What Happens When Employers Can No Longer Discriminate in Job Ads?

When employers’ explicit gender requests were unexpectedly removed from a Chinese job board overnight, pools of successful applicants became more integrated: women’s (men’s) share of call-backs to jobs that had requested men (women) rose by 62 (146) percent. The removal ‘worked’ in this sense because it generated a large increase in gender-mismatched applications, and because those applications were treated surprisingly well by employers, suggesting that employers’ gender requests often represented relatively weak preferences or outdated stereotypes. The job titles that were integrated by the ban, however, were not the most gendered ones, and were disproportionately lower-wage jobs. The data here is provided by a Xiamen-based job board, XMRC. Two samples are used, the main analysis sample and the DiD sample. To construction of the main analysis sample started with the universe of applications that were made on XMRC between January 1, 2018 and October 25, 2019, and the corresponding ads. Only the following job ads were retained: Job ads that received at least one application both before and after the ad ban (March 1, 2019), and Applications to those ads that were made between August 3,1 2018 and August 29, 2019. This results in a sample of 52 complete weeks (26 before and 26 after the ban), in which the first post-ban week begins on Friday March 1, 2019 -- the first day of the ban. This ends with a wide window (almost two years) to make sure that all the job ads that ‘straddled’ the ban were captured. Then only the ads that actually straddled the ban (received at least one application before and after it) were retained. Finally, the analysis sample comprises all applications to those ads that occurred during a one-year window surrounding the ban. To allow comparison of changes on XMRC around the ban with changes in 2018, also a DiD sample was created. To do so, the main estimation sample was replicated -- which comprises applications that were made between September 2018 and August 2019 -- on two different periods: January -- August 2018 and January-August 2019. The latter period contains the date on which the 2019 ban occurred, and the former contains the date on which it would have occurred in 2018. Unfortunately, these two periods cannot be designed to exactly mimic the main analysis sample because no XMRC data from 2017 are available. While this restricts the length of the pre-ban period in both years to just two months, it allows to compare trends in the main outcomes between 2018 and 2019 on both sides of the ban date. This sample was used to conduct a difference-in-difference analysis of the ban’s effect -- which uses equivalent days or weeks from 2018 as controls for 2019 -- as a robustness check of the main results. Notably, since important events affecting China’s labor market -- especially the Spring Festival -- are determined by the lunar calendar, this new DiD sample requires to line up days and weeks between 2018 and 2019 to that they represent the same days and weeks relative to the start of the Spring Festival in both years. The primary dataset comprises 3,130,317 applications made by 204,343 workers (resumes) to 116,725 ads, placed by 15,437 firms, resulting in 348,062 call-backs.

当雇主对性别的明确要求意外地从一家中国招聘平台上在一夜之间被移除时，成功申请者的群体变得更加多元化：要求男性（女性）职位的女性（男性）获得的回电比例分别上升了62%（146%）。
此移除在某种程度上‘奏效’，因为它引发了性别不匹配申请的大量增加，并且这些申请得到了雇主的出乎意料的良好对待，这表明雇主对性别的请求通常代表了相对微弱的偏好或过时的刻板印象。
然而，该禁令整合的职位并非最具性别化的职位，且这些职位工资水平普遍偏低。此数据由厦门的招聘平台XMRC提供。
本研究使用了两个样本，即主要分析样本和双重差分（DiD）样本。
主要分析样本的构建始于2018年1月1日至2019年10月25日期间在XMRC上提交的申请及其相应的广告。
仅保留了以下广告：在广告禁令（2019年3月1日）前后都至少收到一个申请的广告，以及2018年8月3日至8月29日期间对这些广告的申请。
这导致了一个包含52周（禁令前26周和禁令后26周）的样本，禁令后的第一周始于2019年3月1日，即禁令的第一天。这确保了几乎两年的广泛时间窗口，以确保捕捉到所有跨越禁令的招聘广告。
然后，仅保留了实际跨越禁令的广告（在禁令前后都至少收到一个申请的广告）。
最后，分析样本包括在禁令前后一年内对那些广告的所有申请。
为了比较禁令前后XMRC上变化与2018年的变化，还创建了一个DiD样本。
为此，复制了主要估计样本——该样本包括2018年9月至2019年8月期间提交的申请——在两个不同的时期：2018年1月至8月和2018年1月至8月。
后者包含了2019年禁令发生的日期，而前者包含了2018年禁令可能发生的日期。
遗憾的是，这两个时期无法设计得与主要分析样本完全相同，因为无法获取2017年的XMRC数据。
尽管这限制了两年中禁令前的时期仅为两个月，但它允许比较禁令日期两侧2018年和2019年主要结果的趋势。
此样本用于进行禁令效果的差异-差异分析——该分析使用2018年的等效日或周作为2019年的控制变量——作为主要结果的稳健性检验。
值得注意的是，由于影响中国劳动力市场的重要事件——尤其是春节——由农历决定，因此这个新的DiD样本需要将2018年和2019年的天数和周数对齐，以确保它们相对于每年春节的开始日期代表相同的日期和周数。
主要数据集包括204,343名工作者（简历）向15,437家公司发布的116,725个广告提交的3,130,317份申请，导致了348,062次回电。