Policy Research Working Paper 10548 Preferences for Wage Discrimination against Women William Seitz Poverty and Equity Global Practice August 2023 Policy Research Working Paper 10548 Abstract This study demonstrates systematic bias against women in conventional levels in all three countries, among both male public perceptions of the fairness of wages. In nationally and female respondents, and in each of the eight occupa- representative survey experiments across more than 70,000 tions studied. The results also demonstrate the presence individual vignettes posed to 4,500 respondents in three of significant bias in favor or older workers, specifically Central Asian countries, respondents were 13 percent more for white-collar occupations, and the absence of this rela- likely to say wages were “too high” when the randomly tionship for the blue-collar occupations included in the assigned person described in the vignette (subject) was experiment. The findings reinforce the importance of bias a woman, and 34 percent more likely to say they were as a contributing factor to the gender pay gap, and the value “too low” when the subject was a man. The pattern of bias of equal pay regulations to prevent gender discrimination favoring higher wages for men is statistically significant at in wage setting. This paper is a product of the Poverty and Equity Global Practice. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/prwp. The author may be contacted at wseitz@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Preferences for Wage Discrimination against Women William Seitz1 Key Words: Discrimination, gender equality, gender wage gap, equal pay for work of equal value, Central Asia. Acknowledgments: Technical and financial support from the UK Foreign Commonwealth and Development office is gratefully acknowledged. I am grateful for the review and feedback provided by World Bank reviewers: Ana Maria Munoz Boudet and Patricia Fernandes. 1 The World Bank, Poverty and Equity in Central Asia, E-mail: wseitz@worldbank.org. Introduction and country context Although the right to equal remuneration for women and men for work of equal value has been acknowledged in dozens of international accords over the past century, women around the world continue to earn remarkably less than men. The gender pay gap, a summary indicator of this phenomenon, refers to the difference in average wages earned by men and women. According to the International Labour Organization (ILO), the global gap stood at 20 percent in 2022, 2 meaning that women on average earned about 80 cents for each dollar men earned. While differences in the characteristics of workers—such as their level of education, skill types, hours worked, occupational choices, and experience—explain some of this gap, a significant portion remains unexplained by observable worker characteristics other than gender. This paper describes the results of a study using an experimental design to investigate the role of public attitudes in wage setting, and perceptions of wage fairness among nationally representative samples of the populations of three countries in Central Asia. The results identify a systematic negative bias in public perceptions of the value of women’s work. The consequences of the gender pay gap are far-reaching—affecting individuals, families, firms, and society at large. On an individual level, pay gaps reduce the earnings of women over the course of their careers, resulting in reduced monetary welfare throughout a women’s working life, less economic autonomy (due to the fewer resources accumulated), lower retirement savings, and the consequences of greater vulnerability in old age. In contrast, on average more household earnings translate to better living standards for all household members, better health, and improved education outcomes. At the level of firms and employers, unfairness has important effects on productivity. For instance, in their meta-analysis on workplace attitudes, Tirana, et. al. (2019) found perceptions of workplace-level gender pay inequality to be strongly associated with reduced motivation and job satisfaction across a range of studies. Potential negative effects in that study extended to physical health outcomes and behaviors, psychological health, and many work-related outcomes (both job-based and relationship-based). Numerous potential sources of bias in wage setting have been documented, including perceived lower attachment to employment (Barron et al, 1993; Blau and Khan, 2017), stereotypes (Bordalo et al. 2016, Coffman et al. 2021), and differences in negotiating behavior (Azmat and Petrongolo 2014). The prevalence of these forms of bias risks entrenching gender stereotypes and makes social and career advancement more difficult for women. By perpetuating the idea that men's work is more valuable than women's work, imbalances can reinforce broader trends with respect to gender inequality and discrimination, while reducing these gaps tends to generate virtuous cycles, incentivizing higher labor force participation. In aggregate, the gender pay gap is also a first-order economic concern at the national level. When women are empowered and have equal opportunities to men, they can contribute to the economy to their full potential, leading to a more diverse and dynamic workforce. Gender equality in labor market opportunities thus strongly influences productivity growth, innovation, and competitiveness. Higher household incomes can also generate faster economic growth through multiplier effects on aggregate demand. The World Economic Forum estimates that if women participated in the economy identically to men, global GDP in 2025 would increase by as much as 26 percent (WEF 2022). 2 ILO: New data shine light on gender gaps in the labour market. 2 These patterns hold especially true in the context of Central Asia, the setting of the experimental study described in the sections that follow. If women were to participate in equal measure to men, national income would be 27 percent higher in Kazakhstan, 29 percent higher in Uzbekistan, and 39 percent higher in the Kyrgyz Republic. In Uzbekistan, World Bank estimates suggest that equalizing the average wage among women and men who are already working would alone pull more than 700,000 people out of poverty (World Bank, 2023). But although measuring the presence of gender differences in wage rates (and incomes) is routine, clearly identifying systematic bias and discrimination against women in wage setting is more challenging. A common approach to investigating differences in wages and the potential for discrimination is a method introduced by Blinder (1973) and Oaxaca (1973), which measures wage differences between groups using a decomposition technique. The approach identifies the proportion of the wage differential between groups that can be attributed to differences in the levels of characteristics between them (for instance, differing levels of education or experience). Then, assuming identical characteristics in the two groups, the remaining inequality can be attributed to differential effects of the characteristics (such as discrimination) as well as any unobserved factors. Weichselbaumer and Winter-Ebmer (2005) conducted a meta-analysis of 1,535 such estimates and found that although the gender wage gap has been steadily declining for decades—from around 65 in the 1960s to around 20 in recent years—improvements are almost entirely attributable to better labor market endowments of women (such as better education, training, and work attachment). At the same time, the unexplained component of the wage gap—where potential discrimination resides—has not systematically declined since the 1960s. Penner et. al. (2022) showed that although worker characteristics and selection effects play an important role in the pay gap, data that identify women and men who do the same work for the same employer often reveal significant within-job pay differences in the 11 countries they study. There is also strong indirect evidence that the gender wage gap is due to views and attitudes rather than ability. Maloney (2022) shows that misogyny, as proxied by derogatory internet search terms, is an economically meaningful and statistically significant predictor of the wage gap at the local level. Such drivers are often referred to as taste-based discrimination, as opposed to motivations based on differential outcomes such as average group productivity levels, which are often referred to as forms of statistical discrimination (Bertrand and Duflo, 2017). Taken collectively, these and similar studies suggest that discrimination against women remains a consequential factor in wage differences around the world. One commonly cited factor potentially driving lower pay for women than men is often referred to as the “paradox of the contented female worker.” It is supposed that an apparent tendency for women more often to report satisfaction with their work compared to men in the same occupations (and at times the same employer) could contribute to persistent gender pay differences. But as Adriaans and Targa (2023) show, in many countries, overall satisfaction with work does not always translate to satisfaction with compensation. Women are in fact more likely than men to be dissatisfied with their own level of compensation in 15 of the 28 countries studied by the authors. Thus, potential satisfaction differentials appear context specific and are unlikely to explain systematically lower pay for women at the global level. These concerns are especially relevant in the Central Asian countries of Uzbekistan, the Kyrgyz Republic, and Kazakhstan—where gender inequality in labor market outcomes poses significant challenges. According to government statistics, on average, pay among working women is about 39 percent less than men's pay in Uzbekistan, 25 percent less in the Kyrgyz Republic, and 22 percent less in Kazakhstan, higher 3 than the global average in each case. With universal coverage of primary and secondary education, women achieve relatively high levels of education in all three counties (figure 1), making such notable gender pay gaps particularly striking. Figure 1: Summary measures of educational attainment in Central Asia Expected years of schooling Gross tertiary enrollment rate (%) 18 80 15.8 68 16 15.1 70 14 12.2 13.2 12.7 12.6 11.5 60 55 11.9 12 10.7 10.9 47 50 10 37 40 35 8 27 27 29 30 6 20 16 4 13 2 10 0 0 TJK UZB KAZ TKM KGZ TJK UZB KAZ TKM KGZ Women Men Women Men Source: World Development Indicators and National Statistical Offices. Taking Uzbekistan as an example that illustrates challenges across the region, the gender gap in labor force participation is strongly linked to both educational attainment and the number of children living in the household (figure 2). Only 6 percent of women with general secondary education or below, and 12 percent of women with upper secondary education, are active in the labor market compared to 18 percent and 45 percent of men, respectively. However, 52 percent of women with bachelor’s degrees or higher are active in the labor market, which is more comparable to the 60 percent labor participation rate for men with bachelor’s degrees. Furthermore, women (ages 16+) in the bottom two wealth quintiles were more than 50 percent likely to be inactive or unemployed, whereas 39 percent of men in the lowest wealth quintile and 32 percent of men in the second lowest quintile were either inactive or unemployed. Having young children also affect female and male labor force participation differently. Having young children has distinct effects on female and male labor force participation. In the case of men, having more young children is associated with a higher likelihood of working. For women, it is the reverse, women are much less likely to work outside the home if more children are present. 4 Figure 2: Employment's Link to Education and Household Composition in Uzbekistan Employment rate by educational Employment rate by number of children attainment in household 100% 100% 90% 90% 80% 74% 80% 70% 70% 60% 60% 50% 43% 33% 50% 40% 40% 30% 20% 11% 30% 10% 0% 20% 0% 10% primary general high vocational tertiary 0% or less secondary schoo; 0 1 2 3 4 5 6 7+ Male Female Male Female Source: Household Budget Survey of Uzbekistan, 2022 In all three societies, intrahousehold power tends to be imbalanced, and gender norms tend to be more traditional, with women expected to prioritize caregiving responsibilities and men expected to be the primary breadwinners (figure 3). These expectations can limit women's participation in the labor market and their opportunities for career advancement. Additionally, women in all three countries are underrepresented in leadership positions and are more likely to work in lower-paying occupations (Muradova and Seitz (2021) show this is due in part to discriminatory hiring practices). Figure 3: Survey-based perceptions of women's role in work and family life Agree: Women should earn less than men to Agree: Women should spend less time sustain peace in the family working than men to dedicate time at home Urban Urban UZB UZB Rural Rural National 34% National 77% Urban Urban KGZ KGZ Rural Rural National 42% National 73% Urban Urban KAZ KAZ Rural Rural National 20% National 74% 0% 10% 20% 30% 40% 50% 50% 60% 70% 80% 90% Source: Listening to Central Asia surveys, 2022 5 Globally, legal provisions that require equal pay for work of equal value to address the potential for such biased perspectives are common (currently covering workers in at least 95 countries) and becoming more so. The effectiveness of these legal remedies to at least partially reduce the gender pay gap is well established. For instance, Cruz and Rau (2016) find the introduction of equal pay provisions in Chile reduced the firm premium gender gap by 6.1 percent, an effect driven primarily by increased bargaining power for women. All OECD countries have some form of equal pay for work of equal value provision, and in recent years, many advanced economies have strengthened provisions to include mandatory pay transparency to make equal pay provisions more effective. For instance, in 2018 the Government of Iceland introduced a law requiring all employers with more than 25 staff members to publish detailed pay information, and in regulatory filings firms must actively prove they give equal pay for work of equal value. But such equal pay provisions are absent across most of Central Asia. In the Kyrgyz Republic, there is no explicit equal pay for work of equal value provision in the labor code or other regulations focused on employee pay. 3 Likewise, Kazakhstan does not explicitly require equal pay for work of equal value. 4 Uzbekistan alone mandates equal pay for equal work provisions, which were included in Uzbekistan’s labor code signed into law by President Mirzoev in October 2022 and entered into force in April 2023. This is the first explicit provision of this type in a Central Asian country to protect workers from pay discrimination. 5 To shed light on the potential role of beliefs about the value of women’s work compared to that of men in explaining the gender wage gap, a survey experiment was included in three nationally representative surveys in Central Asia in 2022. The experiment took the form of vignettes about workers with randomly assigned characteristics that were read aloud to approximately 4,500 residents of Kazakhstan, Kyrgyz Republic, and Uzbekistan. Each respondent was asked about a set of 8 vignettes per interview, each of the eight focused on a specific occupation. The list of occupations was standardized such that every respondent was told one vignette for the same 8 occupations in the same order. Following each vignette, respondents were asked whether they thought the person described was: “severely underpaid, underpaid, fairly paid, overpaid, or severely overpaid.” Over the course of two months more than 70,000 evaluations of fair earnings were gathered in this form. The results show a clear and systematic bias against the value of women’s work. Even though the prompts were identical in every other respect, respondents were 13 percent more likely to say wages were “severely overpaid” when the person described in the vignette (subject) was a woman, and 34 percent more likely to say they were “severely underpaid” when the subject was a man. The findings suggest that the social context is on average biased against women with the same profile as a man, which likely plays a significant role in perpetuating gender inequality in the labor market and the size of the gender wage gap in Central Asia. The presence of this systematic bias against women reinforces the importance of legal provisions to ensure equal pay for work of equal value, and the importance of challenging and changing the norms that perpetuate gender inequality more broadly. 3 Details from the World Bank’s Women Business and the Law report are summarized here: https://wbl.worldbank.org/content/dam/documents/wbl/2021/snapshots/Kyrgyz-republic.pdf, accessed October 27, 2022. 4 Details from the World Bank’s Women Business and the Law report are summarized here: https://wbl.worldbank.org/content/dam/documents/wbl/2021/snapshots/Kazakhstan.pdf. 5 Details from the World Bank’s Women Business and the Law report are summarized here: https://wbl.worldbank.org/content/dam/documents/wbl/2021/snapshots/Uzbekistan.pdf, accessed October 27, 2022. 6 Data and methods The experiment was embedded in the Listening to Central Asia series of surveys in the Kyrgyz Republic, Uzbekistan, and Kazakhstan, surveys that are designed to continuously monitor wellbeing and views on topical policy issues. 6 The surveys are collected from a single (“omniscient”) household member, and household selection was designed for national representativeness. Fieldwork was conducted in two phases. In the first phase, a national in-person household survey was conducted in each country using regionally stratified two stage cluster sampling procedures. In the second stage, a randomly selected sub-sample from the first stage in each country was recruited in a monthly panel survey using a phone-based method conducted by private survey companies under the direction of World Bank staff. The sub-samples were each nationally representative and for the present study include a total of 6,783 individual responses (in Uzbekistan, the Kyrgyz Republic, and in Kazakhstan). Data collection for the survey experiment was conducted in May-June 2021. For the core experiment, respondents were assigned a set of 8 separate vignettes, each with three randomly generated components of the form: “A -year-old works as . monthly gross earnings total (before taxes and extra charges).” Following each vignette, the respondent was asked: “Is this person severely overpaid, overpaid, fairly paid, underpaid or severely underpaid.” The content of the randomly generated portions (including all components set off by “<>” in the example above, apart from occupation) followed the design described in Table 1. Every respondent was asked similar questions over a fixed set of eight occupations, including: store clerk, farm worker, receptionist, software engineer, manager, medical doctor, taxi driver, and schoolteacher. Table 1: Randomized elements of the experiment design Dimension Values Sex Male/Female Age 25/35/45/55 Pay KZT: (a) 200,000; (b) 300,000; (c) 400,000; (d) 500,000; (e) 600,000 KGS: (a)10,000; (b) 20,000; (c) 30,000; (d) 40,000; (e) 50,000. UZS: (a) 2.5 mln; (b) 3.5 mln; (c) 4.5 mln; (d) 5.5 mln; (e) 6.5 mln. As demonstrated in table 2, randomization was successful across all assigned treatments, as well as for each country individually. There were no statistically significant differences across any treatment arms for key respondent characteristics including gender, (log) income, subjective poverty status at the time of interview, and respondent age. In each country, women were over-represented among respondents (ranging from 59 percent of respondents in Uzbekistan to 70 percent in the Kyrgyz Republic). Average respondent age ranged from 43 in Kazakhstan, to 48 in Uzbekistan. The proposed pay levels were set such that category (b) was the average wage reported by the official statistical authority in each country, while 6 See https://www.worldbank.org/en/country/uzbekistan/brief/l2cu; and https://www.worldbank.org/en/country/tajikistan/brief/listening2tajikistan. 7 option (a) was a rounded increment below the national average, while (c), (d), and (e) were all values above the national average (in terms of fixed local currency increments). Table 2: Balance of respondent characteristics across randomly assigned vignettes Average Kazakhstan Average Kyrgyz Republic Average Uzbekistan Vignette Log Subj. Log Subj. Log Subj. assign Female Inc. poor Age Female Inc. poor Age Female Inc. poor Age Woman 0.60 11.71 0.22 43.68 0.70 9.30 0.23 45.01 0.58 13.92 0.10 48.21 Man 0.59 11.72 0.23 43.39 0.70 9.31 0.23 44.96 0.59 13.95 0.10 48.16 Age 25 0.59 11.72 0.23 43.43 0.70 9.30 0.23 45.41 0.59 13.95 0.10 48.22 Age 35 0.59 11.71 0.23 43.58 0.70 9.31 0.23 45.04 0.59 13.94 0.10 48.07 Age 45 0.60 11.70 0.23 43.84 0.70 9.32 0.22 44.79 0.58 13.92 0.10 48.04 Age 55 0.60 11.73 0.22 43.28 0.69 9.30 0.22 44.70 0.58 13.93 0.10 48.42 Pay 1 0.59 11.69 0.23 43.80 0.69 9.32 0.22 45.11 0.59 13.93 0.10 47.99 Pay 2 0.59 11.73 0.22 43.44 0.71 9.36 0.23 44.94 0.58 13.96 0.09 48.14 Pay 3 0.59 11.72 0.22 43.74 0.70 9.29 0.22 44.81 0.59 13.92 0.10 48.40 Pay 4 0.59 11.72 0.23 43.60 0.71 9.29 0.22 45.04 0.58 13.93 0.10 48.19 Pay 5 0.61 11.71 0.23 43.10 0.70 9.29 0.23 45.02 0.58 13.92 0.11 48.24 Results The results reveal systematic differences in responses about the fairness of randomly assigned wage amounts depending on the sex of the worker described in the vignette. Across the three countries and all eight occupations but without controlling for other characteristics, women were 17 percent less likely to be described as “severely underpaid,” and 9 percent less likely to be described as “somewhat overpaid.” In contrast, they were 7 percent more likely to be “fairly paid,” 8 percent more likely to be described as “somewhat overpaid, and 13 percent more likely to be described as “severely overpaid” (Table 3). These patterns hold in each of the three countries, though with the largest absolute discrepancy in Uzbekistan. Table 3: Summary Response by Gender of Vignette Subject Severely Somewhat Severely Fair Underpaid overpaid overpaid underpaid Total Female 3.3% 11.4% 53.3% 19.0% 12.9% Male 2.9% 10.6% 50.1% 21.0% 15.4% % difference 13% 8% 7% -9% -17% Kyrgyz Republic Female 5.6% 10.1% 44.9% 17.8% 21.6% Male 5.3% 9.4% 44.0% 18.0% 23.2% % difference 5% 7% 2% -1% -7% Kazakhstan Female 4.0% 18.0% 54.5% 19.2% 4.3% Male 3.1% 18.5% 52.3% 21.1% 5.1% % difference 30% -2% 4% -9% -16% Uzbekistan 8 Female 0.5% 6.6% 60.6% 20.1% 12.2% Male 0.5% 4.4% 54.0% 23.8% 17.3% % difference 16% 51% 12% -16% -30% Table 4 reports the results of regressions for which the dependent variable is the categorical response of the respondent regarding the fairness of the wage (with “severely overpaid” coded as 1, “overpaid” coded as 2, “fairly paid” coded as 3, “underpaid” coded as 4, and “severely underpaid” coded as 5). Columns 1- 3 of table 4 use simple OLS regression, with standard errors clustered at the household level. The positive and statistically significant coefficient for “male” signifies that compared to the base category “female,” randomly assigning a male subject in the vignette increases the likelihood the respondent will say the subject is “underpaid” and reduce the likelihood that the respondent will say that the subject is overpaid. The results are fully aligned using ordered logit regression, instead of OLS (columns 4-6). Combining “severe” and “somewhat” responses indicates that respondents were 13 percent more likely to say wages were “too high” when the randomly assigned subject of the vignette a woman, and 34 percent more likely to say they were “too low” when the subject was a man. The coefficient for a female respondent (included in columns 2 and 5) is not significant, indicating that there is no difference between average responses of male and female respondents and the gender of the fictional subject of the vignette. Likewise, restricting the sample to only women (columns 3 and 6) revealed there is almost no difference in average results within this subgroup. Table 4: Summary regression results across all occupations and all countries OLS Ordered Logit (1) (2) (3) (4) (5) (6) Random assignment sex Male 0.094*** 0.094*** 0.094*** 0.238*** 0.238*** 0.239*** (0.007) (0.007) (0.009) (0.017) (0.017) (0.021) Random assignment age Age 35 0.038*** 0.038*** 0.024** 0.092*** 0.092*** 0.063** (0.010) (0.010) (0.012) (0.024) (0.024) (0.029) Age 45 0.056*** 0.056*** 0.058*** 0.144*** 0.144*** 0.153*** (0.010) (0.010) (0.012) (0.024) (0.024) (0.030) Age 55 0.050*** 0.050*** 0.048*** 0.131*** 0.131*** 0.124*** (0.010) (0.010) (0.012) (0.025) (0.025) (0.031) Random assignment pay Pay Group 2 -0.563*** -0.563*** -0.570*** -1.305*** -1.305*** -1.310*** (0.012) (0.012) (0.015) (0.029) (0.029) (0.037) Pay Group 3 -0.980*** -0.980*** -0.997*** -2.360*** -2.360*** -2.372*** (0.013) (0.013) (0.016) (0.034) (0.034) (0.043) Pay Group 4 -1.250*** -1.250*** -1.276*** -3.101*** -3.101*** -3.121*** (0.013) (0.013) (0.017) (0.038) (0.038) (0.049) Pay Group 5 -1.465*** -1.465*** -1.501*** -3.656*** -3.656*** -3.693*** (0.015) (0.015) (0.019) (0.042) (0.042) (0.053) Female Respondent 0.002 0.005 (0.013) (0.033) Constant 4.174*** 4.173*** 4.198*** (0.017) (0.019) (0.021) Sample All All Female All All Female 9 Observations 70,224 70,224 45,960 70,224 70,224 45,960 R2 0.331 0.331 0.334 Adjusted R2 0.331 0.331 0.333 notes: Standard error in parenthesis; .01 - ***; .05 - **; .1 - *; Dependent variable: 1=Severely overpaid, 2= Somewhat overpaid, 3= Fair, 4=Somewhat underpaid, 5=Severely underpaid; Controls for survey round not shown. Base category for gender is female, Base category for age is 25, Base category for pay is the lowest pay. Columns 1-3 report results of ordinary least squares regression, while columns 4-6 report results for ordered logit regression. Interacting the randomized components of treatments allows interpretations of differences on average within subgroups of each randomized component of the survey question. Table 5 reports interaction terms for a female subject of the vignette at each described age level. Coefficients can be interpreted with respect to the excluded category (female, age 25). A positive and statistically significant coefficient signifies that the subject is less likely to be described as overpaid and more likely to be described as underpaid. The results reported in column 1 show that men and women subjects both are more often thought to be underpaid as age increases. The absolute difference is greater for men, ranging from the lowest difference (.185 on the 1–5 scale) to the greatest difference for the most aged group (men aged 55, with .218 difference on the same scale). For women described in a vignette, the difference ranged from .047 for the subject aged 35 to .071 for a woman aged 55. This comparison shows that within treatments that randomly assigned a female name to the subject, the effect of age on the perceived value of work (in terms of the absolute difference between women of differing ages, and men of differing ages) is of similar size to that of men. The effect of age on perceived appropriateness of payment is dominated by the size of the difference between randomly assigned male/female description of the person in the vignette. Column 2 shows that the relationship is nearly unchanged when controlling for the sex of the respondent. Column 3 shows the results when restricting the sample to female respondents. The results show that among women, the size of both the subject gender and age effects is large. This suggest that female respondents on average are more likely to think that men are underpaid compared to women and are more likely to say that older subjects (of either gender) are underpaid compared to younger subjects. Table 5: Randomized characteristics interacted: age (1) (2) (3) Interaction sex#age Female#Age 35 0.047** 0.047** 0.068*** (0.019) (0.019) (0.024) Female#Age 45 0.049*** 0.049*** 0.063*** (0.018) (0.018) (0.024) Female#Age 55 0.071*** 0.071*** 0.078*** (0.019) (0.019) (0.025) Male#Age 25 0.185*** 0.186*** 0.215*** (0.019) (0.019) (0.025) Male#Age 35 0.197*** 0.197*** 0.210*** (0.019) (0.019) (0.025) Male#Age 45 0.223*** 0.223*** 0.268*** (0.020) (0.020) (0.026) Male#Age 55 0.218*** 0.218*** 0.250*** (0.020) (0.020) (0.026) 10 Random assignment pay Pay Group 2 -0.489*** -0.489*** -0.485*** (0.018) (0.018) (0.024) Pay Group 3 -0.868*** -0.868*** -0.871*** (0.018) (0.018) (0.024) Pay Group 4 -1.071*** -1.071*** -1.090*** (0.018) (0.018) (0.024) Pay Group 5 -1.237*** -1.237*** -1.262*** (0.019) (0.019) (0.026) Female Respondent -0.006 (0.018) Constant 4.044*** 4.047*** 4.027*** (0.022) (0.024) (0.029) Sample All All Female Observations 23,880 23,880 14,008 R2 0.302 0.302 0.312 Adjusted R2 0.302 0.302 0.311 notes: Standard error in parenthesis; .01 - ***; .05 - **; .1 - *; Columns 1-3 report results of ordinary least squares regression; Dependent variable: 1=Severely overpaid, 2= Somewhat overpaid, 3= Fair, 4=Somewhat underpaid, 5=Severely underpaid; Controls for survey round not shown. Base category for gender is female, Base category for age is 25, Base category for pay is the lowest pay. Table 6 reports interaction terms for a female subject of the vignette with respect to the perceived appropriateness of pay by random level of pay described. The results reported in column 1 show that within both male and female groups, a higher randomly assigned wage rate led to a lower share of respondents describing the subject as underpaid. The systematic difference for the effect of the wage rate between male and female subjects was again dominated by the randomly assigned gender of the vignette subject. Figure 4 reports the key results from tables 5 and 6 graphically. Table 6: Randomized characteristics interacted: level of pay (1) (2) (3) Interaction sex#level of pay Female#Pay Group 2 -0.505*** -0.505*** -0.512*** (0.025) (0.025) (0.033) Female#Pay Group 3 -0.849*** -0.849*** -0.847*** (0.024) (0.024) (0.032) Female#Pay Group 4 -1.019*** -1.019*** -1.013*** (0.024) (0.024) (0.032) Female#Pay Group 5 -1.178*** -1.178*** -1.167*** (0.024) (0.024) (0.033) Male#Pay Group 1 0.209*** 0.210*** 0.248*** (0.025) (0.025) (0.033) Male#Pay Group 2 -0.263*** -0.263*** -0.208*** (0.026) (0.026) (0.035) Male#Pay Group 3 -0.678*** -0.678*** -0.644*** (0.025) (0.025) (0.034) Male#Pay Group 4 -0.914*** -0.914*** -0.918*** (0.025) (0.025) (0.032) 11 Male#Pay Group 5 -1.088*** -1.088*** -1.108*** (0.025) (0.025) (0.034) Random assignment age Age 35 0.029** 0.029** 0.032* (0.014) (0.014) (0.018) Age 45 0.044*** 0.044*** 0.060*** (0.014) (0.014) (0.017) Age 55 0.052*** 0.052*** 0.057*** (0.014) (0.014) (0.018) Female Respondent -0.007 (0.018) Constant 4.032*** 4.036*** 4.010*** (0.024) (0.026) (0.032) Sample Observations 23,880 23,880 14,008 R2 0.303 0.303 0.314 Adjusted R2 0.303 0.303 0.314 notes: Standard error in parenthesis; .01 - ***; .05 - **; .1 - *; Columns 1-3 report results of ordinary least squares regression; Dependent variable: 1=Severely overpaid, 2= Somewhat overpaid, 3= Fair, 4=Somewhat underpaid, 5=Severely underpaid; Controls for survey round not shown. Base category for gender is female, Base category for age is 25, Base category for pay is the lowest pay. Figure 4: Coefficients of interacted treatments for age and pay group Decrease in "too low," increase in "too high" Increase in "too low," reduction in "too high" 0.40 0.25 0.20 0.00 0.20 -0.20 -0.40 0.15 -0.60 -0.80 0.10 -1.00 -1.20 0.05 -1.40 0.00 Female Male Female Male Notes: change relative to the missing category (the missing blue column). Broad occupational categories corresponding with common groupings of “blue” and “white” collar occupations were deliberately split evenly in the experimental treatment. All respondents were given vignettes for an equal number of both white- and blue-collar occupations. Blue collar occupations included store clerk, farm worker, receptionist, and taxi driver. White collar occupations included software engineer, manager, medical doctor, and schoolteacher. The results provided in table 5 (column 1) show that respondents were significantly more likely to say that vignette subjects were underpaid (or 12 conversely, less likely to say were overpaid) when the occupation was in the white-collar category compared to the blue-collar category. However, including a dummy variable for the occupation being in the white-collar category did not affect the main coefficient of interest (the coefficient for the gender of the randomly assigned vignette subject) to the third decimal place. An insignificant interaction term between the vignette subject being assigned male gender and the occupation being white-collar (or, conversely, the subject being assigned female gender and the occupation being blue-collar) suggests that these two effects are not correlated (column 2). Although respondents were significantly less likely to think workers in blue collar professions were underpaid, this effect did not vary between male and female vignette subjects. The results highlight large differences between occupations (i.e., doctor) vs. positions (i.e. manager) which may suggest power in a work-related responsibilities is a relevant factor, and biases related to views on women’s decision-making may be correlated with views on wages and gender. Table 7: Including Interaction between occupational categories and sex of the vignette subject Across all countries, by blue/white collar occupations (1) (2) (3) Random assignment sex/white collar occupation question Male 0.094*** 0.098*** 0.098*** (0.007) (0.010) (0.010) White collar 0.192*** 0.196*** 0.196*** (0.007) (0.010) (0.010) Male*White collar -0.009 -0.009 (0.014) (0.014) Random assignment age Age 35 0.037*** 0.037*** 0.037*** (0.010) (0.010) (0.010) Age 45 0.055*** 0.055*** 0.055*** (0.010) (0.010) (0.010) Age 55 0.051*** 0.051*** 0.051*** (0.010) (0.010) (0.010) Random assignment pay Pay Group 2 -0.563*** -0.563*** -0.563*** (0.012) (0.012) (0.012) Pay Group 3 -0.980*** -0.980*** -0.980*** (0.012) (0.012) (0.012) Pay Group 4 -1.251*** -1.251*** -1.251*** (0.013) (0.013) (0.013) Pay Group 5 -1.466*** -1.466*** -1.466*** (0.015) (0.015) (0.015) Female Respondent 0.002 (0.013) Constant 4.079*** 4.077*** 4.076*** (0.018) (0.018) (0.020) Sample All All All Observations 70,224 70,224 70,224 Adjusted R2 0.341 0.341 0.341 13 notes: Standard error in parenthesis; .01 - ***; .05 - **; .1 - *; Dependent variable: 1=Severely overpaid, 2= Somewhat overpaid, 3= Fair, 4=Somewhat underpaid, 5=Severely underpaid; Controls for survey round not shown. Base category for gender is female, Base category for age is 25, Base category for pay is the lowest pay. Columns 1-3 report results of ordinary least squares regression, while columns 4-6 report results for ordered logit regression. Blue collar occupations include store clerk, farm worker, receptionist, and taxi driver. White collar occupations include software engineer, manager, medical doctor, and schoolteacher. Nonetheless, there were systematic differences with respect to blue- and white-collar occupations and the randomized age and wage rate of the subject of the vignette. The results reported in table 6 show that the gradient of responses vary by age-cohorts in the vignettes. Comparing the coefficients for the age assignment between columns 1-3 and 4-6 reveals striking differences. The age of vignette subjects in blue- collar occupations are uncorrelated with the perceived appropriateness of the wages they are paid. While in white-collar jobs, there is a strong tendency to say that the vignette subject was underpaid when randomly assigned higher ages (columns 4-6). This pattern suggests that wage increases (and correspondingly perceptions of being underpaid) are much flatter in blue-collar compared to white-collar occupations. This assessment of experience returns in the labor market for occupations that have a career promotion pathway vs. those that tend to have flat income expectations over time tracks real world patterns. Table 8: Within blue- and white-collar occupations Across all countries, by blue/white collar occupations Blue collar White collar (1) (2) (3) (4) (5) (6) Random assignment sex Male 0.098*** 0.098*** 0.093*** 0.090*** 0.090*** 0.095*** (0.010) (0.010) (0.012) (0.009) (0.009) (0.012) Random assignment age Age 35 0.005 0.005 -0.009 0.070*** 0.070*** 0.058*** (0.014) (0.014) (0.017) (0.013) (0.013) (0.016) Age 45 0.014 0.014 0.017 0.099*** 0.099*** 0.098*** (0.014) (0.014) (0.017) (0.013) (0.013) (0.016) Age 55 0.008 0.008 0.001 0.096*** 0.096*** 0.097*** (0.014) (0.014) (0.018) (0.013) (0.013) (0.016) Random assignment pay Pay Group 2 -0.571*** -0.570*** -0.571*** -0.557*** -0.557*** -0.565*** (0.016) (0.016) (0.020) (0.016) (0.016) (0.020) Pay Group 3 -0.937*** -0.936*** -0.948*** -1.024*** -1.024*** -1.046*** (0.017) (0.017) (0.021) (0.016) (0.016) (0.020) Pay Group 4 -1.181*** -1.181*** -1.201*** -1.322*** -1.322*** -1.354*** (0.017) (0.017) (0.022) (0.016) (0.016) (0.020) Pay Group 5 -1.387*** -1.387*** -1.417*** -1.547*** -1.547*** -1.586*** (0.019) (0.019) (0.023) (0.017) (0.017) (0.022) Female Respondent -0.007 0.010 (0.015) (0.015) Constant 4.131*** 4.135*** 4.141*** 4.218*** 4.211*** 4.252*** (0.022) (0.024) (0.027) (0.021) (0.023) (0.025) Sample All All Female All All Female Observations 35,112 35,112 22,980 35,112 35,112 22,980 14 Adjusted R2 0.310 0.310 0.309 0.364 0.364 0.370 notes: Standard error in parenthesis; .01 - ***; .05 - **; .1 - *; Dependent variable: 1=Severely overpaid, 2= Somewhat overpaid, 3= Fair, 4=Somewhat underpaid, 5=Severely underpaid; Controls for survey round not shown. Base category for gender is female, Base category for age is 25, Base category for pay is the lowest pay. Columns 1-3 report results of ordinary least squares regression, while columns 4-6 report results for ordered logit regression. Blue collar occupations include store clerk, farm worker, receptionist, and taxi driver. White collar occupations include software engineer, manager, medical doctor, and schoolteacher. Disaggregating results by occupation reveals further systematic differences on perceptions of the appropriateness of wage rates between men and women. Tables 6-8 describe the results separately for each country and for each occupation included in the study. Regressions including controls for the respondent characteristics variables included in the balance table 2 are included in Appendix B. Table 6 describes the results across all occupations for the Kyrgyz Republic. Column 1 reports the average for the country across all occupations, with the positive and significant coefficient for “male” signifying that compared to the base category “female,” randomly assigning a male subject in the vignette increases the likelihood the respondent will say the subject is “underpaid” and reduce the likelihood that the respondent will say that the subject is “overpaid.” In the Kyrgyz Republic, systematic bias against t women was demonstrated for the occupations of farm workers (column 3), receptionist (column 4), manager (column 6), and medical doctor (column 7). In no case was a there a bias on average against men compared to women (i.e., with a negative and statistically significant coefficient for “male”). Table 7 describes the results across all occupations for Kazakhstan. Column 1 reports the average for the country across all occupations, with the positive and significant coefficient for “male” signifying that compared to the base category “female,” randomly assigning a male subject in the vignette increases the likelihood the respondent will say the subject is “underpaid” and reduce the likelihood that the respondent will say that the subject is “overpaid.” In Kazakhstan, systematic bias against women was demonstrated for the occupations of store clerk, farm worker, receptionist, software engineer and taxi driver. Again, in no case was a there a bias on average against men compared to women (with a negative and statistically significant coefficient for “male”). Table 8 describes the results across all occupations for Uzbekistan. Column 1 reports the average for the country across all occupations, with the positive and significant coefficient for “male” signifying that compared to the base category “female,” randomly assigning a male subject in the vignette increases the likelihood the respondent will say the subject is “underpaid” and reduce the likelihood that the respondent will say that the subject is “overpaid.” In Uzbekistan, systematic bias against women was demonstrated in every occupation studied, and in no case was a bias against men (with a negative and statistically significant coefficient for “male”). 15 Table 9: Results table for Kyrgyz Republic, by occupation All Farm Software Medical Store Clerk Receptionist Manager Taxi Driver Schoolteacher Occupations Worker Engineer Doctor (1) (2) (3) (4) (5) (6) (7) (8) (9) Random assignment sex Male 0.055*** 0.021 0.076** 0.091*** 0.048 0.066* 0.112*** 0.033 -0.000 (0.012) (0.034) (0.035) (0.034) (0.034) (0.036) (0.034) (0.034) (0.034) Random assignment age Age 35 0.036** -0.003 -0.094* 0.038 0.055 0.101** 0.056 0.054 0.110** (0.017) (0.051) (0.049) (0.047) (0.049) (0.049) (0.051) (0.048) (0.046) Age 45 0.049*** -0.004 -0.009 0.026 0.110** 0.096** 0.054 0.009 0.120*** (0.018) (0.050) (0.048) (0.049) (0.050) (0.047) (0.049) (0.049) (0.046) Age 55 0.033* -0.081 -0.041 -0.029 0.040 0.112** 0.143*** 0.028 0.096** (0.018) (0.053) (0.049) (0.050) (0.050) (0.049) (0.048) (0.046) (0.047) Random assignment pay Pay 20000 KGS -0.685*** -0.676*** -0.730*** -0.751*** -0.586*** -0.645*** -0.631*** -0.614*** -0.828*** (0.022) (0.057) (0.055) (0.053) (0.056) (0.053) (0.053) (0.051) (0.052) Pay 30000 KGS -1.204*** -1.119*** -1.196*** -1.255*** -1.112*** -1.155*** -1.195*** -1.173*** -1.436*** (0.022) (0.054) (0.054) (0.052) (0.053) (0.050) (0.051) (0.051) (0.046) Pay 40000 KGS -1.545*** -1.290*** -1.522*** -1.617*** -1.564*** -1.544*** -1.553*** -1.560*** -1.699*** (0.023) (0.056) (0.053) (0.053) (0.052) (0.053) (0.052) (0.051) (0.048) Pay 50000 KGS -1.855*** -1.607*** -1.807*** -1.900*** -1.811*** -1.922*** -1.929*** -1.857*** -1.997*** (0.026) (0.060) (0.057) (0.058) (0.057) (0.057) (0.055) (0.052) (0.053) Constant 4.407*** 4.268*** 4.428*** 4.362*** 4.409*** 4.377*** 4.389*** 4.489*** 4.511*** (0.022) (0.058) (0.054) (0.053) (0.053) (0.052) (0.053) (0.050) (0.049) Sample All All All All All All All All All Observations 23,928 2,991 2,991 2,991 2,991 2,991 2,991 2,991 2,991 R2 0.357 0.273 0.344 0.363 0.350 0.381 0.385 0.389 0.408 Adjusted R2 0.357 0.271 0.342 0.361 0.348 0.379 0.383 0.387 0.406 notes: Standard error in parenthesis; .01 - ***; .05 - **; .1 - *; Ordinary least squares; Dependent variable: 1 = Severely overpaid, 2 = Somewhat overpaid, 3 = Fair, 4 = Somewhat underpaid, 5 = Severely underpaid; Controls for survey round not shown 16 Table 10: Results table for Kazakhstan, by occupation Farm Software Medical All Store Clerk Rec'ist Manager Taxi Driver Schoolteacher Worker Engineer Doctor (1) (2) (3) (4) (5) (6) (7) (8) (9) Random assignment sex Male 0.046*** 0.100*** 0.136*** 0.069* 0.075** 0.027 0.038 -0.061* -0.019 (0.013) (0.036) (0.038) (0.038) (0.032) (0.031) (0.037) (0.031) (0.032) Random assignment age Age 35 0.057*** 0.027 0.044 0.035 0.003 0.099** 0.140*** 0.031 0.106** (0.018) (0.053) (0.050) (0.055) (0.049) (0.043) (0.053) (0.048) (0.045) Age 45 0.085*** 0.019 -0.018 -0.016 0.004 0.192*** 0.197*** 0.122*** 0.163*** (0.018) (0.056) (0.049) (0.057) (0.046) (0.047) (0.048) (0.045) (0.046) Age 55 0.074*** 0.042 0.054 0.019 0.015 0.093** 0.189*** 0.074 0.112** (0.018) (0.054) (0.052) (0.051) (0.046) (0.045) (0.051) (0.046) (0.048) Random assignment pay Pay KZT 300000 -0.485*** -0.556*** -0.480*** -0.489*** -0.447*** -0.465*** -0.334*** -0.475*** -0.656*** (0.021) (0.060) (0.056) (0.074) (0.058) (0.062) (0.056) (0.053) (0.053) Pay KZT 400000 -0.796*** -0.717*** -0.809*** -0.731*** -0.809*** -0.719*** -0.813*** -0.795*** -0.985*** (0.022) (0.057) (0.052) (0.077) (0.057) (0.053) (0.055) (0.053) (0.047) Pay KZT 500000 -1.062*** -0.953*** -0.959*** -1.048*** -1.031*** -0.963*** -1.103*** -1.104*** -1.317*** (0.023) (0.063) (0.057) (0.076) (0.053) (0.050) (0.053) (0.052) (0.047) Pay KZT 600000 -1.204*** -1.070*** -1.070*** -1.167*** -1.144*** -1.148*** -1.277*** -1.251*** -1.518*** (0.025) (0.064) (0.057) (0.078) (0.056) (0.053) (0.057) (0.054) (0.052) Constant 3.672*** 3.394*** 3.571*** 3.459*** 3.725*** 3.602*** 3.897*** 3.758*** 3.976*** (0.029) (0.062) (0.057) (0.083) (0.060) (0.053) (0.055) (0.056) (0.055) Sample All All All All All All All All All Observations 22,416 2,802 2,802 2,802 2,802 2,802 2,802 2,802 2,802 R2 0.264 0.197 0.228 0.246 0.277 0.277 0.325 0.315 0.412 Adjusted R2 0.264 0.194 0.226 0.243 0.275 0.275 0.322 0.313 0.410 notes: Standard error in parenthesis; .01 - ***; .05 - **; .1 - *; Ordinary least squares; Dependent variable: 1 = Severely overpaid, 2 = Somewhat overpaid, 3 = Fair, 4 = Somewhat underpaid, 5 = Severely underpaid; Controls for survey round not shown. 17 Table 11: Results table for Uzbekistan, by Occupation Farm Software Medical All Store Clerk Rec'ist Manager Taxi Driver Schoolteacher Worker Engineer Doctor (1) (2) (3) (4) (5) (6) (7) (8) (9) Random assignment sex Male 0.164*** 0.170*** 0.168*** 0.189*** 0.149*** 0.186*** 0.104*** 0.130*** 0.181*** (0.010) (0.024) (0.027) (0.025) (0.029) (0.026) (0.025) (0.028) (0.026) Random assignment age Age 35 0.029** -0.036 0.017 -0.004 0.065 0.023 0.078** -0.001 0.056* (0.014) (0.035) (0.038) (0.036) (0.042) (0.036) (0.036) (0.040) (0.034) Age 45 0.043*** -0.040 0.053 0.004 0.075* 0.063* 0.130*** 0.036 0.069* (0.014) (0.036) (0.038) (0.036) (0.042) (0.033) (0.037) (0.039) (0.036) Age 55 0.052*** -0.001 0.104*** -0.052 0.078* 0.108*** 0.097*** 0.050 0.064* (0.014) (0.034) (0.037) (0.035) (0.040) (0.036) (0.035) (0.041) (0.034) Random assignment pay Pay UZS 3.5 mln -0.489*** -0.468*** -0.490*** -0.466*** -0.346*** -0.494*** -0.528*** -0.529*** -0.579*** (0.018) (0.042) (0.047) (0.039) (0.049) (0.048) (0.047) (0.047) (0.046) Pay UZS 4.5 mln -0.868*** -0.701*** -0.788*** -0.765*** -0.805*** -0.907*** -0.953*** -0.936*** -1.066*** (0.018) (0.041) (0.045) (0.043) (0.051) (0.043) (0.044) (0.045) (0.042) Pay UZS 5.5 mln -1.071*** -0.802*** -0.967*** -0.952*** -1.028*** -1.105*** -1.256*** -1.135*** -1.357*** (0.018) (0.041) (0.044) (0.040) (0.047) (0.043) (0.040) (0.043) (0.039) Pay UZS 6.5 mln -1.237*** -0.895*** -1.122*** -1.140*** -1.210*** -1.266*** -1.408*** -1.367*** -1.496*** (0.019) (0.042) (0.045) (0.040) (0.043) (0.042) (0.040) (0.043) (0.039) Constant 4.054*** 3.687*** 3.852*** 3.719*** 4.131*** 4.078*** 4.342*** 4.254*** 4.395*** (0.020) (0.044) (0.047) (0.042) (0.048) (0.046) (0.042) (0.047) (0.041) Sample All All All All All All All All All Observations 23,880 2,985 2,985 2,985 2,985 2,985 2,985 2,985 2,985 R2 0.302 0.235 0.273 0.307 0.279 0.350 0.381 0.324 0.431 Adjusted R2 0.302 0.233 0.271 0.305 0.276 0.348 0.379 0.322 0.430 notes: Standard error in parenthesis; .01 - ***; .05 - **; .1 - *; Ordinary least squares; Dependent variable: 1 = Severely overpaid, 2 = Somewhat overpaid, 3 = Fair, 4 = Somewhat underpaid, 5 = Severely underpaid; Controls for survey round not shown. 18 Discussion and conclusions The results of the experiment described in this paper clearly demonstrate significant bias in perceptions of fair (or unfair) wages in nationally representative samples of respondents across three countries in Central Asia, comprising more than 70,000 individual evaluations of the fairness of earnings. The results extend our understanding of potential drivers of patterns of wage differences between men and women. Randomly assigning gender to the subjects of vignettes in a survey experiment (such that half were described as female, and the other half as male) resulted in responses regarding the appropriateness of wages that were systematically biased against women and in favor of men. Other than the key content differences randomized for the purposes of the study (sex, age, and wage rate), the content of the vignettes was standardized such that all subjects (with respect to their age, pay, and occupation) were described the same way. On average, respondents were 13 percent more likely to say wages were too high when the subject of the vignette was a woman, and 34 percent more likely to say they were too low when the subject was a man. However, the scope of the study design places limitations on the interpretation of the results. One such limitation is the narrow set of details provided in the vignettes. As the “story” of the subject is not comprehensive, the lack of details creates the possibility for respondents to “fill-in” details of the story according to their own expectations. As these are unobserved, the study does not isolate what characteristics of women (as described in the vignette) drive the bias, only that the bias exists. Put differently, the study does not identify the precise nature of the implicit assumptions—such as expectations about maternity leave and childcare responsibilities—but only the fact that bias clearly affects the perceived value of women’s work compared to men. Given the salience of gender inequality and the gender pay gap for development outcomes, the results highlight the need for policies and programs that address discriminatory bias against women and improve opportunities for women to equally participate in the labor market. In the context Central Asia where the study was conducted, the lack of legal provisions mandating equal pay for work of equal value is a notable example. However, equal pay regulations alone are unlikely to fully address widespread bias in perceptions, nor the remarkably large gender wage gap in Central Asia. Rather, such policies are likely only the first step in what is likely to be one of many steps needed to correct widespread harmful attitudes. To tackle these inequalities, employers, workers, governments, and the public must confront widespread perceptions about the role and value of women’s work. Without general agreement that discrimination and inequality of opportunity are injustices to be addressed, Central Asia is likely to underperform in reducing poverty and achieving inclusive growth. The study results also highlight the need for further research, especially with a greater focus on the drivers of biased public perceptions. Tackling systemic bias will likely require a better understanding of why it arises in the first place. Though the findings reported here identify the presence of discriminatory attitudes, they do not suffice to establish what interventions would be effective in changing harmful attitudes regarding the appropriate pay for women compared to men. In other contexts, promising 19 interventions have included policies to increase pay transparency (especially at the enterprise level), communication programs highlighting case-studies that challenge misguided perceptions, and positive role-models endorsed by celebrities trusted by the public. Piloting and evaluating such interventions in the context of Central Asia is a promising avenue for further research into addressing the gender inequalities identified in this study. 20 References Adriaans, Jule, and Matteo Targa. "Gender differences in fairness evaluations of own earnings in 28 European countries." European Societies 25.1 (2023): 107-131. Azmat, Ghazala, and Barbara Petrongolo. "Gender and the labor market: What have we learned from field and lab experiments?." Labour economics 30 (2014): 32-40. Barron, John M., Dan A. Black, and Mark A. Loewenstein. "Gender differences in training, capital, and wages." Journal of human Resources (1993): 343-364. Bertrand, Marianne, and Esther Duflo. "Field experiments on discrimination." Handbook of economic field experiments 1 (2017): 309-393. Blau, Francine D., and Lawrence M. Kahn. "The gender wage gap: Extent, trends, and explanations." Journal of economic literature 55.3 (2017): 789-865. Bordalo, Pedro, et al. "Stereotypes." The Quarterly Journal of Economics 131.4 (2016): 1753-1794. Coffman, Katherine, Clio Bryant Flikkema, and Olga Shurchkov. "Gender stereotypes in deliberation and team decisions." Games and Economic Behavior 129 (2021): 329-349. International Labor Organization. Global Wage Report 2022-23: The impact of inflation and COVID-19 on wages and purchasing power. International Labor Organization, 2022. Maloney, Elizabeth Maureen. “The Gender Wage Gap: A Product of Misogyny, Not Just Gender Norms,” (2022) Muradova, Sevilya, and William Hutchins Seitz. "Gender Discrimination in Hiring: Evidence from an Audit Experiment in Uzbekistan." (2021). World Bank Policy Research Working Paper, No. 9784., (2021). Schäfer, Andrea, and Karin Gottschall. "From wage regulation to wage gap: how wage-setting institutions and structures shape the gender wage gap across three industries in 24 European countries and Germany." Cambridge Journal of Economics 39.2 (2015): 467-496. Triana, María del Carmen, et al. "Perceived workplace gender discrimination and employee consequences: A meta-analysis and complementary studies considering country context." Journal of management 45.6 (2019): 2419-2447. Whitehouse, Gillian, and Meg Smith. "Equal pay for work of equal value, wage-setting and the gender pay gap." Journal of Industrial Relations 62.4 (2020): 519-532. Woo, Byung-Deuk, Lindsey A. Goldberg, and Frederick Solt. Forthcoming. “Public Gender Egalitarianism: A Dataset of Dynamic Comparative Public Opinion Toward Egalitarian Gender Roles in the Public Sphere.” British Journal of Political Science. doi: 10.31235/osf.io/37dq9 21 Appendix A: Summary responses by additional randomized components Summary Response by Age of Vignette Subject Severely Somewhat Somewhat Severely Fair overpaid overpaid underpaid underpaid Total Age 25 3.5% 11.3% 52.3% 19.4% 13.5% Age 35 3.0% 11.4% 51.8% 19.9% 13.9% Age 45 3.0% 10.6% 51.7% 20.1% 14.7% Age 55 3.1% 10.8% 50.9% 20.7% 14.5% Kyrgyz Republic Age 25 5.9% 9.6% 44.5% 18.4% 21.6% Age 35 5.1% 10.2% 44.9% 18.1% 21.7% Age 45 5.0% 9.1% 44.8% 18.1% 23.0% Age 55 5.7% 10.1% 43.6% 17.1% 23.4% Kazakhstan Age 25 4.0% 18.7% 54.5% 18.7% 4.1% Age 35 3.3% 18.7% 53.6% 19.5% 4.9% Age 45 3.6% 17.8% 53.1% 20.4% 5.0% Age 55 3.1% 17.8% 52.4% 21.9% 4.8% Uzbekistan Age 25 0.5% 6.1% 58.1% 21.0% 14.2% Age 35 0.5% 6.0% 57.2% 21.9% 14.4% Age 45 0.4% 4.9% 57.4% 21.7% 15.7% Age 55 0.6% 5.0% 56.8% 23.1% 14.5% Summary Response by Pay of Vignette Subject Severely Somewhat Somewhat Severely Fair overpaid overpaid underpaid underpaid Total Pay group 1 0.3% 2.1% 22.1% 33.9% 41.6% Pay group 2 0.8% 4.9% 46.0% 31.5% 16.8% Pay group 3 1.6% 10.2% 63.1% 18.6% 6.5% Pay group 4 4.0% 16.8% 66.4% 9.7% 3.1% Pay group 5 9.1% 21.3% 61.6% 5.9% 2.1% Kyrgyz Republic KGS 10,000 0.4% 2.4% 11.3% 21.0% 64.9% KGS 20,000 1.2% 4.4% 35.0% 32.7% 26.6% KGS 30,000 2.4% 8.8% 58.6% 19.6% 10.6% KGS 40,000 6.6% 14.5% 63.4% 10.3% 5.2% KGS 50,000 16.7% 18.7% 54.3% 6.2% 4.1% Kazakhstan KZT 200,000 0.5% 3.6% 32.4% 47.4% 16.1% 22 KZT 300,000 1.1% 9.2% 57.1% 27.7% 5.0% KZT 400,000 2.1% 17.7% 64.4% 14.2% 1.6% KZT 500,000 4.9% 28.5% 59.9% 6.5% 0.2% KZT 600,000 8.9% 32.7% 53.5% 4.5% 0.4% Uzbekistan UZS 2.5 mln 0.0% 0.5% 23.4% 34.2% 42.0% UZS 3.5 mln 0.0% 1.5% 46.4% 34.0% 18.1% UZS 4.5 mln 0.2% 4.6% 66.4% 21.8% 7.0% UZS 5.5 mln 0.5% 8.0% 75.7% 12.0% 3.7% UZS 6.5 mln 1.7% 13.3% 76.3% 7.0% 1.7% 23 Appendix B: disaggregated regression results including controls for respondent characteristics Table 12: Regression table by occupation with controls for the Kyrgyz Republic Store Farm Software Medical School All Rec'ist Manager Taxi Driver Clerk Worker Engineer Doctor Teacher (1) (2) (3) (4) (5) (6) (7) (8) (9) Random assignment sex Male 0.061*** 0.032 0.072** 0.104*** 0.029 0.077** 0.110*** 0.046 0.030 (0.012) (0.037) (0.036) (0.036) (0.035) (0.037) (0.035) (0.035) (0.035) Random assignment age Age 35 0.027 0.015 -0.124** 0.024 0.066 0.113** 0.035 0.035 0.074 (0.018) (0.053) (0.051) (0.049) (0.050) (0.052) (0.051) (0.050) (0.049) Age 45 0.037** 0.002 -0.018 0.006 0.100* 0.071 0.053 0.013 0.086* (0.018) (0.052) (0.049) (0.050) (0.051) (0.049) (0.050) (0.051) (0.048) Age 55 0.028 -0.058 -0.055 -0.054 0.043 0.110** 0.157*** 0.015 0.072 (0.019) (0.055) (0.050) (0.051) (0.050) (0.052) (0.048) (0.048) (0.050) Random assignment pay Pay 20000 KGS -0.693*** -0.660*** -0.759*** -0.794*** -0.580*** -0.637*** -0.644*** -0.622*** -0.850*** (0.022) (0.058) (0.056) (0.054) (0.058) (0.055) (0.057) (0.053) (0.054) Pay 30000 KGS -1.213*** -1.132*** -1.212*** -1.301*** -1.097*** -1.145*** -1.222*** -1.160*** -1.440*** (0.023) (0.057) (0.056) (0.053) (0.055) (0.053) (0.052) (0.054) (0.049) Pay 40000 KGS -1.551*** -1.281*** -1.550*** -1.635*** -1.572*** -1.533*** -1.584*** -1.560*** -1.692*** (0.024) (0.059) (0.054) (0.054) (0.055) (0.055) (0.053) (0.054) (0.050) Pay 50000 KGS -1.871*** -1.604*** -1.852*** -1.951*** -1.814*** -1.926*** -1.953*** -1.868*** -2.002*** (0.027) (0.062) (0.059) (0.059) (0.059) (0.059) (0.055) (0.055) (0.056) Controls Female Respondent 0.006 -0.070 -0.021 0.025 0.032 0.083** 0.081* -0.050 -0.027 (0.025) (0.044) (0.041) (0.041) (0.043) (0.041) (0.043) (0.041) (0.040) Ln(per capita inc) 0.031** 0.021 0.003 0.035* 0.051*** 0.035* 0.034** 0.024 0.045** (0.013) (0.019) (0.015) (0.019) (0.019) (0.019) (0.017) (0.019) (0.018) Self Assessed Poor 0.056* 0.053 0.060 0.097** 0.101** -0.003 -0.018 0.054 0.108** (0.029) (0.051) (0.047) (0.047) (0.050) (0.046) (0.050) (0.047) (0.048) Respondent Age -0.001 -0.002 0.001 -0.001 -0.003** -0.001 -0.002* 0.002 0.001 (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) Constant 4.151*** 4.162*** 4.397*** 4.069*** 4.042*** 4.060*** 4.161*** 4.227*** 4.066*** (0.143) (0.206) (0.161) (0.211) (0.208) (0.212) (0.188) (0.207) (0.198) Sample 21,984 2,748 2,748 2,748 2,748 2,748 2,748 2,748 2,748 R2 0.364 0.278 0.356 0.380 0.358 0.380 0.401 0.395 0.411 Adjusted R2 0.363 0.274 0.353 0.377 0.355 0.377 0.398 0.392 0.408 notes: Standard error in parenthesis, clustered at the household level; .01 - ***; .05 - **; .1 - *; Dependent variable: 1 = Severely overpaid, 2 = Somewhat overpaid, 3 = Fair, 4 = Somewhat underpaid, 5 = Severely underpaid; Controls for survey round not shown 24 Table 13: Regression table by occupation with controls for Kazakhstan Store Farm Software Medical School All Rec'ist Manager Taxi Driver Clerk Worker Engineer Doctor Teacher (1) (2) (3) (4) (5) (6) (7) (8) (9) Random assignment sex Male 0.033** 0.084** 0.148*** 0.043 0.059* 0.012 0.017 -0.045 -0.025 (0.013) (0.038) (0.040) (0.038) (0.034) (0.032) (0.036) (0.033) (0.033) Random assignment age Age 35 0.039** -0.026 0.074 0.034 -0.031 0.119*** 0.116** -0.016 0.059 (0.018) (0.054) (0.052) (0.060) (0.052) (0.044) (0.050) (0.051) (0.048) Age 45 0.075*** -0.023 -0.016 0.049 -0.040 0.233*** 0.140*** 0.121** 0.119** (0.018) (0.055) (0.052) (0.055) (0.048) (0.048) (0.045) (0.048) (0.048) Age 55 0.061*** 0.000 0.082 0.030 -0.017 0.120** 0.151*** 0.056 0.083* (0.018) (0.056) (0.054) (0.055) (0.048) (0.047) (0.050) (0.048) (0.050) Random assignment pay Pay KZT 300000 -0.500*** -0.523*** -0.499*** -0.548*** -0.458*** -0.528*** -0.389*** -0.458*** -0.642*** (0.022) (0.062) (0.058) (0.062) (0.062) (0.064) (0.058) (0.057) (0.056) Pay KZT 400000 -0.808*** -0.689*** -0.810*** -0.789*** -0.785*** -0.783*** -0.831*** -0.778*** -0.987*** (0.022) (0.061) (0.053) (0.066) (0.063) (0.053) (0.056) (0.056) (0.051) Pay KZT 500000 -1.079*** -0.926*** -0.975*** -1.099*** -1.031*** -1.026*** -1.114*** -1.114*** -1.334*** (0.022) (0.065) (0.061) (0.065) (0.057) (0.050) (0.054) (0.055) (0.051) Pay KZT 600000 -1.215*** -1.049*** -1.084*** -1.222*** -1.166*** -1.190*** -1.288*** -1.256*** -1.492*** (0.026) (0.066) (0.060) (0.066) (0.058) (0.054) (0.054) (0.058) (0.053) Controls Female Respondent 0.026 0.053 0.028 -0.020 0.051 0.022 0.053 -0.017 0.047 (0.029) (0.046) (0.045) (0.046) (0.039) (0.039) (0.040) (0.041) (0.039) Ln(per capita inc) 0.056*** 0.036* 0.046* 0.025 0.087*** 0.051** 0.104*** 0.039* 0.069*** (0.015) (0.022) (0.025) (0.025) (0.021) (0.023) (0.022) (0.021) (0.021) Self-Assessed Poor 0.115*** 0.118** 0.116** 0.120** 0.091** 0.094** 0.111*** 0.148*** 0.110** (0.033) (0.054) (0.053) (0.049) (0.042) (0.040) (0.042) (0.047) (0.043) Respondent Age 0.000 -0.002 0.002 -0.002 -0.001 0.001 0.001 0.001 0.002 (0.001) (0.002) (0.002) (0.002) (0.001) (0.001) (0.001) (0.001) (0.001) Constant 2.983*** 3.033*** 2.875*** 3.283*** 2.748*** 2.970*** 2.628*** 3.229*** 3.068*** (0.189) (0.265) (0.292) (0.308) (0.259) (0.274) (0.264) (0.261) (0.258) Sample 20,056 2,507 2,507 2,507 2,507 2,507 2,507 2,507 2,507 R2 0.276 0.199 0.236 0.269 0.291 0.302 0.346 0.329 0.416 Adjusted R2 0.275 0.195 0.232 0.265 0.288 0.299 0.343 0.325 0.413 notes: Standard error in parenthesis, clustered at the household level; .01 - ***; .05 - **; .1 - *; Dependent variable: 1 = Severely overpaid, 2 = Somewhat overpaid, 3 = Fair, 4 = Somewhat underpaid, 5 = Severely underpaid; Controls for survey round not shown 25 Table 14: Regression table by occupation with controls for Uzbekistan Farm Software Medical School All Store Clerk Recep. Manager Taxi Driver Worker Engineer Doctor Teacher (1) (2) (3) (4) (5) (6) (7) (8) (9) Random assignment sex Male 0.165*** 0.171*** 0.164*** 0.190*** 0.147*** 0.187*** 0.108*** 0.133*** 0.182*** (0.010) (0.024) (0.027) (0.025) (0.029) (0.026) (0.025) (0.028) (0.026) Random assignment age Age 35 0.030** -0.027 0.024 0.005 0.056 0.025 0.074** -0.004 0.050 (0.014) (0.035) (0.038) (0.036) (0.042) (0.036) (0.036) (0.040) (0.034) Age 45 0.043*** -0.039 0.055 0.004 0.071* 0.065* 0.130*** 0.035 0.069* (0.014) (0.036) (0.039) (0.035) (0.043) (0.034) (0.037) (0.039) (0.036) Age 55 0.054*** 0.002 0.108*** -0.048 0.070* 0.120*** 0.103*** 0.058 0.057* (0.014) (0.034) (0.037) (0.034) (0.041) (0.036) (0.036) (0.041) (0.034) Random assignment pay Pay UZS 3.5 mln -0.493*** -0.477*** -0.497*** -0.461*** -0.333*** -0.499*** -0.535*** -0.524*** -0.591*** (0.018) (0.042) (0.047) (0.039) (0.050) (0.048) (0.047) (0.048) (0.047) Pay UZS 4.5 mln -0.872*** -0.709*** -0.791*** -0.765*** -0.806*** -0.911*** -0.957*** -0.935*** -1.069*** (0.018) (0.041) (0.045) (0.043) (0.051) (0.043) (0.044) (0.046) (0.043) Pay UZS 5.5 mln -1.075*** -0.809*** -0.977*** -0.954*** -1.029*** -1.110*** -1.254*** -1.134*** -1.361*** (0.018) (0.041) (0.045) (0.040) (0.047) (0.043) (0.040) (0.044) (0.039) Pay UZS 6.5 mln -1.238*** -0.893*** -1.128*** -1.137*** -1.209*** -1.265*** -1.406*** -1.369*** -1.498*** (0.019) (0.042) (0.045) (0.040) (0.043) (0.042) (0.040) (0.043) (0.039) Controls Female Respondent 0.007 0.037 0.035 0.033 -0.070** 0.035 -0.013 -0.009 -0.001 (0.018) (0.026) (0.030) (0.027) (0.033) (0.027) (0.029) (0.032) (0.028) Ln(per capita inc) 0.028*** 0.012 0.015 0.052*** 0.036** 0.011 0.035** 0.032** 0.025** (0.009) (0.012) (0.012) (0.014) (0.015) (0.013) (0.014) (0.016) (0.012) Self-Assessed Poor -0.098*** -0.156*** -0.098* -0.028 -0.142** -0.098** -0.120** -0.072 -0.093** (0.029) (0.043) (0.051) (0.046) (0.056) (0.041) (0.052) (0.056) (0.044) Respondent Age 0.002** 0.002** 0.003** 0.003*** -0.001 0.001 -0.000 0.003** 0.001 (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) Constant 3.597*** 3.402*** 3.509*** 2.814*** 3.735*** 3.887*** 3.882*** 3.684*** 4.006*** (0.135) (0.175) (0.190) (0.202) (0.222) (0.196) (0.213) (0.244) (0.173) Sample 23,560 2,945 2,945 2,945 2,945 2,945 2,945 2,945 2,945 R2 0.307 0.242 0.279 0.318 0.288 0.353 0.385 0.331 0.435 Adjusted R2 0.306 0.239 0.276 0.315 0.285 0.351 0.383 0.328 0.432 notes: Standard error in parenthesis, clustered at the household level; .01 - ***; .05 - **; .1 - *; Dependent variable: 1 = Severely overpaid, 2 = Somewhat overpaid, 3 = Fair, 4 = Somewhat underpaid, 5 = Severely underpaid; Controls for survey round not shown 26