Breaking Bad: When Being Disadvantaged Incentivizes (Seemingly) Risky Behavior

Abstract

We investigate how variation in initial conditions, which assign individuals into advantaged or disadvantaged positions, alters behavior. We illustrate the problem within a labor market context and consider the impact of accumulated debt on wage selectivity. Using a two-period model, we show that debt exerts a non-monotonic effect on wage selectivity, with agents assigned low and high levels of debt being significantly more likely to reject an initial wage offer than agents with moderate debt. This prediction is supported by our experiment, which finds a statistically significant dip in wage selectivity for subjects assigned moderate levels of debt.

This is a preview of subscription content, access via your institution.

Fig. 1
Fig. 2
Fig. 3
Fig. 4
Fig. 5
Fig. 6

Notes

  1. 1.

    A similar finding is observed in Zhang et al. (2016) which shows that Chinese banks engage in riskier lending as their non-performing loan (NPL) ratio (NPLs over total loans) increases.

  2. 2.

    To focus attention on how agents respond to their accumulated debt balances, we choose to make an agent’s initial debt level exogenous. As such, we ignore an agent’s decision to accumulate debt, and we avoid potential concerns related to moral hazard and loss minimization which may arise in such a context.

  3. 3.

    Respondents were asked about money owed on mortgages, school loans, automobiles, and outstanding credit card balances.

  4. 4.

    Lusardi et al. (2011) reported that roughly half of Americans could not raise $2000 dollars in 30 days.

  5. 5.

    One could interpret our bankruptcy protection as either implementing complete wage garnishment or as allowing agents who receive a sufficiently low-wage offer in the second period to “walk away” from their debt. See "Theoretical Model" section for a more detailed discussion of these interpretations.

  6. 6.

    An agent is risk averse if \(\alpha \in (0,1)\), risk neutral if \(\alpha =1\), and risk loving if \(\alpha >1\).

  7. 7.

    We have considered more severe bankruptcy penalties (possible negative values) and find that our theoretical predictions are robust to moderate variations in this burden (results available upon request).

  8. 8.

    This distribution of risk preferences assumes that most agents will be moderately risk averse. However, it also allows for the possibility of risk-neutral and risk-loving agents. These assumed risk parameters are consistent with preferences observed in laboratory studies such as Holt and Laury (2002). See Fig. 7 in Appendix for the histogram of agents’ risk preferences used in our simulation.

  9. 9.

    Our results are robust to small variations in the size of this range. However, increasing or decreasing the range significantly, say by ± 5 percentage points, can significantly alter our results.

  10. 10.

    We have considered several alternative methods for introducing errors into the decision process and our results are consistent. The important issue is to introduce a degree of randomness into the wage acceptance decision and for this randomness to decline as the offered wage moves further away from the agent’s reservation wage.

  11. 11.

    While it is possible, in theory, to personally identify subjects by connecting their worker IDs to their individual Amazon profiles (Lease et al. 2013), we have not engaged in such a practice.

  12. 12.

    One benefit of experiments conducted through AMT is that observations can be gathered at a relatively low cost. For example, our entire experiment, which required the collection of over 500 observations, costs slightly less than $500. Conducting this same experiment within a laboratory setting would cost upward of $10,000. While lower subject payment helps facilitate the collection of more data, it also raises suspicion regarding the salience of incentives. To this end, we would like to stress two points: first, subjects are free to participate in our experiment own their own time and within the comfort of their homes. Second, and more importantly, our experiment represents a very small time commitment. (Median time to completion is only 4.15 min.) Thus, even though the total average payout is only $0.53, this extrapolates out to an hourly wage of $7.66.

  13. 13.

    It should also be noted that the SOEP question has been shown to be behaviorally relevant and highly correlated with subjects’ decisions in incentivized lottery experiments. See Dohmen et al. (2011) for results from a laboratory-in-field design and Gibson and Johnson (2019) for results from on online experiment under low stakes.

  14. 14.

    See Appendix for exact wording of questions. Note some questions are reverse coded.

  15. 15.

    We chose \(\$0.47\) as our upper limit as we want to ensure that all subjects could still receive a positive payout from accepting the initial wage offer.

  16. 16.

    We deliberately chose to have subjects make binary decisions rather than have them state a reservation wage or use a Becker–DeGroot–Marschak method (BDM) (Becker et al. 1964) to elicit reservation wages for two reasons. First, people are not often approached with BDM-like mechanisms or give stated reservation wages during job interviews. Instead, they are told what the job pays and can walk away if the pay is not high enough. Second, the instructions for BDMs are long and are confusing for some subjects. While it is tempting to make the argument that these types of mechanisms work well in the laboratory and thus should do well outside of the laboratory, this argument fails to take into account that subjects in the laboratory are (i) more educated than the AMT population (see the educational levels of the sample analyzed in Holt and Laury (2002) for example) and (ii) having the ability to cheaply ask clarifying questions. Thus, these more precise but also more complicated mechanisms have reinforcing drawbacks that make them less suitable for the population from which our sample is drawn, that is, the mechanisms are unfamiliar and it is difficult to ask questions. Having said that, it should be noted that the existing literature suggests binary decisions and strategy methods produce similar results (Brandts and Charness 2009).

  17. 17.

    A total of 589 subjects attempt the experiment but there are a few incomplete submissions because of accidental submissions.

  18. 18.

    The high average time spent on the HIT is driven by a handful of outliers. This is not surprising given that subjects can get up and do other things while working on the HIT. As previously discussed, the median time spent is only 4.15 min.

  19. 19.

    Note that the number of subjects in treatment blocks is slightly different. This is for two reasons: (1) debt was randomly assigned and (2) despite the safeguards some subjects managed to participate twice due to an unforeseen lag in assigning qualifications. While we were careful to not post a new batch before qualifications were assigned from our end, on a few occasions it turned out that there was a significant time lag between when we uploaded the qualification scores and when they were actually assigned to the subjects. We drop the second observation from subjects who participated twice. However, these subjects were paid for both observations because their participation a second time was not their fault.

  20. 20.

    Debt is controlled for using four dummy variables corresponding to each of the debt treatments. Full results available upon request.

  21. 21.

    We conducted another experiment to test the primary predictions of the model but in a somewhat more realistic scenario, with multiple wage offers and a real effort task. The alternative experiment is omitted to maintain consistency with our model but details and results (consistent with what is presented here) relating to this experiment are discussed in a previous version of the manuscript which is available upon request.

  22. 22.

    These cutoffs ensure that we have roughly an even number of subjects at each debt level, and that the range of subjective risk scores of the ER=0 group are two risk units off of the median and average reported subjective risk preference score. It should also be noted that the results displayed below are robust to small variations in these limits (results available from authors upon request).

  23. 23.

    Similar results are observed using other specifications and models and are available upon request.

  24. 24.

    Recall that higher attention scores actually reflect lower levels of attention by the subject. Also, as before, the results presented below are robust to small variations in these limits. Interestingly, the average reported attention score for the LA group is 14.38 which is more attentive than people diagnosed with AD/HD but much less attentive than a standard laboratory subject pool (Stanford et al. 2009).

  25. 25.

    We are grateful to an anonymous referee for making this point.

  26. 26.

    To save space, we omit the marginal effects for Models 2 and 3. However, these are similar to what is shown in panel a of Fig. 9 and are available upon request.

References

  1. Agranov, Marina, and Pietro Ortoleva. 2017. Stochastic choice and preferences for randomization. Journal of Political Economy 125 (1): 40–68.

    Article  Google Scholar 

  2. Barratt, Ernest S., J. Patton, and M. Stanford. 1975. Barratt impulsiveness scale. Austin: Barratt-Psychiatry Medical Branch, University of Texas.

    Google Scholar 

  3. Becker, Gordon M., Morris H. DeGroot, and Jacob Marschak. 1964. Measuring utility by a single-response sequential method. Behavioral Science 9 (3): 226–232.

    Article  Google Scholar 

  4. Bernstein, Asaf. 2015. ‘Household debt overhang and labor supply’

  5. Blanchflower, David G., and Andrew J. Oswald. 2013. Does high home-ownership impair the labor market?. Technical Report, National Bureau of Economic Research.

  6. Brandts, Jordi, and Gary Charness. 2009. ‘The strategy versus the direct-response method: A survey of experimental comparisons.’ Technical Report

  7. Dohmen, Thomas, Armin Falk, David Huffman, Uwe Sunde, Jürgen Schupp, and Gert G. Wagner. 2011. Individual risk attitudes: Measurement, determinants, and behavioral consequences. Journal of the European Economic Association 9 (3): 522–550.

    Article  Google Scholar 

  8. Gibson, John, and David Johnson. 2019. Are online samples credible? Evidence from risk elicitation tests. Atlantic Economic Journal 47: 377–379.

    Article  Google Scholar 

  9. Gruener, Sven. 2017. Correlates of multiple switching in the holt and laury procedure. Economics Bulletin 37 (1): 297–304.

    Google Scholar 

  10. Holt, Charles A., and Susan K. Laury. 2002. Risk aversion and incentive effects. The American Economic Review 92 (5): 1644.

    Article  Google Scholar 

  11. Jacobson, Sarah, and Ragan Petrie. 2009. Learning from mistakes: What do inconsistent choices over risk tell us? Journal of risk and uncertainty 38 (2): 143–158.

    Article  Google Scholar 

  12. Johnson, David and John Ryan. 2020. Amazon mechanical turk workers can provide consistent and economically meaningful data. Southern Economic Journal Forthcoming.

  13. Lease, Matthew, Jessica Hullman, Jeffrey P Bigham, Michael S Bernstein, Juho Kim, Walter Lasecki, Saeideh Bakhshi, Tanushree Mitra, and Robert C Miller. 2013. ‘Mechanical turk is not anonymous.’ Technical Report

  14. Lusardi, Annamaria, Daniel J. Schneider, and Peter Tufano. 2011. Financially fragile households: Evidence and implications. Technical Report, National Bureau of Economic Research

  15. McCall, John Joseph. 1970. ‘Economics of information and job search.’ The Quarterly Journal of Economics pp. 113–126.

  16. Oswald, Andrew J. 1996. ‘A conjecture on the explanation for high unemployment in the industrialized pantions: Part 1.’ Working paper. Warick Economic Research Papers (No. 474). Coventry: University of Warwick, Department of Economics.

  17. Pew Charitable Trust. 2015. The complex story of American debt. Philadelphia: Pew Charitable Trust.

    Google Scholar 

  18. Pew Charitable Trust. 2015. What resources do families have for financial emergencies? The role of emergency savings in family financial security. Philadelphia: Pew Charitable Trust.

    Google Scholar 

  19. Stanford, Matthew S., Charles W. Mathias, Donald M. Dougherty, Sarah L. Lake, Nathaniel E. Anderson, and Jim H. Patton. 2009. Fifty years of the barratt impulsiveness scale: An update and review. Personality and individual differences 47 (5): 385–395.

    Article  Google Scholar 

  20. Williams, Barry. 2014. Bank risk and national governance in Asia. Journal of Banking & Finance 49: 10–26.

    Article  Google Scholar 

  21. Zhang, Dayong, Jing Cai, David G. Dickinson, and Ali M. Kutan. 2016. Non-performing loans, moral hazard and regulation of the Chinese commercial banking system. Journal of Banking & finance 63: 48–60.

    Article  Google Scholar 

Download references

Author information

Affiliations

Authors

Corresponding author

Correspondence to John Gibson.

Additional information

Publisher's Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Appendices

Supplemental Appendix

Agents’ Simulated \(\alpha\) Parameter (Risk Preferences)

In Fig. 7, we present the simulated risk preferences for 50,000 agents. Each agent draws their risk, \(\alpha\), from N(0.8, 0.03) and thus assumes that most agents will be risk averse while also allowing for risk-loving agents. These preferences were selected to match the distribution of risk preferences often observed laboratory studies.

Fig. 7
figure7

Simulation results: histogram of risk preferences

Risk Preferences and Attention

In Fig. 8 we present the CDFs of subjective risk preferences (Fig. 8a) and reported attention scores (Fig. 8b) for subjects participating in the experiment by assigned debt.

Fig. 8
figure8

CDF of subjective risk preferences and reported attention scores

In Table 5, we present the p-values from a set of t-tests and U-tests, testing for differences in the means and distributions of risk preferences and reported attention scores across each of the assigned debts. While subjects with an assigned debt of $0.25 on average tend to be slightly less risk averse, this is being driven by the fact that only a single subject with an assigned debt of $0.25 indicated a subjective risk preference of 0, compared to an average of six for the other debt treatments. When we compare average subjective risk preferences while omitting subjects who report a subjective risk preference of 0 (p-values shown in brackets in Table 5), the average subjective risk preferences and distribution of subjective risk preferences across debts are not significantly different. However, as can be observed in Table 5, the means and distributions of reported attention scores are not statistically different across any of the assigned debts.

Table 5 Means and distribution tests results for risk preferences and attention across debts

Robustness Check of Result 2

As a robustness check of Result 2, we run a series of probits estimating the impact of assigned debt and classifications on the likelihood of rejecting the initial wage offer (Table 6). Model 1 only controls for the subjects’ assigned debt (DEBT) and its square (\(\hbox {DEBT}^2\)) and provides a basic robustness check for proportions tests discussed in Result 1. Models 2 and 3 introduce indicator variables to control for “extreme risk” (ER) and “low attention” (LA) among subjects, while Models 4 and 5 also control for interactions between these classifications and assigned debt. Lastly, Model 6 is a full specification with controls for the combined classifications (HANR, HAER, LANR, and LAER, with HANR being the omitted group), as well as interactions with debt.

Generally, the results presented in Table 6 correspond with what is discussed in the main text. While we find that the coefficient estimates on DEBT and \(\hbox {DEBT}^2\) are statistically significant and have their expected signs across all specifications, we focus our attention on the computed marginal effect of debt on the probability of rejection. Figure 9 presents the marginal effect of debt on the probability of rejecting the initial wage for Models 1, 4, 5, and 6.Footnote 26 Inspection of panel a of Fig. 9 indicates that debt exerts a statistically significant non-monotonic effect on wage selectivity. Specifically, at low debt levels, the marginal effect is found to be negative and statistically different from zero. However, the marginal effect eventually crosses zero and becomes positive and significantly different from zero at high-debt levels. Furthermore, panels b, c, and d of Fig. 9 provide additional support for Hypothesis 2. Specifically, the marginal effect is found to be steepest and most significant for agents with moderate risk preferences (ER = 0), high attention (LA = 0), and the combination of both previous characteristics (HANR).

Table 6 Debt responses by type
Fig. 9
figure9

Marginal effect of debt on probability of rejecting the first wage

Experiment Instructions

Below we present the experimental instructions used in all sessions of the experiment. “XX” indicates the assigned the debt which is either 0, .13, .25, .35, or 0.47. Brackets contain text that is not seen by subjects.

Informed Consent

This research is being conducted by Dr. XXX and Dr. XXX who are professors at the XXX and XXX, respectively.

I chose to voluntarily participate in this research study. I have been recruited for this study through Amazon Mechanical Turk. Only persons 18 years of age or older may participate. I affirm that I am 18 years of age or older. Only individuals who read and write English may participate. I affirm that I can read and write in English. This study has been approved by the XXX Research Ethics Board.

This study involves completing a demographic survey in the first stage and a short task in the second stage. The survey will take less than 5 min. The task will take less than 2 min. I will be paid 25 cents for completing the survey and a bonus. The amount of my bonus will be based on my decisions and the outcome of a randomly generated number.

I am free to withdraw from the study at any time without incurring the ill will of the researchers. If I withdraw, my data will not be used and will be deleted by the researchers as early as possible. If I wish to withdraw, I must do so within 20 days of completing the study. There are no known risks or benefits from this study beyond those from any typical activity I might do in an online environment. This study will benefit society by helping researchers better understand how individuals respond to incentives in labor markets. The confidentiality of any personal information will be protected to the extent allowed by law.

My name or AMT account number will not be reported with any results related to this research. I can obtain further information from Dr. XXX. If I have any questions about this study, I can contact Dr. XXX at XXXX.edu. If I have any questions about my rights as a participant, I should contact the Human Subjects Protection Program at (XXX) XXX-XXXX. I may ask questions at any time via email (XXXX.edu).

Should new information become available during the course of this study, about risks or benefits that might affect my willingness to continue in this research project, it will be given to me as soon as possible.

By clicking on the start button below, I am indicating my consent to participate in this study.

If you do not wish to participate, please return the HIT.

Survey

Please complete the survey below. Doing so you will earn your 25 cents participation fee in addition to your bonus that will be determined by your decisions in the second stage.

What is your gender?

How do you see yourself: Are you generally a person who is fully prepared to take risks or do you try to avoid taking risks? Please tick a box on the scale, where the value 0 means: “risk averse” and the value 10 means: “fully prepared to take risks.” You can use the values in between to make your estimate.

DIRECTIONS: People differ in the ways they act and think in different situations. This is a test to measure some of the ways in which you act and think. Read each statement and click on the appropriate circle on the right side of this page. Do not spend too much time on any statement. Answer quickly and honestly.

[Barratt Impulsiveness Scale Questions]

  • I plan tasks carefully.

  • I do things without thinking.

  • I make-up my mind quickly.

  • I am happy-go-lucky.

  • I don’t “pay attention.”

  • I have “racing” thoughts.

  • I plan trips well ahead of time.

  • I am self controlled.

  • I concentrate easily.

  • I save regularly.

  • I “squirm” at plays or lectures.

  • I am a careful thinker.

  • I plan for job security.

  • I say things without thinking.

  • I like to think about complex problems.

  • I change jobs.

  • I act “on impulse.”

  • I get easily bored when solving thought problems.

  • I act on the spur of the moment.

  • I am a steady thinker.

  • I change residences.

  • I buy things on impulse.

  • I can only think about one thing at a time.

  • I change hobbies.

  • I spend or charge more than I earn.

  • I often have extraneous thoughts when thinking.

  • I am more interested in the present than the future.

  • I am restless at the theatre or lectures.

  • I like puzzles.

  • I am future oriented.

Stage 2:

The second stage of the experiment has 2 parts. In the first part, you will be offered a wage of 25 cents. During the first part, you will also be assigned a debt of XX cents.

You can either accept or reject the wage you are offered. If you accept your first wage you will “work” for two periods and will earn two times your wage (because you worked two periods) minus your assigned debt. Consequently, if you accept, your earnings will be cents (that is, 2*25) minus XX cents (plus your 25 cents participation fee).

Please click the “Next” button to continue.

Stage 2 (continued):

You also have the option to reject the wage of 25 cents. If you reject, you will draw another wage. There is a fifty percent chance you will be offered a wage of 25 cents and a fifty percent chance you will be offered a wage of 75 cents. Your debt of XX cents will be the same as what you were assigned in the first part.

If you reject the first wage, you only work 1 period. Therefore, if you reject the first wage you are offered, your earnings will be either 25 or 75 cents minus XX cents (plus your 25 cents participation fee).

Under certain circumstances it is possible for you to have negative earnings. If this happens, you will earn a bonus of 0 cents but will still be paid your participation fee of 25 cents.

Please click the “Next” button to continue.

Decision Screen

You have been assigned a debt of XX cents.

You have been offered a wage rate of 25 cents. If you accept, you will earn a bonus of XX cents.

If you reject, there is a fifty percent chance you will earn a bonus of XX cents and a fifty percent chance you will earn a bonus of XX cents.

Do you accept your wage of 25 cents?

Rights and permissions

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

Gibson, J., Johnson, D. Breaking Bad: When Being Disadvantaged Incentivizes (Seemingly) Risky Behavior. Eastern Econ J 47, 107–134 (2021). https://doi.org/10.1057/s41302-020-00172-6

Download citation

Keywords

  • Debt
  • Wage acceptance
  • Amazon Mechanical Turk

JEL classification

  • C9
  • J31
  • D21