6

Estimating Community Wide Effects of FC Adoption_supplementary materials

. psmatch2 fcinfn0 know sex age fcsee stinohepc hivseek fcblfrc mcsex30 fcmediaS hiv , outcome (fconeve) common logit ate

------

Variable Sample | Treated Controls Difference S.E. T-stat

------+------

fconeve Unmatched | .391891892 .18 .211891892 .046920282 4.52

ATT | .37593985 .203007519 .172932331 .074264372 2.33

ATU | .214285714 .285714286 .071428571 . .

ATE | .11627907 . .

------+------

Note: S.E. does not take into account that the propensity score is estimated.

psmatch2: | psmatch2: Common

Treatment | support

assignment | Off suppo On suppor | Total

------+------+------

Untreated | 32 168 | 200

Treated | 15 133 | 148

------+------+------

Total | 47 301 | 348

repeat analysis without strong predictors, as suggested by reviewer:

w/o fcsee and fcmediaS

. psmatch2 fcinfn0 know sex age stinohepc hivseek fcblfrc mcsex30 hiv , outcome (fconeve) common

> logit ate

Logistic regression Number of obs = 348

LR chi2(8) = 45.46

Prob > chi2 = 0.0000

Log likelihood = -214.58581 Pseudo R2 = 0.0958

------

fcinfn0 | Coef. Std. Err. z P>|z| [95% Conf. Interval]

------+------

know | -.2202235 .6493483 -0.34 0.735 -1.492923 1.052476

sex | .5037462 .2396115 2.10 0.036 .0341163 .9733762

age | -.0281476 .0109574 -2.57 0.010 -.0496237 -.0066715

stinohepc | .6094463 .2446059 2.49 0.013 .1300275 1.088865

hivseek | .8911438 .2455329 3.63 0.000 .409908 1.372379

fcblfrc | .5395844 .4034027 1.34 0.181 -.2510704 1.330239

mcsex30 | .3509394 .2363423 1.48 0.138 -.1122831 .8141619

hiv | .187482 .1420411 1.32 0.187 -.0909135 .4658774

_cons | -2.353749 1.256718 -1.87 0.061 -4.816871 .1093728

------

There are observations with identical propensity score values.

The sort order of the data could affect your results.

Make sure that the sort order is random before calling psmatch2.

------

Variable Sample | Treated Controls Difference S.E. T-stat

------+------

fconeve Unmatched | .391891892 .18 .211891892 .046920282 4.52

ATT | .391891892 .162162162 .22972973 .068224726 3.37

ATU | .178947368 .384210526 .205263158 . .

ATE | .215976331 . .

------+------

Note: S.E. does not take into account that the propensity score is estimated.

psmatch2: | psmatch2: Common

Treatment | support

assignment | Off suppo On suppor | Total

------+------+------

Untreated | 10 190 | 200

Treated | 0 148 | 148

------+------+------

Total | 10 338 | 348

repeat analysis without strong predictors, as suggested by reviewer:

w/o fcsee and fcmediaS and hivseek

. psmatch2 fcinfn0 know sex age stinohepc fcblfrc mcsex30 hiv , outcome (fconeve) common logit ate

Logistic regression Number of obs = 348

LR chi2(7) = 32.06

Prob > chi2 = 0.0000

Log likelihood = -221.28309 Pseudo R2 = 0.0676

------

fcinfn0 | Coef. Std. Err. z P>|z| [95% Conf. Interval]

------+------

know | -.8272496 .6156692 -1.34 0.179 -2.033939 .3794398

sex | .5733057 .2344222 2.45 0.014 .1138466 1.032765

age | -.0294798 .0107454 -2.74 0.006 -.0505403 -.0084193

stinohepc | .5511488 .2386236 2.31 0.021 .083455 1.018843

fcblfrc | .5939295 .3964567 1.50 0.134 -.1831114 1.37097

mcsex30 | .390728 .2317195 1.69 0.092 -.0634338 .8448898

hiv | .2359905 .1377846 1.71 0.087 -.0340624 .5060435

_cons | -1.842781 1.219155 -1.51 0.131 -4.23228 .5467181

------

There are observations with identical propensity score values.

The sort order of the data could affect your results.

Make sure that the sort order is random before calling psmatch2.

------

Variable Sample | Treated Controls Difference S.E. T-stat

------+------

fconeve Unmatched | .391891892 .18 .211891892 .046920282 4.52

ATT | .394557823 .19047619 .204081633 .069676038 2.93

ATU | .180412371 .463917526 .283505155 . .

ATE | .249266862 . .

------+------

Note: S.E. does not take into account that the propensity score is estimated.

psmatch2: | psmatch2: Common

Treatment | support

assignment | Off suppo On suppor | Total

------+------+------

Untreated | 6 194 | 200

Treated | 1 147 | 148

------+------+------

Total | 7 341 | 348

. attnd fconeve fcinfn0 know sex age hivseek fcblfrc stinohepc mcsex30 fcsee fcmediaS hiv, comsup boot reps(100) dots logit detail

Estimation of the ATT with the nearest neighbor matching method

Random draw version

****************************************************************

Note: the common support option has been selected

The region of common support is [.11038721, .92812903]

The outcome is fconeve

Variable | Obs Mean Std. Dev. Min Max

------+------

fconeve | 321 .2959502 .4571814 0 1

The treatment is fcinfn0

fcinfn0 | Freq. Percent Cum.

------+------

no one spoke about FC | 168 53.16 53.16

someone spoke about FC positively or b | 148 46.84 100.00

------+------

Total | 316 100.00

The distribution of the pscore is

Pr(fcinfn0)

------

Percentiles Smallest

1% .1148199 .1103872

5% .1300056 .110692

10% .1603485 .111217 Obs 321

25% .3009036 .1148199 Sum of Wgt. 321

50% .4297598 Mean .4620596

Largest Std. Dev. .2242147

75% .619251 .9200741

90% .8168924 .9227483 Variance .0502722

95% .8728809 .9262863 Skewness .3600555

99% .9200741 .928129 Kurtosis 2.199449

The program is searching the nearest neighbor of each treated unit.

This operation may take a while.

****************************************************

Forward search

****************************************************

Backward search

****************************************************

Choice between backward or forward match

****************************************************

Display of final results

****************************************************

The number of treated is 148

The number of treated which have been matched is 148

Average absolute pscore difference between treated and controls

Variable | Obs Mean Std. Dev. Min Max

------+------

PSDIF | 148 .0063397 .0107588 4.65e-06 .0552482

Average outcome of the matched treated

Variable | Obs Mean Std. Dev. Min Max

------+------

fconeve | 148 .3918919 .4898304 0 1

Average outcome of the matched controls

Variable | Obs Weight Mean Std. Dev. Min Max

------+------

fconeve | 69 148 .1824324 .3890299 0 1

ATT estimation with Nearest Neighbor Matching method

(random draw version)

Analytical standard errors

------

n. treat. n. contr. ATT Std. Err. t

------

148 69 0.209 0.083 2.524

------

Note: the numbers of treated and controls refer to actual

nearest neighbour matches

Bootstrapping of standard errors

command: attnd fconeve fcinfn0 know sex age hivseek fcblfrc stinohepc mcsex30 fcsee fcmediaS hiv , pscore() logit comsup

statistic: attnd = r(attnd)

note: label truncated to 80 characters

Bootstrap statistics Number of obs = 398

Replications = 100

------

Variable | Reps Observed Bias Std. Err. [95% Conf. Interval]

------+------

attnd | 100 .2094595 -.0723294 .082463 .0458349 .373084 (N)

| -.0214286 .2875 (P)

| .1102941 .3178295 (BC)

------

Note: N = normal

P = percentile

BC = bias-corrected

ATT estimation with Nearest Neighbor Matching method

(random draw version)

Bootstrapped standard errors

------

n. treat. n. contr. ATT Std. Err. t

------

148 69 0.209 0.082 2.540

------

Note: the numbers of treated and controls refer to actual

nearest neighbour matches

repeat analysis without strong predictor, as suggested by reviewer:

w/o fcsee and fcmediaS

Analytical standard errors

------

n. treat. n. contr. ATT Std. Err. t

------

148 81 0.230 0.068 3.367

------

Note: the numbers of treated and controls refer to actual

nearest neighbour matches

Bootstrapped standard errors

------

n. treat. n. contr. ATT Std. Err. t

------

148 81 0.230 0.073 3.159

------

Note: the numbers of treated and controls refer to actual

repeat analysis without strong predictor, as suggested by reviewer:

w/o fcsee and fcmediaS and hivseek

Analytical standard errors

------

n. treat. n. contr. ATT Std. Err. t

------

148 83 0.203 0.069 2.933

------

Note: the numbers of treated and controls refer to actual

nearest neighbour matches

Bootstrapped standard errors

------

n. treat. n. contr. ATT Std. Err. t

------

148 83 0.203 0.074 2.740

------

sensatt fconeve fcinfn0 know sex age hivseek fcblfrc stinohepc mcsex30 fcsee fcmediaS hiv, p( unprsx) reps(100) comsup logit bootstrap

as suggested by reviewer: [2ND]

The probability of having U=1 if T=1 and Y=1 (p11) is equal to: 0.37

The probability of having U=1 if T=1 and Y=0 (p10) is equal to: 0.61

The probability of having U=1 if T=0 and Y=1 (p01) is equal to: 0.42

The probability of having U=1 if T=0 and Y=0 (p00) is equal to: 0.60

The probability of having U=1 if T=1 (p1.) is equal to: 0.51

The probability of having U=1 if T=0 (p0.) is equal to: 0.57

The program is iterating the ATT estimation with simulated confounder.

You have chosen to perform 100 iterations. This step may take a while.

ATT estimation with simulated confounder

General multiple-imputation standard errors

------

ATT Std. Err. Out. Eff. Sel. Eff.

------

0.141 0.100 0.537 0.922

------

Note: Both the outcome and the selection effect

are odds ratios from logit estimations.

So the unobserved confounder U accounts for (.209-.141 )/.209 = .32% of the baseline estimate, but the effect on both outcome and selection are negative, so we try another dummy.

sensatt fconeve fcinfn0 know sex age hivseek fcblfrc stinohepc mcsex30 fcsee fcmediaS hiv, p11(0.40) p10(0.40) p01(0.20) p00(0.20) reps(100) comsup logit

The probability of having U=1 if T=1 and Y=1 (p11) is equal to: 0.40

The probability of having U=1 if T=1 and Y=0 (p10) is equal to: 0.40

The probability of having U=1 if T=0 and Y=1 (p01) is equal to: 0.40

The probability of having U=1 if T=0 and Y=0 (p00) is equal to: 0.40

The probability of having U=1 if T=1 (p1.) is equal to: 0.40

The probability of having U=1 if T=0 (p0.) is equal to: 0.40

ATT estimation with simulated confounder

General multiple-imputation standard errors

------

ATT Std. Err. Out. Eff. Sel. Eff.

------

0.135 0.099 1.133 1.035

------

Note: Both the outcome and the selection effect

are odds ratios from logit estimations.

So the unobserved confounder U accounts for (.209-.135 )/.209 = .35% of the baseline estimate, which seems sizeable in fact.

pscore fcinfn0 know sex age hivseek fcblfrc stinohepc mcsex30 fcsee fcmediaS hiv, pscore( psb4c10) blockid( bl4cov10) comsup numblo(4) level(0.005) logit

*the solution with 5 blocks yielded a 5th group of 6 in non-exposed (30 in exposed), so tried 4 blocks;

The region of common support is [.11038721, .92812903]

Description of the estimated propensity score in region of common support

Estimated propensity score

------

Percentiles Smallest

1% .1148199 .1103872

5% .1300056 .110692

10% .1603485 .111217 Obs 321

25% .3009036 .1148199 Sum of Wgt. 321

50% .4297598 Mean .4620596

Largest Std. Dev. .2242147

75% .619251 .9200741

90% .8168924 .9227483 Variance .0502722

95% .8728809 .9262863 Skewness .3600555

99% .9200741 .928129 Kurtosis 2.199449

******************************************************

Step 1: Identification of the optimal number of blocks

Use option detail if you want more detailed output

******************************************************

The final number of blocks is 4

This number of blocks ensures that the mean propensity score

is not different for treated and controls in each blocks

**********************************************************

Step 2: Test of balancing property of the propensity score

Use option detail if you want more detailed output

**********************************************************

The balancing property is satisfied

This table shows the inferior bound, the number of treated and the number of controls for each block

Inferior |

of block | fcinfn0

of pscore | no one s someone | Total

------+------+------

.1103872 | 47 14 | 61

.25 | 86 50 | 136

.5 | 27 47 | 74

.75 | 8 37 | 45

------+------+------

Total | 168 148 | 316

Note: the common support option has been selected

. logit fcinfn0 know sex age hivseek fcblfrc stinohepc mcsex30 fcsee fcmediaS hiv, nolog

Logistic regression Number of obs = 348

LR chi2(10) = 89.75

Prob > chi2 = 0.0000

Log likelihood = -192.4429 Pseudo R2 = 0.1891

------

fcinfn0 | Coef. Std. Err. z P>|z| [95% Conf. Interval]

------+------

know | .0723086 .6987278 0.10 0.918 -1.297173 1.44179

sex | .4650407 .255435 1.82 0.069 -.0356028 .9656842

age | -.0173805 .0119044 -1.46 0.144 -.0407126 .0059516

hivseek | .7305862 .2642011 2.77 0.006 .2127614 1.248411

fcblfrc | .4206706 .4329957 0.97 0.331 -.4279853 1.269326

stinohepc | .4436606 .2616602 1.70 0.090 -.069184 .9565053

mcsex30 | .2827335 .2529966 1.12 0.264 -.2131307 .7785976

fcsee | 1.506495 .332786 4.53 0.000 .8542462 2.158743

fcmediaS | 1.415616 .3826396 3.70 0.000 .6656558 2.165575

hiv | .1551879 .1512385 1.03 0.305 -.141234 .4516099

_cons | -3.767882 1.381353 -2.73 0.006 -6.475284 -1.06048

------

. linktest, nolog

Logistic regression Number of obs = 348

LR chi2(2) = 89.78

Prob > chi2 = 0.0000

Log likelihood = -192.42592 Pseudo R2 = 0.1892

------

fcinfn0 | Coef. Std. Err. z P>|z| [95% Conf. Interval]

------+------

_hat | .9934277 .1310549 7.58 0.000 .7365647 1.250291

_hatsq | -.0158106 .0858107 -0.18 0.854 -.1839965 .1523752

_cons | .0144738 .1501139 0.10 0.923 -.2797439 .3086916

------