Ethnic Discrimination in China's Internet Job Board Labor Market

We conduct a large�?scale field experiment to investigate how Chinese firms respond to job applications from ethnic minority and Han applicants for jobs posted on a large Chinese Internet job board. We denote ethnicity by means of names that are typically Han Chinese and distinctively Mongolian, Tibetan, and Uighur. We find significant differences in the callback rates by ethnicity and that these differences vary systematically across ethnic groups. Not all firms discriminate – approximately half treat all candidates equally. State-owned firms are significantly less likely than privately�?owned firms to discriminate against minorities by calling only candidates with Han names and much more likely to treat candidates equally.


Introduction and Context
Laborers shall not be discriminated against in employment due to their nationality, race, sex, or religious belief. Article 12: People's Republic of China Labor Law Workers enjoy the right to employment on equal footing and to choice of jobs on their own initiative in accordance with law. In seeking employment, workers shall not be subject to discrimination because of their ethnic backgrounds, race, gender, religious beliefs, etc.
Article 3: People's Republic of China Law on Promotion of Employment There is a nascent but growing academic literature focused on how China's ethnic minorities have fared in China's economic transition. Researchers have addressed disparities in rural and urban income (Gustafsson & Li 2003) (Ding & Li 2009), rural poverty (Gustafsson & Ding 2008), occupational attainment and job segregation (Hannum & Xie, 1998), job search, hiring, and promotion (Hasmath 2011(Hasmath , 2012bHowell and Fan 2011), and labor force participation (Maurer--Fazio, Hughes, & Zhang 2007, 2010. Although the specific research questions asked, and methodologies applied, have varied considerably, most of the above papers find that China's ethnic minorities have not fared well in comparison to the majority Han during China's transition to a market economy. However, some of the differences in Han and minority wellbeing are attributable to differences in education levels and residential location. It is particularly important to sort out whether and to what extent observed differences in the labor market outcomes of majority Han and ethnic minority participants are due to differences like these in Han and ethnic minority attributes (differences in productive characteristics) or to the treatment of those attributes, that is, to discrimination. A number of the above mentioned studies apply decomposition techniques to try to discover what share of the differences in labor market outcomes can be "explained" by differences in productive characteristics. The "unexplained" differences, that is, the residual differences in income or labor force participation or occupational attainment are then often attributed to discrimination but may in reality be due to model 4 misspecification, omitted variables, and/or unobservable aspects of productivity differences.
This study takes a more direct approach to studying discrimination. Using a resume audit approach we directly measure the extent of ethnic discrimination in the hiring practices of Chinese firms. The resume audit methodology allows us to experimentally strictly control aspects of job candidates' productivity while varying their ethnicity (as denoted by ethnically differentiated names). 1 Resume audit studies have been used to study discrimination in numerous country and labor market contexts. 2 In our resume audit study, conducted in the summer of 2011, we submitted 21,592 applications to 10,796 positions advertised on Internet job boards in 6 Chinese cities. The positions varied over three types of occupations. We randomly paired Han applications with one of three minorities with ethnically distinct names: Mongolian, Uighur, and Tibetan. We found significant discrimination against each of the minorities except in Hohhot and Urumqi----two autonomous minority areas.
Discrimination was less prevalent in tight labor markets. Although each minority experienced discrimination, not all firms discriminated. Close to half the firms treated their applicants equally. State--owned firms were significantly more likely to treat candidates equally and much less likely to discriminate in favor of Han candidates than private firms. This is the first resume audit study to test for discrimination in hiring against ethnic minorities in China.

5
In the following sections of this paper, we first provide some context on China's ethnic minorities and then review the literature that examines how they have fared during China's economic transformation to a market economy. We then briefly review the literature on ethnicity--focused resume audit studies and go on to describe our own experimental design. We then present the analysis of our results-first from the perspective of individual applicants and then from the perspective of the firms selecting applicants for callbacks. We summarize our findings in the concluding section.

China's Ethnic Minorities
According to China's 2010 population census, ethnic minorities constitute 8.4 % of the Chinese population (112 million people). We use the term ethnic minority here to refer to members of the 55 minority groups who, along with the Han majority, 6 economic well being of China's minority peoples. Even though overall economic indicators show a rising standard of living in ethnic minority regions (Mackerras, 2003: 56-76), these improvements have not kept pace with developments in the national economy. Sautman (2010) reports that the material gap between Han and minority areas continues to widen.
According to a recent government White Paper: The Constitution stipulates, "The state does its utmost to promote the common prosperity of all ethnic groups in the country." The Law on Regional Ethnic Autonomy stipulates that it is a legal obligation of the higher--level state organs to help the minority areas accelerate their development. ( (Bhattacharji, 2012). For example, Xinjiang's economy has grown at double--digit rates. The rising number of jobs has led to substantial in--migration of Han to the area. This has caused Uighur residents to fear they are being excluded from the good jobs and to believe that the Han migrants are grabbing these jobs as well as other resources (the Economist, June 30, 2011), which in turn has led to a rising ethno--religious consciousness (Hasmath 5 Barabantseva (2009) notes that although the rhetoric surrounding the Western Development Plan suggests that it was designed specifically for ethnic minorities and minority areas, one of its core elements is the opening of the areas natural resources for the benefit of the rest of the country. 7 2012). There are reports of rising resentment over job discrimination (Bequelin 2009, Fallows 2009 According to Sautman (2010), there are two main types of preferential policies aimed at China's ethnic minorities: subsidies to minority areas and affirmative action for minority people. The extent of the latter varies by particular minority. The government's preferential policies include preferential treatment in school admissions, hiring and promotion. Sautman (2010) points out that the preferential policies were designed at a time when the state sector was the dominant sector in the Chinese economy and that few of these policies apply to the private sector, which has grown in importance rapidly throughout the reform period. Whether, and to what extent, these policies are implemented are important questions, as is the question of whether the policies are differentially applied in state--sector and privately owned firms.
While ethnic minorities in western China resent the in--migration of substantial numbers of Han, there is also rising Han resentment of minorities who are viewed as receiving unwarranted advantages (Sautman, 2010 to interview-a critical first step in the hiring process.

Literature Review-Ethnic Minorities in China's Economic Reform
There is a growing interest in understanding how China's economic transition of the last three decades has affected ethnic minority people. Gustafsson and Li (2003) find that the gaps in rural income between the Han and ethnic minorities widened from 19.2% to 35.9% between 1988 and 1995 (based on survey data from 19 provinces). They decompose the income differentials and attribute the lion's share to differences in endowments-particularly to location, given that China's minorities are clustered in provinces with low per capita GDP. Gustaffson and Ding (2008) build on this earlier work to focus on poverty. Based on a 2002 survey of 22 provinces, they argue that poverty in rural China has a very strong spatial dimension-that ethnic minorities have higher rates of both persistent and temporary poverty because minorities are concentrated in western China, home to most of China's poor. They report that that ethnic minority status has little independent effect in explaining poverty and, rather, that factors such as the education level of the household head, village mean income, and whether the village is located in a mountainous area are much more important factors than ethnicity.
Hannum and Xie (1998) employ population census data to examine the effects of market reform on differences in occupational attainment of Xinjiang's (mainly Turkic) minorities in comparison to the Han. They find that the ethnic gap in occupational attainment between the Han and the minorities widened between the 1982 and 1990 censuses. They also conclude that the gap is not directly attributable to discrimination but rather to growing differences in productive characteristics, in particular to an increased gap in the educational attainment between the Han and the minorities and a presumed strengthening of the relationship between educational attainment and higher--status occupations.
In contrast to the above--mentioned studies, when Ding and Li (2009) analyze differences in income determination for Hui (minority) and Han urban residents in Ningxia based on survey data gathered in 2007 and decompose the differences in earnings into treatment and endowment effects, they find that the treatment effects are more important than endowment effects in explaining the incomes differences.
They point out, however, that the treatment effects do not always favor the Han. In their study, the returns to education are somewhat higher for the Han than the Hui, while the returns to experience are higher for the Hui than the Han. They find that state ownership of the workplace favors the Han. Maurer--Fazio, Hughes, and Zhang (2007) find that minorities were affected more adversely than the Han by reductions in urban sector employment and exited the labor force more rapidly than Han. Maurer--Fazio, Hughes, and Zhang (2010) estimate urban labor force participation rates of women. They focus on the experience of six ethnic minorities and the Han and find sizable differences. In pair--wise comparisons between Han and ethnic women, they find that the treatment of women's characteristics, whether in the market or socially, tends to increase the Han advantage in labor force participation, while the levels of those characteristics tend to reduce the Han advantage. After conducting interviews and analyzing census data, Hasmath (2011) argues that in practice ethnic minorities are disadvantaged relative to the Han in terms of hiring and promotions, particularly for well--paying, high--skill jobs.
The economic literature to date suggests that China's ethnic minorities are not faring particularly well, relative to the Han, in terms of labor market outcomes.
However, there is no clear consensus as to whether the differences in income, poverty, occupational attainment, and related factors are due to differences in residential location and productive characteristics or to discriminatory treatment.

Review of the Literature-Ethnicity--Focused Resume Audit Studies
The resume audit methodology is well suited for measuring discrimination in hiring. 7 Resume audit studies combine the advantages of controlled experiments and realistic field settings. In a resume audit study, sets of resumes are created for fictitious applicants. They are carefully crafted to ensure that candidates will appear, on paper, to be equally productive with similar work and educational histories. The researchers then control a particular attribute under study such as gender or age (or in our case ethnicity). Discrimination is measured by the differences in the rates of callbacks for interviews received by individuals whose applications typically vary only in terms of the attribute under study.
Resume audit studies have been used to investigate the role of ethnicity in a number of different national settings. Bertrand and Mullainathan (2004) study race in US labor markets by randomly assigning White--sounding and African American-sounding names to resumes and submitting those resumes to positions advertised in Boston and Chicago newspapers. They find that candidates with White names receive 50% more callbacks than candidates with African American names. Siddique (2008) explores caste--based discrimination in India's white--collar labor market. She finds that low--caste applicants (signaled by typical low--caste names) need to put in 20% more applications than high--cast applicants to get the same number of callbacks for interviews. She finds that high--caste applicants received even higher callback rates when the recruiting is carried out by men or by Hindu recruiters. Kaas and Manger (2011) develop a study of German firms' responses to applications for student internships from applicants with Turkish--and German--sounding names.
They are particularly careful to create resumes that do not conflate immigration status with ethnicity. The applicants they create with Turkish--sounding names "are" 7 See for example, Riach and Rich 2002, Bertrand and Mullainathan 2004, and Pager 2007 German citizens who were both born and educated in Germany and whose mother tongue is German. The only difference between these applicants and the others lies in the ethnicity of their names. They find, all else equal, that the applicants with German sounding names get 14% more callbacks than those with Turkish--sounding names, in general, but 24% more callbacks from small firms. However, they are able to eliminate the differences in callback rates by including letters of reference that contain favorable information about the personality of the applicant. They take this to indicate that the discrimination is statistical in nature, since supplying more information could eliminate it. Carlsson and Rooth (2007), Bursell (2007), and Carlsson (2010)  Their male and female applicants were assigned roles as either Beijing residents or applicants from outside Beijing with rural household registrations. These applicants audited 81 employers and found that rural and female applicants were much more likely to receive negative responses from recruiters than male and urban applicants.
We believe that our study is the first resume audit study to focus on potential ethnic discrimination in China's contemporary labor markets.

Experiment Design
This experiment focuses on one very dynamic segment of China's labor market-the Internet job board sector. We employ the largest of the Chinese job boards, 51job.com, to investigate whether Chinese firms respond differently to job applications from ethnic minority and Han applicants. We denote ethnicity by using applicant names for ethnic minority job candidates that are recognizably different from those of the Han majority but typical of Mongolian, Uighur, and Tibetan individuals, respectively.
On a daily basis, millions of jobs are advertised on China's Internet job boards and many millions of applications are submitted in response. Some of these job boards are nationally recognized, while others focus on jobs in a particular region or city or on jobs of a particular type. Table 1 presents data on the number of page views for the three most popular job boards in China as of spring 2011.
Our experiment was large in scale-in the summer of 2011, we submitted 21,592 on--line applications for 10,796 advertised positions. We applied for positions in three different occupations: accountants, administrative assistants/specialists, and sales representatives. These occupations were chosen because they differ quite extensively in the degree to which their incumbents typically interact with firms' customers. The diversity of occupations allows us to explore whether firms might discriminate in response to perceived customer preferences.

13
We submitted applications in response to job ads posted for six cities that vary in terms of the size of their populations, geographic locations, prosperity, and ethnic composition. Nanjing, a historically prominent and prosperous eastern seaboard city located in the lower reaches of the Yangtze River, is the commercial, industrial, and political/administrative capital of Jiangsu Province. Shenzhen is similarly located in a prosperous coastal region, but in the southern province of Guangdong on the Pearl River Delta. Shenzhen, a quiet backwater prior to its 1979 development as a special economic zone, has experienced extremely rapid growth and the in-migration of many millions of residents from many parts China. Chengdu and Kunming were chosen for this study due to their location in less prosperous western provinces. Chengdu, located in the heart of an agricultural, densely populated, To focus on ethnicity, we limited the range and scope of the resumes we created.
Each of our resumes represented a 24--year old, single, currently employed, university--educated woman. Each resume was designed to be realistic in terms of job experience, education, certificates and training, and other personal information.
The content of our fictitious resumes was based on observation of a very large number of resumes for the same types of occupations at the same levels of experience that were posted on a competing Internet job board, chinahr.com. The descriptions of tasks carried out by an applicant on her first and/or second job after college graduation were taken from these real resumes and appropriately edited and reassigned to our fictional resumes. The company names associated with our applicants' work history were altered to represent real companies operating in each of our target cities. After the resumes were built, they were randomly assigned a Han, Mongolian, Uighur, or Tibetan name. 8 Each resume was assigned an email address, mobile phone number, ID number, and regular home address.
Four versions of each person's resume were created that varied along two dimensions. In two of the four, the candidate had only one job/one employer since college graduation, and in the other two versions the candidate had already worked for two different companies. The variation in work history allows us to examine whether employers prefer candidates whose resumes seem to demonstrate stability (just one previous job) to those who might seem more ambitious/driven (those with two previous jobs and a history of increasing job responsibilities). For each of these two types of work history (one or two jobs), we created two types of educational experience-one in which the candidate was educated at a local university and one in which the candidate was educated at a similarly ranked university in another province. We wanted to eliminate, or at least mitigate, potential employer concern about language issues and/or minority language education, and thus ensured that each of our job candidate's university education took place either in the same city 8 Although in China, an individual's official ethnicity is recorded on his/her identity card, we did not explicitly state our candidate's official ethnicity on resumes since 1) there were no fill--in blanks or check boxes for this purpose on the up--loadable forms and 2) we did not find any examples of candidates stating their ethnicity in the sample of real resumes we examined.

15
where the job was posted 9 or in a province not known as, nor thought of, an ethnic minority area/minority autonomous region. We paid close attention to university rankings to ensure that each fictitious candidate within an occupational category had an education that would be viewed by employers to be of comparable quality.
We attempted to create sets of resumes that were equivalent in terms of the applicants' productive characteristics. The registration process for the 51job.com job board allowed us to upload the four different versions of each fictitious person's resume under one name with one email address and one mobile phone contact number.
Given China's hukou (residential registration) system, which limits where people may live and work, we worried that employers might be concerned about whether applicants with ethnic minority names had the right to live and work in the areas for which the jobs were posted. That is, we feared that a firm might pass over an application from a minority candidate based on presumptions about her hukou status if the resume failed to indicate that the applicant was very likely to be a local resident or had in some way already overcome any potential hukou limitations. To mitigate this concern, we designed our applicants' resumes such that each of the applicants already had several years of work history in the same city as that of the company's job posting. In half of our cases, the applicants had, additionally, attended a local university.

Application Process
For a two--month period from mid--June to mid--August, we submitted a pair of applications for each suitable job posting in each of our three occupations in each of our six locations. By "suitable," we simply mean that our candidates' characteristics were a good match for the advertised position, that is, they appeared well qualified for the positions. For the cities of Chengdu, Kunming, Nanjing, and Shenzhen, we submitted one Han and one randomly chosen minority for each advertised position. 9 For job postings in the city of Shenzhen, we counted education obtained in the nearby city of Guangzhou as local.
Once the ethnicity composition of the pair of applications was determined, we randomly chose which of the four versions of her resume was to be submitted for each applicant. The two resumes were submitted on the same day, usually within an hour or so of each other. For the city of Urumqi, we submitted one Han and one Uighur application and similarly for the city of Hohhot, we submitted one Han and one Mongolian application. The order of submission was randomly determined and tracked. Resume versions were also carefully tracked. As mentioned above, we submitted close to 22,000 applications over the course of the summer. However, Internet job boards were not as widely used in Urumqi and Hohhot as in our other cities----few jobs ads fitting our candidates' criteria were posted in these locations.
We were therefore only able to submit approximately 800 applications for the roughly 400 job postings in ethnic minority autonomous regions.
We also collected as much information about the firm advertising each position as seemed feasible, given the scale of our experiment. We recorded each firm's name and, if available, the size of the firm (measured in terms of the number of its employees), and the type of firm ownership. The vast majority of firms that pursued/contacted our candidates did so by calling the mobile phones associated with each of our fictitious candidates. A very small number responded by email and an even smaller number responded by means of sending a text message to one of the mobile phones. We registered and recorded as a callback any action by a firm that indicated they were interested in the candidate and wanted to follow up with an interview or further contact of one sort or another. The student research assistants answering the calls were trained to immediately inform callers that the candidate had just accepted another position and was no longer interested in that firm's job opening. They responded in like manner to emails and followed up appropriately, usually with a phone call, to text messages.

Analysis of Results from the Perspective of Individual Applicants
There are large and statistically significant differences in the rates of interview callbacks received by applicants with Han--, Mongolian--, Uighur--, and Tibetan--sounding names. 10 1,389 of our 21,592 candidates received callbacks, yielding a callback rate of 6.43%. That is, on average our candidates were contacted to set up interviews (or carry out the next step(s) in the hiring process) by six or seven firms for each hundred applications that they submitted. The callback rates varied by candidate ethnicity from a high of 8.15% for those with typical Han names to a low of 3.69% for those with typical Tibetan names.
(Insert Table 2 here.) The bottom row of Table 2 expresses the callback rates of those with Han--sounding names as a ratio relative to those with minority--sounding names. What these ratios reveal is that an equally qualified woman with a Mongolian--sounding name would need to put in 36% more applications than a woman with a Han--sounding name to get the same number of callbacks. Women with Uighur--and Tibetan--sounding names would need to put in 83% and 121% more applications, respectively than women with a Han--sounding names in order to get the same number of callbacks for interviews. There appears to be a significant degree of discrimination (on the part of firms that participate in the Internet job board labor market) against women with ethnic minority names (names that are clearly non--Han), relative to those with Han names. Of course, the resume audit methodology can only detect discrimination in one stage of the hiring process, but this early stage is the critical one of reviewing resumes in order to choose candidates for interviews.
The data in Table 3, indicate how the callback rates vary by ethnicity (as indicated by ethnically--differentiated names) across occupations. As discussed above, we expect that occupation may come into play as an explanatory variable in a couple of 10 Hasmath (2011) argues that a fairly large share of urban jobs in China are obtained through network searches and that there is a higher propensity for ethnic minority disadvantages when jobs are obtained through such channels. If so, the results presented here may rather represent a lower bound for the degree of discrimination faced by job candidates of these ethnic groups. different ways. First, firms may treat candidates with ethnically--differentiated names differently if they believe their own customers have discriminatory preferences. If profit--maximizing firms try to accommodate their customers' (discriminatory) tastes, we'd expect that minority candidates in occupations that have more exposure/more interactions with the firm's customers are more discriminated against than those in occupations with low degrees of direct interaction with the firm's customers. We expect that sales representatives are likely to interact more with a firm's customers than administrative assistant/specialists and, in turn, that adminstrative assistant/specialists are likely to interact more with a firm's customers than accountants. Thus, we might expect the callbacks ratios for those applying for positions as sales representatives to vary more by ethnicity than those applying for accounting positions.
Second, it might be the case that the degree of slack or tightness in labor markets varies by occupation and that this will also affect the differences in callback rates (as oposed to callback ratios) by ethnicity across occupations. In tight labor markets firms could not as easily indulge either their own or their customers' discriminatory preferences (Bursell, 2007). We can get a feel for the tightness of the labor markets by the overall callback rates by occupation.
(Insert Table 3 here.) Similarly, the population of the Inner Mongolian Autonomous Region is predominantly Han, that is only 21% of its population is classified as minority, and 20 the population of its capital Hohhot is only 13% minority. It seems that both differences in the ethnic composition of the cities in our study plus differences in local cultural norms could lead to differences in the treatment of ethnic minorities in the labor market by location. Callback rates and ratios by city are presented in Table   4, below.
In Chengdu, Kunming, Nanjing, and Shenzhen, our application procedure dictated that for each opening we submit pairs of applicants that consisted of one Han  Table 4 here.)

Multivariate Probit Analysis -Individual Experience
In Tables 2 through 4 above, we have sequentially discussed some of the factors that influence the callback rates received by candidates who are distinguished from each other only by whether their names sound as if they belong to one of China's more prominent ethnic minority groups (Mongolian, Uighur, or Tibetan) or to the Han majority. In this section of the paper, we explore how these factors interact. We estimate a probit regression with the dependent variable indicating whether or not a candidate received a callback for an interview. The independent variables include controls for a candidate's ethnicity, occupation, and location. 12 We also created 11 We were unable to tell from the job postings whether the firms to which we submitted applications were minority owned. We were also unable to determine anything about the ethnic status of the human resource professionals who screened the resumes.
control variables to track resume characteristics such as whether the candidate had one or two previous jobs in the years since graduating from university and whether the candidate had been university educated in the same city as the job posting or in another province.
The results in Table 5a  and Urumqi in combination with the small number of job postings in these two cities left several cells either empty or with too few responses to be able to calculate marginal effects. We estimated a model with a full set of interactions of the ethnicity, occupation, and location variables for the subset of the data that excluded the 810 observations from Hohhot and Urumqi. We found the marginal effects to be virtually identical to those presented in Table 5a below. (The coefficients that differed from those presented above differed only in the third decimal place. See Appendix Table 1a) 23 The marginal effects presented in Table 5a reveal that firms had clear preferences on resume characteristics. All else equal, the callback rate for those who had just one previous job was 1 percentage point higher than those who had two previous jobs and were applying for their third. (Recall that all of our candidates are 24 years old and had graduated from university approximately 2 to 3 years before applying for the posted positions.) Similarly, firms seemed to value local education. All else equal, those who attended university in the same city as the job posting received callbacks rates almost 1 percentage point higher than those educated at universities in other provinces. These results are interesting in and of themselves and they also provide evidence that the reviewers are carefully reading resumes and not making random calls. This evidence that resume screeners are carefully reading resumes adds more weight to the findings of discrimination on ethnic sounding names.
There are also significant differences by both occupation and location, suggesting that the labor market conditions differ substantially across these dimensions. In the summer of 2011, the callback rate for accounting positions was over 2 percentage points lower than that of sales representative while that for administrative assistant/specialists was almost 2 percentage points higher. The labor market conditions in Kunming seemed to facilitate the chances of getting interview callbacks. The callback rate in Urumqi was quite low; perhaps, given the paucity of job postings there, Internet job boards were not yet in common use and/or when used, not fully trusted as a good source of job candidates. Of course, the low callback rate may also simply reflect local labor market conditions.
We explore the effects of location further by interacting controls for the cities in autonomous minority areas (Urumqi and Hohhot) with controls for their own minorities (Uighurs in Urumqi and Mongolian in Hohhot). (See Table 5c imply that all firms discriminate. We turn now to an analysis of our firm data. (Insert Table 5c here.)

Results----Analysis of Firms Making Callbacks
Our job candidates received a total of 1,389 interview callbacks. 944 different firms made these callbacks. Some of our firms contacted only candidates with Han names, some contacted only candidates with minority names, and some contacted both. In this section of the paper, we examine the characteristics of firms that appear to treat candidates of differing ethnicity equally and those that seem to favor those of one ethnicity over another. We have organized our results according to two sets of firm characteristics: firm ownership (Tables 6a and 6b) and firm size (Tables 7a and 7b) and by the ethnicities of the pair of candidates submitted to each firm (Table 8a and 8b). In each of these sets of tables, the a--table contains the results for all 944 firms making at least one callback. The b--tables repeat the analysis for firms located only in Urumqi and Hohhot, the two autonomous minority areas.
(Insert Tables 6a and 6b here.) The first row of In Table 6b, we see that firms located in minority areas are much more likely to callback both candidates than is the case for the whole sample of firms (61% vs. Table 6a). We also observe a higher share of firms making callbacks to minority only candidates (23% vs. 7%) and a lower share of firms making callbacks to Han only candidates (16% vs. 46%). Firms in minority areas are most likely to callback both candidates, but when they make only one call, it is more likely to be to a candidate with a minority name than a candidate with a Han name (23% vs. 16%).

47% in
The sample of firms making callbacks in minority areas is too small to make any definitive statements on differentiation by ownership type In the next pair of tables, we explore the hypothesis that firm size may be related to firms' decisions about whom to call back. Employers may choose to hire candidates of one ethnicity over another because they prefer to have employees that are culturally similar to them. That is, firms may be making choices that are based simply on employer tastes or preferences and that are not related to firm productivity (Bursell, 2007). If so, it is possible that we would observe less discrimination in large companies than small ones. That is, it may be the case that social interactions between owners/managers and their employees are less important in large companies than small ones.
(Insert Tables 7a and 7b here.) In Table 7a, a mixed picture emerges. While the share of small firms' callbacks, those with 50 or fewer employees, going to minority only candidates (10.8%) is significantly higher than the shares of callbacks to minority only candidates at medium (5.2%) and large firms (4.6%), the share of small firms' callbacks to both candidates (42.9%) is significantly lower than the share of callbacks to both candidates at medium--sized firms (50.1%). Although small companies seem slightly less likely than large companies to discriminate against ethnic minority candidates in the sense of having higher callback shares for Han only candidates (47% vs. 49%) and slightly more likely to callback Han only candidates than medium--sized companies (47% vs. 45%), these differences are not statistically significant. When we separate out the callbacks made by firms in minority autonomous regions, we see that the share of callbacks to both candidates by firms of all sizes is 50% or greater. (See Table 7b.) We also observe that shares of callback by firms in minority areas to minority only candidates is as least as high as the share of callbacks going to Han only candidates. Firms in minority areas appear to treat minority candidates equitably. (Insert Tables 8a and 8b here.) In Table 8a, we see that the ethnic composition of candidate pairs matters.
Mongolian--Han candidate pairs have a lower share of Han only callbacks than do Uighur--Han or Tibetan--Han pairs. In addition, Mongolian--Han pairs have higher shares of callbacks for minority only and both called than do Uighur--Han or Tibetan--Han pairs of applicants. In Table 8b, we see that firms in Hohhot, which received only Mongolian--Han pairs, were highly likely to contact both their Han and Mongolian candidates for interviews. Firms in Urumqi, which received only Uighur--Han pairs, seemed quite inclined to contact only Uighur candidates for interviews.
Of course, the number of firms in minority areas making callbacks is very small.
Therefore, the results for the autonomous regions should be viewed as suggestive only.

Multinomial Probit ----Analysis of Firms That Make Callbacks
In the above analysis, we've examined tables of firm callback patterns by firm ownership, firm size, and the ethnic composition of candidate pairs. In each case there are some significant factors that help to explain differences in firm behavior with respect to choosing which candidates to callback for interviews. We now combine these factors in a multinomial probit regression. The dependent variable is a categorical variable that indicates whether firms call back only candidates with These characteristics were shown to be preferred in the individual level analysis.
In the multinomial probit, we use equal treatment as the base outcome, that is, that firms callback both candidates for interviews. Table 9 consists of three panels that reveal the marginal effects of our control variables on the predicted outcomes: a) that firms make callbacks only to Han candidates; b) that firms make callbacks only to minority candidates; or c) that firms make callbacks to both candidates. 14 What factors help to explain whether a firm will call back only the Han candidate from the pair of applications it received? Table 9 Panel--a reveals that all else equal, firms that receive Han--Tibetan pairs are 8.4 percentage points more likely to call only the Han candidate than firms that receive applications from Han--Mongolian pairs. In addition, it appears that human resource managers are paying attention to 13 The base case was that the Han and minority candidates had equally desirable or equally undesirable characteristics. For example, both were educated locally or both were educated out of province or both had just one previous job or both had two previous jobs.
14 The marginal effect of each control variable in a) and b) sum up to the marginal effect of the same variable in c). Our analysis at both the individual and the firm levels clearly indicates that human resource managers screening resumes are paying attention to resume characteristics. For these positions, advertised to those with 1 to 3 years work experience, they clearly prefer candidates who have not previously changed jobs.