A field experiment on the malleability of willingness to compete

2017-01-11

2017-06-30

The primary outcomes of interest are the following: the gender difference in willingness to compete, and boys' willingness to compete and girls' willingness to compete.
The secondary variables of interest are the factors via which the effect on willingness to compete could operate. These are the performance measures in the piece rate and tournament round (see explanation below), risk attitudes, confidence and questions on children's aspirations.

To measure willingness to compete, we follow the design of Niederle and Vesterlund (2007) closely (see Niederle, Muriel and Lise Vesterlund, 2007, “Do women shy away from competition? Do men compete too much?” Quarterly Journal of Economics 122(3): 1067-1101). Children perform a simple math task (adding up sets of 3 two-digit numbers) in three rounds. In Round 1 they do so under a piece-rate payment scheme. In Round 2 they do so under a tournament payment scheme, in which children's performance is compared to the performance of three other randomly selected children. The winner receives four times as much per correct answer as under the piece-rate scheme, while the losers do not receive anything. In Round 3, children can choose which of the two payment schemes they prefer. This choice is our measure of willingness to compete.
Explanation to the secondary variables of interest: the performance measures are the number of correctly solved exercises in Round 1, and the number of correctly solved exercises in Round 2. Risk attitudes are measured by an incentivized lottery choice question in which children make a single choice between a sure payoff and three lotteries of increasing expected payoff and variance (following Catherine C. Eckel and Philip J. Grossman, 2002, ‘‘Sex Differences and Statistical Stereotyping in Attitudes toward Financial Risk,’’ Evolution and Human Behavior, 23(4): 281–295), and we also ask a self-assessment question about risk attitudes in general (following Thomas Dohmen, Armin Falk, David Huffman, Uwe Sunde, Jurgen Schupp, and Gert G. Wagner, 2011, ‘‘Individual Risk Attitudes: Measurement, Determinants, and Behavioral Consequences,’’ Journal of the European Economic Association, 9:522–550). Confidence is measured as guessed rank in Round 2 (elicited in an incentivized way), and we also ask a self-assessment question from the children on how confident they are in general. The questions on children's aspirations are subjective and are related to the treatment instruments (we ask whether children want to be like those who were described in the treatment instruments).

The experiment will be conducted in primary schools. There will be two different treatment conditions, and within each classroom children will be randomized to the different conditions.

The aim of the research is to examine whether gender roles in children's stories influence willingness to compete and the gender difference therein. Therefore children will be first randomly allocated into one of the two treatment conditions. About half of the children will read a story in which the girl is trying hard to surpass other children's performance and get a prize, and she eventually succeeds. The story also features a boy who is interested in the prize but he chooses not to compete for it eventually. In the other treatment children read the exact same story except that the names of the characters (and of course the corresponding pronouns) are reversed, and hence the gender roles are reversed.
After children read the stories, they have to answer some incentivized questions about it. Children know this in advance, so they have an incentive to read the stories carefully.
After they finished this reading task, they go on to the math tasks. In particular, they have to solve simple math exercises in three rounds, and one of the rounds will be randomly selected for payment. As mentioned by the explanation to the outcome variables, we follow the seminal paper of Niederle and Vesterlund (2007) closely in this part. In Round 1 there is a piece-rate payment scheme and in Round 2 there is a tournament payment scheme. After they are done with these rounds, children are asked to choose which type of payment scheme they would like to have in Round 3. This choice is our measure of willingness to compete. Then we let children do Round 3 as well.
After children finished Round 3, they receive a questionnaire. The questionnaire asks them about their confidence, risk attitudes, whether they would like to be like the boy in the story and like the girl in the story, basic demographic variables, family background, hobbies and whether they have any additional remarks.
At the end of the experiment, dice throws determine which of the math round will be payed and which options will be paid out by the risk question. Afterwards the experimenters leave the room and calculate the earnings elsewhere. Later they return and let children get their rewards based on their earnings. Since participants are primary school pupils of around age 10, children do not receive actual money but get toys instead. That is, they earn tokens with their answers and then they can buy various toys from the experimenters from their tokens, as if it was a marketplace.

Randomization will be done by a deck of playing cards. That is, the treatment of each child will depend on the suit of the card that the child gets randomly.

Individuals are randomized to the different conditions within each classroom. Thus, in each classroom about half of the pupils will get Treatment A, and the others will get Treatment B.

No

Children in each classroom are randomized into two different treatment conditions, so in principle we are thinking about the planned number of observations, not the planned number of classrooms. Thus, the number of clusters that we need depends on the average number of pupils per classroom. Recruitment is in progress so we do not know yet how many pupils there will be per classroom on average. Once we have a good estimation of that number, the planned number of classes will be the planned number of observations divided by the average number of pupils per class. See also answer to planned number of observations below.

Since our intervention is rather minimal, we do not expect a large effect. A small effect in the 0.2-0.25 standard deviation range sounds more reasonable. Hence we would like to have a large sample size to have enough power to detect small effects. This is also why we have only two treatment conditions. Although there are many more potential treatments that could be interesting to look at, we decided to have only two different treatment conditions because we did not want to reduce the sample size per treatment condition. The recruitment of schools has started and is still in progress. Since the final number of pupils will in the end depend on the number of schools that are willing to participate, it is difficult to pin it down to an exact number. There will be two data collection rounds because of the schedule of our data collectors. Our aim is to have around 600-700 pupils from the two rounds altogether, because we need enough power to detect a small effect size. For example, with a sample size of 665 pupils we would have 80 percent power to detect an effect of 0.2 standard deviations if we can get an R-squared of about 0.15 with our background variables. (Our background variables are age, school year, the classroom identifiers, parental background and sibling composition.)

There will be approximately the same number of pupils in each treatment arm.