Representation through Information: Bringing Politician Actions Closer to Citizen Preferences

The interventions involve (a) providing local politicians with accurate information on citizen preferences and (b) giving them information about how accurate their existing beliefs about citizen preferences are.

2018-07-10

2018-10-31

Policy recommendations to political party

Demand for information, justifications for policy recommendations, willingness to take up a government scheme, proxy for honesty, perception of representatives

Individual local elected representatives are cross-randomized into the following treatment arms:

(1) Receiving information about the preferences of (one of six possible subsets of) citizens from within their National Assembly constituency.

(2) Receiving information on how accurate their prior beliefs about citizen preferences are.

The design includes two further cross-randomizations at the politician level, aimed at testing the sensitivity of treatment effects to potential design effects.

Individual local elected representatives are cross-randomized into the following treatment arms:

(1) Citizen Preferences Information Treatment:

The politicians randomized into this treatment group will be provided with accurate information about citizen preferences on a set of nine salient policy issues. Data on preferences has been gathered through a random survey of voters in the same National Assembly constituency as the local politician, which includes the local politician's jurisdiction. The purpose of this treatment is to mimic a `preference gathering' exercise by the politician. They will be further randomized into one of six treatment groups, with each group receiving the preferences of a different subgroup of the sample.

- Full sample, including both male and female adult citizens of all political affiliations
- Adult male citizens of all political affiliations
- Adult female citizens of all political affiliations
- Both male and female adult citizens who support the local politician's party
- Male citizens who support the local politician's party
- Female citizens who support the local politician's party

Within each politician, 6 of the 9 policy issues will be assigned to treatment and the remaining three will serve as within-politician controls. On each of the six treatment issues, they will receive information on citizen preferences from the same subset of the population they have been assigned to.

The nine policy issues are:

A1: Allocation of additional local funds across waste or sanitation
A2: Allocation of additional local funds across filter plants or street lights
A3: Allocation of additional local funds across local streets or water supply
B1: Focus of provincial health department across smaller general health centers or specialized healthcare centers
B2: Focus of national government on eradicating corruption or addressing unemployment
B3: Allocation of federal government resources towards fixing electricity issues or water issues
C1: Initiating development projects in the presence of environmental costs
C2: Support for Women on Wheels scheme
C3: Level of taxes and services in general in the economy

(2) Accuracy Treatment: In this treatment, local representatives will be told how accurate their prior beliefs about citizen preferences
are. Once their priors about citizen preferences have been elicited, the survey software will automatically calculate an `accuracy score' based on the politician's responses and the preferences expressed by voters. This score will range from 0 (equal to a random guess) to 100 (entirely accurate). In pilot data, the mean accuracy score was 34 with a standard deviation of 22. The purpose of this treatment is to demonstrate that the existing information that politicians have about citizen preferences is not as accurate as they tend to believe.

The design includes two further cross-randomizations at the politician level, aimed at testing the sensitivity
of treatment effects to potential design effects. These are:

(1) Whose Preferences are Important?
There is a potential concern that in addition to providing information on citizen preferences, the treatment may also prime the importance of these preferences and de-emphasize the importance of politicians' own views. To test whether this is the case, politicians in each of the experimental groups described above will be cross-randomized into one of three groups. In each group, politicians will be read a different script before these policy recommendations are elicited:

- Script emphasizing the importance of citizens' preferences: It is important for elected representatives
such as yourselves to take the preferences of citizens into account when making decisions. Please keep
this in mind when making your recommendations.

- Script emphasizing the importance of politicians' own views regarding good policies: You have been
elected by the people and your views are important. Please keep this in mind when making your
recommendation.
- Script emphasizing the need to balance citizen preferences with politicians' own views regarding good
policies: Citizens' preferences are important but so are your own views. Please consider both of these
when making your recommendations

(2) Who is Requesting Recommendations?

Another potential concern is that whether preferences are being requested by the party's senior leadership or
not may be consequential for whether local politicians respond to citizen preferences. To test whether this
is the case, the party will issue local politicians one of the following two variations of the letter at random:

- Letter stating the party is requesting their preferences and will incorporate these in the policy making
process

- Letter stating the party president is requesting their preferences and will incorporate these in the policy
making process

Randomization done in office on a computer

Individual local representatives randomized into experimental groups as described in the experimental design section.
Policy issues within each treatment local representative randomized into treatment or within-treatment-cluster controls.

Yes

720 local representatives

6480 policy recommendations

Preferences information treatment:
2/3 local representatives to receive preferences information. Within this group, 1/6 each to receive preferences of a different subset.

Accuracy treatment:
1/2 local representatives to receive accuracy information.