009681 .6163268
yobe | -.0157629 .1642379 -0.10 0.924 -.3378553 .3063295
borno | .0095577 .1618743 0.06 0.953 -.3078993 .3270147
adamawa | -.2877648 .1631481 -1.76 0.078 -.6077198 .0321902
gombe | .0768753 .1462841 0.53 0.599 -.2100072 .3637578
bauchi | .3715389 .1514305 2.45 0.014 .0745636 .6685142
kano | .0891833 .1387662 0.64 0.520 -.1829557 .3613223
kaduna | 0 (omitted)
kebbi | -.0766893 .1538785 -0.50 0.618 -.3784654 .2250869
niger | -.3141513 .1534187 -2.05 0.041 -.6150257 -.0132768
abuja | -.4034342 .1915664 -2.11 0.035 -.7791213 -.027747
nasarawa | -.0566398 .2138995 -0.26 0.791 -.4761251 .3628455
plateau | -.32141 .1756258 -1.83 0.067 -.6658354 .0230155
taraba | .1213225 .1683685 0.72 0.471 -.2088706 .4515155
benue | -.0591002 .1748634 -0.34 0.735 -.4020306 .2838302
kogi | .432422 .2098261 2.06 0.039 .0209253 .8439188
kwara | -.4152658 .19115 -2.17 0.030 -.7901363 -.0403952
oyo | -.0257761 .1844987 -0.14 0.889 -.3876026 .3360503
osun | -.6566143 .1752572 -3.75 0.000 -1.000317 -.3129118
ekiti | .0372496 .2635448 0.14 0.888 -.4795967 .5540958
ondo | -.2915482 .2125072 -1.37 0.170 -.708303 .1252067
edo | -.3771684 .2374049 -1.59 0.112 -.8427508 .088414
anambra | .0657958 .1944728 0.34 0.735 -.3155912 .4471828
enugu | -.2516204 .1999433 -1.26 0.208 -.6437357 .1404949
ebonyi | -.0660226 .1995881 -0.33 0.741 -.4574413 .3253961
cross river | .4106171 .2347734 1.75 0.080 -.0498048 .8710389
akwa ibom | .0735537 .213957 0.34 0.731 -.3460444 .4931518
abia | -.3390893 .2490943 -1.36 0.174 -.8275962 .1494177
imo | -.618356 .2192515 -2.82 0.005 -1.048337 -.1883747
rivers | -.0649557 .2182536 -0.30 0.766 -.49298 .3630687
bayelsa | .6023436 .1887087 3.19 0.001 .2322609 .9724263
delta | .0971328 .2320194 0.42 0.676 -.357888 .5521536
lagos | -.5527324 .2374296 -2.33 0.020 -1.018363 -.0871015
ogun | -.3748587 .1993162 -1.88 0.060 -.7657442 .0160268
|
_cons | 1.085254 .1637543 6.63 0.000 .7641102 1.406398
-----------------------------------------------------------------------------------Code:
. margins, dydx(*) vce(unconditional)
Average marginal effects Number of obs = 14,387
Expression : Pr(teenpreg), predict()
dy/dx w.r.t. : 1.BH 1.post2009TP 1.muslim 1.urban 1.kanuri 1.hhheadmale 1.literacy 2.wealthindex 3.wealthindex 4.wealthindex 5.wealthindex 1.edulevel eduyears 1.edulevelpartner eduyearspartner 1.polygamoushh 1002.surveystate 1003.surveystate
1004.surveystate 1005.surveystate 1006.surveystate 1007.surveystate 1008.surveystate 1009.surveystate 1010.surveystate 1011.surveystate 1012.surveystate 1013.surveystate 1014.surveystate 1015.surveystate 1016.surveystate
1017.surveystate 1018.surveystate 1019.surveystate 1020.surveystate 1021.surveystate 1022.surveystate 1023.surveystate 1024.surveystate 1025.surveystate 1026.surveystate 1027.surveystate 1028.surveystate 1029.surveystate
1030.surveystate 1031.surveystate 1032.surveystate 1033.surveystate 1034.surveystate 1035.surveystate 1036.surveystate 1037.surveystate
-----------------------------------------------------------------------------------
| Linearized
| dy/dx Std. Err. t P>|t| [95% Conf. Interval]
------------------+----------------------------------------------------------------
1.BH | . (not estimable)
1.post2009TP | -.0649622 .0087651 -7.41 0.000 -.0821517 -.0477727
1.muslim | .0665069 .0163947 4.06 0.000 .0343547 .0986591
1.urban | .0157526 .011258 1.40 0.162 -.0063258 .037831
1.kanuri | -.0462524 .031588 -1.46 0.143 -.1082006 .0156958
1.hhheadmale | .0235927 .0162881 1.45 0.148 -.0083504 .0555359
1.literacy | .0123473 .0153524 0.80 0.421 -.0177609 .0424555
|
wealthindex |
poorer | .0175353 .0125973 1.39 0.164 -.0071697 .0422404
middle | -.0173779 .0150855 -1.15 0.249 -.0469626 .0122067
richer | -.0586335 .0176686 -3.32 0.001 -.093284 -.0239831
richest | -.1865903 .0255393 -7.31 0.000 -.2366762 -.1365044
|
1.edulevel | -.0350133 .0214133 -1.64 0.102 -.0770077 .0069811
eduyears | -.0127354 .0022819 -5.58 0.000 -.0172106 -.0082602
1.edulevelpartner | .0281589 .0192538 1.46 0.144 -.0096004 .0659182
eduyearspartner | -.0043545 .0021267 -2.05 0.041 -.0085253 -.0001837
1.polygamoushh | -.0063313 .0110293 -0.57 0.566 -.0279613 .0152987
|
surveystate |
zamfara | . (not estimable)
katsina | . (not estimable)
jigawa | . (not estimable)
yobe | . (not estimable)
borno | . (not estimable)
adamawa | . (not estimable)
gombe | . (not estimable)
bauchi | . (not estimable)
kano | . (not estimable)
kaduna | . (not estimable)
kebbi | . (not estimable)
niger | . (not estimable)
abuja | . (not estimable)
nasarawa | . (not estimable)
plateau | . (not estimable)
taraba | . (not estimable)
benue | . (not estimable)
kogi | . (not estimable)
kwara | . (not estimable)
oyo | . (not estimable)
osun | . (not estimable)
ekiti | . (not estimable)
ondo | . (not estimable)
edo | . (not estimable)
anambra | . (not estimable)
enugu | . (not estimable)
ebonyi | . (not estimable)
cross river | . (not estimable)
akwa ibom | . (not estimable)
abia | . (not estimable)
imo | . (not estimable)
rivers | . (not estimable)
bayelsa | . (not estimable)
delta | . (not estimable)
lagos | . (not estimable)
ogun | . (not estimable)
-----------------------------------------------------------------------------------
Note: dy/dx for factor levels is the discrete change from the base level.
0 Response to Logit estimation - Margins - not estimable
Post a Comment