US workers' willingness to accept meatpacking jobs amid the COVID‐19 pandemic

Abstract We implement a discrete choice experiment to examine the impact of COVID‐19 exposure risk, unemployment risk, enhanced and extended unemployment benefits, and job attributes on low‐skilled workers' willingness to accept (WTA) meatpacking jobs. With a sample average WTA wage of $22.77/h, the current national average meatpacking wage of approximately $15/h is too low for these workers to consider this employment opportunity. Enhanced layoff risk and exposure to COVID‐19 further deterred respondents, while health insurance, retirement benefits, and a signing bonus enhanced respondents' WTA. The additional unemployment benefits of the CARES Act neither deterred nor encouraged respondents WTA.

As nonpartisan researchers, we care about the quality of our survey data and hope to receive the most accurate measures of your opinions. So, it is important that you carefully read all the information provided, and that you thoughtfully give your best answer to each question in the survey. To maintain the integrity of the study, certain measurements have been put in place throughout the survey so as to flag respondents who do not provide quality or complete responses. Do you commit to carefully reading and providing your thoughtful and honest answers to the questions in this survey? I will read carefully and provide my best answers I will not read carefully and provide my best answers I can't promise either way
pandemic? (ii) Did the respondent benefit from the CARES act? (iii) Are you or a family member immunocompromised? For respondents that answered these questions, the sample was further broken down based on whether the answer was "yes" or "no".
First, we divide the sample based on whether unemployment benefits were collected at the time of the survey or during the pandemic (2 nd and 3 rd columns of these job attributes were not as effective in attracting workers from this subsample, as the magnitude of the mean WTA is lower than those of the main results in Table 3. Second, for respondents who answered the question regarding additional benefits from the CARES Act, we divided the subsample into respondents that received the CARES benefit at any point and those that did not receive additional benefits (4 th and 5 th columns of Table   1). The additional benefits of the CARES Act neither deterred nor encouraged respondents to accept meatpacking jobs because the mean WTA for ASC is not statistically different from zero. This finding is in contrast to the negative mean WTA results of Table 3 and is consistent with the literature that FPUC under the CARES Act did not impact employment gaps amid the pandemic (Dube, 2021;Bartik et al., 2020;Marinescu et al., 2020;Finamor and Scott, 2021).
The mean WTA for layoff rate and risk of exposure are similar in magnitude to those of the main results. Additionally, relative to the main results in Table 3, additional benefits through the CARES act dampened the effect of health insurance, retirement policy, and signing bonuses on enhancing workers' WTA meatpacking jobs between 39% and 62%.
Therefore, respondents that received additional benefits under the CARES were less willing to substitute health insurance, retirement policy, and signing bonuses for lower wage rates.
These results could be because health insurance and retirement policy do not alleviate of the severe negative income shock that many experienced due to COVID-19 (Chen and Chaney, 2021).
The subset of respondents who did not receive additional benefits under the CARES Act were the most willing to accept a meatpacking job, as they would have been indifferent between a $7.242/hour pay cut and the status quo at the time of the survey. However, the mean WTA for layoff risk and exposure were again fairly similar in magnitude to the main results. Additionally, as expected, health insurance and retirement benefits enhanced their WTA. However, the signing bonus did not influence this group's decisions, as the mean WTA is insignificant.
Third, we divided the sample into respondents who were and were not immunocompromised (6 th and 7 th columns of Table 1). A priori, we would have expected immunocompromised respondents to avoid employment that involves a high risk of contracting COVID-19.
Therefore, it is counterintuitive that when the sample is divided between respondents that were or had a family member who was immunocompromised, respondents were indifferent between their current status and a meatpacking job (i.e., the estimated mean WTA for ASC is insignificant). However, this result could occur because our sample includes low-income people, 1 which implies that alternative employment opportunities for these respondents might also have a relatively high COVID-19 exposure risk. The results show that the mean WTA for layoff, exposure, insurance, and retirement bonuses for the immunocompromised subsamples are again similar in magnitude to the main results. To provide additional insight into the counterintuitive result, Appendix B, Table 3, we rerun analysis for immunocompromised respondents while including an unemployment interaction. The results indicate that including the unemployment interaction had minimal effects on the results the main attributes (ASC, layoff, exposure, insurance, and bonus) as the mean WTA in Table 3 are similar to those from Table 1 without the interaction. However, the key result is the mean WTA value of -$17.405 for the ASC-Unemp term, which indicates that immunocompromised workers that are unemployed are highly adverse to leaving their status quo (unemployment) to accept meatpacking jobs. Thus, it is likely that their unemployment status is a result of their health risk, which contributes to their unwillingness to accept the meatpacking job.
1 Annual income was 31,671 on average, with a standard deviation of 21,218.

B.2 Unemployed Workers
Because unemployed workers are the likely to enter the meatpacking industry and expand the labor supply, we consider a subset of 93 respondents that were unemployed and actively looking for work at the time of the survey. With 93 respondents and 8 choice sets, the sample size is 744. Of the 93 respondents in this subsample, 66 were actively looking for work and 27 were not looking for work (see Appendix C for sensitivity analysis where we further limit the sample to unemployed that are actively looking for work). In addition to being unemployed, we asked the respondents to indicate the probability (0% to 100%) of their perception of obtaining a desirable, high-paying job before the survey, at the time of the survey, and in the future. As reported in Table 2, these 93 respondents indicated, on average, a 29.54% change before the survey, which increased to 36.56% at the time of the survey and 46.26% by January 2021. Thus, while respondents believed that the probability of obtaining a desirable, high-paying job was below 50%, their outlook improved over time.
In three separate analyses, we interacted the answers to each of the three job-outlook questions with the job attributes. Table 2 reports the mean WTA for this unemployed subsample. The results for this subgroup are striking-the respondents were considerably less willing to accept a meatpacking job compared to the full sample, as the ASC value is substantially more negative at -15.269, -15.999, and -17.819 for the past, present, and future, respectively, relative to -2.857 for the full sample in Table 3. These results could have occurred because approximately 30% of the subgroup is not looking for work; therefore, a large wage premium would be required for these respondents to enter the workforce in a meatpacking job. Additionally, the respondents who were unemployed but looking for work had a relatively better outlook on future employment at 50.5%, compared to 46.26% for all unemployed workers and 45.36% for employed workers. Therefore, it is possible that these respondents could be holding out for better employment opportunities.
Interacting the job attributes with the respondents' perception (evaluated as a probability from 0% to 100%) of obtaining a desirable, high-paying job influences the mean WTA for ASC, but only very slightly (0.173, 0.195, 0.159 for past, present, and future, respectively).
Again, the impact of layoff and exposure have a very similar impact on mean WTA for "past" and "present," as in the main results in Table 3. However, when the job attributes are interacted with unemployed respondents' future outlook on employment opportunities, the impact of layoffs becomes statistically insignificant, which is consistent with these respondents' improvement in employment outlook over time. The mean WTA for exposure to COVID-19 and insurance are also similar to the main results in Table 3. However, the retirement attribute and bonus are statistically insignificant, which suggests that respondents in this subsample have either left the workforce or are holding out for better jobs. Note that, for the subsample of respondents that were employed at the time of the survey, the mean WTA values are very similar to the main results in Table 3. Given the similarity in the results, we do not report these results in the tables. only other interaction term that is statistically significant is layoff rate, which suggest that a higher layoff rate has a small positive impact on immunocompromised workers that are unemployed willingness to accept a meat packing job. Table 4 reports the results for unemployed workers that are actively looking for work.

B.3 Sensitivity Analysis for Mean WTA
The signs and magnitudes of the coefficient estimates are similar to those from Table 2 in the main text, although the interaction terms are largely insignificant. Notes: Standard errors are in parenthesis. *** p < 0.01, ** p < 0.05, * p < 0.1. Variables ending with "-Prob" indicate interaction between the attribute and an indicator variable equal to one if the respondent viewed the information set and zero otherwise. Notes: Standard errors are in parenthesis. *** p < 0.01, ** p < 0.05, * p < 0.1. Variables ending with "-Unemp" and "-Immuno" indicate interaction between the attribute and an indicator variable equal to one if the respondent were unemployed or immunocomprimized, respectively. Notes: Standard errors are in parenthesis. *** p < 0.01, ** p < 0.05, * p < 0.1. Variables ending with "-Prob" indicate interaction between the attribute and an indicator variable equal to one if the respondent viewed the information set and zero otherwise.

C RPL Regression Results
The RPL models are estimated using the "gmnl" package in R (Version 3.4.3) with 500 Halton draws such that the simulation accounts for the panel structure of the data (Sarrias et al., 2017).