Think at London Business School: fresh ideas and opinions from LBS faculty and other experts direct to your inbox
There is a chronic shortage of graduates going into science or engineering (STEM) around the world. That shortage is even more acute among women. In the US, only one in seven engineers is female, while in the UK that number drops to a meagre 6% of the STEM workforce.
So what’s driving the gender gap in engineering?
Hiring practices are an unlikely culprit. Indeed, research suggests that women are more likely to be recruited than their male counterparts.
One theory explored in recent research by Anja Lambrecht, Professor of Marketing at London Business School, is that part of the problem may lie in how information about STEM opportunities is disseminated among women in the first place.
"While advertising algorithms are not inherently gender-biased, the economic forces that govern them might well be driving involuntarily uneven outcomes"
Together with Catherine Tucker, Distinguished Professor of Management Science at MIT Sloan School of Management, Lambrecht has conducted a study to see how internet and social media algorithms determine whether advertising content gets seen more by men or women.
They have found that, while advertising algorithms are not inherently gender-biased, the economic forces that govern them might well be driving involuntarily uneven outcomes that platforms and policymakers may wish to address.
“We were curious to understand a possible relationship between the lack of women in the STEM hiring pool and the way that women were being told about opportunities in the space,” says Lambrecht. “We wanted to know how online adverts were reaching women, and whether the algorithms that decide who sees what were somehow discriminating against them.”
Lambrecht and Tucker conducted a field study using Facebook ads to send messages about STEMS careers to a cross-section of men and women aged between 18 and 65 in 191 countries. Analysis of the data revealed what they had suspected: despite the campaign targeting both genders, 20% more men than women were seeing the ad.
A number of possible explanations for this were examined, and debunked by the researchers.
Economics, algorithms and online ad delivery
“We discounted the possibility that the algorithm had learned its behaviour from women simply not clicking ads as much as men as this is not the case,” explains Lambrecht. “The data shows that when women saw the ad, they were more likely to click than men. Nor is it the case that there are simply more male users on Facebook.”
The researchers also ruled out the possibility that the algorithm was somehow learning its behaviour from other data, such as underlying patterns of discrimination against women in specific countries. Data from the World Bank revealed no relationship between the educational and labour market opportunities for women and whether STEM ads were displayed to them in their study.
“A more likely explanation was that the fact that more men than women were seeing the ad had to do with the economic mechanisms behind online ad delivery,” says Lambrecht.
Online advertisers compete with each other to get their content in front of the same set of eyeballs by placing “bids.” When a user loads a page, Facebook runs an instantaneous real-time auction in the background to determine which ad gets shown. In simple terms, the advertiser who bids highest is the most likely to win.
Advertisers who want to make sure that a target user sees their ad, should set their bidding price to reflect a user’s approximate value to them. For example, if an online retailer expects a particular segment to be spending much with them after seeing an ad and clicking on it, they are likely to bid a higher price to display their ad to this consumer rather than to a consumer who is unlikely to purchase. In fact, they might not even advertise to the consumer who is unlikely to purchase. As a result, the price for displaying an ad varies across different consumers.
And, as Lambrecht explains, it turns out that displaying ads to women is more expensive than to men.
“We looked at Facebook’s advertising recommendations on how to make bids across different gender and age segments by country. It turns out that when you’re targeting women, you need to bid higher. So women are, in effect, more expensive to the advertiser than men, especially in the age group where we find the strongest effect.”
This is likely due to the fact that women are more likely to make purchasing decisions than men. Research suggests that they drive as much as 90% of all consumer purchasing. This makes women simultaneously more valuable and more expensive to online advertisers.
“So algorithms are doing the maths and making decisions about the cost-benefit ratio in showing content to male and female users. And this is the problem,” says Lambrecht. “They don’t intend to be discriminatory, but the outcome is that in minimising costs, advertisers who set up their campaign to not discriminate by gender are delivering their content to one gender more than another. The same effect is true across different platforms that we have tested, like Google Adwords, Instagram and Twitter.”
"Women drive as much as 90% of all consumer purchasing. This makes them simultaneously more valuable and more expensive to online advertisers."
Getting better about targeting
There is plenty here, say Lambrecht and Tucker, that should worry policymakers and platforms alike. Getting important content in front of both genders matters not only in STEM but across a range of opportunities from housing, finance and more.
Resolving this is challenging, not only because of the trade-off with the economic complexities in resource allocation using algorithms, but also because employment laws in most countries do not yet adequately stipulate how targeted advertising fits within existing discrimination frameworks.
One solution, says Lambrecht, could in theory be for advertisers to run separate campaigns for men and women to make sure they can target both demographic groups.
“However, in the US, Federal Law prevents you from targeting employment ads to only one gender. So a law that is designed to avoid discrimination actually makes it harder for advertisers to fix unintentional uneven outcomes. Perhaps one solution would be for platforms to offer a button that would allow advertisers to evenly distribute impressions across demographic groups – even if this means they have to pay more for reaching some of the individuals.”
And with forecasts showing that in the UK alone at least 100,000 new STEM graduates will be needed every year through to 2020, the need to reach women and other demographic groups to bridge the looming skills gap is becoming acute.