Young adult substance use following involuntary job loss

Table 1 presents descriptive statistics (means and proportions) of selected variables based upon displacement status. We use only the first observation for each individual in the calculations. Traditionally in the displacement literature, researchers calculate descriptive statistics for the displaced sample using only pre-job loss observations, and they calculate statistics for the non-displaced sample using all person-year observations. Because the TAS is a cohort-based survey, the displaced sample artificially appears to be younger and less educated when compared to their non-displaced counterparts when calculating the descriptive statistics in that manner. Therefore, we present calculations once using only each person's first observation.

Table 1 shows that displaced and non-displaced young adults are similar in regard to age, current drinking status, the number of drinks per drinking session, proportion employed, proportion responsible for their own household (i.e. not living with family/friends), and the proportion that report using amphetamines, cocaine, barbiturates, or tranquilizers. However, the displaced tend to have less education relative to the control group of non-displaced workers, and job losers are also less likely to be female and white. Those who eventually experience involuntary job loss work more hours per week and have higher rates of cigarette and marijuana use along every dimension. Furthermore, those who eventually experience displacement are more likely to report using pain relievers/narcotics. Given that the calculations in Table 1 restrict the displaced group to an observation that occurs pre-job loss, these statistics highlight the importance of using individual-level fixed-effects in the estimated equations presented above.

When using all person-year observations to calculate averages across all substances, we find that usage increases at the extensive margin but not at the intensive margin regardless of displacement status. In other words, as young adults get older, they are more likely to begin using different substances; however, the intensity of usage remains relatively constant over time. This finding is in line with Kelly and Vuolo (2018) who mention that a characteristic of young adulthood is self-exploration, where part of this exploration process involves substance use.

The analysis starts by examining the relationship between displacement and earnings and employment probabilities for young adult workers. Doing so is important. The earlier literature on displacement and substance use for older adults attributes the increase in alcohol and cigarette consumption to the financial stress caused by persistent earnings losses and employment instability associated with job loss. Therefore, it is important to understand if these young individuals experience the same types of labor market outcomes associated with displacement as with the traditionally older samples. If they do, then poor labor market outcomes associated with job loss could explain any increase in substance use among this younger demographic.

Results in panel A of Table 2 come from equation (1) when using real annual labor earnings, the log of annual earnings, and an employment binary variable as dependent variables. The table only presents the coefficient associated with the displacement dummy variable. The row labeled After refers to all post-displacement periods, including the year of the first reported job loss. The other coefficients are available upon request. We use both the level and natural log of earnings to allow for labor market withdrawal in response to job loss. Estimates show that displacement does lead to significant declines in earnings and the probability of being employed. Earnings losses amount to approximately $4,300. The corresponding loss is 53 percent when using the natural log of earnings. The displaced are also 13 percentage points less likely to be employed following job loss relative to their non-displaced counterparts.

The bottom panel of Table 2 presents estimates from the event history analysis shown in equation (2). As in panel A, only the coefficients associated with the displacement variables are shown. Here, the rows labeled At least t-6 through At least t+6 represent time relative to the first reported job loss, where period t is the year of displacement. Here, two years before displacement is the reference period. For ease of visualizing these results, we present graphs of these coefficients, along with a 95-percent confidence interval, in Figure 1.

Figure 1

Results suggest that earnings losses and declines in the probability of employment occur during the year of reported separation and remain significant for the entire follow-up period. The only exception to this is the coefficient for period t+6 in the log earnings equation. These results do suggest that job loss at younger ages is a true labor market shock as the pre-displacement coefficients for log earnings and employment are statistically insignificant. The pre-displacement coefficients associated with the real earnings regression are statistically significant. However, the coefficients are stable over time leading up to job loss and positive, suggesting that the group of job losers earn more than the comparison group pre-displacement.

Table 3 presents estimates of displacement's potential relationship with substance use from equation (1). Panel A focuses on the probability of using particular substances. Results indicate that young adult job loss is statistically unrelated to the probability of using marijuana, narcotics/pain relievers, and other drugs such as amphetamines, cocaine, barbiturates, or tranquilizers. Each coefficient is relatively small in magnitude. In fact, the estimates for marijuana and pain relievers are negative. Displacement is statistically related to the probabilities of smoking and drinking. Involuntary job loss increases the probability of being a current smoker by 3.2 percentage points and a current drinker by 3.1 percentage points. Each estimate is significant, and each is economically large. The last row in panel A shows the average of the dependent variable during the pre-displacement period. Using this average as a base, the coefficients represent a 14.5 percent increase in the probability of smoking and a 5.6 percent increase in the probability of being a current drinker.

Panel B provides estimates of substance use at the intensive margin using equation (1). During this part of the analysis we restrict the sample to those who report using the substance in question. Recall that we do not estimate the effect of displacement on the intensity of using the Other Drug category because this category combines several drugs together, and the PSID measures usage of these drugs using interval responses. Results suggest that young adult job loss is unrelated to the number of cigarettes smoked per day, the probability of drinking several times per month, and the number of times one uses pain relievers/narcotics during the previous 12-month period. In fact, the estimates for drinking several times per month and narcotics are negative. Displacement is statistically related to increases in marijuana usage and the number of drinks consumed per drinking session. The estimates are economically large, representing a 13 percent increase in the number of drinks per session and a 14 percent increase in marijuana usage when using the pre-job loss averages presented in the final row of panel B.

The results in Table 3 show that young adult job loss is statistically related to the probability of smoking, the probability of drinking, the number of drinks consumed while drinking, and the number of times marijuana is consumed during the previous 12 months. To explore the dynamic relationship between job loss and these four measures of substance use, we re-estimated equation (2) using these four measures as dependent variables. Results are presented graphically in Figure 2. Tables of the coefficients are available in appendix Table A-1.

Figure 2

Focusing on the probability of smoking, the results in Figure 2 suggest that the effect of displacement is immediate, occurring during the year of job loss. While the estimates during periods t + 2 and t + 4 lose precision, the coefficients remain elevated relative to those in the pre-job loss period, suggesting a medium-run effect of displacement on smoking. Figure 2 shows a similar result for the number of times marijuana is used. The effect of displacement is immediate, with the estimate during the year of job loss being significant at the 10-percent level. Unlike the case for smoking, the post-displacement coefficients associated with marijuana consumption grow over time, suggesting more of a long-run effect.

The causal evidence of the relationship between displacement and the probability of drinking is less convincing. The coefficients in periods t and t + 2 are significant. However, they are similar in magnitude to the coefficient in period t – 4. We observe stronger evidence with regards to the number of drinks consumed per drinking session. There is a statistically significant increase in the number of drinks consumed during the year of job loss. The post-displacement coefficients all remain high relative to those in the pre-job loss period, and they are significant until period t + 4. Therefore, while displacement may increase the probability of drinking, the evidence suggests that job loss does increases the intensity of alcohol consumption.

Thus far, the results show that young adult involuntary job loss is associated with long-run reductions in the probability of employment, averaging 13.4 percentage points. Results further show that displacement among young adults increases the probability of smoking and drinking, and the intensity of alcohol and marijuana consumption. It is natural to ask whether substance use after job loss leads to larger reductions in the probability of employment. We investigate this in two ways. First, we split the displaced sample into five separate groups based upon substance use at the time of job loss: those who smoke cigarettes, those who drink, and those who report ever using marijuana, pain relievers/narcotics, and other drugs. We then estimate equation (1) once for each treated group. The control group is the same for each regression. The dependent variable is the employment dummy variable. Results are available in panel A of appendix Table B-1 and provide mixed evidence with regards to substance use leading to larger employment losses. Post-displacement employment losses are larger than 13.4 for those who smoke during the survey of the reported job loss, who have ever used pain relievers, and for those who have ever used other drugs. Employment losses are smaller than average for current drinkers and those reporting ever consuming marijuana. Importantly, however, the estimates in appendix Table B-1 are all close in magnitude to those in Table 2. To facilitate meaningful comparisons to non-users at the time of job loss, we present estimates in panel B of appendix Table B-1 for those who report not using substances during the year of displacement.

The second way we investigate whether substance use leads to larger employment losses is by examining changes in substance use patterns. Specifically, we limit the treatment group to those who did not use substances during the survey wave prior to job loss but began using substances during the year of displacement. Results are in panel A of appendix Table B-2 and, again, offer mixed evidence. Those who start smoking, using marijuana, and pain relievers/narcotics have larger declines in the probability of employment relative to the average reported in Table 2. However, those who start drinking or using other drugs report smaller employment losses. As with appendix Table B-1, however, the estimates are similar to that in Table 2. Again, to facility meaningful comparisons to those who did not start using substances at the time of job loss, we present estimates in panel B of appendix Table B-2 for those who remain non-users during the wave prior to job loss and the wave of initial displacement.

The purpose of this subsection is to investigate the potential mechanisms through which involuntary job loss may lead to increases in substance use among young adults. The first mechanism investigated here is financial support. It is possible that workers have personal savings and checking accounts that they can access in the event of a job loss. If job losers have an established work history, then they may be able to access government support in the form of unemployment insurance. Finally, young adults may receive financial support from parents or other relatives during times of financial hardship. In fact, 88 percent of the displaced workers in this sample receive some sort of financial support during the survey wave of the first reported job loss. The relationship between financial support and substance use post-displacement is ambiguous. On the one hand, access to additional financial resources may lower the stress associated with job loss, which would decrease the likelihood of substance use as a coping mechanism. On the other hand, additional financial resources may make it easier to obtain cigarettes, drugs, and alcohol after displacement leading to an increase in usage.

To investigate the relationship between financial assistance and post-displacement substance use, we separate the sample of displaced workers into groups based upon whether they received aid during the year of job loss from parents/relatives, the government, or their own resources. Here, relatives provide assistance by helping to pay tuition, rent, or a mortgage, providing a personal loan, or paying for expenses.

This category also includes other forms of financial assistance not specifically defined by the PSID.

Panel A in Table 4 focuses on displaced workers who receive help from parents/relatives. Results show that those who receive financial assistance from relatives have higher probabilities of smoking after displacement. The coefficient associated with the probability of drinking loses precision; however, this is most likely due to the smaller sample size of the treated group. Those who receive aid from the government are significantly more likely to begin smoking and using amphetamines, cocaine, barbiturates, or tranquilizers. Those who rely on their own resources during job loss are significantly more likely to begin smoking and drinking. When estimates in Table 4 are statistically significant, they are somewhat larger in magnitude when compared to the main results in panel A of Table 3. The only exception to this is the probability of smoking for those who rely on their own financial resources. This provides slight evidence that outside financial assistance may make it easier to afford different drugs and alcohol post-job loss.

Results in Table 5 show that displaced young adults who receive aid from relatives significantly increase marijuana consumption and decrease the probability of drinking several times per month. Receiving aid from the government leads to increases in the number of drinks consumed per session, and displaced young adults relying on their own resources significantly increase the number of drinks per session and their consumption of marijuana. As before, when estimates are significant, they are slightly larger in magnitude compared to the main findings in panel B of Table 3. This is particularly true for marijuana consumption. Again, this provides transient evidence that financial assistance from outside resources can make affording substances easier after displacement, leading to increased usage. The exceptions to this are drinking after displacement when receiving aid from relatives or relying on one's own savings.

The final mechanism investigated here is the state of the economy at the time of displacement. Prior research shows that those who experience involuntary job loss during declining economic activity experience worse labor market outcomes relative to those who suffer displacement during expansions (e.g. Couch, Jolly, & Placzek 2011). Greater labor market instability may lead to higher rates of substance use for the reasons discussed in the introduction. To this end, we re-estimate equation (1) after limiting the treatment group to those displaced workers who experience an increase in their state's unemployment rate between the period two years before job loss to the year of displacement. We do not provide estimates for using pain relievers here. Recall that data on the usage of pain relievers only exists for three surveys, 2015, 2017, and 2019. During this period, there are only 14 displaced workers who experience an increase in the state unemployment rate and have data available on using pain relievers. Results from this analysis are in Table 6. When focusing on the extensive margin, the coefficients are somewhat larger than the main results, yet they are insignificant given the smaller size of the treated sample. When focusing on the intensive margin, it appears that displacement during poor economic times leads to a reduction in the probability of drinking several times per month and an increase in the number of drinks consumed when drinking.

The displacement literature suggests that gender and racial differences exist in individual outcomes after involuntary job loss. Jahromi and Callaway (2020) show that there is significant heterogeneity in earnings losses between men and women using the Displaced Worker Survey and a sample of workers with an average age of 42. Furthermore, the authors present evidence suggesting that gender differences are unequal through the earnings distribution. The authors do note that racial differences in earnings losses exist; however, these differences tend to be relatively small. In contrast, FairlieKletzer and (1998) show evidence that employment rates among displaced Black workers are significantly lower when compared to their white counterparts. The gender and racial differences in post-displacement experiences may lead to gender and racial differences regarding substance use after involuntary job loss.

Similarly, it is possible that the relationship between job loss and substance use is different between those who are responsible for their own household at the time of displacement and those who reside with relatives or friends. Krolikowski et al. (2020) show that relatively younger displaced workers (those between 25 and 35 years old) experience smaller earnings losses after job loss when they live within the same Census tract as their parents.

This subsection explores heterogeneity in substance use across gender, race, and whether workers are responsible for their own household at the time of job loss. Here, we classify respondents as being responsible for their own household if the PSID identifies them as being a household head or the spouse/cohabitor of the household head. To investigate any potential differences in substance use post-job loss, we alter equation (1): (3) yit=xit′β1+δ1Ditk≥0+φ1Ditk≥0*groupi+αi+γt+uit. {y_{it}} = x_{it}^\prime{\beta _1} + {\delta _1}D_{it}^{k \ge 0} + {\varphi _1}D_{it}^{k \ge 0}*grou{p_i} + {\alpha _i} + {\gamma _t} + {u_{it}}.

Here, the variable group defines the group to which a person belongs. When investigating differential effects of displacement by gender, group equals one if the person is a female. When examining differences by race, group equals one if the respondent reports being white. Finally, to see if the effect of displacement differs by whether or not someone is responsible for their own household, group equals one if the individual is a household head or spouse/cohabiter with the household head at the time of job loss.

Results when investigating heterogeneity by gender are in Table 7, by race are in Table 8, and by whether someone lives on their own in Table 9. Panel A in each table investigates substance use at the extensive margin, whereas panel B focuses on the intensive margin. The rows of estimates in each panel represent the total effect of displacement on substance use for the group in question. For example, in Table 7, the row labeled Men shows estimates of δ₁, and the row labeled Women presents estimates of δ₁ + φ₁. The final rows in each panel indicate whether the estimates are statistically different across the groups and each group's pre-treatment average.

We begin this subsection with a brief discussion of differences in labor market outcomes associated with involuntary job loss by gender, race, and living arrangements. Results from estimating equation (3) using various labor market outcomes as the dependent variable are in appendix Table C-1. The results suggest that non-white job losers experience significantly larger earnings losses than their white counterparts do regardless of whether the natural log of earnings or the level of earnings is the dependent variable. While the racial difference in employment probabilities is not significant, the point estimates suggest that non-whites do experience larger declines in the probability of being employed post displacement. When focusing on the natural log of earnings, female earnings losses tend to be statistically similar to male earnings losses. However, women suffer statistically larger earnings losses when examining the level of earnings. This is most likely caused by women leaving the labor market at higher rates relative to men after job loss. The results support this notion. While the gender difference is insignificant, women do have lower probabilities of employment after job loss than men do. The results based on whether displaced workers are responsible for their own household are interesting. Those who live on their own at the time of job loss experience smaller earnings losses relative to those who live with friends or relatives. However, those responsible for their own households are statistically much more likely to be non-employed.

These differing labor market outcomes post displacement may lead to differences in post-job loss substance use patterns. When focusing on cigarettes, Table 7 shows that there is slight evidence that male job losers experience higher probabilities of smoking relative to female job losers. The coefficient for men is double that for women; however, the difference is insignificant. The same can be said for those who are responsible for their own home at the time of job loss. These individuals are slightly more likely to become smokers relative to those who live with friends or relatives; still, the difference is insignificant. Non-white displaced workers, however, are statistically more likely than white job losers (5.3 versus −0.5 percentage points) to become smokers. Results in panel B of Tables 7 through 9 suggest that there are no differences in smoking intensity by gender, race, or living arrangements.

The degree of heterogeneity is higher when focusing on alcohol consumption. At the extensive margin, displaced women have a higher probability of drinking compared to male job losers. While the gender gap is insignificant, the point estimate for women is 14 times larger than for men (a significant 5.7 percentage point increase for women job losers versus an insignificant 0.4 percentage point increase for male job losers). A similar finding occurs with race. The racial gap in drinking probabilities is insignificant; however, it is economically large. The point estimate for white job losers is 0.006 and insignificant, whereas the corresponding estimate for non-whites is 0.045 and significant at the 5-percent level. We do find significant differences in drinking probabilities by living arrangement. Those who live with friends and relatives at the time of job loss are significantly more likely to drink relative to those who are responsible for their own household.

Results show no differences across groups with respect to the probability of drinking several times or more per month. However, a significant racial gap exists with the number of drinks consumed per session. Non-white job losers increase their intensity of drinking by 0.67 drinks per session, whereas displaced white workers drink an insignificant 0.21 drinks more.

For the results focusing on marijuana consumption at the extensive margin, there appear to be no differences in the probability of usage across race or living status. A gender gap does exist. Young adult men experience a significant reduced probability of consuming marijuana, whereas the point estimate for female job losers is statistically insignificant. Therefore, job loss does not seem to increase the probability of usage for young adult women. However, job loss seems to have a negative relationship with the probability of marijuana consumption for men. At the intensive margin, displacement seems to have no differential effects across groups.

The final set of results focuses on responses to displacement with regards to use of harder drugs. Focusing on the other drug category, which contains amphetamines, cocaine, barbiturates, and tranquilizers, we see that men, whites, and those who live with friends and relatives are significantly more likely to use these harder drugs after job loss relative to women, non-whites, and those responsible for their own home at the time of separation, respectively. The same can be said for the probability of using non-over-the-counter pain relievers/narcotics. Each of the estimates based on living status is significant. However, the estimates by gender and race are not. The probability of using pain relievers/narcotics after job loss increases by six percentage points for men and nine points for whites. This contrasts with a decrease in the probability of usage for women and non-whites by five percentage points each. These economic differences are rather large. The lack of statistical precision is most likely due to the small sample sizes for this drug. Recall that data on narcotics usage is only available for three surveys. It would be interesting to re-estimate these regressions as more data become available. At the intensive margin, women and non-white job losers use pain relievers significantly less after displacement, whereas the estimates are insignificant for the remaining groups. The gender gap in the number of times a person uses pain relievers is significant.

To summarize the findings presented in this subsection, women and non-white young adults are more likely to drink, and non-whites increase their drinking intensity after involuntary job loss relative to the respective comparison groups. Non-whites and men are more likely to report being current smokers. Finally, young adult men, whites, and those living with friends and relatives are much more likely to use hard drugs such as amphetamines, cocaine, barbiturates, tranquilizers, and narcotics/pain relievers after displacement relative to their respective comparison groups.

It is possible that differences in substance use at the intensive margin arise because of differences in rates of usage prejob loss. We replicated the analysis after replacing the dependent variable with the ratio of the yearly measure to the amount consumed during the period two years before job loss. Results are in appendix Table D-1 and are qualitatively similar to those presented in Tables 7 through 9. See the appendix for further detail.

There are a few potential reasons for these differences in substance use patterns. First, evidence presented here suggest that men and whites experience smaller earnings losses relative to women and non-whites. Furthermore, men and whites experience smaller reductions in the probability of employment. Harder drugs tend to be more expensive relative to alcohol. Therefore, it may be the case that whites and men are able to afford these substances after job loss more easily relative to women and non-whites. While there is some evidence that those living with friends and relatives at the time of displacement experience larger earnings losses when compared to those responsible for their own household, those living with friends/relatives may not need to pay for housing and food to the same extent. Therefore, these individuals may still be able to afford these hard drugs because overall disposable income may be higher. The second potential explanation relates specifically to gender. Ringel, Ellickson, and Collins (2007) note how society may view female hard drug use negatively. Therefore, women may be more likely to turn to alcohol as a coping mechanism relative to the hard drugs examined here in order to avoid a negative social stigma associated with female hard drug use.

Individuals must respond to at least two surveys to be in the sample. Further, displaced workers must be present in the sample during the survey wave preceding the first reported job loss. Therefore, the sample used here is an unbalanced panel. It is possible that non-random attrition could bias the results from equations (1) and (2) for three reasons. First, the sample used in this paper contains young adults between the ages of 17 and 28. Individuals in this age range may experience high rates of mobility as they transition from living with parents/relatives to living on their own and vice versa. Changes in living status could lead to differential rates of attrition since geographic mobility may lead to an inability of the PSID to follow respondents. What matters here is if job displacement leads to differential rates of moving into or out of relatives’ homes. To this end, we estimated equation (1) twice. In the first estimation, the dependent variable equals one if the respondent is responsible for his/her own household and zero otherwise. In the second regression, the dependent variable equals one if the respondent moved into or exited from a relative's house from the previous survey and zero otherwise. The results are in appendix Table E-1 and show that job displacement does not lead to differential probabilities of either living on one's own or transitioning from one living arrangement to another.

Non-employment and substance use are two additional channels that potentially lead to non-random attrition. Non-employment may lead to survey non-response due to the stress involved with labor market transitions. Substance use may lead to non-random attrition if it leads to an inability or lack of desire to respond to surveys. Above we show that displacement leads to reductions in the probability of employment and increased rates of using some substances. Therefore, displaced workers may face differential rates of attrition if non-employment/substance use lead to differential rates of attrition. To this end we estimated several regressions. The dependent variable in each is binary, equaling one if the respondent attrites from the survey in the following period and zero otherwise. The independent variable of interest in each regression is a dummy for either being employed, being a current smoker, being a current drinker, ever using marijuana, ever using pain relievers/narcotics, and ever using the other drug category described above. Extra control variables include a quadratic in age, year dummies, and individual fixed effects. We do not have data after 2019 and do not know if people attrite from the survey after 2019. Therefore, we do not use observations in 2019 in these regressions. Results appear in appendix Table E-2 and show that employment status and substance use status are statistically unrelated to the probability of attrition. Therefore, even if displacement leads to increased probabilities of substance use and lower probabilities of employment, these differential probabilities do not seem to lead to an increased odds of sample attrition.

As a final way to investigate displacement's effect on the probability of attrition, we estimated equation (1) when defining the dependent variable to equal one if the respondent attrites from the sample in the following survey and zero otherwise. As with the regressions described in the previous paragraph, we exclude observations from 2019. Results are in appendix Table E-3 and show displacement is statistically unrelated to the probability of attrition. Admittedly, this is not a perfect test of the effect of displacement on the probability of attrition. If displacement does lead to higher odds of attrition, then it is reasonable to expect that the displaced workers who remain in our sample may experience better labor market outcomes or smaller instances of substance use relative to those job losers who attrite. This would tend to reduce the magnitude of our estimates and make it harder for us to find an effect.

It is possible that substance use, particularly intense substance use, leads to displacement. While the type of job loss analyzed here is involuntary, it does include layoffs/firings. Firms have discretion over who to lay off/fire. It is reasonable to expect that those who choose to use drugs and alcohol exhibit excessive absenteeism because of either poor health associated with substance use or incapacitation because of being currently under the influence of drugs and alcohol or recovering from a previous day's usage. Excessive absenteeism is a potential cause of termination. Two characteristics of the analysis should help mitigate this concern. First, equation (1) does include individual fixed-effects. These fixed effects should control for any unobservable characteristics that correlate with substance use and the probability of displacement. Furthermore, Stevens (1997) notes how approximately 16 percent of those reporting lay-off/fired are actually fired for cause. It is reasonable to expect that those who are fired for cause exhibit the excessive absenteeism (and other unobservable characteristics that would lead to job loss and substance use) described here.

To the extent that individual fixed-effects and the potentially low proportion of firings for cause do not eliminate this reverse causality, we re-estimate equation (1) after making two important changes. First, we change the dependent variable to a binary variable equaling one if the respondent reports a job loss in the following survey wave. Second, our independent variable of interest becomes one of the measures of usage in the current survey wave. In other words, equation (1) becomes the following: job lossit+1=xit′β1+δ1substance useit+αi+γt+uit job\;los{s_{it + 1}} = x_{it}^\prime{\beta _1} + {\delta _1}{{substance}}\;us{e_{it}} + {\alpha _i} + {\gamma _t} + {u_{it}} . The results from estimating this equation are in appendix Table F-1 and show that current usage at either the extensive, or intensive margins is statistically unrelated to the probability of reporting a job loss in the following survey wave. These results provide evidence that the findings in Table 3 are not capturing substance use leading to job loss.

This subsection analyses the sensitivity of the main results in Tables 2 and 3 to changes in sample selection criteria. Results are in appendix Tables F-2 through F-4. All estimates are from equation (1). Table F-2 focuses on labor market outcomes associated with displacement, Table F-3 investigates substance use at the extensive margin, and Table F-4 analyzes usage at the intensive margin. Each table contains four panels, one for each dimension of sensitivity being investigated. We describe each of these dimensions in turn.

First, the PSID asks about job loss for up to five different jobs. We define a displaced worker as one who reports losing a job due to plant closure or layoff/firing from any of these five jobs. To the extent that young adults consider job 1 more important than job 5, we limit the treated sample to those who only report a displacement from job 1. Typically, the displacement literature only focuses on job loss from the main job. Therefore, this adjustment makes the treated sample appear more similar to those from the literature focusing on relatively older samples. Results from this appear in panel A of each table.

Second, the sample here contains young adults ranging in age from 17 to 28. Teenagers and those in their early 20s may have different labor market attachments than those in their middle- and late-20s. To make certain that treatment and comparison groups are similar, we restrict the entire sample to only those who report being employed during the first survey. Results from this analysis are in panel B of each table. Additionally, we further restrict the sample to those who report working at least 25 hours per week during their first observation. The results from this change in sample selection criteria are in panel C.

Each respondent can attrite from the survey at any time. In fact, the TAS only follows survey participants until the age of 28, after which the PSID moves the individual to the main family interview. Therefore, the sample contains forced attrition. To make certain that identification of the treatment effect is not coming from a particular subset of the treated sample, we limit the treated group to those who are in the treatment window at least between periods t-2 and t+2. Results from this are in panel D.

Results in the appendix show that altering the sample selection criteria does not affect the estimated effect of displacement on the various labor market outcomes investigated here. Each estimate remains significant and of the same sign as reported in Table 2. When investigating the sensitivity of the estimates for substance use at the extensive margin, results in the appendix show that the effect of job loss on the probability of smoking is quite similar to the main results in Table 3. The only difference is that the estimate is imprecise when focusing on displacements from the main job. However, the point estimate is the same as that reported in Table 3. The relationship between job loss and the probability of drinking is lower and insignificant in each estimation when compared to the main estimate. However, given that the main estimate is marginally significant at the 10-percent level, this variability is unsurprising. Finally, results in the appendix regarding the number of drinks consumed per drinking session are consistent with the main findings. However, the relationship between job loss and the number of times the respondent uses marijuana is no longer significant, and the point estimate, while consistently positive, varies in size when compared to the main results in Table 3.

This final sub-section examines the sensitivity of the main results to altering the definition of job displacement and the use of an alternative estimator. Specifically, we focus on layoffs that occur through plant closures only and the use of propensity score matching along with equation (1). These analyses are provided to help in further identifying causality.

The displacement literature generally views plant closures as more exogenous than layoffs and firings. This is due to firms not having discretion over who to fire during a plant closure. Therefore, it is common in the literature to provide estimates of the effect of job loss separately by layoffs and plant closures. Unfortunately, in this instance, the sample of young adults losing their jobs because of a firm closure is quite small, equaling 172. This small sample makes it very difficult to draw meaningful conclusions from any analyses. However, for completeness, we replicated most of the analysis after re-defining job loss as that resulting from plant closures only. The results are in appendix G. To summarize the main findings, job loss from a firm closure still results in significant earnings losses and declines in the probability of employment. Plant closures result in no statistically significant positive effects on substance use at either the extensive or intensive margins. However, there is a significant negative effect on the number of times the job loser consumes pain relievers. Female job losers are more likely to drink relative to their male counterparts; however, men consume more alcohol after job loss compared to women. Non-white job losers are more likely to smoke cigarettes, and they are more likely to drink more per session and consume marijuana more intensely relative to white job losers. Finally, those who live on their own at the time of job loss have lower probabilities of not only using pain relievers/narcotics, but also using them less intensely after displacement.

The final sensitivity analysis we conduct is propensity score matching similar to that in Schmeider, von Wachter, and Heining (2022). Displaced and non-displaced workers may be inherently different. Not controlling for these differences could lead to biased estimates of the effect of displacement on substance use. When these differences are time-invariant, using individual fixed-effects is appropriate. However, as Schmeider et al. (2022) discuss, propensity score matching can be useful if displaced and non-displaced individuals have differential trends. To construct a matched comparison group, define a displacement year as d. Since all displaced workers must be present in the dataset in the survey wave prior to the first reported job loss, then we have seven displacement years, 2007, 2009, 2011, 2013, 2015, 2017, and 2019. For each year d, we match each displaced worker to a non-displaced worker without replacement. We match individuals based on propensity scores calculated from a probit model, where we include as control variables dummies for baseline survey year, gender, and race, and the following characteristics in period d – 2: labor earnings, employment status, a quartic in age, and whether or not someone consumes cigarettes, alcohol, marijuana, or other drugs, and whether someone lives on their own versus with family/friends. As in Schmeider et al. (2022), the control group of matched workers can include those who experience displacement in the future. We then alter equation (1) as follows: yitd=xitd′β1+δ1Ditdk≥0+αi+γt+uitd {y_{itd}} = x_{itd}^\prime{\beta _1} + {\delta _1}D_{itd}^{k \ge 0} + {\alpha _i} + {\gamma _t} + {u_{itd}} . Here, the subscript d ensures that we are specifically focusing on different displacement-year cohorts. Results from this analysis are in appendix H and are similar to the main results presented in Tables 2 and 3. Relative to the matched control group, displaced workers still experience an increase in the probability of smoking and consume more alcohol per drinking session. The coefficients associated with the probability of drinking and the number of times using marijuana decline slightly and lose significance. However, job losers are now more likely to use amphetamines, cocaine, barbiturates, and tranquilizers post displacement.

With the matched sample of treated and control units described above, it is possible to estimate event history models such as: yitd=xitd′β1+∑k≥−mDitdkδk+∑k≥−mcitdkψk+αi+γt+uitd {y_{itd}} = x_{itd}^\prime{\beta _1} + {\sum _{k \ge - m}}D_{itd}^k{\delta _k} + {\sum _{k \ge - m}}c_{itd}^k{\psi _k} + {\alpha _i} + {\gamma _t} + {u_{itd}} . Here, all variables are as before. The citdk c_{itd}^k are dummy variables representing time relative to displacement year d and are created for all displaced and non-displaced workers. Schmeider et al. (2022) note how these citdk c_{itd}^k are important for controlling for potential trends in outcome variables that may exist around cohort years for all workers, displaced and non-displaced. We re-estimated the event history analyses from above. The results are in Figures H-1 and H-2 and are qualitatively similar to the main analysis.

Schmeider et al. (2022) include relative year fixed effects because the authors impose a tenure restriction on their sample of treated and control units. Our sample here does not impose any tenure restriction.