For the following model under the IV framework:
Code:
ivreg2 cognition (infl_pneumonia_rate1000=inter) female agesurvey black postsulfa `postAtheen' i.birth_year i.birth_state, cluster(birth_state) partial(i.birth_year i.birth_state)
I used two methods to address this problem. The first method is by including DNAMGRIMAGE_accl as an interaction term with infl_pneumonia_rate in the IV model. I followed some previous discussions about IV interactions in the forum and write the following code:
Code:
ivreg2 cognition (infl_pneumonia_rate1000 c.infl_pneumonia_rate1000#i.DNAMGRIMAGE_accl=inter c.inter#i.DNAMGRIMAGE_accl) female agesurvey i.DNAMGRIMAGE_accl black postsulfa `postAtheen' i.birth_year i.birth_state, cluster(birth_state) partial(i.birth_year i.birth_state)
Code:
IV (2SLS) estimation
--------------------
Estimates efficient for homoskedasticity only
Statistics robust to heteroskedasticity and clustering on birth_state
Number of clusters (birth_state) = 48 Number of obs = 13385
F( 21, 47) = 163.82
Prob > F = 0.0000
Total (centered) SS = 242728.7548 Centered R2 = 0.1287
Total (uncentered) SS = 242728.7548 Uncentered R2 = 0.1287
Residual SS = 211487.5393 Root MSE = 3.975
------------------------------------------------------------------------------------------------------------
| Robust
cognition | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------------------------------------+----------------------------------------------------------------
infl_pneumonia_rate1000 | -2.700423 1.641164 -1.65 0.100 -5.917046 .5161992
|
DNAMGRIMAGE_accl#c.infl_pneumonia_rate1000 |
1 | 1.388179 .8338651 1.66 0.096 -.2461667 3.022525
|
female | .76827 .1373375 5.59 0.000 .4990933 1.037447
agesurvey | -.1910098 .0051996 -36.74 0.000 -.2012008 -.1808187
1.DNAMGRIMAGE_accl | -1.780273 .7225659 -2.46 0.014 -3.196476 -.3640699
black | -2.721402 .2830009 -9.62 0.000 -3.276073 -2.16673
postsulfa | 7.26792 9.512314 0.76 0.445 -11.37587 25.91171
post_complete | -8.505998 7.04445 -1.21 0.227 -22.31287 5.300871
post_b_mmr_1930_36 | -.3231463 .3937424 -0.82 0.412 -1.094867 .4485745
post_b_diar_1930_36 | -.0887637 .0945664 -0.94 0.348 -.2741104 .096583
post_b_malaria_1930_36 | -.0032673 .0021566 -1.51 0.130 -.0074942 .0009596
post_b_tbr_1930_36 | -.2205102 3.32944 -0.07 0.947 -6.746093 6.305072
post_base_scarfever_rate | -46857.07 36927.85 -1.27 0.204 -119234.3 25520.18
post_base_meningitis_rate | 14000.31 38008.46 0.37 0.713 -60494.91 88495.53
post_b_physician_cap_193036 | -1.39203 .6091681 -2.29 0.022 -2.585978 -.1980829
post_b_pharm_cap_193036 | 1.719759 1.593879 1.08 0.281 -1.404185 4.843704
post_b_nb_hos_imp_pc_193036 | -13.21998 18.62931 -0.71 0.478 -49.73275 23.2928
post_b_pci_193036 | -.0029973 .0042377 -0.71 0.479 -.011303 .0053084
post_b_urban_st_193036 | .4362025 5.122835 0.09 0.932 -9.60437 10.47678
post_b_illit_st_193036 | .9197056 11.53224 0.08 0.936 -21.68308 23.52249
post_b_nb_sch_imp_pc_193036 | -.0892432 .3536831 -0.25 0.801 -.7824494 .6039629The second method is just by stratifying the initial IV model by the binary variable DNAMGRIMAGE_accl. I used the following code:
1. When DNAMGRIMAGE_accl = 1
Code:
ivreg2 cognition (infl_pneumonia_rate1000=inter) female agesurvey black postsulfa `postAtheen' i.birth_year i.birth_state if DNAMGRIMAGE_accl == 1, cluster(birth_state) partial(i.birth_year i.birth_state)
Code:
IV (2SLS) estimation
--------------------
Estimates efficient for homoskedasticity only
Statistics robust to heteroskedasticity and clustering on birth_state
Number of clusters (birth_state) = 48 Number of obs = 5838
F( 19, 47) = 146.57
Prob > F = 0.0000
Total (centered) SS = 101303.6092 Centered R2 = 0.1305
Total (uncentered) SS = 101303.6092 Uncentered R2 = 0.1305
Residual SS = 88081.7182 Root MSE = 3.884
---------------------------------------------------------------------------------------------
| Robust
cognition | Coef. Std. Err. z P>|z| [95% Conf. Interval]
----------------------------+----------------------------------------------------------------
infl_pneumonia_rate1000 | -.4627608 2.066673 -0.22 0.823 -4.513366 3.587844
female | .7953502 .2299685 3.46 0.001 .3446202 1.24608
agesurvey | -.1996929 .0075702 -26.38 0.000 -.2145302 -.1848557
black | -2.322097 .394588 -5.88 0.000 -3.095476 -1.548719
postsulfa | 7.610768 14.96973 0.51 0.611 -21.72936 36.95089
post_complete | -.674218 9.90113 -0.07 0.946 -20.08008 18.73164
post_b_mmr_1930_36 | -.0718856 .58801 -0.12 0.903 -1.224364 1.080593
post_b_diar_1930_36 | -.175774 .1387585 -1.27 0.205 -.4477357 .0961876
post_b_malaria_1930_36 | -.0012827 .0037815 -0.34 0.734 -.0086943 .0061289
post_b_tbr_1930_36 | -4.834991 5.406693 -0.89 0.371 -15.43192 5.761933
post_base_scarfever_rate | 40192.7 42631.75 0.94 0.346 -43364 123749.4
post_base_meningitis_rate | 41752.66 89054.39 0.47 0.639 -132790.7 216296
post_b_physician_cap_193036 | -3.246845 1.390723 -2.33 0.020 -5.972613 -.5210775
post_b_pharm_cap_193036 | 2.52376 2.686741 0.94 0.348 -2.742156 7.789676
post_b_nb_hos_imp_pc_193036 | -17.48562 38.14572 -0.46 0.647 -92.24986 57.27861
post_b_pci_193036 | .0084787 .0077569 1.09 0.274 -.0067246 .0236819
post_b_urban_st_193036 | -11.44327 8.021479 -1.43 0.154 -27.16508 4.27854
post_b_illit_st_193036 | 11.93194 19.74517 0.60 0.546 -26.76788 50.63175
post_b_nb_sch_imp_pc_193036 | -.6371579 .6122712 -1.04 0.298 -1.837187 .5628715
---------------------------------------------------------------------------------------------2. When DNAMGRIMAGE_accl = 0
Code:
ivreg2 cognition (infl_pneumonia_rate1000=inter) female agesurvey black postsulfa `postAtheen' i.birth_year i.birth_state if DNAMGRIMAGE_accl == 0, cluster(birth_state) partial(i.birth_year i.birth_state)
Code:
IV (2SLS) estimation
--------------------
Estimates efficient for homoskedasticity only
Statistics robust to heteroskedasticity and clustering on birth_state
Number of clusters (birth_state) = 48 Number of obs = 7547
F( 19, 47) = 76.47
Prob > F = 0.0000
Total (centered) SS = 131605.8729 Centered R2 = 0.1198
Total (uncentered) SS = 131605.8729 Uncentered R2 = 0.1198
Residual SS = 115841.1524 Root MSE = 3.918
---------------------------------------------------------------------------------------------
| Robust
cognition | Coef. Std. Err. z P>|z| [95% Conf. Interval]
----------------------------+----------------------------------------------------------------
infl_pneumonia_rate1000 | -3.189662 2.171009 -1.47 0.142 -7.444761 1.065437
female | .7863143 .1631763 4.82 0.000 .4664946 1.106134
agesurvey | -.1859477 .0064833 -28.68 0.000 -.1986546 -.1732407
black | -2.846336 .448716 -6.34 0.000 -3.725803 -1.966868
postsulfa | 5.667493 10.68912 0.53 0.596 -15.28279 26.61778
post_complete | -16.86658 9.549301 -1.77 0.077 -35.58286 1.849707
post_b_mmr_1930_36 | -.5785419 .5686961 -1.02 0.309 -1.693166 .536082
post_b_diar_1930_36 | .2194986 .1302678 1.68 0.092 -.0358216 .4748189
post_b_malaria_1930_36 | -.0059695 .0033666 -1.77 0.076 -.0125679 .0006288
post_b_tbr_1930_36 | -3.152445 3.633162 -0.87 0.386 -10.27331 3.968422
post_base_scarfever_rate | -116179.3 47454.37 -2.45 0.014 -209188.2 -23170.43
post_base_meningitis_rate | 40355.72 36380.03 1.11 0.267 -30947.82 111659.3
post_b_physician_cap_193036 | .0375361 1.289499 0.03 0.977 -2.489836 2.564909
post_b_pharm_cap_193036 | 2.416589 2.204938 1.10 0.273 -1.90501 6.738188
post_b_nb_hos_imp_pc_193036 | 17.58173 24.02155 0.73 0.464 -29.49963 64.6631
post_b_pci_193036 | -.0144308 .0065389 -2.21 0.027 -.0272469 -.0016148
post_b_urban_st_193036 | 14.76539 7.294849 2.02 0.043 .467744 29.06303
post_b_illit_st_193036 | 16.74622 17.71698 0.95 0.345 -17.97842 51.47087
post_b_nb_sch_imp_pc_193036 | .2122467 .3956725 0.54 0.592 -.5632571 .9877505
---------------------------------------------------------------------------------------------For example, in the subgroup analysis, when DNAMGRIMAGE_accl = 1, the coefficient of infl_pneumonia_rate1000 is -0.46276. But in the IV model, the point estimate for infl_pneumonia_rate1000 when DNAMGRIMAGE_accl = 1 is (-2.700423+1.388179) = -1.312244
And in the subgroup analysis, when DNAMGRIMAGE_accl = 0, the coefficient of infl_pneumonia_rate1000 is -3.189662. But in the IV model, the point estimate for infl_pneumonia_rate1000 when DNAMGRIMAGE_accl = 0 is -2.700423.
Are there any mistakes in my code for either interaction or subgroup IV analysis? If there is no mistake, how should we explain the difference in the point estimate between IV interaction analysis and IV subgroup analysis?
Many thanks!!!
0 Response to Difference between instrumental variable interaction and instrumental variable subgroup analysis
Post a Comment