Penalty contracts: is it all about paying the cash upfront?

We experimentally investigate the relevance of a cash upfront payment for the effort-enhancing effect of penalty contracts. We find that penalty contracts where participants receive the upfront payment physically before working on a real effort task induce more effort than penalty contracts where participants are only informed about the upfront payment. When comparing penalty contracts with economically-equivalent bonus contracts, we find that penalty contracts lead to higher effort provision than bonus contracts, but only if participants physically receive the upfront cash payment in advance. We suggest that the higher salience of the cash upfront payment might be a core driver of the detected framing effect. Our findings emphasize the importance of experimental design choices when studying framed incentive contracts.


Introduction
Incentive contracts are commonly used to motivate employees (Prendergast 1999), and research shows that framing may affect their effectiveness. In particular, experimental evidence reveals that negatively framed incentive contracts that penalize poor performance, can induce more effort than economically-equivalent, positively framed incentive contracts rewarding good performance (e.g., Hannan et al. 2005). One explanation for the effort-enhancing effect of negatively framed incentives is loss aversion around a reference point, formalized in prospect theory (Kahneman and Tversky 1979).
However, while some experiments find large and statistically significant framing effects in the laboratory with Hedges' g statistics around 0.50 (e.g., Armantier and Boly 2015;Hannan et al. 2005;Imas et al. 2017;von Bieberstein et al. 2020) other studies report smaller and only marginally significant effects (e.g., Brooks et al. 2012), and still others do not find a statistically significant effect of contract framing on effort at all (e.g., Essl and Jaussi 2017;Grolleau et al. 2016). 1 Similarly, a recent meta-analysis on framed incentive contracts points out a high variability of effect sizes found in laboratory experiments and much smaller effect sizes for field than for lab studies (Ferraro and Tracy 2022). The meta-analysis also documents larger effect sizes in studies where workers get their pay handed out in advance compared to studies that only verbally describe the payment but factually pay it out at the end of the task (0.24 SD vs. 0.08 SD), but the difference is not statistically significant. However, the studies included in this comparison also differ in many other design choices, making it hard to assess the impact of the timing of the payment in isolation.
Thus, the aim of this study is to systematically examine two questions: First, do penalty contracts with cash upfront payment elicit significantly more effort than penalty contracts without cash upfront payment? Second, does the penalty contract have a performance-enhancing effect compared to the bonus contract if the cash payment is physically paid out up front, but not if the payment is only announced and paid later? To answer these questions we conducted a between-subject laboratory experiment with two stages. In both stages, participants work on a real effort task. In stage 1, they receive a fixed wage independent of performance and in stage 2, they work under a framed incentive contract. Depending on the treatment, the incentive contract is either a penalty contract with cash upfront payment, a penalty contract without cash upfront payment, or a bonus contract. All contracts are economically equivalent and participants receive a substantially higher payoff when meeting an announced performance target (15 CHF vs. 5 CHF). 2 Comparing both types of penalty schemes, we find that penalty contracts with cash upfront payment elicit significantly more effort than penalty contracts without cash upfront payment. However, if we control for performance in stage 1, the effect is only significant at the 10-percent level. With respect to the framing of the contract, we find that penalty 1 An overview of laboratory experiments analyzing the effectiveness of penalty contracts is available in Table 4 in Appendix A. 2 At the time of the experiment, 1 CHF equaled about 1 US$. contracts with physical cash upfront payment elicit significantly more effort than economically-equivalent bonus contracts. In contrast, we do not find a statistically significant effort difference between the penalty contract without cash upfront payment and the bonus contract.
There are several potential reasons why the cash upfront payment could be important for producing the effort-enhancing effect of penalty contracts (see the discussion in Sect. 4). Our preferred interpretation is that our results are due to an increase in the salience of the upfront payment, which in turn reinforces the feeling of ownership and induces a greater fear of losing the money. This is in line with a study of Reb and Connolly (2007), who find that people who physically possess a good have a higher monetary valuation for keeping it compared to people who only factually own the good without physical possession.
Up to this point, there are only a few attempts to bring more clarity in the experimental design choices researchers face when studying framed incentive contracts in the lab. One exception searching for systematic differences is de Quidt et al. (2017), who suggest that whether participants can check if they meet a performance target while performing the task, or not, influences the effectiveness of contract framing. However, they did not find causal evidence supporting this conjecture and the authors acknowledge that there are many other differences between the studies they considered in their review of the literature. In addition to the timing of payment, Ferraro and Tracy (2022) also consider the type of payment (piece-rate vs. threshold contract design) in their meta-analysis, identifying comparable effect sizes for both types of payment. In this paper, we experimentally vary the timing of the payment as a driver of differences in the framing effect. While most lab experiments with cash upfront payment find an effort-enhancing effect of penalty contracts as compared to bonus contracts (with the exception of Grolleau et al. (2016), who use a cash upfront payment but do not find a statistically significant framing effect), the results of laboratory studies without cash upfront payment are more mixed with some studies documenting an effort-enhancing effect and others failing to do so (see Table  4 in Appendix A).

Experimental design and procedure
In a lab experiment, we examine the importance of physical possession of the upfront payment as a potential driver of framing effects. 3 We conducted two penalty treatments, where participants received 15 CHF upfront and had to pay a penalty of 10 CHF in case of missing the target. In the Penalty Cash treatment, participants were physically paid out their upfront payment in cash prior to the task, whereas in the Penalty No Cash treatment, they were only informed about the upfront payment, but did not physically receive the money until the end of the experiment. As a control treatment, we implemented an economically-equivalent Bonus treatment, where participants were paid out 5 CHF as an upfront payment and when meeting the predefined target, they earned an additional bonus of 10 CHF.
At the beginning of the lab experiment, we assessed individual loss aversion (Gächter et al. 2022). Subjects saw a list of six lotteries and decided for each lottery, whether to accept or reject it. Each lottery had a 50% probability of winning 6 CHF and a 50% probability of losing 2 CHF up to 7 CHF, with the losing price increasing in increments of 1 CHF. At the end of the experiment, one lottery was randomly chosen and paid out. This first part was followed by two stages, where participants worked on a Word Encryption task with Double Randomization (WEDR task) (Benndorf et al. 2019). 4 Participants had to encode words as numbers and could only proceed with the next word if they encoded all letters correctly. Effort was measured as number of solved words. We chose this task for several reasons: it requires no special knowledge or cognitive abilities, learning possibilities are trivial, there is no scope for guessing, and it is gender-neutral (Benndorf et al. 2019). In stage 1, participants worked on the encryption task for 3 min and received a fixed wage of 5 CHF, irrespective of the number of solved words. Participants familiarized themselves with the task and we obtained a measure of baseline performance capturing motivation and ability.
In stage 2, participants had 4 min to encode a maximum of 20 words. Payment was performance-based and tied to an announced target. If participants solved at least 12 words, they received 15 CHF, otherwise their payment was 5 CHF. Solving 12 words corresponds to the 80th performance percentile under piece-rate incentives in a similar experiment (von Bieberstein et al. 2020), such that we chose it as challenging, but feasible target. In our experiment, participants solved on average 10.09 words (S.D.=1.7), and 19% reached the target of 12 words. While working in stage 2, participants could see a display of the number of correctly coded words so far. To keep payment procedures consistent over both stages of the experiment, the fixed wage of 5 CHF in stage 1 was always executed in the same way as the upfront payment in stage 2. This means that participants in the Penalty Cash treatment and the Bonus treatment physically received their fixed wage before working on the real effort task. In the Penalty No Cash treatment, the fixed wage was not distributed in advance, but only at the end of the experiment. Given the differences in the payment procedure of stage 1, we will check in our analysis for performance differences in this stage depending on the treatment (and find none, see Sect. 3). Moreover, to ensure confidentiality and comparability across all treatments, the returned penalty payments and the bonus payments in the treatments with physical possession were administered via envelopes. This way neither the experimenter distributing and collecting the payments, nor any other participant could infer if the target was met. 5 The experiment was conducted in the Aare-Lab of the University of Bern between December 2019 and February 2020. Subjects were recruited via Sona Systems, and the experiment was computerized using z-Tree (Fischbacher 2007). In total, 195 students from various disciplines participated. Based on prior research, we expected a medium sized effect of contract framing on productivity (Cohen's d = 0.50). Thus, with at least 67 participants per treatment, we have 80% power to detect such an effect at a 5% level of significance. The participants were randomly assigned to one of the three treatments, which in turn were randomized over morning, midday, and afternoon sessions. Participants received written instructions for the real effort encryption task and had to answer control questions to ensure that they understood the performance-based incentives in stage 2. Each session lasted about 45 min and average earnings were 17 CHF including a show-up fee of 4 CHF. The experiment concluded with a short questionnaire on demographics. 6

Results
To analyze the effect of physical possession of the base pay on effort provision, we compare the number of correctly solved words across treatments. Table 1 reports descriptive statistics for the number of words solved in stage 2 and the baseline performance measured by the number of words solved in stage 1. First, given the differences in the payment procedure of stage 1, we check for performance differences in this stage depending on the treatment. Baseline performance in stage 1 is similar in all three treatments, ranging from an average of 6.44 to an average of 6.69 correctly solved words. None of the pairwise treatment comparisons for stage 1 is statistically significant (Mann-Whitney rank sum tests, see Table 1 below for p-values). 7 Thus, paying out the cash upfront when working under a fixed wage does not lead to statistically different performance levels across the treatments compared to paying the cash at the end of the experiment.
Looking at performance levels in stage 2 and comparing the two penalty contracts that only differ in terms of the timing of the physical payment reveals a significant difference with the cash upfront payment outperforming the penalty contract without physical upfront payment (p = 0.025, Mann-Whitney rank sum test). Results further reveal that participants who worked under a penalty contract with physical upfront payment solved significantly more words than those who worked under a bonus contract (p = 0.003). In contrast, participants in the Penalty No Cash treatment did not exert more effort than participants in the Bonus treatment (p = 0.566).
In addition to the descriptive results, we conducted an OLS regression analysis. First, we investigate the question of whether the money at stake is paid out in cash upfront is a driver of the effort-enhancing effect under penalty framed incentive contracts. In Specification 1 of Table 2 we regress the number of solved words in the encryption task of stage 2 on the treatment dummy variable Penalty Cash, which is 1 if the individual participated in the Penalty Cash treatment and 0 for participants 6 Our experiment was always followed by one of two other experiments such that the total session duration was either 60 or 90 min. As expected, neither the number of words solved in stage 1, nor in stage 2 differ statistically significantly between the shorter and longer sessions (stage 1: p = 0.332, stage 2: p = 0.853; Mann-Whitney rank sum tests, two-sided.) 7 All statistical tests are two-sided.
in the Penalty No Cash treatment. We also include baseline performance, which is the number of solved words under the fixed wage in stage 1. As expected, participants' baseline performance in stage 1 positively predicts the number of solved words under performance-based incentives. Results show that penalty contracts with physical upfront payment induce more effort than penalty contracts without physical upfront payment. However, compared to the descriptive statistics, the inclusion of baseline performance weakens the significance level of the Penalty Cash coefficient (p < 0.10). In Specification 2, we further include loss aversion, which is represented by the number of rejected lotteries in the loss aversion test, and whether the participant is female, or not. Including these controls does not alter the results. Since we chose a challenging target, it could be that low performers as measured by their performance in stage 1 have given up pursuing the target irrespective of the treatment. 8 When conducting the regression analysis excluding the lowest ~ 10% or the lowest 20% of the performers, 9 the coefficient for Penalty Cash increases in magnitude and is statistically significant on the 5% level (model 2, coefficient: 0.491, p-value = 0.042 excluding the lowest ~ 10%; coefficient: 0.621, p-value = 0.016, excluding the lowest 20%).
Next, we analyze the effort-enhancing effect of penalty contracts with and without a cash upfront payment compared to the bonus contract. Specifications 1 and 2 of Table 3 report the corresponding estimates. Note that both treatment dummies Penalty Cash and Penalty No Cash are 0 for the Bonus treatment and take on the value 1 if the participant worked under a penalty contract with a cash upfront payment or without a cash upfront payment, respectively. In line with the descriptive statistics, the results show that when a cash upfront payment is in place, penalty contracts induce significantly more effort than economically-equivalent bonus contracts. The exclusion of the lowest ~ 10% or lowest 20% of performers for this analysis leads to an increase in magnitude of the treatment effect and shows a (highly) statistically significant difference comparing the Penalty Cash to the Bonus treatment (model 2, coefficient: 0.513, p-value = 0.033, excluding the lowest ~ 10%; coefficient: 0.748, p-value = 0.003, excluding the lowest 20%). In contrast, the exclusion of low performers has no effect on the significance levels when comparing the Penalty No Cash to the Bonus treatment. In sum, without the cash upfront payment, there is no effort-enhancing effect of the penalty contract compared to the bonus contract. We therefore conclude that in our experiment cash upfront payment is crucial for enhancing effort under penalty contracts.

Discussion and conclusion
Experimental evidence on the effectiveness of penalty contracts is ambiguous (see Table 4 in Appendix A). A closer look at the literature reveals that existing studies differ considerably in their design choices. We contribute to the ongoing debate about the differing methods by examining the relevance of the cash upfront payment as a driver of contract framing effects. First, we compare effort provision when working under two penalty contracts that only differ in terms of the timing of the physical payment. In the Penalty Cash treatment, participants solved significantly more tasks as compared to the Penalty No Cash treatment. Second, we find that in comparison to the Bonus treatment, the Penalty Cash treatment leads to significantly higher effort provision, but not the Penalty No Cash treatment. Based on our findings, we conclude that the effort-enhancing effect of penalty contracts is considerably driven by a cash upfront payment handed out prior to working on the task.
There are several potential reasons as to why the cash handout in the Penalty Cash treatment fosters higher effort provision. First, one potential driver of the effort-enhancing effect of the cash payment could be positive reciprocity (Falk and Fischbacher 2006;Rabin 1993). If people regard the cash upfront payment as a kind action of the experimenter, they might return more effort. Therefore, one possibility is that positive reciprocity in response to receiving an upfront cash payment induces higher effort irrespective of any fear of losing the money. To analyze this possibility, we use stage 1 of the experiment and examine the effect of upfront cash payments on effort under a fixed wage. Pairwise comparisons of the effort in stage 1 reveal that there are no statistically significant differences in baseline performance across the treatments (see Table 1 in the Results Section). This suggests that the effortenhancing effect of the physical upfront payment under penalty contracts is likely not due to a positive reciprocation of the cash upfront payment. Second, handing out the cash upfront could affect participants' beliefs that it is highly likely or the norm to reach the target. However, while we cannot rule out this possibility, we believe that it is rather unlikely, given that participants got well acquainted with the task in stage 1 and received feedback about their own performance, allowing them to use this knowledge in their assessment of stage 2. Finally, a more likely driver behind our results could be that holding the cash upfront payment in one's hands increased its salience and in turn induced a stronger feeling of ownership. Reb and Connolly (2007) experimentally examined the effect of physical possession of a good in contrast to factual ownership on participants' subjective feeling of ownership. They concluded that the endowment effect (Thaler 1980), which states that people tend to value a good more highly when they own it, results from perceived ownership induced by physical possession. Physical possession could thus induce a stronger shift of the reference point (Kahneman and Tversky 1979) and therefore, given loss aversion, foster effort provision. 10 Following this argumentation, our results might be due to an increase in the salience of the upfront payment, which in turn reinforces the feeling of ownership, shifts the reference point, and thus induces a greater fear of losing the money. Alternatively, physical possession could also change the salience of an otherwise identical shift of the reference point. For instance, in the study of mixed feelings, Kahneman (1992) argues that people can have multiple reference points at a given point in time and salience is one factor determining which of these reference points receives a higher weight for the perceptions of the decision maker. Further research is warranted to provide a deeper understanding of the underlying mechanisms.
Furthermore, future research could deepen our understanding of the driving factors behind framing effects and could broaden our understanding of the interplay with different factors. For example, Czibor et al. (2022) find that penalty contracts do not lead to an increase in performance, but they observe an increase in theft, suggesting that there are important behavioral spillover effects of framed contracts. This could further be extended to the interplay between supervisor remuneration and employee engagement (e.g., Hendriks et al. 2022 for CEO compensation and employee effort). In addition, the threat of employees having to actually pay back money could have an effect on the social sustainability of the firm (Tipu 2022), that future research could investigate. Related to our design choices, we believe that the implemented real effort task (Benndorf et al. 2019), as well as the choice and parametrization of the incentive contracts conditioning a substantial fraction of the maximum payoff on a challenging, yet achievable, performance target might be conductive to inducing a relatively large framing effect. Learning more about framing effects in different tasks or with different targets could prove to be an important avenue for future research.
Our findings emphasize the importance of experimental design choices when studying framed incentive contracts. With respect to the implementation of framing experiments in the laboratory our contribution is the following: In order to use scare resources efficiently, we recommend including cash upfront payments when aiming for eliciting a contract framing effect or analyzing its consequences. Our findings suggest physical possession of the money as a simple and valid tool to successfully induce framing effects in a laboratory setting.