In this week’s WAPPP seminar, Professor Katherine Coffman discussed her research on a lab experiment designed to isolate the effectiveness of two key features of sponsorship. (Professor Coffman is being heavily recruited for a position at HKS, so we’re hoping that she enjoyed the talk as much as we did!) While working in a lab setting allows researchers to resolve the selection problem, it does narrow the focus to certain measurable aspects of sponsorship. In this case, Professor Coffman focused on what she calls the Belief Channel – that being sponsored is a vote of confidence in the protégé’s abilities that may increase their competitiveness – and the Payment-Tying Channel, which illustrates the link between the protégé’s success and the sponsor’s. These two features, competitiveness and earnings, are important for real-world outcomes, are easy to incentivize, and are easy to measure in a lab setting. It could be that the observed positive effects of sponsorship are due to something other than these two factors. However, if we see an effect of the Belief Channel and the Payment-Tying Channel, we may want to formalize these in sponsorship programs.
Professor Coffman’s experimental design parallels a famous study on competitiveness. Participants go through four rounds of a basic math task. In each round, they are given four minutes to solve as many addition problems as possible. The incentives for solving these problems vary across each of the rounds. In the first round, participants receive 50¢ for each problem they solve correctly. In the second round, researchers compare the performance of each participant. The top 25% earn $2 per problem solved, while the bottom 75% get nothing.
In round 3, participants get a series of choices about their incentives for the round. Over 9 binary-choice questions, participants are asked whether they’d rather get a guaranteed 50¢ per solved problem, or whether they’d rather get a certain dollar amount per problem for being in the top 25%. By varying the dollar amount, researchers are able to see what incentive level is enough to get participants to opt into the riskier competition-style incentive. Critically, participants have not gotten feedback on their performance in round 1 and round 2 and don’t know whether they placed in the top 25% in round 2.
Up to this point, all participants have had the same experience. After round 3, three participants are chosen at random to be sponsors, and all other participants become potential protégés. The key challenge here is balancing the selection effect (maybe only the best are chosen to be sponsored) with wanting sponsorship to be a meaningful vote of confidence. Professor Coffman strategically addressed this problem by creating a series of “matched pairs” for sponsors to choose from. This way, the sponsored and unsponsored groups had relatively equal ability, but the impact of being “chosen” was still meaningful.
All in all, there were four protégé treatments:
- Belief Signal participants were chosen by sponsors, but their sponsors did not earn money based on their performance.
- Payment-Tying participants were randomly assigned to a sponsor, but the sponsors earned 25¢ for every problem their protégé answered correctly if their protégé chose to compete and was in the top 25%.
- Belief Signal and Payment-Tying participants were chosen by sponsors, and their sponsors earned money based on the protégé’s performance.
- Unsponsored participants were either never presented to a sponsor to be chosen or were not chosen by a sponsor.
Participants were then asked to repeat the procedure from round 3—making a series of judgments about the necessary wage rate for them to opt into competition and then completing the task. The researchers examined differences in competitiveness and performance between rounds 3 and 4 to better understand the effects of sponsorship.
The three key metrics in the findings were the cutoff, the lowest wage at which a person was willing to compete, their performance, and their earnings.
In round 3, the average men's cutoff was $1.86, compared to $2.02 for women, which indicates a greater appetite for competition for men. Indeed, men were more willing to compete than women at every given wage rate. Men also outperformed women in round 3, solving about one more problem on average. As such, men outearned women in round 3 (average $11.25 versus $7.94). Conditional on performance, men outearned women by $1.03, which suggests that 1/3 of the gender gap is driven by choices about competition rather than performance.
In round 4, the unsponsored exhibited a smaller gender gap than in round 3. However, for each of the sponsored groups, the gender gap grew wider in round 4 than in round 3. While each group decreased their cutoff by about 20 cents, which shows that they became more willing to compete, cutoffs changed more for men than for women. Indeed, sponsorship increases the gender gap from 13¢ in round 3 to 37¢ in the Belief Signal group, 27¢ in the Belief Signal and Payment-Tying group, and 44¢ in the Payment-Tying group.
What makes this gender gap bigger? Sponsorship is intended to encourage talented but underconfident participants to make more competitive choices. It could be that this version of sponsorship isn’t having the intended effect for this target population. To test this, Professor Coffman examined participants’ beliefs about their abilities in rounds 3 and 4.
In the baseline data, both men and women are somewhat overconfident. More than half of men rate themselves in the top quartile! (In fairness, so do 34% of women). In general, the confidence gap between men and women is driven by women being underconfident, more likely to rate themselves below their true ability level.
Sponsorship treatments with the Belief Signal, the vote of confidence that a sponsor chose you, do increase participants’ confidence. Compared to the unsponsored groups, participants in the Belief Signal group rank themselves 0.2 quartiles higher, and participants in the Belief Signal and Payment-Tying Group rank themselves 0.17 quartiles higher. These changes in beliefs don’t vary with gender—everyone reacts to a vote of confidence in a similar way.
However, the group that was most likely to change their willingness to compete in round 4 was overconfident men. Women showed very little movement in their willingness to compete, whether they were overconfident, properly calibrated, or underconfident. This finding is a bit discouraging—improving confidence doesn’t impact those who need it most, but instead concentrates on those who already have plenty of confidence.
Professor Coffman also examined the effect of the sponsorship treatment on performance. In the real world, sponsorship is intended to have the biggest impact on the best performers. However, in this data, the strongest increase in willingness to compete came from men in the bottom three quartiles. There was very little change in willingness to compete for top-quartile men or for women at any performance rank. Participants in a Payment-Tying condition showed additional increases in their performance: having their sponsor’s compensation tied to their performance provides an additional incentive to do well. For women, the effect on performance was small, but for men it was significant. Not only was this group more willing to compete, but they were also improving their performance. This improvement in performance increased men’s earnings, while women’s earnings were not significantly impacted by any sponsorship treatment.
Sponsorship, therefore, is mainly reaching overconfident, low-ability men, and in this lab setting sponsorship fails to close the gender gap in competitiveness or earnings. Talk about unintended effects! The key takeaway, according to Professor Coffman, is that we’re not studying the parts of sponsorship that really matter for women. Other channels (like access to professional networks or sponsor advocacy in promotional meetings) may have a much greater impact. What features of sponsorship programs haven’t been examined that may be critical for closing gender gaps?