voter will have a payoff chart which has a maximum payoff. Payoffs decrease sym- metrically as candidate positions move away from the maximum in either direction as in the sample chart. However, different voters’ maximums may be at different points on the number line, and their payoffs may decrease at different rates.
One important rule in the experiment is that the information on your payoff chart is private information. None of the other voters should know the information on your payoff chart. Please do not talk with other participants during the experiment. Are there any questions about the payoff chart?
In the experiment there will be two groups of voters (drawn from a uniform distri- bution) voting in each poll and election, uninformed and informed. All voters in this room are uninformed. This means that throughout each period the positions of the two candidates will not be made public. You will be given limited information about positions of the candidates and the other voters. Candidate A is always furthest to the left (closest to 0) and candidate B is always furthest to the right (closest to 100).
You will also know, from your payoff chart, the percentage of all voters who have maximums to the left and to the right of your most preferred position.
The informed voters will be generated by the experimenters. The informed voters will know candidate positions and thus will always vote for their most preferred candidate. Every poll and election will include the true preferences of these voters.
These voters are also included in the information you have about the percentage of voters who have maximums to the left and to the right of your most preferred position.
To review, the sequence of events will be as follows: At the start of each period the first poll will be taken and the results announced. Remember that both uninformed and informed voters will participate in all polls. The second and third polls will be taken and results announced. Then the final election will take place. After the final election the candidate positions from that period will be announced. After the last period the experiment will end. At this point, voters will be paid the sum of their payoffs for the position of the winning candidate in each election. Your monetary payoffs are increasing in the number of francs you earn.
Please take a moment to fill out the quiz below using your true payoff chart and
not the sample payoff.
Quiz:
1. My point of maximum payoff on my payoff chart is: (blank) 2. At this maximum point I will get a payoff of (blank) francs.
3. There are (blank) percent of voters to the left and (blank) percent of voters to the right of my maximum payoff point.
4. If Candidate A is at position 60, then Candidate B must be located between num- bers (blank) and (blank).
5. True or False: By announcing the candidate that is closest to my maximum payoff in poll 1, I will get additional francs regardless of whether or not that candidate wins:
(blank)
Are there any questions?
Figure 6.1: Sample Payoff
SAMPLE Payoff Chart
0 100 200 300 400 500 600
1 5 9 13 17 21 25 29 33 37 41 45 49 53 57 61 65 69 73 77 81 85 89 93 97 Candidate Positions
Francs
SAMPLE Maximum Payoff is at Candidate Position = 45
30% of ALL Voters are to your left
<---
70% of ALL Voters are to your right
--->
Chapter 7
Summary of Survey Data and Political Networks
This section examines three distinct data-sets and uses a variety of statistical tech- niques to examine the potential consequences of political networks on an individual’s party identification and presidential vote choice. Conditional upon a series of assump- tions about the data-generating processes, political networks play highly influential roles in determining an individual’s political behavior.
The first chapter in this section uses the 2000 American National Election Survey to determine whether reported political discussants’ presidential choices are likely to influence the respondent. This chapter controls for a large number of covariates which would alleviate biases which might be introduced due to the respondent’s selection of particular discussants. The chapter finds statistically significant effects on presidential vote choice and develops a statistical technique to make inferences about the role of discussant presidential choices using the minimal amount of parametric assumptions.
The second chapter in this section uses an internet survey conducted in October and November of 2006. Again the respondent provides information about political discussants which is used to examine the possibility that respondent’s choice of party identification is influenced by the discussants choices. Controlling for the respondent’s characteristics and the nature of the relationship with the discussant, there remains an effect of discussant party identification on respondent’s choice of party identification.
The third chapter in this section uses the 2006 American National Election Pilot
Study to determine whether the discussant’s partisan identification plays a role in determining the respondent’s choice of party identification. Here each respondent is asked twice to identify their party — once in November 2004 and again in Novem- ber 2006. In the second instance each respondent is also asked to provide a series of discussants and to describe the nature of those relationships. Controlling for the respondent’s characteristics and the discussant relationships, there is an increased probability that each respondent will change the party identification towards agree- ment with the discussants’.
One unique opportunity which emerges from the second and third chapters in this section is the opportunity to examine the responses to identical questions asked on both an internet and an in-person survey. Each survey asked respondents to generate discussants, and both asked respondents about their discussant’s political preferences. Both surveys were in the field during identical timeframes. However, the second chapter uses an internet survey (which attempts to gather a national probability sample) while the third chapter is based upon a survey that uses an in-person door-to-door method. We observe a fairly high number of responses who indicate that they have discussion partners who are geographically dispersed, we observe more disagreement in the in-person survey than the internet survey, and in both instances we observe correlation between the respondent’s preferences and those of the discussion partners.
One assumption made across these analyses is that voters are able to accurately determine the partisanship of their social connections. Studies of the ability of voters to determine this have found that there is a bias for the respondent to over-report agreement with the discussant. In Huckfeldt and Sprague’s 1996 Indianapolis-St.
Louis Study they find that the respondents are 80% accurate when asked about the partisanship of their political discussants and that the respondents are more likely to be accurate if they agree with the discussant’s true political preferences (Huckfelt, Johnson, and Sprague 2004).
This section provides a rare and limited opportunity to gain some insight into the direction of causality by incorporating additional control variables and new statistical
technologies but, standing alone, does not resolve whether or not the discussants in- fluence the respondent or vice versa. Furthermore, it leaves unaddressed the questions of motivation — do respondents use discussant information to update their priors and thus make better choices, or are they simply motivated by peer pressure to appear like those around them?
This chapter does validate a set of assumptions about voter communication found in the theoretical framework as well as demonstrates findings which are consistent with some of the theoretical chapter’s results. In particular, each of these chapters demonstrates that voters are likely to communicate about politics and that there will be correlation between social connection and candidate/party choice, controlling for preferences.