More Women in Tech? Evidence from a field experiment addressing social identity

De-biasing job application messaging can remove perceived barriers to success that prevent women from entering the technology workforce.

Introduction

Despite evidence that suggests women have made gains in most sectors of the global economy, significant gender wage gaps remain. A large portion of the wage gap can be explained by the different occupational and industry choices made by men and women. For example, in Peru women only make up 7% of the coding industry. The authors argue that, in addition to considering marginal returns for their skills, women select into a career based on their beliefs about their expected success in that field. These beliefs are shaped by existing gender and social norms, such as the stereotype that men are more successful in STEM careers.

There are also perceived social identity costs for women who select into STEM careers. Society often dictates norms of behavior for certain social groups, and non-adherence to the social norm can cause psychological costs for individuals who do not fit within the norm. For example, in a traditional society where women are considered mainly responsible for household and family care, women who choose to work in STEM may incur a social identity costs because they do not fit the norm.  These beliefs then influence self-selection patterns of women in that society and prevent them from applying for higher paying technology jobs in male-dominated fields.

This paper investigates the role of social identity and social norms in choosing an occupation. The authors’ explored the relationship of identity and the choice to work in STEM through a randomized field experiment of low-income women applying to a 5-month women’s only software coding and leadership program. They primarily investigated whether a message that counteracts the social and gender stereotypes associated with software development explicitly displayed in the program application influenced the applicant pool and the rate at which women applied to the training program. The second study sought to explain what aspect of social identity could explain the occupational segregation that happens between men and women in the technological industry.

Findings

The de-biasing (counter-stereotypical) message in this study’s manipulation gave examples of female role models, provided information on the expected monetary returns from the bootcamp, and showed that women can succeed in STEM. The study found that de-biasing messaging can increase the number of women applicants and the number of high skill woman applicants for technology related opportunities. The study in Peru aimed to test the impact of de-biasing messaging on application rates and analyzed the changes in the applicant pool under different messages. The study in Mexico aimed to test what part of the de-biasing messaging was most effective.

From the study conducted in Lima, Peru:

  • When women received the de-biasing message, application rates to the programming bootcamp doubled from 7% to 15%.

Among those who applied:

  • Women who received the debiasing message reported a greater average bias against women in tech (.29 SD versus .19 SD) compared to women applicants who did not receive the debiasing message. Gender bias was measured through an implicit association test (IAT).
  • Women who received the debiasing message had a more traditional view of the role of women in society (.38 SD versus .25 SD) compared to women who did not receive the debiasing message. Identification with traditional gender roles was measured through a survey on how much each woman identified with traditional female gender roles.
  • Even though on average cognitive skills in the treatment group were lower, there were also significantly more women with very high cognitive skills applying to the training. These are the women that the training provider wanted to attract.
  • There was no significant difference in levels of interest in tech between women who received the debiasing message and women who were in the control group (51.6%), indicating that lack of interest in technology cannot explain the disparity in application rates.

From the study conducted in Mexico City:

  • Role models and stating that women can be successful in the STEM sector are key components of a successful de-biasing message. But so are emphasizing the presence of other women in the sector as well as the higher expected returns to women from a career in technology.

Social identity can help to explain persistent gender differences in the technology field. De-biasing messaging in recruitment can increase the number of women applicants as well as the number of high skilled women applicants.

Methodology

In the first experiment, women from low-income backgrounds in Lima, Peru were randomly exposed to different information on the application website for a software training and leadership program.  These recruitment messages either contained generic information about the program or the same information with an additional de-biasing paragraph. The de-biasing messages consisted of information on women’s aptitude for tech, access to a network of women, and an example of a successful female role model. 5,387 people followed the link to the website (originally approached through social media and local media) and 605 applied. Visitors to the registration website were randomly selected to see either the control or target message. Researchers interviewed applicants and conducted tests on their cognitive skills, self-identification with traditional gender roles, and implicit gender bias around occupation choice. This experiment did not study the effect of de-bias messaging on men applicants.

In the second experiment in Mexico City, researchers aimed to narrow down the most impactful element of the de-biasing message. Application website visitors were shown one of four different messages at the application stage. In this experiment, the control contained all three parts of the debiasing message: information on success and returns in the field, access to a network and an example of a role model. The three treatment messages contained two of the three elements to test the impact of removing an element.

Related GAP Studies