|
Field
Last Published
|
Before
May 29, 2016 02:21 PM
|
After
June 06, 2016 03:54 PM
|
|
Field
Primary Outcomes (End Points)
|
Before
Our key outcomes of interest are:
- Whether or not driver has a receipt (both conditional on reporting a trip and not)
- Amount paid by driver at the toll (both conditional on not receiving a receipt and not), according to multiple measurements
- Number of trips taken
|
After
Our key outcomes of interest are:
- Whether or not driver has a receipt (both conditional on reporting a trip and not)
- Amount paid by driver at the toll (both conditional on not receiving a receipt and not), according to multiple measurements
- Number of trips taken
- Views of government corruption
|
|
Field
Primary Outcomes (Explanation)
|
Before
Our outcomes are constructed as follows:
- Whether or not driver has a receipt: dummy that is 1 if the driver brings a valid receipt (valid indicates that it is printed by the toll receipt printers, it is for a motorcycle trip, it is on the date the driver reports for his trip, it has the driver’s name, and it is from the toll the driver reported passing) and 0 otherwise
- Amount paid by driver at the toll: this will be measured 3 ways (direct self-report, writing an amount on a piece of paper that is not viewed by the enumerator, and using an “amount experiment” that allows us to estimate the driver’s reported bribe without him revealing it directly)
- Number of trips taken: integer count of trips, measured by self-report and for a sub-sample by an alternative method of verification involving direct observation
|
After
Our outcomes are constructed as follows:
- Whether or not driver has a receipt: dummy that is 1 if the driver brings a valid receipt (valid indicates that it is printed by the toll receipt printers, it is for a motorcycle trip, it is on the date the driver reports for his trip, it has the driver’s name, and it is from the toll the driver reported passing) and 0 otherwise
- Amount paid by driver at the toll: this will be measured 3 ways (direct self-report, writing an amount on a piece of paper that is not viewed by the enumerator, and using an “amount experiment” that allows us to estimate the driver’s reported bribe without him revealing it directly)
- Number of trips taken: integer count of trips, measured by self-report and for a sub-sample by an alternative method of verification involving direct observation
- Views of government corruption: measured by perceptions based questions about government corruption
|
|
Field
Planned Number of Clusters
|
Before
We will have approximately 1200 drivers and we will cluster at the level of the driver. For analyses using toll officer treatments, we will do a second clustering by toll-day.
|
After
We will have approximately 1000 drivers and we will cluster at the level of the driver. For analyses using toll officer treatments, we will do a second clustering by toll-day.
|
|
Field
Planned Number of Observations
|
Before
Our total number of observations will be 4800 driver-visits, of which 3600 will involve randomized treatments. The initial baseline will give us 1200 data points, but will not involve a randomized treatment.
|
After
Our total number of observations will be 4000 driver-visits, of which 3000 will involve randomized treatments. The initial baseline will give us 1000 data points, but will not involve a randomized treatment.
|
|
Field
Sample size (or number of clusters) by treatment arms
|
Before
Of the 3600 visits, we plan to assign:
1200 visits to control
1200 visits to cash rebates (600 to 1000FC, 300 to 2000FC, 300 to 2500 FC)
600 visits to pure charity donation
600 visits to donation via government
1200 visits to social norms “wall of pride” treatment (overlapping assignment with the final two visits by drivers to the office)
600 drivers to receive information about tax payment behavior of drivers in pilot
15% of toll-days are assigned to cash incentives for toll officers.
|
After
Of the 3000 visits, we plan to assign:
1000 visits to control
1000 visits to cash rebates (500 to 1000FC, 500 to 2000FC)
500 visits to pure charity donation
500 visits to donation via government
1000 visits to social norms “wall of pride” treatment (overlapping assignment with the final two visits by drivers to the office)
500 drivers to receive information about tax payment behavior of drivers in pilot
15% of toll-days are assigned to cash incentives for toll officers.
|
|
Field
Power calculation: Minimum Detectable Effect Size for Main Outcomes
|
Before
Our power depends on several variables besides effect size. These variables are: the proportion of total variation that is constant within individual (i.e. would be absorbed by a fixed effect), the rate at which drivers complete trips, and whether we use individual fixed effects or instead do pooled OLS. All of these calculations ignore the role of covariates.
In terms of our main outcomes, the standard deviations are:
1) Receipt rate: 30 percentage points
2) Amount paid at toll: 828 FC (41% of the value of the toll)
Under an assumption that 30% of the variation is constant within individual and an assumption that 65% of drivers complete trips (based on piloting), we simulated our outcomes. In the simulations, our power to detect effects is approximately the following (these are not exact due to the fact that we used simulations, not standard power calculations, because we are not aware of standard calculations that incorporate our design elements):
3) Linear effect of 0.05 standard deviations per 25% of the value of the toll cash rebate:
a. Pooled OLS (1), FE (1)
4) Specific dummies for each rebate:
a. 50% of toll: 0.16 standard deviations (OLS); 0.3 standard deviations (FE)
b. 100% of toll: 0.19 standard deviations (OLS); 0.33 standard deviations (FE)
c. 125% of toll: 0.19 standard deviations (OLS) ; 0.33 standard deviations (FE)
5) Charity: pooled OLS of 0.17 standard deviations, FE of 0.22 standard deviations
6) Government: pooled OLS of 0.17 standard deviations, FE of 0.22 standard deviations
7) Wall of pride: pooled OLS of 0.12 standard deviations, FE of 0.17 standard deviations
8) Social norms treatment: pooled OLS of 0.13 standard deviations; cannot be evaluated using FE regressions
We believe that these effects are plausible, since even 1/3 of the standard deviation of receipts received is only 10 percentage points and the incentives are large. We found similar sized effects to those we are powered to detect here in pilot activities.
|
After
Our power depends on several variables besides effect size. These variables are: the proportion of total variation that is constant within individual (i.e. would be absorbed by a fixed effect), the rate at which drivers complete trips, and whether we use individual fixed effects or instead do pooled OLS. All of these calculations ignore the role of covariates.
In terms of our main outcomes, the standard deviations are:
1) Receipt rate: 30 percentage points
2) Amount paid at toll: 828 FC (41% of the value of the toll)
Under an assumption that 30% of the variation is constant within individual and an assumption that 65% of drivers complete trips (based on piloting), we simulated our outcomes. In the simulations, our power to detect effects is approximately the following (these are not exact due to the fact that we used simulations, not standard power calculations, because we are not aware of standard calculations that incorporate our design elements):
3) Linear effect of 0.05 standard deviations per 25% of the value of the toll cash rebate:
a. Pooled OLS (1), FE (1)
4) Specific dummies for each rebate:
a. 50% of toll: 0.16 standard deviations (OLS); 0.3 standard deviations (FE)
b. 100% of toll: 0.16 standard deviations (OLS); 0.3 standard deviations (FE)
5) Charity: pooled OLS of 0.17 standard deviations, FE of 0.22 standard deviations
6) Government: pooled OLS of 0.17 standard deviations, FE of 0.22 standard deviations
7) Wall of pride: pooled OLS of 0.12 standard deviations, FE of 0.17 standard deviations
8) Social norms treatment: pooled OLS of 0.13 standard deviations; cannot be evaluated using FE regressions
We believe that these effects are plausible, since even 1/3 of the standard deviation of receipts received is only 10 percentage points and the incentives are large. We found similar sized effects to those we are powered to detect here in pilot activities.
|
|
Field
Intervention (Hidden)
|
Before
We plan to offer the following:
- Cash rebates varying between 25% and 125% of the value of the toll
- Donations either directly to a well-known social institution (e.g. widow’s group) OR to the same social institution, but via the government
- Information on tax compliance rates reported during pilot research activities
- Opportunity for drivers to be placed on a “wall of pride” if they bring a valid receipt
|
After
We plan to offer the following:
- Cash rebates of either 50% or 100% of the value of the toll
- Donations either directly to a well-known social institution (to support widows) OR to the same social institution, but via the government
- Information on tax compliance rates reported during pilot research activities
- Opportunity for drivers to be placed on a “wall of pride” if they bring a valid receipt
|