"If you're not a liberal when you're 25, you have no heart. If you're not a conservative by the time you're 35, you have no brain."

* Francois Guisot, French premier and historian (1787-1874) (1)

"I want candy!"

* Every kid, ever.

I. INTRODUCTION

We exploit the second premise to test the first premise.

We conducted trick-or-treat experiments to learn about political attitudes before the 2008 and 2012 presidential elections. (2) We set up two tables on the porch of a home for Halloween, one festooned with McCain campaign props in 2008 (Romney in 2012) and the other with Obama props. Children, at the stairs leading up to the porch, were told they could choose which side to go. Half of the children were randomly assigned to be offered twice as much candy for the McCain table (Romney in 2012), and half were offered an equal amount. (3,4) We can compare the change in the overall support for Obama in 2008 versus 2012 by comparing the proportion that choose Obama (i.e., the external margin), as well as the change in the intensity of the support for President Obama by comparing the elasticity with respect to candy (i.e., the intensive margin). (5,6)

The experimental setup allows us to measure not just what proportion of children who trick-or-treat in this neighborhood support each candidate (as indicated by their choice of table), but also how elastic their support is, or, more precisely, how elastic their desire is to make a public statement of their support. In 2008, 79% of children chose the Obama table with no incentives, whereas 71 % of children chose the Obama table when offered double the candy at the McCain table. In 2012, 82% of children chose the Obama table with no incentive, whereas 78% of children chose the Obama table when promised double candy at the Romney table.

Children's responses to the candy incentives varied by age, however. Younger children's preference for Obama was sticky with respect to price, but older children's preference for Obama was elastic. This result was first observed in 2008 and then successfully replicated in the 2012 experiment. We discuss several interpretations in the conclusion, including a differential response to symbolic versus monetary rewards, a parental-contamination story for the younger children, and the simplest possibility: that younger children just did not understand the task as much and so made the political choice they did understand.

II. EXPERIMENTAL DESIGN-2008 ELECTION (OBAMA AND MCCAIN)

We conducted the experiment on the front porch of a home in the East Rock neighborhood of New Haven, Connecticut. The neighborhood has closely located homes, and a high proportion of residents pass out candy. It is popular with trick-or-treaters, both from the immediate neighborhood and from neighboring communities. A typical home in this neighborhood receives more than 500 trick-or-treaters each Halloween. We used this large flow of individuals to test several aspects of political choice. (7)

Two tables were set up on the front porch of the home. The right side of the porch was decorated with John McCain election material and the left side of the porch was decorated with Barack Obama election material. In addition, life-size cutouts of each candidate, clearly visible, stood on their respective sides.

Trick-or-treaters were met in front of the porch and told to draw a number from a bowl that assigned them randomly to one of two groups (8):

1. Equal Candy: Children were asked, "You can get two pieces of candy from the Obama table, or two pieces of candy from the McCain table. Which do you prefer?"

2. Double Candy for Voting Republican: Children were asked, "You can get two pieces of candy from the Obama table, or four pieces of candy from the McCain table. Which do you prefer?"

A total of 322 individuals participated in the experiment. If a child was too young to understand the initial choice between the Obama and McCain tables, they did not participate in the experiment and are not included in our analysis (they were, however, still given candy). (9) We limited the trick-or-treaters included in our analysis to those between and including age 4 and 15. (10)

The experimental design allows us to calculate the "candy elasticity" of McCain. The sample for this analysis is all children who were assigned to Treatment 1 or Treatment 2 (322 children). In order to avoid potential gender confound, one female and one male experimenter sat behind each candidate's table and alternated speaking. The gender of the experimenter had no effect on any of the treatments.

Although many children arrived in groups, raising issues of social conformity influencing individual decisions and thus creating lack of independence across observations, we strived to interact with the children one-on-one, without interference by others, by separating the queue from the decision area. Specifically, the queue to the porch started about 3 feet in front of the steps to the porch. We placed several chairs there to block the flow of traffic. We then brought each child one at a time to the steps of the porch, explained their options to them, and then let them go up to the top of the stairs and to one of the sides of the porch. We cannot claim that this method was perfect. We posit that any influence of social conformity would lead to incorrect (underestimated) standard errors, but would less likely create biased estimates. However, if social contagion leads those with minority views (i.e., in this environment, supportive of Republicans) to hide their views and publicly show support for Obama, this would upwardly bias the proportion supporting Obama.

The summary statistics for the sample population are given in Table 1 and brief descriptive statistics are given in Table 2.

III. RESULTS--2008 ELECTION (OBAMA AND MCCAIN)

In the "equal candy" treatment, 79% of children chose the Obama table, reflecting the high level of support for the Democratic Party in New Haven, Connecticut. (11) When offered twice the amount of candy to go to the McCain table, 71% of the children still chose the Obama table, though the difference is not statistically significant (Table 1).

Note that the preference for Obama is likely understated, as there was consistently a longer line at the Obama table than at the McCain table. This means that even in the "equal candy" treatment, the "cost" of acquiring candy from the Obama side was slightly higher, as a long line is costly for subjects intent on maximizing their total candy intake for the evening. (12)

Children ages eight and under did not respond to the additional candy incentives--approximately 30% of children chose the McCain table in both treatment groups. Children ages nine and older however, were much more responsive to the candy incentive. The percentage of older children that visited the McCain table increased from 10% without the incentives to 30% with the incentives.

In the conclusion, we posit several explanations for this heterogeneity.

IV. EXPERIMENTAL DESIGN-2012 ELECTION (OBAMA AND ROMNEY)

We conducted a variant of our 2008 experiment l week before the 2012 presidential election, substituting McCain material with Romney material. The purpose of the replication was to determine whether or not the overall support and the enthusiasm of support for Obama had dwindled since 2008. Trick-or-treaters were met in front of the porch and told to draw a number from a bag that assigned them randomly to one of two groups:

1. Equal Candy: Children were asked, "You can get one piece of candy from the Obama table, or one piece of candy from the Romney table. Which do you prefer?"

2. Double Candy for Voting Republican: Children were asked, "You can get one piece of candy from the Obama table, or two pieces of candy from the Romney table. Which do you prefer?"

The summary statistics for the sample population of 157 people are given in Table 3 and brief descriptive statistics are given in Table 4. The sample size is lower than in 2008 because Hurricane Sandy affected the New Haven community a week before Halloween, and lingering effects adversely affected turnout for trick-or-treating. (13)

V. RESULTS--2012 ELECTION (OBAMA AND ROMNEY)

Our results are largely consistent with the results from 2008, suggesting that support for Obama in this context has not declined since 2008. Eighty-two percent of children chose Obama in the "equal candy" treatment, whereas 78% of children chose Obama when twice as much candy was offered at the Romney table.

As in 2008, for children ages nine and older, the double candy incentive appeared to encourage some Obama supporters to choose Romney. While 17 % of older children chose Romney when offered equal candy, 31 % of older children chose Romney when offered double candy. For children ages eight or under, the double candy incentive had the opposite effect: 18% chose Romney when offered equal candy, whereas 14% chose Romney when offered more candy.

VI. CONCLUSION

While the choice of tables is entirely benign, it appears strikingly inelastic. Despite the strictly symbolic nature of choosing one table over another, most children chose the Obama table even when offered twice as much candy at the McCain/Romney table. Seventy-nine percent of children chose the Obama table with no incentives, whereas 71% of children chose the Obama table even when incentivized to choose the McCain table. In 2012, 82% of children chose the Obama table with no incentives, whereas 78% of children chose the Obama table when incentivized to go to the Romney table. These results suggest that the level of overall support and the enthusiasm of support for Obama remained roughly the same from 2008 to 2012.

We observe important heterogeneity: children ages nine and older were much more likely to be swayed by the double candy incentives than children ages eight and under. We posit several possible interpretations: (1) Younger children may have seen the choosing of a table as deeply symbolic of their choice for President, whereas older children may have seen it more as a simple game. (14) (2) Younger children may have been less capable of processing the trade-off. Given their confusion, they may have focused on the easier political choice. Although we posit this as a viable explanation, in past Halloween experiments we did not find differential responses by age even when the exercise was more complicated. (3) Children eight and under may perceive this as their chance to state publicly their political preferences and take the process more literally, whereas older children may see this public statement as having less meaning. This may be the case even if younger children are less politically engaged or possess less political knowledge. (4) Younger children were more likely to have their parents nearby, and thus may have been reluctant to change their choice for candy. (15) The political science and child development literature typically documents the formation of political preference in the mid-teen years (Beck and Jennings 1982; Murray and Mulvaney 2012; Tedin 1974), suggesting the heterogeneity by age we find may be a by-product of differential obedience to parental preference and not inelastic versus elastic political preferences.

Several sample selection and experimental interpretation issues are important to note. First, East Rock is an enclave of Yale University and borders lower socioeconomic neighborhoods. In both election years, the proportion of children in the "equal candy" treatments that chose the Obama table was greater than the proportion of New Haven County voters that voted for Barack Obama. Thus, even those who support McCain do so knowing they are in the minority in their community, and this could lead to differences in behavior relative to how McCain supporters in more conservative neighborhoods would behave. Second, for some younger children, it was not possible to get them to decide without the parent observing them, or making a suggestion to them. However, on the rare occasions when a pushy parent clearly chose on behalf of the child, we noted this and dropped the data point. Finally, as with any empirical exercise, questions of external validity remain. In particular, it is unclear whether the results of this experiment were particular to the 2008 and 2012 elections, particular to Obama, or particular to some other circumstance surrounding these two experiments.

doi: 10.1111/ecin.12233

REFERENCES

Anagol, S., S. Bennett, G. Bryan, T. Davenport, N. Hite, D. Karlan, P. Lagunes, and M. McConnell. "There's Something about Ambiguity." Working Paper, Yale University, 2008. Accessed May 13, 2015. http://karlan.yale .edu/p/halloween_final.pdf.

Beck, P. A., and M. K. Jennings. "Pathways to Participation." American Political Science Review, 76(1), 1982, 94-108.

Murnighan, J. K., and M. S. Saxon. "Ultimatum Bargaining by Children and Adults." Journal of Economic Psychology, 19(4), 1998, 415-45.

Murray, G. R., and M. K. Mulvaney. "Parenting Styles, Socialization, and the Transmission of Political Ideology and Partisanship." Politics & Policy, 40(6), 2012, 1106-30.

Read, D., and G. Loewenstein. "Diversification Bias: Explaining the Discrepancy in Variety Seeking between Combined and Separated Choices." Journal of Experimental Psychology, 1(1), 1995, 34-49.

Shapiro, F. R. The Yale Book of Quotations. New Haven, CT: Yale University Press, 2006.

Tedin, K. L. "The Influence of Parents on the Political Attitudes of Adolescents." American Political Science Review, 68(4), 1974, 1579-92.

(1.) The exact quote: "Not to be a republican at twenty is proof of want of heart; to be one at thirty is proof of want of head." (Shapiro 2006). This quote is often falsely attributed to Winston Churchill.

(2.) We also replicated results from a 2007 Halloween experiment, in which children who wore the most common costumes exhibited more aversion to ambiguity; see Anagol et al. (2008) for those results. Note that experiments on trick-or-treaters have a history in the psychology literature--see Read and Loewenstein (1995) for an example regarding choice bracketing, albeit with a much smaller sample size.

(3.) Providing an incentive to go to the Obama table would have been logistically infeasible; because this neighborhood heavily supported Obama, we had to provide an opposing incentive to even length of processing time on each side of the porch.

(4.) We also conducted an additional experiment in each year (see notes 5 and 6) in which trick-or-treaters did not get a choice of sides, and were randomly assigned. Those observations are not included in the analysis here.

(5.) The 2008 experiment contained a decision upon collecting their candy, to accept a clear bag or paper bag with candy. This was part of a second study on trust and comfort with ambiguity. Trick-or-treaters assigned to a side (rather than choosing a side) are omitted from the analysis here. See http://karlan.yale.edU/p/HalloweenPolitics_v4.pdf for results related to the trust component.

(6.) The 2012 experiment also contained a decision upon collecting their candy, to choose candy or fruit. Again, as with the 2008 experiment, trick-or-treaters assigned to a side (rather than choosing a side) are omitted from the analysis here. See http://karlan.yale.edU/p/Michelle-2012.pdf for results related to the candy or fruit experiment.

(7.) Because no individual identifying information was recorded about each child, and because no harm (beyond that of the socially acceptable excessive sugar intake) was imposed on the children, this was approved by Yale University's Institutional Review Board without needing explicit signed consent by the parents of each child.

(8.) There were actually two more groups for a total of four: half of the trick-or-treaters were assigned to one table or the other for the trust experiment described in note 4. As we are not using these other two treatment groups for any of the analysis, we are describing the study throughout as just having two experimental conditions.

(9.) We judged 21 children to be too young to understand the question related to the additional candy bag treatment (see note 2), so although they arrived at either the McCain or Obama table, they are omitted from all experimental results (and are not considered part of our sample). Age was not recorded for 25 children, who were omitted from our age group analyses.

(10.) We excluded from the analysis six trick-or-treaters 3 years old or less and 17 trick-or-treaters 15 years old or older.

(11.) In the actual 2008 presidential election, 61 % of New Haven County voters voted for Barack Obama and 38% voted for John McCain (source: http://elections.nytimes.com/2008/ results/president/map.html).

(12.) Of course the subjects may have desired to spend time with their friends, so a popular table is not inherently costly even if the wait is longer.

(13.) The immediate neighborhood was not affected seriously by Hurricane Sandy, unless one considers intermittent buffering of streaming Netflix videos during the storm as a serious adverse consequence. Regardless, the mayor of New Haven announced the "suggestion" to delay trick-or-treating by a week, thus creating confusion for some, lower turnout on Halloween, and lower statistical power for this experiment.

(14.) Such a result for young children would not be unprecedented: Mumighan and Saxon (1998) find that children age eight to nine display a strong sense of fairness in the ultimatum game, even if it entails forgone rewards.

(15.) Note that had any parents explicitly influenced the child, the data were not collected. The child merely received candy and we moved on to the next participant. Naturally the mere presence of a parent may influence the child's choices.

JULIAN JAMISON and DEAN KARLAN *

* The views expressed are those of the authors and do not necessarily reflect those of the Consumer Financial Protection Bureau or the United States. This article is based on previous drafts circulated under the titles "Candy We Can Believe In: A Halloween Experiment on Trust and Political Symbolism" and "Candy We Still Believe In: Using Halloween to Measure Change in Extensive and Intensive Support for Barack Obama." We thank Yale University for the research support. Human Subjects Approval provided by Yale University IRB # 1210010995 and 0710003195. The authors gratefully acknowledge the editor Yoram Bauman and three anonymous referees, Treb Allen, Gharad Bryan, Snaebjorn Gunnsteinsson, Melanie Morten, Scott Nelson, Bram Thuysbaert, Jane Bang, Zachary Groff, Sakshi Kumar, Kelsey Larson, Adele Rossouw, and Martin Sweeney for written contributions and research assistance; Muneeza Alam, Iva Rashkova, and Katie Wilson for assistance running the 2008 field experiment: and Emek Basker, Dan Keniston, and Ryan Knight for insightful suggestions.

Jamison: Senior Economist, Office of Research, Consumer Financial Protection Bureau, Washington, DC 20552. Phone 202-281-9759, Fax 855-237-2392, E-mail [email protected]

Karlan: Professor, Department of Economics, Yale University, New Haven, CT 06520. Phone 203-432-4479, Fax 203-432-5591, E-mail [email protected]

TABLE 1
Summary Statistics (2008-McCain vs.
Obama)

                         Same # of    2x Candies at
                          Candies      McCain Table    Total

Panel A: selection treatments and candidate choices
  McCain                    33             48           81
                            21%            29%          25%
  Obama                    123            118          241
                            79%            71%          75%
  Total                    156            166          322

Panel B: selection treatment and candidate choices, age
eight and under
  McCain                    27             24           51
                            30%            29%          29%
  Obama                     64             58          122
                            70%            71%          71%
  Total                     91             82          173

Panel C: selection treatment and candidate choices, age
nine and above
  McCain                     6             24           30
                             9%            29%          20%
  Obama                     59             60          119
                            91%            71%          80%
  Total                     65             84          149

Note: p values for the f-test of equality of proportion
choosing Obama in Column 1 versus Column 2 is [.044],
[.797], and [.001] for Panels A, B, and C, respectively.
Analysis similar to Panel B and Panel C but for males and
females yielded no significant heterogeneity.

TABLE 2
Sample Frame Description (2008)

                                          Cumulative
                    Frequency   Percent    Percent

Age
  4-8 years           173        52.11       52.11
  9-10 years           75        22.59       74.70
  11-14 years          74        22.29       96.99
  Missing              10         3.01      100.00
  Total               332       100.00
Time of evening
  5.30-6.30 p.m.       38        11.80       11.80
  6.30-7.30 p.m.      157        48.76       60.56
  7.30-8.30 p.m.      127        39.44      100.00
  Total               322       100.00
Gender
  Male                158        49.07       49.07
  Female              164        50.93      100.00
  Total               322       100.00

TABLE 3
Summary Statistics (2012-Romney vs.
Obama)

                      Same # of    2x Candies at
                       Candies      Romney Table    Total

Panel A: selection treatments and candidate choices
  Romney                14              17           31
                        17.7%           22%          20%
  Obama                 65              61          126
                        82.3%           78%          80%
  Total                   79            78          157

Panel B: selection treatment and candidate choices, age
eight and under
  Romney                  9              6           15
                         18%            14%          16%
  Obama                  40             36           76
                         82%            86%          84%
  Total                  49             42           91

Panel C: selection treatment and candidate choices, age
nine and above
  Romney                  5             11           16
                         17%            31%          24%
  Obama                  25             25           50
                         83%            69%          76%
  Total                  30             36           66

Note: p values for the f-test of equality of proportion
choosing Obama in Column 1 versus Column 2 is [.500],
[.603], and [.165] for Panels A, B, and C, respectively.
Analysis similar to Panel B and Panel C but for males and
females yielded no significant heterogeneity.

TABLE 4
Sample Frame Description (2012)

                                       Cumulative
                 Frequency   Percent    Percent

Age
  4-8 years         91        57.96       57.96
  9-10 years        33        21.02       78.98
  11-13 years       33        21.02      100.00
  Total            157       100.00
Gender
  Male              76        48.41       48.41
  Female            81        51.59      100.00
  Total            157       100.00