It is important to note that employers cannot ask about applicants’ criminal backgrounds according to California’s 2018 ”Ban the Box” policy. However, they can run background checks. Furthermore, using self-disclosed criminal history in the experimental design is justified for the following reasons. First, according to correspondence studies, it is not uncommon to self-disclose criminal history . Previous studies show that self-disclosing criminal history is recommended to ex-offenders applying for jobs because they can control disclosure rather than waiting for their history to be disclosed in a background check. Thus, self-disclosure of criminal history is a realistic experimental setting. Treatment 2 is nearly identical to Treatment 1, the only difference being the applicants’ conviction status. The resumes, rather than stating the applicant has been convicted of a misdemeanor, state that they have been arrested but not convicted. Thus, Treatment 2 includes those with misdemeanor arrest records but not convictions. Treatment 3 has the same characteristic feature as Control but includes applicants with a different racial background. Treatment 3 contains resumes for black job applicants with clean records. Race is indicated by using black-sounding names. Names are borrowed from Bertrand and Mullainathan . Treatment 4 is identical to Treatment 1, greenhouse tables except that resumes are for black applicants who have been convicted of misdemeanors. Treatment 5 is identical to Control, except that resumes are for white female applicants.
Treatment 6 includes the same characteristics as Treatment 1 but with white female applicants. Table 2.1 presents the main characteristics of each group. After rating all the resumes, subjects were asked to rank them from the best to worst . Subjects did not have to explain their decisions in this phase. This different approach enforces the idea of there being competition between resumes. In the field experiments studying employment such as Leasure’s study, applicants received callbacks if they met the requirements of the employers, and there is a cut-off point: if applicants are above the points required by the companies’ hiring standard. Ranking does not provide such a clear cut-off point. However, the better the ranking, the more likely it is that applicants will move to the next round of interviews, which means they are closer to the cut-off point. Thus, ranking serves a similar purpose. In the rating task, the competition among applicants is not obvious, and the rating is based on individual resumes. Additionally, I held a treatment session for the same resumes but with the expungement treatment included. In each of the groups of those with records – “convicted white male,” “convicted white female,” “convicted black male” – and the “arrest-only white male” group, one resume is randomly selected to have a clean record. I simply removed the self-disclosure of their criminal background to demonstrate the effect of expungement. In the treatment session, subjects performed the same tasks in relation to these modified resumes as with the other resumes. The goal was to study whether these modified resumes received better ratings after any criminal background was expunged. Thus, the experiment is divided into a control session and a treatment session. One concern in this experiment is whether the quality of data is as good as that collected by field experiments. Thus, to safeguard the quality of the data, I implemented numerous steps to ensure that subjects understood their task to rate and rank these applicants as if it were part of a real hiring process. Another concern regarding the rating task was that, on average, some subjects might consistently award low scores and others consistently award high scores; for example, if subject A’s average rating for all resumes is 5.8 and subjectB’s average rating is 7.6.
After ranking all resumes, the subjects completed a short survey to disclose their own race and gender. Their demographic information made it possible to check whether they might have racial or gender preferences in relation to applicants that could compound or affect the impact of the applicant having a misdemeanor record. There are some drawbacks and limitations to using lab experiments to study hiring decisions. As mentioned above, the quality of lab data and how realistic they are compared to field experiments are still under investigation. The experimental design lacks a connection between subjects’ earnings and their ratings and rankings. However, laboratory experiments require less time to collect data and allow easy modification of the research design to study race, gender, and expungement simultaneously.The following tables present the results of the experiment. Table 2.2 presents the linear regression results when comparing resumes with different conviction statuses. Column 1 shows that, when pooling all the resumes, as long as a resume indicates that the applicant has a record , the rating is negatively affected at a 1% significance level. The result allows Hypothesis H1 to be rejected and demonstrates that having a misdemeanor record has a strong negative impact on employment outcomes. Thus, it is worth studying whether expungement also affects the ratings. The results also suggest that black applicants are discriminated against, and this result is significant at the 1% level. Female applicants receive slightly lower ratings, but this result is not statistically significant. All the resume data for female and black applicants are then dropped to compare the impact of having a misdemeanor conviction on record compared to having only an arrest record. White male applicants with clean, arrest-only, or convicted status were kept in the sample to control for race and gender.
Column 2 suggests that being arrested without conviction lowers an applicant’s ratings by 1.971 points, and being convicted lowers the rating by 1.652 points. Both results are significant at the 1% level. This shows that subjects do not differentiate between a conviction or an arrest when dealing with applicants with records. Although the results suggest that applicants with arrest-only records receive lower points than those with misdemeanors, flood tray the difference is small. Subjects tend to pay some attention to job experience and score accordingly, but the dominant impact still comes from the criminal record. Hypothesis H2 can thus be rejected, contradicting the results in Uggen et al. . Uggen et al. find that being arrested has almost no effect on employment access. However, Uggen et al. use an audit study and increase contact with employers. For example, applicants are always asked to have a conversation with the hiring manager when they fill out the in-person application. Uggen et al. conclude that contact with hiring managers is significantly related to increased callbacks. However, applicants applying for jobs online do not have such close contact. Thus, the present experimental design is closer to an online job application process, which does not facilitate direct contact between employers and applicants. The results demonstrate that although marijuana is legalized in California, subjects still consider marijuana-related misdemeanors as a negative component in the job hiring processes. Thus, both Columns 1 and 2 suggest that marijuana-related misdemeanors have a negative impact on employment access, implying that automatic expungement should improve employment outcomes. Table 2.3 presents the linear regression results with the inclusion of expungement. All resumes with clean records were dropped other than those for which records were expunged. The expungement process involves removing the record from an applicant’s background. I removed the self-disclosure of applicants’ criminal backgrounds to mimic this result. I then divided the control session into three groups consisting of around 30 participants and calculated the average rating for all resumes that included a criminal background. Because of limited funding, I only conducted one treatment session consisting of 30 participants.I calculated the average ratings for all resumes with criminal backgrounds and those with expungement. Column 1 of Table 2.3 shows that the difference in the average scores between the control and treatment sessions are significantly affected by the expungement process. Without expungement, the difference between the two sessions for the resumes with criminal records is almost neglectable. With expungement of records, however, the ratings for the resumes increased by 1.173 points, and this result was statistically significant at the 1% level. Hypothesis H3 can thus be rejected, demonstrating that the same resumes perform much better in the ratings after expungement, which further supports the idea that automatic expungement should improve ex-offenders’ employment outcomes. Table 2.4 presents the linear regression results when comparing the resumes of applicants of different races. Black and white applicants with clean records or a conviction were kept in order to control for gender and conviction status. Both Columns 1 and 2 suggest that black applicants receive lower scores. The results show that the discriminatory attitudes are statistically significant at the 1% level. The interaction term of black applicants and a misdemeanor record in Column 2 suggests that black applicants receive ratings that are 0.246 points lower than those of white applicants with misdemeanor records. Thus, Hypothesis H4 can be rejected, which is in line with many previous research findings. It is important to pay attention to those with marijuana-related misdemeanors who are black because of the racial disparity in ratings . The results here also demonstrate that black applicants are double-penalized for having a marijuana-related record, and automatic expungement will improve their employment outcomes. Table 2.5 presents the linear regression results for the comparison of resumes for applicants of different genders.
Applicants’ racial background was controlled, and only the resumes of white male and white female applicants with either clean or misdemeanor records were kept. Column 1 shows that female applicants received slightly lower ratings, with this result being significant at the 5% level. Thus, Hypothesis H5 can be rejected. Column 2includes the interaction term of female applicants and misdemeanor records. As the results suggest, female applicants receive lower scores than male applicants with similar misdemeanor records, but the effect is not statistically significant at the 5% level. However, the results still suggest that female applicants face a similar double penalty as black applicants in the job market. Both gender and race play important roles when applicants have marijuana related misdemeanor records. Thus, it is safe to conclude that automatic expungement will significantly improve ex-offenders’ employment outcomes, particularly for racial minorities and women. To summarize the results of the resume-rating process, contrary to Uggen et al.’s findings, having a record for a misdemeanor, regardless of conviction status, has a significant negative effect on employment access . The results also suggest that race significantly impacts employment access, and gender has some impact. The results are largely consistent with Leasure . Leasure finds that employers did not distinguish between applicants with records on the basis of the crime’s severity. A misdemeanor record had the same negative impact on employment access as a felony record. These results resonate with my findings here that arrest and conviction have an almost equal negative effect on employment access. Furthermore, black and female applicants receive worse ratings than other applicants with misdemeanor records and are double penalized. The following tables present the regression results for the rankings. Table 2.6 presents the linear regression results comparing the rankings of resumes with different conviction statuses. Column 1 shows that, as predicted, having a record, regardless of conviction status, significantly increases the ranking. Note that the higher the ranking, the worse subjects think the applicants are in terms of employ ability. Race and gender are then controlled to compare only the effects of a conviction and of an arrest only. Column 2 shows results consistent with the rating. Resumes with arrest records are ranked much worse than resumes with no records. The impact of only being arrested is almost the same as having a misdemeanor conviction. Table 2.7 presents the linear regression results when comparing the resumes of applicants of different races. Black and white applicants with clean records or a conviction are kept to control all other variables. All columns show that black applicants receive a slightly worse ranking than white applicants. Including the interaction term of black applicants and a misdemeanor record, the results in Column 2 suggest that among applicants with misdemeanor records, black applicants receive a ranking of 1.12 points worse than that of white applicants. The results correspond with the ratings, but having a criminal background still has a larger impact on a resume’s ranking than race. Table 2.8 presents the linear regression results comparing the resumes of applicants of different genders.