Also available in: Español, Français
A nice new paper from Oriana Bandiera, Greg Fischer, Andrea Prat and Erina Ytsma give us an answer. Before getting to that answer (and since it’s International Women’s Day after all), it’s worth asking: why would one ask that question in the first place?
Bandiera and co. motivate the question by pointing out that in a number of psychological outcomes relevant to responses to incentives, women do seem to respond differently than men. The first is risk aversion. A number of papers have found that women exhibit higher risk aversion than men, particularly in experiments (see for example Charness and Gneezy’s 2012 paper looking at financial risk aversion). And folks who are more risk averse may not respond well to performance based incentives that could increase the variability of their income.
The second dimension is overconfidence. A number of experiments show men to be more likely to be confident (see for example, the review in Croson and Gneezy). With lower confidence, it is possible that individuals would not try the different ways to improve performance that is likely to make performance pay pay off.
Third is altruism. Evidence from games tends to show that women behave more altruistically than men (see reviews in Cronson and Gneezy or Bertrand’s handbook chapter). Hence, Bandiera and co. posit, women are already likely to be internalizing their employers welfare to a greater extent and additional performance incentives will have a weak effect.
Finally, there is competition. There is some evidence that women are less likely to exhibit competitive behavior (see the reviews or the very nice cross-cultural differences in Gneezy, List and Leonard), but this isn’t always true (e.g. some work I was part of on adolescents in Uganda finds no differences).
All told, these might portend that women would respond less strongly to performance based compensation than men. So Bandiera and co. take to the world, and gather up a range of studies to see if this might be the case. They canvass published studies and a select set of working papers. They are looking for work that compares at least two levels of (cash) incentives, one of which is higher powered than the other. And they restrict the search to studies (of both the field experiment and lab experiment varieties) that require real effort and produce real output (no hypotheticals here). Lots are excluded – compensation that comes in the form of recognition, studies that use different incentive structures (e.g. piece rates versus tournaments), and team incentives.
They end up with 37 studies. Now, they’re after the gender differences. And some of these studies didn’t look at this. So Bandiera and co. need the data (it’ll also be important for the method they apply – more on this below). They end up getting the data for 16 papers which contain a total of 18 experiments for them to look at. And it’s a pretty diverse set of studies – ranging from US college students stuffing envelopes to health workers selling condoms in Zambia to Kenyan school teachers being paid based on test scores.
OK, now it’s time to add this all up. Most meta-analyses use random effects models. But Bandiera and co. bring a Bayesian hierarchal model (BHM) to bear. The paper has a nice explanation of the BHM and how it works relative to other approaches, but I am going to skip that here in the interests of some semblance of brevity.
So…do women respond differently? Let’s turn to the authors: “despite significant variation in context, including task, location, and the structure of pay for performance, the differences between men and women in the response to incentives appear to be relatively consistent and consistently close to zero across contexts. This implies that these studies have external validity, that is knowing that the gender differential is zero implies that the next, hypothetical study is also very likely to find a zero effect.” So no significant difference between men and women. And both sexes show a significant positive response (of about 0.28 standard deviations) to incentives. So, incentives are effective, and equally so.
Before wrapping up, there are two interesting things to note about this paper. The first is that the way Bandiera and co. approach things, they are able to ask a question related to, but not directly asked by some of the original studies. They take the existing data, apply the BHM and are able to look at gender differences. Second, as Bandiera and co. point out, this approach will only give us results generalizable to settings where experiments are run. So, here’s to further diversification.
In the end, this paper shows us that we can learn new things about different questions by aggregating evidence and data that we already have. But the key thing is that the data be available. Bandiera and co. tell me that their main challenge was to find the relevant papers and the information they needed within those – so let’s keep making that available.
A nice new paper from Oriana Bandiera, Greg Fischer, Andrea Prat and Erina Ytsma give us an answer. Before getting to that answer (and since it’s International Women’s Day after all), it’s worth asking: why would one ask that question in the first place?
Bandiera and co. motivate the question by pointing out that in a number of psychological outcomes relevant to responses to incentives, women do seem to respond differently than men. The first is risk aversion. A number of papers have found that women exhibit higher risk aversion than men, particularly in experiments (see for example Charness and Gneezy’s 2012 paper looking at financial risk aversion). And folks who are more risk averse may not respond well to performance based incentives that could increase the variability of their income.
The second dimension is overconfidence. A number of experiments show men to be more likely to be confident (see for example, the review in Croson and Gneezy). With lower confidence, it is possible that individuals would not try the different ways to improve performance that is likely to make performance pay pay off.
Third is altruism. Evidence from games tends to show that women behave more altruistically than men (see reviews in Cronson and Gneezy or Bertrand’s handbook chapter). Hence, Bandiera and co. posit, women are already likely to be internalizing their employers welfare to a greater extent and additional performance incentives will have a weak effect.
Finally, there is competition. There is some evidence that women are less likely to exhibit competitive behavior (see the reviews or the very nice cross-cultural differences in Gneezy, List and Leonard), but this isn’t always true (e.g. some work I was part of on adolescents in Uganda finds no differences).
All told, these might portend that women would respond less strongly to performance based compensation than men. So Bandiera and co. take to the world, and gather up a range of studies to see if this might be the case. They canvass published studies and a select set of working papers. They are looking for work that compares at least two levels of (cash) incentives, one of which is higher powered than the other. And they restrict the search to studies (of both the field experiment and lab experiment varieties) that require real effort and produce real output (no hypotheticals here). Lots are excluded – compensation that comes in the form of recognition, studies that use different incentive structures (e.g. piece rates versus tournaments), and team incentives.
They end up with 37 studies. Now, they’re after the gender differences. And some of these studies didn’t look at this. So Bandiera and co. need the data (it’ll also be important for the method they apply – more on this below). They end up getting the data for 16 papers which contain a total of 18 experiments for them to look at. And it’s a pretty diverse set of studies – ranging from US college students stuffing envelopes to health workers selling condoms in Zambia to Kenyan school teachers being paid based on test scores.
OK, now it’s time to add this all up. Most meta-analyses use random effects models. But Bandiera and co. bring a Bayesian hierarchal model (BHM) to bear. The paper has a nice explanation of the BHM and how it works relative to other approaches, but I am going to skip that here in the interests of some semblance of brevity.
So…do women respond differently? Let’s turn to the authors: “despite significant variation in context, including task, location, and the structure of pay for performance, the differences between men and women in the response to incentives appear to be relatively consistent and consistently close to zero across contexts. This implies that these studies have external validity, that is knowing that the gender differential is zero implies that the next, hypothetical study is also very likely to find a zero effect.” So no significant difference between men and women. And both sexes show a significant positive response (of about 0.28 standard deviations) to incentives. So, incentives are effective, and equally so.
Before wrapping up, there are two interesting things to note about this paper. The first is that the way Bandiera and co. approach things, they are able to ask a question related to, but not directly asked by some of the original studies. They take the existing data, apply the BHM and are able to look at gender differences. Second, as Bandiera and co. point out, this approach will only give us results generalizable to settings where experiments are run. So, here’s to further diversification.
In the end, this paper shows us that we can learn new things about different questions by aggregating evidence and data that we already have. But the key thing is that the data be available. Bandiera and co. tell me that their main challenge was to find the relevant papers and the information they needed within those – so let’s keep making that available.
Join the Conversation