Knowing Thy Neighbor: What Information Neighbors Have and How Best to Elicit It

Our main project questions are as follows

1. What do people in peri-urban communities in low-income countries know about one another regarding
characteristics relevant to the distribution of scarce resources such as business grants?
(a) Do people distort their reports when they know they will be utilized for allocative decisions?
(b) Does the distortion of information vary by whether individuals are asked to report alone or in public?
(c) Can monetary incentives for accuracy induce more accurate reports about community members?
(d) How does the ranking mechanism (asking respondents to rank peers relative to one another in a small group versus relative to microentrepreneurs in the larger community) affect the quality and manipulability of the information provided?
(e) How does network proximity and “social capital” between network members affect the accuracy of the information provided?
(f) Can a cross-reporting mechanism (eg. asking respondents to state who their peers are likely to misreport about and who amongst their peers will have the most accurate rankings) enhance our ability to discern which community members’ reports are more valuable?

2. How does the information we collect compare to other methods of identifying entrepreneurs, such as demographic predictiveness or psychometric tests?
2.1 Description of the Experiment
Our main intervention is to randomly allocate $100 grants to microentrepreneurs, a shock of 60% of average microfirm capital in our sample. By randomly allocating grants, we can induce differential business growth in treatment firms. We will collect detailed business information from both grant recipients and non-recipients to test if community reports predicted whose businesses grew (Q1). As these grants are highly desirable, community members may lie about their peers if they believe their reports will be used to allocate grants.
We can therefore experimentally test whether community members manipulate their reports to benefit their friends and whether our mechanisms can reduce this perverse incentive. We answer the questions posed above using the following treatments:

1. Rankings Used for Allocation Decisions (R0 vs R1) – A random subset of households will be informed that their reports will influence the probability that their peers receive cash grants. The control group, whose rankings do not influence grant allocation, will provide an estimate of the quality of information embedded in communities in the absence of incentives to distort responses. Comparing
the two groups’ rankings will tell us whether community members strategically misreport information to affect allocative decisions (i.e. the possibility of receiving a grant). (Q2)

2. Public Rankings (P0 vs P1): Respondents will be randomized to make reports in public (visible to the entire group) or in private (group members do not see reports). The difference in the accuracy of reports between these two groups will tell us whether privacy is an important criterion for elicitation of peer information. (Q2a)

3. Incentives (I0 vs I1): We will randomize whether or not respondents receive monetary incentives for
the accuracy of their reports. (Q2b)

In addition to the main treatments, other features of the experiment will help us make richer predictions
about how to elicit truthful answers more accurately.

• All subjects will be asked to rank their peers relative to one another within the group and also relative
to the community at large. Specifically, we will ask respondents to give a relative ranking of the
respondents in the group (a unique rank of 1 to 5 for a 5-member group) and a quintile ranking in
which respondents report which quintile of the distribution of the ranking variable the rankee belongs to
vis-a-vis other similar entrepreneurs in the community. So for income, for example, we ask a respondent
to place a rankee in 1 of 5 quintiles for average monthly income in the community. Unlike the relative
rankings, quintile rankings are not unique such that the respondent could, in theory, place all rankees
in the top quintile. (Q2c)

• We will map out detailed social relations to understand how informational quality and magnitude of
informational distortions vary with social distance. (Q2d)


• We will ask respondents whom their peers are likely to lie about and whom amongst their peers will be
able to do each ranking exercise most accurately. We will test if we can use this information to weigh
each person’s reports. (Q2e)

2016-03-01

2016-04-30

Our outcomes of interest are whether people can predict

1. Education

(a) We only ask for quintile rankings

2. Marginal returns to a Rs.6000 grant

(a) We ask for relative and quintile rankings

3. Average monthly income over the past year

(a) We ask for relative and quintile rankings

4. Projected monthly profits with an Rs.6000 grant

(a) We ask for relative and quintile rankings

5. Total value of household assets

(a) We ask for relative and quintile rankings

6. Number of hours worked by business owner

7. Medical expenses

8. Loan repayment trouble

9. Digitspan

We also ask who deserves the grant (according to any criteria the respondent herself chooses).

We cross-randomize 3 treatments arms:

1. Rankings Used for Allocation Decisions (R0 vs R1) – A random subset of households will be informed that their reports will influence the probability that their peers receive cash grants.

2. Public Rankings (P0 vs P1): Respondents will be randomized to make reports in public (visible to the entire group) or in private (group members do not see reports).

3. Incentives (I0 vs I1): We will randomize whether or not respondents receive monetary incentives for the accuracy of their reports.

After all baseline data is collected in a neighborhood, groups will be invited to conduct the ranking exercises in a large community hall. One group will be invited to conduct the exercise at a time. Once in the hall, group members will be paired with a surveyor at a surveying station. The station is completely private with a section divider so that respondents cannot observe each others’ rankings while sitting with the surveyor. To minimize variation across surveyors in implementation of the treatments, we have created animated videos to show to respondents. In the videos we explain: what is a quintile and how to do a quintile ranking, what are marginal returns, profits, income, and assets.

The videos also explain the treatments. For the incentive treatments, the videos explain how incentives are paid. The payment rule is difficult to explain in a seminar, let alone to person with low levels of literacy. Following Prelec (2004) and Rigol Roth (2016), respondents are told that their payments will depend on their rankings and the rankings of their peers. The message that is emphasized is that truthful answers are rewarded more highly than untruthful answers. We also explain what second order beliefs are and collect this data along with the first order rankings. Incentives are paid at the end of each question (7 times in total). In the public treatment, after each ranking, respondents are asked to gather in the center of the room while the surveyors process the data. They take their ranking sheets along with them and these sheets are visible to all group members. In the private treatment, respondents are assured that their individual
rankings will never be observed by anyone other than the researchers. The respondents remain behind their privacy screens.

Lastly, in the revealed treatments, the videos explain how the rankings can affect who receives the grant. Respondents are told when they arrive in the hall that at the end of the exercise, a lottery will be conducted to randomly select the lottery winner. They are each given 20 tickets. In the revealed treatment, the video explains that there are extra lottery tickets that will be awarded after each IAP round (Q4-Q7).
The lottery tickets will be awarded to the person who was most highly ranked in each round. Once all rankings are completed, a lottery is conducted. Group members will put their lottery tickets in a bucket and one or two winners will be selected, depending on the randomization status of the group.

Public lottery

The randomization unit is the ranking group. There are 8 treatments so we created clusters of groups and randomized the 8 treatments within those clusters at the neighborhood leve.

Yes

36

1500

187 for 8 treatment arms

We are powered to detect treatment effects on marginal returns to a grant on par with De Mel, McKenzie, Woodruff (2008).