Stereotypes of appearance, non-cognitive characteristics and labor market chances

Transcription

1 Stereotypes of appearance, non-cognitive characteristics and labor market chances Mahmood Arai, Marie Gartell, Magnus Rödin and Gülay Özcan January 27, 2016 Abstract Using an experimental setup involving 436 case workers at Swedish Public Employment Services (SPES) as subjects and the profile pictures and recorded voices of 75 jobseekers at the employment offices as treatments, we report results indicating that perception of stereotypical Swedish appearance is highly correlated to non-cognitive attributes such as being orderly, mature, ambitious, agreeable and attractive. Case workers tend to favor job seekers with a foreign accent when choosing candidates to recommend for labor market programs (LMP). Furthermore, male case workers act in favor of those perceived to have stereotypical Swedish appearance. This bias represents roughly a 50 per cent higher chance of being selected when comparing the candidate with the highest stereotypical Swedish looking score (8/10) and the candidate who has the lowest stereotypical Swedish looking score (3/10) in our sample. 1 Introduction Non-cognitive attributes such as attractiveness and agreeableness are reported to matter for perceived ability and labor market achievements. 1 Stockholm University. Swedish Public Employment Services (SPES). The experiments were performed while Marie Gartell was at SPES. Gartell s current affiliation is Swedish National Audit Office. Financial support by IFAU is gratefully acknowledged. We are grateful to seminar participants at at IFAU, Linnaeus university, and BFH Conference on Discrimination in the Labor Market for helpful comments. 1 See Hamermesh & Biddle (1994), Jackson et al. (1995), Averett & Korenman (1996), Heckman (1999), Nyhus & Pons (2005), Borghans et al. (2006), Mobius & Rosenblat (2006), Rooth (2009), Mocan & Tekin (2010), Ruffle & Shtudiner (2010), Rödin & Özcan (2011), López Bóo et al. (2013) and Dechter (2015). 1

2 These attributes as well as many others can, however, be perceived differently depending on the social and ethnic belonging of interacting agents and whether a person is considered to be an in-group or an out-group member. 2 Group belonging signaled by a person s name, look or speech can trigger intergroup stereotypes activation. There is a vast literature in social psychology that deal with stereotyping, prejudice and discrimination (See Katz and Braly (1933) for early work on racial and ethnic stereotypes and Dovidio et al for an overview of the literature). Results from laboratory experiments indicate that ethnic stereotypes imply that individuals might be associated with different levels of trust or cooperation (Fershtman and Gneezy 2001; Castillo and Petrie 2010; Ahmed 2010; Rooth 2010). Laboratory, field and quasi-experiments have mainly focused on names as signal of group affiliation and associated beliefs about group characteristics. 3 We use pictures, recorded voices and real names of job seekers to signal perceived group affiliation. The main aim of this study is to examine how non-cognitive attributes associated with a person vary with perceived group belonging indicated by beliefs about stereotypical Swedish look affect choice probabilities when agents are involved in process of selecting between a pair of candidates. We focus on job seekers at the Swedish Public Employment Service (SPES) offices. Jobseekers might be treated differently at the SPES offices and in the labor market in general because of their ethnic background. We examine the two following questions. (i) Are productivity-related noncognitive characteristics correlated to probability of being perceived as having a stereotypical Swedish look? (ii) Do persons with identical relevant cognitive characteristics face different chances of being selected into labor market programs (LMP) depending on their look and non-cognitive characteristics perceived to be embedded in their pictures? We ran a computerized experiment involving 436 case workers at SPES as subjects. We recruited 75 jobseekers at the employment offices, took their profile pictures and recorded their voices. The experiment has two sequences. In the first sequence involving 160 case workers, we measure scores regarding various attributes associated with the pictures or the recorded voices of the job seekers. Case workers were asked to guess how these portrait pictures and recorded voices were scored by employers. Guesses that matched the scores reported by employers who had previously assigned 2 See e.g Tajfel and Turner (1979) for social identity theory. In economic theory, such differential treatment is usually in terms of employers (or coworker/customers) taste (Becker 1957) or differences in group statistics (Arrow, 1973; Phelps, 1972) 3 See Riach & Rich (2002), Bertrand & Mullainathan (2003)). 2

3 scores to all pictures and voices were rewarded with lottery tickets. Case workers scores yield us a measurement of stereotype beliefs about attributes that might be perceived to be embedded in these portrait pictures and recorded voices. We constructed 75 jobseeker profiles with these pictures and recorded voices in accordance with the layout template that is used for registered job seekers at SPES. These profiles also include randomized information about education level, previous occupations and unemployment spells. In the second sequence, job seekers ability to successfully complete a LMP and chances of success after completing a program were measured using evaluations by 260 case workers that had not participated in the first sequence. Case workers were asked to guess how their colleagues had previously assigned scores to these profiles. Finally, they were asked to guess which candidates their colleagues had chosen to recommend for a LMP when choosing between two profiles, when restricted to choosing only one out of two candidates for a LMP. Guesses were rewarded if they matched the results obtained in our pilot study. Our results indicate that there is a clear bias in evaluating non-cognitive characteristics in favor of those whose physical appearance corresponds to the stereotypes of Swedish look. Using data of case workers perception of how employers would score job seekers portrait pictures, we find that stereotypical Swedish look is correlated to our measured non-cognitive attributes: Orderly (ρ = 0.66), Mature (ρ = 0.66), Ambitious (ρ = 64), Agreeable (ρ = 0.51) and Masculine/Feminine (ρ = 0.41). The weakest correlation is found for Attractive (ρ = 0.14). Stereotypical Swedish look is also correlated to being perceived as Educated (ρ = 0.63). Similar pattern, though with a much weaker correlations, is observed for records voices. Having no foreign (non-swedish) accent is positively correlated to attributes such as Agreeable Voice, Extrovert, Motivated, Secure and Smart. We find that that those who have foreign accent have higher chances of being assigned to a labor market program. Men tend to favor job seekers that more closely resemble to stereotypes of Swedish look. Female case workers favor those with higher scores on stereotypical Swedish look only when they make judgements about attributes associated with profile pictures but do not care about look when when choosing between competing candidates for an employment program. The general policy of the SPES aims at giving priority to job seekers that have the lowest chances of finding employment. This implies that those who are perceived to have foreign accent might be associated with lower job chances due to observed higher risks of unemployment for foreign born. 3

4 SPES has, however, no policy about risk of bias associated with stereotypes of physical appearance. In absence of such policy/awareness, there is a room for discrimination in favoring job seekers that more closely resemble to stereotypes of Swedish look. The remaining of the paper is organized as follows. Next section describes the experimental setup. Section 3. discusses the determinants of the choice problem when facing two candidates and have to choose one candidate to recommend for a labor market program. The data generated in the experiment is described in Section 4. Section 5. presents results of estimating the impact of stereotypical look and other attributes of candidates on the probability of being admitted to a labor market program. Finally Section 6. summarizes the paper. 2 The Experiment Design We limit our attention to two types of employment training; nursing education and warehousing and logistics with the forklift driver license training. In this way we study a male-dominated and a female-dominated occupation. Job seekers We recruited 75 unemployed job seekers at SPES offices. Only male job-seekers were recruited for the warehouse program and only female jobseekers for the nursing program. We took portrait pictures of each job seeker. These portrait pictures were later harmonized in light, contrast etc to avoid variation in technical details of the picture quality. We asked job seekers to read 8 short messages saying that they ask to be admitted into a LMP. Sequence I: The classification game for measuring stereotype beliefs about job seekers In the first sequence, we organize a classification game in order to measure stereotype beliefs about attributes that might be perceived to be embedded in these portrait pictures and recorded voices. To prepare for this game, we asked two employers to assign values (1,2,...,10) in various dimensions for how they think employers in general would perceive these portrait pictures and recorded voices. A classification game was then organized engaging SPES case workers as subjects where they were asked to guess how these portrait pictures and recorded voices are scored by employers. A 4

5 correct guess was rewarded with lottery tickets. The case workers listen to messages that are randomly drawn from a pool of 8 different prewritten messages for each occupation. The classification game yields around 30 observations on each of 75 portrait pictures and recorded voices for 9 picture attributes and 8 voice attributes. The picture attributes were: Trustworthy, Orderly, Masculine/Feminine, Mature, Educated, Ambitious, Agreeable, Attractive, Stereotype Swedish Look and Age, and the attributes for recorded voice were Agreeable Voice, Extrovert, Motivated, Insecure, Tired, Sounds smart, Grown up in Stockholm and Grown up abroad. Constructing Job seekers profiles Portrait pictures and job seekers real names were then used to construct profiles for all our 75 job seekers. In constructing the profiles we randomly assign education level (completed / not completed high school), date of registration at the SPES office indicating short or long latest unemployment spell and previous work experience (6 different typical low-skilled occupations). A job seeker is then represented with these randomly assigned characteristics together with portrait pictures, name, recorded voice message and a date of birth based on average age associated with each picture from the classification game. Sequence II: The labor market program game In the second sequence we are first interested in how case workers would assign scores to a profile in the following three dimensions: i) ability to successfully complete a program, ii) chances of success in the labor market after completed program and iii) the need to participate in a program. The last measure is added because the case workers are supposed to prioritize job seekers with highest needs for LMPs. We use our 75 profiles to run a LMP game in two steps. In the first step workers are asked to guess how their colleagues had assigned scores to profiles in the three above mentioned dimensions. We used the assigned scores from our pilot as correct answers to reward the case workers. Correct guesses are rewarded with lottery tickets. Case workers considered 8 profiles, one at a time, that are randomly drown from our pool of 75 profiles. Note that no case worker saw the picture of a job seeker or listened to a voice message more than once. 5

6 Choice probabilities associated with job seeker profiles Finally, in the second step of the second sequence, we are interested to know how case workers choose between two profiles if they are restricted to recommend one job seeker for a LMP but not the other one. Due to limited number of LMP slots, such situations are quite common in case workers everyday practice. This gives us choice probabilities associated with each profile. The case workers are asked to guess who had the highest probability of being selected by their colleagues when comparing profiles in 4 pairs of profiles that are randomly drawn from the pool of our 75 candidates not drawn in the first step. This is repeated four times involving in total 4 pairs of profiles. As before correct guesses are rewarded with lottery tickets. 3 Probability of being selected in competition This setup allows us to use data from the classification game in the first sequence of our experiment, to estimate the probability of a profile to be selected for a LMP in competition with other profiles as a function of the profile characteristics. We can estimate the effect of perceived stereotypical Swedish look while we can control for not having foreign accent, and having a Swedish-sounding name, ability to successfully complete a LMP and chances of labor market success after completed program associated with the profiles. Notice that education as well as work and unemployment experience are randomly assigned to the profiles. Given that various attributes of a profile can be highly correlated, we focus on perceived stereotypical Swedish look and then examine the effects of embedded attributes by replacing the variable for stereotypical Swedish look with the variables measuring other attributes to assess what is seen in a look. The equation of interest is then as follows: y (i j) = α + δ 1 ij S LOOK + δ 2 ij S SP EECH + δ 3 ij S NAME + γ ij X + ε Where y is a dummy variable measuring if a profile i is selected when competing with another profile j for participating in a labor market program (LMP) or not. Data for y are from the second step in the second sequence of our experiment where case workers are instructed to choose one candidate when comparing pairs (i, j) of profiles. The pairs are draws from the set of profiles that are not drawn and shown previously to the case worker. The variable ij S measures between pair difference in average scores of 6

7 stereotypical Swedishness of the look, scores of not having foreign accent generated in the classification game in the first sequence of the experiment and whether the name is Swedish sounding or not. 4 The variable ij X is a vector of between pair difference in average scores of perceived ability of completing a LPM and average score of chances of success after completing the LMP. Data for X are generated in the first step in the LMP-game of the experiment. We do not have to include other characteristics of a profile since they are all randomly assigned to the portrait pictures and recorded voices. In such a setup our estimate of δ 1 will identify the effect of physical appearance on probability of being selected. Then we replace this variable with each of non-cognitive scores associated with the profile portrait pictures and recorded voices to extract information about what is seen in a look and how various non-cognitive attributes influence the choice probability of a profile through observing a picture in a profile. 4 Description of the data We had 436 case workers who participated in different sequences of the experiment. In the first sequence, 99 case workers classified voices and 61 case workers classified pictures. In the second sequence we had 276 case workers. Due to answer alternative I do not know for scores associated with pictures and voices we had less than one percent missing values for portrait pictures and 1.7 percent for voices. For scores related to ability to complete a LMP, we had two missing value and for chances of success in the labor market after completed program, we had 7 missing values due to the answer alternative I do not know. These missing values are so few and therefore neglected when computing average scores. Since we examine gender differences in behavior, we need to know if the case worker was a Woman or a Man. The information for this variable is extracted from the survey to the case workers after the experiment. We miss the response on sex for one case worker. The observations for this case worker are deleted implying that we lose four of 1104 observations in evaluation and program decision part. The estimations in the second sequence have 1100(= 275 4) observations including 304 observations for 4 Whether names would be considered as Swedish sounding or not is assigned subjectively and independently by all 4 of us. A name is regarded here as Swedish sounding if at least 2 of 4 considered it as Swedish sounding. In 16 percent of cases 2 of 4 classified the name as Swedish sounding. This classification yields 43/75 as Swedish sounding names. 7

8 76 male case workers and 796 observations for 199 female case workers. Figure 1. & Figure 2. give the distribution of scores for various attributes associated with profile pictures and recorded voices. Table 1. & Table 2. give the correlation for picture attributes and voice attributes. The correlations are listed under the diagonal and the p values for zero correlation are found above the diagonal. Results indicate that stereotypical Swedish look is strongly correlated with all attributes that are usually perceived as being positively correlated with productivity. The weakest correlations are found for Age and Attractive. Similar pattern though with much weaker correlations are observed for voice attributes. 5 Table 3. gives the main characteristics of the case workers in the two sequences regarding classification game and the LMP game related part in the second sequence. We see that men in the program decision sequence are slightly older, have slightly longer seniority and have in a lesser degree Swedish as mother tongue. The differences are not large but we will check whether these differences have any effects on our results. Our main assumption is that case workers in the second sequence have similar perceptions of how job seekers are perceived by employers as those who classified the portrait pictures in the first sequence of the experiment. We also checked job assignments of the case workers (no shown here) in the two sequences and the pattern is quite similar for both sequences. Figure 3. depicts the distribution of the evaluation scores related to ability to successfully complete a LMP, the chances of success in finding a job after completing the program and the need for a LMP associated with job seekers profile. This implies that there is a variation in how profiles are perceived. The question is what explains this variation. Notice that the profiles picture are randomly matched with information about duration of last unemployment spell, previous occupation and having a high school degree. To describe what attributes are correlated to these perceived ability and success scores we run regressions explaining each of these variable each with picture and voice attributes. We do not need to include other characteristics of the profiles since they are randomly assigned and thus orthogonal to the attributes. Results in Table 4. indicate that compared with male case workers, female case workers, are more generous in assigning scores to female job seekers. Moreover, case workers consider the ability of those who have no 5 We checked the correlation between scores for perceived to be born abroad based on the recorded voice and the scores for the stereotypical look. There is a negative correlation ( 0.08) but the p value is as high as This means that we can separate the effect of these two characteristics on our measured outcomes. 8

9 accent to be higher than those who have foreign accents. Having a foreign accent might signal lower level of Swedish language proficiency but can also be associated to a non-swedish ethnic belonging. Regarding our main variable of interest, male case workers tend to assign higher ability score to those with stereotypical Swedish look when it comes to women applying for nursing program. Female case workers do this for male job seekers. There is a cross-gender positive bias in estimating ability of individuals with stereotypical Swedish look. To better understand how scores for ability to complete LMPs are related to other profile attributes, we estimate separately for female and male, a model replacing the score for stereotypical look with other measured attributes, one at a time, controlling for accent and foreign sounding name. This approach is chosen due to very high correlations for various attributes implying that we will not be able to separate the effect of individual attributes in a meaningful way. However, we can measure an aggregate effect of physical appearance using an additive index of scores of all the measured attributes. Results of these estimations for various attributes for male and female case workers are presented in Table 5. Let us first look at the results for the female case workers presented in panel (a) and (b). These results indicate that ability of completing the program for forklift driving which are all men are estimated to be higher for those with stereotypical Swedish look. The same is observed for Agreeableness and Trustworthiness and Ambitious. For assistant nurses, a non-cognitive additive index seem to have stable positive effect for perceived ability of profiles that aim at assistant nurse occupation. The same is true if the candidate is perceived to be attractive. Female case workers seem to assign lower ability score to men who are less stereotypically Swedish looking but seem to put stress on other attributes when it comes to female job seekers. Results for male case workers are reported in panels (c) and (d) in Table 5. These results do not give much insight about how male case workers estimate the ability of male job seekers. However, male case workers seem to be sensitive to almost all of attributes of female job seekers except attractiveness when estimating their ability of completing the assistant nursing program. The results for probability of success in the labor market presented in Table 6. have a similar message as in the case of estimated ability score. Using the need for LMP gives very little insight in how the scores of need is related to profile attributes. To sum up, our results indicate that there is a cross gender pattern of 9

10 overstating the ability of completing a program and probability of labor market success after completing the program for those who have a stereotypical Swedish look. It should be noted that while the probability of success in the labor market might partly reflect case workers anticipated discrimination based on stereotypical Swedish look, the perceived ability of successfully completing the LMP should not involve expectations about labor market discrimination. After this description of data and the correlation patterns in attributes and estimated ability and success measures, in the next section we examine the effect of profile picture attributes and voice attributes on probability of being recommended for a LMP. 5 Impact of profile attributes on probability of being selected Results reported in Table 8. indicate that female case workers choose those who have foreign accents when choosing between two candidates. Female case workers are not sensitive to other attributes given the program completion ability and probability of labor market success. Male case workers are sensitive to accent in a similar way as female case workers, but they also put a weight on look. Results clearly indicate that though there is a cross gender favoring of job seekers with stereotypical Swedish look in evaluation of ability and chances of success, it is only men that prefer those with Stereotypical Swedish look. This is the case both when they choose between two male candidates or between two female candidates. This represents roughly 50 per cent higher chance of being selected for the most stereotypical Swedish looking (8/10) candidate as compare to the candidate who has the lowest stereotypical Swedish look score (3/10) in our sample. Furthermore, we include case worker characteristics, age, seniority and Swedish as mother tongue in our specifications in Table 9 and find that results are robust for including these characteristics and none of these characteristics are systematically correlated to the choice decisions. 6 In order to further examine what the case workers see in the stereotypic Swedish look, we estimated the probability of being selected to participate in a LMP as a function of differences in scores of various profile picture characteristics controlling for accent and perceived ability. Inspecting the 6 We cannot run regressions for male and female case workers separately as the number of case workers with other languages than Swedish is small for male case workers (less than 10). 10

11 results in Table 10 we find that, male case workers positive appreciation of stereotypical Swedish look for female candidates is not related to any other profile picture attribute. For male candidates, however, many other positive profile picture attributes exhibit similar effect on the choice probability as the stereotypical Swedish look. Men see productivity related attributes in stereotypical Swedish look when they face men but when they face women candidates, they do not see much in the female candidates look, except its stereotypical Swedishness. We do similar estimations as in Table 10 focusing on voice attributes instead of look. Results are reported in Table 11. The probability of being assigned to labor market program is lower for those who have not foreign accent for both male and female case workers. Results indicate that female case workers are sensitive to male candidates voice attributes in a compensating manner. Male case workers seem to react only to one attribute of the female candidates voice. They favor female candidates with agreeable voice. No other effects are found for the male case workers data. 6 Summary Using data of case workers perception of how employers would score job seekers portrait pictures, stereotypical Swedish look is correlated to our measured non-cognitive attributes: Orderly (ρ = 0.66), Mature (ρ = 0.66), Ambitious (ρ = 64), Agreeable (ρ = 0.51), Masculine/Feminine (ρ = 0.41) and Attractive (ρ = 0.14). Stereotypical Swedish look is also correlated to being perceived as Educated (ρ = 0.63). This means that being perceived to have stereotypical Swedish look is positively associated with many attributes that employers value on the labor market. Moreover, having no accent is correlated with higher perceived ability of completing a LMP. Having a foreign (non-swedish) accent is a signal of being foreign born. Case workers tend to favor job seekers with foreign accent when choosing candidates to recommend for LMP. This is in line with the official policies of the Swedish Public Employment Service of giving priority to foreign born. Male case workers are positively sensitive to stereotypical Swedish look when choosing between two candidates. This bias represents roughly 50 per cent higher chance of being selected for the candidate with the highest stereotypical Swedish looking score (8/10) as compared to the candidate who has the lowest stereotypical Swedish look score (3/10) in our sample. Such a differences is similar to call-back differences in corresponding tests using Swedish-sounding and arabic-sounding names on resumés (see e.g. 11

12 Bursell 2007, Carlsson & Rooth, 2007, Arai et al., 2015). These names could possibly send ethnic belonging signals similar to those signals sent by stereotypical view of the physical appearances. Male case workers see productivity-related attributes in stereotypical Swedish look when they face men but when they face women candidates, they do not see anything else than stereotypical Swedishness of the look. Female case workers choice pattern can be interpreted as a compensating behavior when they observe voice attributes of male candidates. Male case workers are on the other hand insensitive to voice attributes except when they favor female candidates with agreeable voice. Case workers in public employment offices have instructions to employ a positive bias in favor of those who are perceived to have lower job chances when offering labor market program. Our results indicate that both men and women succeed to follow these instructions using foreign accent in speech as an indication of being associated with lower job chances. When it comes to look, female case workers do not seem to care about look but male case workers do and favor those perceived to have stereotypeical Swedish look. It is not easy to find an information story consistent with the different behavior of male and female case workers reported above. We interpret these results as an indication of discrimination based on stereotypes of appearances without any relevant information content. 12

13 References Ahmed, Ali, 2009, What is in a surname? The role of ethnicity in economic decision making, Applied Economics, 42 (1): 1-9. Arai, Mahmood, Bursell, Moa and Lena Nekby, 2015, The Reverse Gender Gap in Ethnic Discrimination: Employer Stereotypes of Men and Women with Arabic Names, International Migration Review, DOI: /imre Arrow, Kenneth, 1973, The theory of discrimination, Princeton University Press, 30(2): Averett S. and Korenman S., 1996, The economic reality of beauty myth, Journal of Human Resources 31: Aigner, Dennis. J. and Glen G. Cain, 1977, Statistical theories of discrimination in labor markets, Industrial and Labor Relations Review, Becker, Gary S.,1971, The economics of discrimination, University of Chicago Press. Bertrand Marianne and Sendhil Mullainathan, 2004, Are Emily and Greg More Employable than Lakisha and Jamal? A Field Experiment on Labor Market Discrimination, American Economic Review, 94: Borghans, Lex, Duckworth, Angela L., Heckman, James. J. and ter Weel, Bas, 2006, The Economics and Psychology of Personality Traits, Journal of Human Resources,43: Bursell, Moa, 2007, What s in a name? A field experiment test for the existence of ethnic discrimination in the hiring process, SULCIS Working Paper 2007:7. Carlsson, Magnus and Dan-Olof Rooth, Evidence of ethnic discrimination in the Swedish labor market using experimental data, Labour Economics, Elsevier, 14: Castillo, Marco and Petrie, Ragan, 2010, Discrimination in the lab: Does information trump appearance?, Games and Economic Behavior, Elsevier, vol. 68(1):

14 Dechter, Evgenia K., 2015, Physical appearance and earnings, hair color matters, Labour Economics, 32: Fershtman, Chaim and Uri Gneezy, 2001, Discrimination in a segmented society: An experimental approach, Quarterly Journal of Economics, 116 (1) : Hamermesh, Daniel S. and Biddle, Jeff E., Beauty and the Labor Market., American Economic Review, 84: Jackson Linda A., John E. Hunter and Carole N. Hodge, 1995, Physical Attractiveness and Intellectual Competence: A Meta-Analytic Review, Social Psychology Quarterly, 58: Heckman, James 1999, Policies to forster Human Capital NBER Working paper Katz, D.; Braly, K, 1933, Racial stereotypes of one hundred college students, The Journal of Abnormal and Social Psychology, Vol 28(3): López Bóo Florencia, Rossi Martín A., Urzúa Sergio S., 2013, The labor market return to an attractive face: Evidence from a field experiment, Economics Letters, Elsevier, 118: Mobius, M. M. and Rosenblat, T. S., 2006, Why beauty matters, American Economic Review, 96: Mocan, N. and Tekin, E., 2010, Ugly criminals, The Review of Economics and Statistics, 92: Nyhus, Ellen K. & Pons, Empar, 2005, The effects of personality on earnings, Journal of Economic Psychology, 26(3): Phelps, Edmund. S.,1972, The statistical theory of racism and sexism, The American Economic Review, 62: Riach, Peter A. and Rich Judith, 2002, Field experiments of discrimination in the market place, The Economic Journal 112:

15 Rooth, D., 2009, Obesity, attractiveness and differential treatment in hiring - a field experiment, Journal of Human Resources, 44: Rooth, Dan-Olof, 2010, Automatic associations and discrimination in hiring : Real world evidence, Labour Economics, 17: Rödin, M. and Özcan, G., 2011, It How You Look or Speak That Matters? - An Experimental Study Exploring the Mechanisms of Ethnic Discrimination, Research Papers in Economics, Department of Economics, Stockholm University, 2011(12). Ruffle, B. and Z. Shtudiner, 2014, Are Good- Looking People More Employable? Management Science, 61: Tajfel, Henri and Turner, John C., 1979, An Integrative Theory of Intergroup Conflict. In W. G. Austin & S. Worchel (Eds.), The Social Psychology of Intergroup Relations. Monterey, CA: Brooks-Cole. 15

16 Figure 1: Average scores av 75 profile picture attributes. Stereotypic Swedish look Trustworthy scores Orderly Masculine/Feminine Mature Educated Ambitious Agreeable Attractive Age

17 Figure 2: Average scores av 75 recorded voice attributes. Grown up abroad Agreeable Extrovert Motivated Unsecure Tired Inteligent

18 Table 1: Correlations of the picture attributes. Above-main-diagonal cells represent p values for H0 : r = 0 of corresponding below-main-diagonal cells. Trustworthy Orderly MasculFem Trustworthy Orderly Masculine/Feminine Mature Educated Ambitious Agreeable Attractive Stereotype Swedish look Age Mature Educated Ambitious Agreeable Attractive StypeSwed Age 18

19 Table 2: Correlations of the voice attributes. Above-main-diagonal cells represent p values for H0 : r = 0 of corresponding below-main-diagonal cells. AgreeableVoice Extrovert Agreeable Voice Extrovert Motivated Unsecure Tired Sounds smart Grown upp in Stockholm Grown up abroad Motivated Unsecure Tired Intelligent GrownSt GrownAbroad 19

20 Table 3: Characteristics of case workers participating in the program decision and participating in the classifications. Program Decision Classifications Men Wom Men Wom Age Seniority Mother Tongue Swedish

21 Figure 3: Average evaluation scores av entire profile of candidates. Fig 3a: Case workers evaluation of candidates need for a program score Fig 3b: Case workers evaluation of candidates ability to succesfully finnish a program score Fig 3c: Case workers evaluation of candidates cahnces of success after completion of a program score 21

22 Table 4: Ability, no accent and look. Dependent variable is ability score average. Standard errors in parenthesis. Forklift Driver Assistant Nurse (1) (2) (3) (1) (2) (3) Stereotype Swedish look * 0.262* 0.230* (0.080) (0.075) (0.075) (0.079) (0.078) (0.081) Look X Female Case Worker 0.200* 0.198* 0.198* * * * (0.095) (0.091) (0.092) (0.094) (0.093) (0.092) Female Case Worker * 1.173* 1.169* (0.408) (0.386) (0.391) (0.484) (0.477) (0.475) No accent 0.313* 0.317* 0.371* 0.449* (0.048) (0.111) (0.052) (0.073) Foreign-Sounding Name (0.227) (0.134) NOTE: Swedish look and no accent are measured as mean score and are standardized. Case worker cluster-robust standard errors. p < 0.05 indicated by *. 22

23 Table 5: Ability score evaluated by case workers and standardized profile picture attributes. Dependent variable is ability score average. Each column represents the estimate of the impact of each attribute at the time in where we also control for accent and foreign-sounding name. Stereotype Swedish look Agreeable Trustworthy Educated Orderly Attractive Mature Ambitious Non-cognitive Look Index Panel (a) Female Case Workers Evaluations of Forklift Driver candidates 0.122* 0.106* 0.173* * 0.153* (0.057) (0.051) (0.058) (0.054) (0.055) (0.065) (0.054) (0.055) (0.052) Panel (b) Female Case Workers Evaluations of Assistent Nurse candidates * (0.052) (0.054) (0.046) (0.055) (0.056) (0.068) (0.059) (0.057) (0.063) Panel (c) Male Case Workers Evaluations of Forklift Driver candidates (0.075) (0.075) (0.078) (0.080) (0.073) (0.114) (0.075) (0.077) (0.076) Panel (d) Male Case Workers Evaluations of Assistent Nurse candidates 0.263* 0.412* 0.270* 0.310* 0.388* * 0.371* 0.463* (0.090) (0.112) (0.085) (0.103) (0.101) (0.096) (0.107) (0.102) (0.102) NOTE: Case worker cluster-robust standard errors. p < 0.05 indicated by *. 23

24 Table 6: Chances of success in the labor market, no accent and look. Dependent variable is average score for chances of success after completing a LMP. Standard errors in parantheses. Forklift Driver Assistant Nurse (1) (2) (3) (1) (2) (3) Stereotype Swedish look * 0.195* 0.212* (0.085) (0.083) (0.083) (0.068) (0.070) (0.072) Look X Female Case Worker * * * (0.099) (0.097) (0.098) (0.083) (0.083) (0.083) Female Case Worker * 1.231* 1.233* (0.411) (0.402) (0.409) (0.412) (0.411) (0.410) No accent 0.182* * 0.154* (0.043) (0.104) (0.050) (0.068) Foreign-Sounding Name (0.204) (0.128) NOTE: Swedish look and no accent are measured as mean score and are standardized. Case worker cluster-robust standard errors. p < 0.05 indicated by *. 24

25 Table 7: The degree of need of participating in a LMP, no accent and look. Dependent variable is degree of need score average. Standard errors in parantheses. Forklift Driver Assistant Nurse (1) (2) (3) (1) (2) (3) Stereotype Swedish look (0.089) (0.090) (0.090) (0.094) (0.095) (0.097) Look X Female Case Worker (0.111) (0.111) (0.112) (0.116) (0.117) (0.117) Female Case Worker (0.484) (0.489) (0.493) (0.546) (0.551) (0.549) No accent * * (0.046) (0.138) (0.060) (0.077) Foreign-Sounding Name (0.287) (0.148) NOTE: Swedish look and no accent are measured as mean score and are standardized. Case worker cluster-robust standard errors. p < 0.05 indicated by *. 25

26 Table 8: Probability of being selected to participate in a LMP and the differences in accent and look. Dependent variable measures selected/not selected when compared with another candidate. Standard errors in parantheses. Forklift Driver Assistant Nurse (1) (2) (3) (1) (2) (3) Stereotype Swedish look 0.089* 0.085* 0.080* 0.061* 0.054* 0.057* (0.018) (0.018) (0.019) (0.022) (0.022) (0.022) Look X Female Case Worker * * * * * * (0.022) (0.022) (0.022) (0.026) (0.025) (0.025) Female Case Worker (0.046) (0.045) (0.046) (0.047) (0.046) (0.046) Ability * * (0.050) (0.064) (0.066) (0.047) (0.061) (0.064) Chances of Success (0.060) (0.061) (0.066) (0.063) (0.063) (0.067) No accent * * * (0.007) (0.013) (0.007) (0.009) Foreign-Sounding Name (0.078) (0.042) NOTE: Variables are measured as difference in mean score. Case worker cluster-robust standard errors. p < 0.05 indicated by *. 26

27 Table 9: Probability of being selected to participate in a LMP and differences and look including case worker characteristics. Dependent variable measures selected/not selected when compared with another candidate. Standard errors in parantheses. Forklift Driver Assistant Nurse (1) (2) (3) (1) (2) (3) Stereotype Swedish look 0.087* 0.088* 0.088* 0.084* 0.085* 0.085* (0.023) (0.023) (0.023) (0.029) (0.028) (0.028) Look X Female Case Worker * * * (0.022) (0.022) (0.022) (0.026) (0.026) (0.026) Look X CW Mother Tongue Swedish (0.024) (0.024) (0.024) (0.027) (0.027) (0.027) Female Case Worker (0.046) (0.047) (0.046) (0.047) (0.047) (0.048) CW Mother Tongue Swedish (0.046) (0.046) (0.046) (0.048) (0.048) (0.048) Seniority of Case Worker (0.013) (0.019) (0.013) (0.017) Age of Case Worker (0.002) (0.002) NOTE: Regressions include control for job seekers ability, chances of success, accent and name. Variables regarding job seekers are measured as difference in mean score. Case worker cluster-robust standard errors. p < 0.05 indicated by *. 27

28 Table 10: Look: Probability of being selected to participate in a LMP and the differences in attributes associated the picture of the candidates. Dependent variable measures selected/not selected when compared with another candidate. Each row represents the estimates for an attribute controlling for ability to complete a LMP. Standard errors in parantheses. Stereotype Swedish look Trustworthy Orderly Educated Attractive Agreeable Mature Masculine/Feminine Non-Cognitive Look Index Panel (a) Male case workers choosing Forklift Driver candidates 0.086* 0.093* 0.086* 0.059* * 0.080* 0.119* 0.085* (0.019) (0.027) (0.019) (0.02) (0.016) (0.024) (0.019) (0.033) (0.024) Panel (b) Male case workers choosing Assistent Nurse candidates 0.047* (0.023) (0.019) (0.026) (0.022) (0.014) (0.034) (0.026) (0.023) (0.033) Panel (c) Female case workers choosing Forklift Driver candidates (0.012) (0.017) (0.013) (0.012) (0.010) (0.016) (0.012) (0.019) (0.015) Panel (d) Female case workers choosing Assistent Nurse candidates (0.014) (0.012) (0.016) (0.015) (0.009) (0.022) (0.016) (0.014) (0.021) NOTE: Variables are measured as difference in mean score. Case worker cluster-robust standard errors. p < 0.05 indicated by *. 28

29 Table 11: Voice: Probability of being selected to participate in a LMP and the differences in attributes associated with the recorded voice of the candidates. Dependent variable measures selected/not selected when compared with another candidate. Each row represents the estimates for an attribute one at a time controlling for ability to complete a LMP. Standard errors in parantheses. No accent Motivation Extrovert Agreeable Voice Unsecure Voice Tired Voice Sounds Smart Non-Cognitive Voice Index Panel (a) Male case workers choosing Forklift Driver candidates (0.014) (0.033) (0.028) (0.038) (0.033) (0.026) (0.038) (0.034) Panel (b) Male case workers choosing Assistent Nurse candidates * * (0.015) (0.030) (0.028) (0.036) (0.026) (0.025) (0.044) (0.032) Panel (c) Female case workers choosing Forklift Driver candidates * * * * 0.042* * (0.008) (0.019) (0.016) (0.020) (0.018) (0.014) (0.022) (0.019) Panel (d) Female case workers choosing Assistent Nurse candidates * (0.008) (0.019) (0.017) (0.023) (0.016) (0.016) (0.026) (0.020) NOTE: Variables are measured as difference in mean score. The voice index is computed by adding all the voice scores where scores for Tired and Unsecure are included with a negative sign. Case worker cluster-robust standard errors. p < 0.05 indicated by *. 29