Sie sind auf Seite 1von 16

Learning HIV Status: Evaluating the Use of

Incentives and Their Impact


Quinn Annand, Trenton Herriford, and Logan McDonald
June 12, 2015

The Experiment

In the last twenty years, the HIV/AIDS epidemic has affected a large population in
African countries like Malawi. Many governments and organizations have asked for increasing investments for counseling and testing for HIV. Testing for HIV can be an effective way to stop the spread of this disease, but it is only effective if the patients volunteer
to counseling and testing. The governments of African countries have spent millions of
dollars on this issue. For example, in South Africa, 55 percent of all HIV/AIDS program
spending in 2000 was for counseling and testing, including emphasis on door-to-door testing, a costly way of ensuring people receive their test results. In her paper, The Demand
for, and Impact of, Learning HIV Status, Rebecca Thornton evaluates an experiment
in which individuals in rural areas of Malawi were randomly given monetary incentives
to induce the demand to learn HIV results rather than sending nurses door-to-door to
deliver the news. Through this experiment, Thornton finds that the barriers to obtaining
HIV test results are relatively low. Furthermore, because receiving a HIV-positive diagnosis has a significant effect on the future purchase of condoms, learning the outcome
of your test has a sizable impact on the transmission of HIV/AIDS. Thus, monetary incentives rather than door-to-door HIV testing may be a more efficient and effective HIV
prevention strategy.
Thorntons paper randomizes individual incentives to learn HIV status and the location of the testing centers where the results are available in order to avoid the problems of
selection and reporting bias. The design of the experiment, in this way, relies on the fact
1

that the factors influencing the decision to learn HIV results correlates with behavioral
outcomes, which presents the potential bias that the impact of learning HIV results may
have on the demand for safe sex. Even though it would be unethical to exclude any person
from the random testing and counseling, the experiment can examine the effectiveness
of incentives by stratifying the dollar amounts. To avoid this dilemma, nurses went to
randomly assigned respondents in rural Malawi and offered a free door-to-door HIV test.
They then gave the respondents a voucher valued between zero and three dollars that
one could redeem when obtaining the results at their nearby center. The values of the
vouchers were randomized by letting each respondent draw tokens from a bag. The author notes that receiving no incentive or a zero from the bag may have resulted in a
demotivating effect on the individuals wanting to pick up their results already; however,
this effect was impossible to measure, and the author believes the effect was probably
minimal.
Two to four months after the testing, the results were available for pick-up at the
temporary centers placed at random spots based on participants clustered geospatial
coordinates. When the participants came to receive their results, they also received
counseling lasting for an average of 30 minutes, and those who were positive were referred
for even more counseling. Two months after the results were available, respondents tested
for HIV in Balaka and Rumphi districts were reinterviewed as a follow-up. They were
asked about their behavior over the two month period and their attitudes towards sexual
health. At the end of the interview, the respondents were both offered thirty extra cents
for their participation and the opportunity to purchase condoms at a subsidized rate
(thus from their thirty cents). All of the responses and condom purchases were recorded
by the interviewer.
There were several other considerations in undertaking this experiment and making
sure it was valid. For instance, Thornton paid close attention to the fact psychological
costs associated with learning HIV results may be an unseen variable preventing patients
from learning their scores. For this reason, Thornton made sure to include if the patient
was HIV positive in her analysis. To further support internal validity, Thornton shows
that there was a lack of differential attrition associated with incentive levels or distance
in the experiment. In terms of external validity, Thornton points to the fact that baseline
characteristics in 2004 are similar to those of a population-based survey also conducted

in Malawi in 2004. For example, 72.8 percent of women living in rural Malawi were
married (71 percent in Thorntons data) and 4.3 percent of women and 13.1 percent of
men reported using a condom at last intercourse (16.4 and 27.3 percent respectively in
Thorntons data).

Summary Statistics

We were able to successfully replicate the summary statistics used in Thorntons paper
through the data provided. In order to replicate the observations Thornton used, we took
complete cases of any incentive offered or not, complete cases of demographic data, and all
HIV tests completed that were not inconclusive (fourteen respondents were inconclusive).
Table 1 shows our replicated results.
Some things we can notice from the summary statistics include the fact 76 percent of
respondents reported having sex twelve months before the baseline, the average amount
of times each respondent reported having sex each month was four, and only 21 percent
reported using a condom in the past 12 months. Men comprised a little under half of the
group and were more likely to purchase condoms than women. The average number of
years of education at the baseline was 3.6 years, a number that will become important
for our evaluation of this study later in the paper. Most of the respondents owned land,
although slightly more people in the control were land-owners than in the treatment
group.
Table 2 and table 3 compare the control and treatment groups using summary statistics. From these results, we can see that the means for the treatment and controls are
very similar in terms of the respondents age, years of education, HIV results, distance
from the reported centers, and reported sexual behavior. Intuitively, we can confirm that
the control and treatment are similar in order to draw conclusions from their comparison;
that is, the control group is a good counterfactual.

Measures and Methodology

Again, Thornton is evaluating the impact of monetary incentives on whether or not an


individual retrieves HIV test results. She does so by constructing a linear probability

Table 1: Main Sample Summary Statistics


(a) Main sample

Statistic

Mean

St. Dev.

Sex (Male)

0.46

0.50

Age

33.38

13.65

Married

0.71

0.45

Years of Completed Education at the Baseline

3.64

3.73

Owned Land at the Baseline

0.73

0.44

HIV Results (positve)

0.06

0.24

Had a HIV Test Before Baseline

0.18

0.38

Reported Having Sex 12 Months Before Baseline

0.76

0.43

Reported using a Condom in Past 12 Months

0.21

0.41

Times per Month Reported Having Sex

3.90

10.84

Distance From Randomly Assigned Center

2.02

1.27

(b) Main Sample Units that Completed Follow-Up Survey

Statistic

Mean

St. Dev.

Sex (Male)

0.46

0.50

Age

34.59

14.32

Married

0.72

0.45

Years of Completed Education at the Baseline

3.83

3.84

Owned Land at the Baseline

0.74

0.44

HIV Results (positve)

0.05

0.21

Had a HIV Test Before Baseline

0.22

0.41

Reported Having Sex 12 Months Before Baseline

0.76

0.43

Reported using a Condom in Past 12 Months

0.21

0.41

Times per Month Reported Having Sex

3.65

9.48

Distance From Randomly Assigned Center

2.20

1.34

Number of Condoms Purchased in Follow-Up Survey

0.88

1.89

Indicator of Any Condoms Purchased in Follow-Up Survey

0.24

0.43

Table 2: Control Group Summary Statistics


(a) Control Group

Statistic

Mean

St. Dev.

Sex (Male)

0.47

0.50

Age

32.11

12.81

Married

0.71

0.45

Years of Completed Education at the Baseline

4.53

3.91

Owned Land at the Baseline

0.66

0.47

HIV Results (positve)

0.06

0.24

Had a HIV Test Before Baseline

0.21

0.41

Reported Having Sex 12 Months Before Baseline

0.78

0.41

Reported using a Condom in Past 12 Months

0.25

0.43

Times per Month Reported Having Sex

2.47

7.21

Distance From Randomly Assigned Center

1.95

1.22

(b) Control Group Units that Completed Follow-Up Survey

Statistic

Mean

St. Dev.

Sex (Male)

0.46

0.50

Age

33.38

13.48

Married

0.73

0.45

Years of Completed Education at the Baseline

4.88

3.90

Owned Land at the Baseline

0.63

0.48

HIV Results (positve)

0.05

0.22

Had a HIV Test Before Baseline

0.27

0.44

Reported Having Sex 12 Months Before Baseline

0.76

0.43

Reported using a Condom in Past 12 Months

0.28

0.45

Times per Month Reported Having Sex

2.53

7.54

Distance From Randomly Assigned Center

2.04

1.25

Number of Condoms Purchased in Follow-Up Survey

0.70

1.92

Indicator of Any Condoms Purchased in Follow-Up Survey

0.18

0.39

Table 3: Treatment Group Summary Statistics


(a) Treatment Group

Statistic

Mean

St. Dev.

Sex (Male)

0.46

0.50

Age

33.74

13.86

Married

0.71

0.45

Years of Completed Education at the Baseline

3.39

3.64

Owned Land at the Baseline

0.75

0.43

HIV Results (positve)

0.06

0.24

Had a HIV Test Before Baseline

0.17

0.38

Reported Having Sex 12 Months Before Baseline

0.76

0.43

Reported using a Condom in Past 12 Months

0.20

0.40

Times per Month Reported Having Sex

4.30

11.61

Distance From Randomly Assigned Center

2.03

1.28

(b) Treatment Group Units that Completed Follow-Up Survey

Statistic

Mean

St. Dev.

Sex (Male)

0.46

0.50

Age

34.92

14.53

Married

0.72

0.45

Years of Completed Education at the Baseline

3.54

3.77

Owned Land at the Baseline

0.76

0.42

HIV Results (positve)

0.05

0.21

Had a HIV Test Before Baseline

0.20

0.40

Reported Having Sex 12 Months Before Baseline

0.76

0.43

Reported using a Condom in Past 12 Months

0.20

0.40

Times per Month Reported Having Sex

3.97

9.95

Distance From Randomly Assigned Center

2.24

1.36

Number of Condoms Purchased in Follow-Up Survey

0.93

1.88

Indicator of Any Condoms Purchased in Follow-Up Survey

0.26

0.44

model with the corresponding binary response variable (retrieving test results). The
advantage of a probability model is its ability to estimate unobservables (e.g., utility,
etc.) effect on the response, as these unobservable, or latent, variables become proxied
by the specified explanatory variables.
There are certainly many factors that will influence the individuals decision to retrieve
their test results. This study is primarily interested in opportunity cost, i.e., the cost of
retrieving the test results. Theoretically, a monetary incentive will reduce this cost, but
there still is the issue of the effort to travel a certain distance and the time it takes to get
there. To measure this, Thorton includes the distance from the temporary test results
(VCT) centers as a variable as well as obviously including the monetary incentive.
As stated above, there may also be psychological costs, warranting the inclusion of a
variable to proxy for this. One may argue that Thorton included actually having HIV in
the model for this reason those with HIV may actually fear receiving results a priori.
Alternatively, this may also simply be seen as demographic data, which would also have
an effect: Thorton considers age, sex, and location of residency by district.
Thornton reports five OLS regressions and their respective marginal probit models,
utilizing the variables described above. To measure the demand for HIV results in a
regression framework, Thornton estimates:
GotResults =0 + 1 AnyIncentive + . . .
+ 2 AmountIncentive + 3 AmountIncentive2 + . . .

+ 4 Distance + 5 Distance2 + X 0 + .
The dependent variable, GotResults, is a binary variable that indicates whether or

not individual received HIV results. The independent variable X is a vector that includes
covariates of gender, age, age-squared, HIV status, and district dummies, as well as a
control for a simulated average distance in each VCT zone. AnyIncentive is a dummy
variable indicating whether or not any monetary incentive was given to an individual
at the time of the HIV testing. The five OLS models that she reports are variations of
specifications with the listed variables.
In addition to her variables, we also consider the effects of education, which might also
add to the opportunity cost of retrieving test results. This is to say that someone who is
more educated is presumably able to earn more income in the time that it takes to pick up
the test results than someone with less education, other things equal. Additionally, there
7

are different income effects one also should consider. Of course, one would ideally use
household income or expenditures, but neither of these variables were provided. However,
education might be a better proxy for rationality or maturity, which may affect the
psychological costs such that they are lower.
Figure 1: Histogram showing the frequency of different years of education in the main
sample

Because education often does not have continuous returns to income but instead
discrete returns based on levels of education (e.g., primary school, secondary school, etc.),
we seek to create an appropriate dummy variable. To do so, we observe the frequency of
the years of education of units in the main sample, which are shown in Figure 1. One
will see the sharp decline right after 7 years. For this reason, we note that the effects on
income are likely different between those with 1-7 years of education and those with 8 or
more and construct or models accordingly because there is probably some infrastructure
supporting school up to the seventh year.

Results

Using the variables and regression framework that Thornton uses to estimate the impact
of incentives on learning HIV status, we were able to replicate her results. Just as
Thornton reports, the estimates for AnyIncentive compare the treatment group (those
who received any incentive, n=2194) to the control group (those who did not, n=618).
The main result from the regression analysis is that those who receive a monetary
incentive are significantly more likely to retrieve HIV test results than those who do not.
In Thorntons most basic model (column 1, Table 4), the estimated effect of receiving any
8

Table 4: Impact of Monetary Incentives and Distance on Learning HIV Test Results
(Dependent variable: Attendance at HIV results centers)
GotResults
got
OLS

Any Incentive

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

0.431

0.309

0.218

0.220

0.219

1.187

0.762

0.488

0.504

0.495

(0.020)

(0.024)

(0.031)

(0.031)

(0.031)

(0.063)

(0.079)

(0.102)

(0.103)

(0.103)

0.093

0.277

0.277

0.276

0.343

0.934

0.929

0.927

(0.011)

(0.042)

(0.042)

(0.042)

(0.041)

(0.147)

(0.147)

(0.147)

0.064

0.065

0.064

0.212

0.210

0.209

(0.014)

(0.014)

(0.014)

(0.050)

(0.051)

(0.050)

Amount of Incentive (USD)

Amount of Incentive Squared

HIV Status

Distance (KM)

probit

(1)

0.055

0.052

0.050

0.058

0.055

0.175

0.179

0.171

0.196

0.188

(0.033)

(0.033)

(0.032)

(0.032)

(0.033)

(0.108)

(0.109)

(0.110)

(0.110)

(0.110)

0.010

0.013

0.015

0.015

0.014

0.031

0.048

0.055

0.055

0.053

(0.016)

(0.016)

(0.016)

(0.016)

(0.016)

(0.053)

(0.054)

(0.054)

(0.054)

(0.054)

Distance-Squared

0.076

0.278

(0.024)

(0.083)

Over 1.5 KM

0.010

0.039

(0.005)

(0.016)

Male

Age (years)

Age-Squared

0.004

0.004

0.004

0.004

0.015

0.014

0.013

0.014

0.015

(0.003)

(0.003)

(0.003)

(0.003)

(0.010)

(0.010)

(0.010)

(0.010)

(0.010)

0.00004

0.00003

0.00003

0.00003

0.00003

0.0001

0.0001

0.0001

0.0001

0.0001

(0.00004)

(0.00004)

(0.00004)

(0.00004)

(0.00004)

(0.0001)

(0.0001)

(0.0001)

(0.0001)

(0.0001)

0.005

0.007

0.016

0.024

(0.011)

(0.010)

(0.039)

(0.033)

0.137

0.155

(0.020)

0.156

(0.020)

0.123

0.151

(0.020)

0.118

0.161

(0.020)

Constant

(0.061)

0.005

Balaka

0.130

(0.018)

(0.003)

Simulated Average Distance

Rumphi

0.037

(0.020)

0.122

0.114

0.109

(0.020)

(0.020)

(0.020)

(0.023)

(0.023)

0.346

(0.056)

0.427

0.383

0.404

(0.070)

(0.071)

(0.082)

(0.082)

0.322

0.302

0.036

0.216

(0.187)

(0.188)

(0.209)

(0.193)

2,812

2,812

2,812

2,812

2,812

1,489.970

1,451.964

1,443.222

1,433.702

1,439.314

2,995.939

2,921.928

2,906.444

2,893.404

2,902.629

2,812

2,812
0.210

0.179

0.199

0.205

0.209

0.206

Log Likelihood
Akaike Inf. Crit.
0.413 (df = 2803)

88.451

(0.072)

(0.185)

0.213

(df = 7; 2804)

(0.056)

2,812

(df = 8; 2803)

0.412 (df = 2802)

81.494

(df = 9; 2802)

0.411 (df = 2799)


62.989

(df = 12; 2799)

0.381

(0.072)

0.395

0.207

88.742

0.429

0.543

0.528

(0.071)

(0.061)

2,812

F Statistic

(0.070)

0.557

0.442

0.202

0.412

(0.071)

(0.055)

2,812

0.418 (df = 2804)

0.534

0.370

0.181

Residual Std. Error

(0.070)

(0.055)

R2
Adjusted R

0.462

0.365

Observations

0.411 (df = 2800)


67.493

(df = 11; 2800)

Note:

p<0.1;

p<0.05;

p<0.01

monetary incentive is an increase of 43 percentage points in the probability of retrieving


results, statistically significant at 1%. This result shows a large effect on retrieving results.
Recall that, because of randomization of the monetary incentives, this study meets the
requirements of internal validity, so this result should be taken seriously.1
The effect of incentive on probability of retrieving scores increases as amount of incentive increases at a rate of 9.3 percentage points for every additional dollar (column
2, Table 4) and displays diminishing marginal returns (column 3, Table 4). The estimates from the second OLS regression (column 2, Table 4) predict that holding all else
constant, an incentive of 50 cents, would increase the probability an individual retrieves
their results by 35.5 percentage points. These results show that very small incentives can
have a large effect on demand for knowledge of HIV status.
The second major result is the effect of distance from the nearest testing center on
probability of retrieving HIV results. Thornton finds that as distance increases, there is a
negative impact on the probability of retrieving results (column 4, Table 4), significant at
1%. This result is consistent with the effect of monetary incentives. As distance increases,
the cost of retrieving results also increases due to opportunity and transportation costs.
The higher the cost, the less likely an individual will retrieve their results. Both shorter
distances to the testing centers and monetary incentives lower the cost, and increase
the demand for knowledge of HIV status. The simulated average distance is included
to ensure internal validity, given that the locations of the VCT centers were randomly
assigned and not the distances individuals must travel to get to the testing centers.
Other potentially interesting results include the estimated effects of gender and age,
which remained statistically insignificant and near zero through out. Additionally, those
with HIV were 5.5 percentage points less likely to retrieve results than those who did not
have HIV, statistically significant at 10% (column 5, Table 4). Finally, the marginal probit
effects (reported in the appendix), were consistent with the OLS estimates, eliminating
1

Thorton notes that in the pilot study nurses would have respondents redraw the incentives if they

received a zero voucher, removing the randomization. Nurses were subsequently told for this study
that they would loose there jobs if they did not follow the rules of randomization. Some geographical areas showed that this sympathy might still exist. When using observables that may proxy for
this sympathy as instruments for receiving an incentive in this area, we found that the results were
fairly robust to the regressions above. Given this and the rigor of Thortons analysis in the area of
randomization, we did not explore this issue any further.

10

any biasing due to the binary nature of the dependent variable.


In regards to education, Thorton runs a regression that is not shown, but we include it
in our study. For our treatment of this variable, we use the OLS model shown in column
4 in Table 4 as our base model because of its straightforward treatment of distance,
which is reproduced in column 1 in Table 5. The effect of having an education and
the effect compared to those without in the second model (column 2, Table 5) and the
effect of 1-7 years of education in the third model (column 3, Table 5) both were not
statistically significant at any appropriate significance level. However, having 8 more
years of education (column 3, Table 5) reduced the probability of retrieving test results
by 7.6 percentage points, statistically significant at 1%. As one can see, this model also
shows the robustness of the base models estimates given the addition of the education
variable.
When comparing our results with Thortons, one will notice that our standard errors
are not the same. Thorton uses robust standard errors based on village location. This
data and method was not available us, and our standard errors were close enough so that
it did not affect the statistical significance of the variables.

Conclusion

In summary, we were able to replicate Thortons results, showing that there is a positive,
substantial and statistically significant effect on retrieving test scores from providing a
monetary incentive. We also included an education variable, which also had a statistically
significant (yet negative) effect on getting test results. Further, the estimates were robust
to this addition. In the process of replicating this data, many other models were tested,
but the education addition was the only one that warranted coverage in this paper, given
its nice heuristic Thortons models seemed robust to any addition.
That said, there are several things we thought would improve this experiment. First
of all, we found the second part of the experiment on changing behavior to be problematic
and, accordingly, did not present this here. It is hard to tell if increased condom purchases
in a single moment is an externally valid way of indicating sexual behavior for a group.
Secondly, we had some ethical and internal validity questions about giving some participants zero incentives. The randomization via pulling numbers out of a hat may ensure

11

Table 5: Impact of Monetary Incentives and Distance on Learning HIV Test Results with
education (OLS models)

GotResults

Any Incentive

Amount of Incentive (USD)

Amount of Incentive Squared (USD)

HIV Status

Male

Distance (KM)

Distance Squared

Age (years)

Age Squared

Simulated Average Distance

Rumphi

Balaka

(1)

(2)

(3)

0.220

0.229

0.230

(0.031)

(0.032)

(0.032)

0.277

0.279

0.281

(0.042)

(0.043)

(0.043)

0.065

0.066

0.067

(0.014)

(0.015)

(0.015)

0.058

0.061

0.060

(0.032)

(0.033)

(0.033)

0.015

0.011

0.008

(0.016)

(0.019)

(0.019)

0.076

0.071

0.074

(0.024)

(0.024)

(0.024)

0.010

0.009

0.009

(0.005)

(0.005)

(0.005)

0.004

0.003

0.003

(0.003)

(0.003)

(0.003)

0.00003

0.00002

0.00002

(0.00004)

(0.00004)

(0.00004)

0.005

0.015

0.016

(0.011)

(0.012)

(0.012)

0.151

0.152

0.135

(0.020)

(0.022)

(0.023)

0.114

0.117

0.116

(0.023)

(0.024)

(0.024)

Any education

0.032
(0.023)

0.014

1-7 years education

(0.024)

0.076

8 years or more education

(0.028)

0.442

0.437

0.470

(0.061)

(0.067)

(0.077)

Observations

2,812

2,530

2,530

R2

0.213

0.224

0.227

Adjusted R2

0.209

0.220

0.222

0.411 (df = 2799)

0.402 (df = 2516)

0.401 (df = 2515)

62.989 (df = 12; 2799)

56.027 (df = 13; 2516)

52.637 (df = 14; 2515)

Constant

Residual Std. Error


F Statistic

Note:

12

p<0.1;

p<0.05;

p<0.01

randomized results, but it could lead to some disappointment on the part of those that
received no incentive. In a way, the disappointment of not getting an incentive could act
as a disincentive, so if any positive incentives had such a powerful effect to push people
one way or another, it is reasonable to think that rejection might have the same effect.
Alternatively, this effect could also simply give the the estimate of incentive effect an
upward bias. Thus, a different method of randomization might be considered to solve
this problem, and it should be implemented in further studies. Given this method, It one
might also question the ethics of giving a person who is HIV-positive no incentive to pick
up their results.
One should also note that even though the outcomes of this experiment show us that
incentives work, an argument for continuing the door-to-door approach to inform patients
could be the levity of HIV-status. That is, ethically, even though money could be saved
with using incentives, door-to-door work will always be more effective. Future program
officers should consider this before implementing incentive programs in the future.
Finally, while education may serve as a proxy for an output, it would be nice to see
how these results compare across income or consumption and if that is all a factor in
the effect of the incentive. Overall, we found this to be a highly effective paper, but
when dealing with something as sensitive as HIV results, we believe ethical factors in
the construction of the experiment should be taken into account regardless of other
estimates robustness.

13

Appendix

The following tables are the analogous marginal effects tables from the probit models
presented in Table 4. Each marginal effect corresponds to the change in probability
of retrieving test results given the increase in the variable in question when all other
variables are held at their mean.
Marginal effects corresponding to column 6, Table 4

dF/dx

Std. Err.

P>|z|

Any Incentive

0.438

0.022

19.834

HIV Status

-0.062

0.040

-1.563

0.118

Male

-0.011

0.018

-0.586

0.558

Age (years)

0.005

0.003

1.534

0.125

Age-Squared

-0.00004

0.00004

-0.901

0.368

Rumphi

-0.163

0.025

-6.549

Balaka

-0.144

0.025

-5.849

Marginal effects corresponding to column 7, Table 4

dF/dx

Std. Err.

P>|z|

Any Incentive

0.279

0.030

9.223

Amount of Incentive (USD)

0.116

0.014

8.506

HIV Status

-0.063

0.040

-1.582

0.114

Male

-0.016

0.018

-0.895

0.371

Age (years)

0.005

0.003

1.400

0.162

Age-Squared

-0.00003

0.00004

-0.697

0.486

Rumphi

-0.187

0.025

-7.427

Balaka

-0.149

0.025

-6.022

14

Marginal effects corresponding to column 8, Table 4

dF/dx

Std. Err.

P>|z|

Any Incentive

0.175

0.039

4.546

0.00001

Amount of Incentive (USD)

0.315

0.049

6.385

Amount of Incentive Squared

-0.071

0.017

-4.212

0.00003

HIV Status

-0.060

0.040

-1.504

0.133

Male

-0.018

0.018

-1.007

0.314

Age (years)

0.004

0.003

1.352

0.176

Age-Squared

-0.00003

0.00004

-0.646

0.518

Rumphi

-0.195

0.025

-7.682

Balaka

-0.148

0.025

-5.962

Marginal effects corresponding to column 9, Table 4

dF/dx

Std. Err.

P>|z|

Any Incentive

0.181

0.039

4.673

0.00000

Amount of Incentive (USD)

0.313

0.049

6.329

Amount of Incentive Squared

-0.071

0.017

-4.152

0.00003

HIV Status

-0.069

0.040

-1.718

0.086

Male

-0.018

0.018

-1.002

0.316

Distance (KM)

-0.094

0.028

-3.349

0.001

Distance-Squared

0.013

0.005

2.428

0.015

Age (years)

0.005

0.003

1.418

0.156

Age-Squared

-0.00003

0.00004

-0.693

0.489

Simulated Average Distance

0.005

0.013

0.399

0.690

Rumphi

-0.184

0.026

-7.177

Balaka

-0.139

0.029

-4.872

0.00000

15

Marginal effects corresponding to column 10, Table 4

dF/dx

Std. Err.

P>|z|

Any Incentive

0.178

0.039

4.604

0.00000

Amount of Incentive (USD)

0.313

0.049

6.330

Amount of Incentive Squared

-0.070

0.017

-4.149

0.00003

HIV Status

-0.066

0.040

-1.650

0.099

Male

-0.018

0.018

-0.974

0.330

Over 1.5 KM

-0.044

0.020

-2.147

0.032

Age (years)

0.005

0.003

1.469

0.142

Age-Squared

-0.00003

0.00004

-0.742

0.458

Simulated Average Distance

-0.008

0.011

-0.734

0.463

Rumphi

-0.189

0.026

-7.397

Balaka

-0.132

0.029

-4.618

0.00000

16

Das könnte Ihnen auch gefallen