The gift of being chosen

Abstract

We report evidence from an experiment where an employer selects one of two workers to perform a task for a fixed compensation. Workers differ in their ability. The employer’s payoff depends on the worker’s ability and on a non-contractible effort that the worker exerts once employed. We find that selected workers exert an effort higher than the minimum enforceable one. When the employers can send a free-text form message to the selected worker, workers with low ability exert significantly higher effort than the workers with high ability. The difference in effort overcompensates the difference in ability.

1 Introduction

The gift-exchange hypothesis is one of the central themes of behavioral contract theory. It asserts that an employer could find it profitable to offer a compensation higher than the market clearing wage (i.e., a ‘fair’ or ‘kind’ wage) to trigger a reciprocal effort by a worker. Originally formulated by Akerlof (1982), this theory has generated an extensive theoretical and experimental literature that analyzes the relation between the level of workers’ compensation and their productivity.¹ Thereby, research has focused on the employer–worker relation once the worker has been selected, when the decision of the employer only regards the level of compensation to offer. Recently, Baron contributed to the debate on reciprocity and gift exchange, formulating the notion of ‘empathy wage’. Baron (2013) documents that a gift offered by the employer to the worker is likely to elicit more gratitude among workers who are relatively disadvantaged and in the lower part of the performance distribution (defined as ‘non-stars’). Baron evidences how, in some cases, the magnitude of the difference in gratitude between the star and non-star workers is sufficiently large that it offsets the difference in productivity, suggesting an economic convenience of hiring non-star versus star workers.

In this paper, we aim to extend the literature on reciprocity and gratitude by investigating the selection phase of a job relation. Specifically, we study whether the choice of one worker over another influences the effort decision of the selected worker by eliciting gratitude. In fact, most job offers do not only include a certain scope of action for the employer to decide about the wage level, but they also include a stage in which the worker is chosen among a set of candidates and is informed about the employer’s expectations. We provide experimental evidence supporting the hypothesis that a competitive selection phase can motivate the (chosen) worker to exert a level of costly effort larger than the minimum enforceable one. We also show that, when the employer can send a message to the selected worker, the notion of ‘empathy wage’ can explain our evidence. The low ability workers exert significantly higher effort than the high ability ones. Their effort choice overcompensates the difference in ability, translating into higher profits for the employers.

In order to focus on the worker’s gratitude in the selection stage we design a simple experiment where an employer selects a worker among a set of candidates offering a fixed flat wage. The candidates differ in the level of observable ability, and the employer’s payoff positively depends on both the ability and the effort exerted by the selected worker, once hired. In this setting, the gift received by the worker is ‘being chosen by the employer’, so we expect the hired workers to exert effort in order to express their gratitude. According to the ‘empathy wage’ notion, low ability workers should express more gratitude than high ability ones. The rationale supporting this prediction relies on the psychological theory of counter-factual thinking: workers’ gratitude is also affected by their status in relation to other workers and by an assessment of how much worse, or better, things might have turned out. Finally, when the employer can communicate his intentions to the hired worker, we expect that she is able to reinforce the worker’s gratitude and induce higher effort provision.

We compare three main treatments. In the first two, the ‘No Communication Treatment’ and the ‘Communication Treatment’, the employer selects one worker out of two for a chosen effort task. In the Communication Treatment, simultaneously to the selection, the employer can send a free-form text message to the chosen candidate. In the third treatment, denoted ‘Random Device Treatment’, the worker is selected via a random device.

Average effort exerted by workers is statistically different than the minimum enforceable level in all treatments. When looking at effort exertion depending on workers’ abilities, in both the No Communication Treatment and the Random Device Treatment the average effort exerted is not significantly different both within and between treatments. When communication is allowed, workers with low ability exert, on average, a significantly higher effort than (i) workers with high ability in the Communication Treatment and (ii) workers (with high and low ability) in the No Communication Treatment and in the Random Device Treatment. High ability workers exert, on average, the same effort in all treatments.

In line with Baron (2013), our results evidence that choosing the worker who has the greater ability is not always the best choice for the employer. Even though in our setting the differences in ability among the workers are small, once we include communication, the effect on the profit of the employers is remarkable: when a low ability worker is hired, profits are on average 41 % higher compared to the case in which a high ability worker is hired. When analyzing the content of the messages sent, we find that most messages contain a suggestion of an effort level and refer to some kind of fairness. By means of a control treatment denoted ‘Suggestion Treatment’ (where simultaneously to the selection, the employer is allowed to send a numerical suggestion to the chosen candidate, without additional words), we show that the increase in effort is not mainly driven by facilitated coordination or signaling of the employer’s expectation, but by a positive effect on low ability worker’s motivation.

Our experiment does not conclusively answer the question of which behavioral motivation induces low ability workers to exert more effort than high ability workers. The average effort exerted does not differ depending on the hired workers’ ability when comparing the Random Device Treatment to the No Communication Treatment. Alternative theories, such as guilt aversion (Charness and Dufwenberg 2006) or social esteem (Ellingsen and Johannesson 2008) may explain this result, yet our experiment was not designed to test these theories against each other. Results are consistent with the ‘empathy wage’ when looking at the Communication Treatment. We show that, once selected, low ability workers do actually respond with a significantly higher effort than high ability ones, but only when employers can make their choice salient through communication. Communication is present in almost all real-life interactions resembled by our experiment. Therefore, behavioral responses may be underestimated in studies that focus only on the possibility for the employer to choose the wage level but not which worker to hire and do not include, or limit, the possibility of communication between parties.

Our paper contributes to the experimental literature that investigates the role of communication. Communication has been shown to foster both efficiency and pro-social behavior in many different games. In public good games, communication reduces free-riding behavior, increasing voluntary contributions (Isaac and Walker 1988; Koukoumelis et al. 2012). In commons dilemma situations, it helps in having a more effective sanctioning system (Dawes et al. 1977; Ostrom et al. 1992), and in coordination games it facilitates coordination (Cooper et al. 1992; Blume and Ortmann 2007; Brandts and Cooper 2007). In ultimatum games, the possibility for a second mover to send a reply to the proposer in which she can express her emotions lowers the rate of unfair offers that are rejected (Xiao and Houser 2005). In trust games, one-way communication from second movers to first movers increases the level of trust (Charness and Dufwenberg 2006): when second movers may signal their intentions before the trust game starts, their promises enhance trustworthy behavior. In gift exchange games, one-way communication from employees to managers promotes reciprocal behavior, because employees advice managers to set higher wages that eventually increase managers’ payoffs, too (Cooper and Lightle 2013). In our experiment, one-way communication goes in the other direction, from first movers to second movers: employers hire an employee at a fixed wage and simultaneously can send a message. Messages therefore can be used to make explicit employers’ expectations regarding employees’ behavior, and motivate employees to exert more effort, fostering their gratitude for being chosen.

Finally, our work provides an alternative explanation for the hiring of candidates who have a lower ability than others. Up to now, this issue has been investigated by placing a focus on the economic origins and consequences of favoritism in groups (Bramoulle and Goyal 2011; Prendergast and Topel 1996), or on nepotism (Levine et al. 2010). However, while research on nepotism and favoritism considers the choice of workers who have a low ability as leading to inefficiency in organizations, our analysis provides a different perspective: our experimental evidence identifies situations where it is beneficial to the employer to hire a worker with lower ability than others available.

The rest of the paper is organized as follows. Section 2 describes our experimental design. Section 3 clarifies our predictions. Section 4 illustrates our procedures. Section 5 presents our main results. Section 6 concludes.

2 Experimental design

We consider a game of complete information where an employer has to select one employee out of two candidates, workers 1 and 2. The employer chooses a worker and uses all her endowment to pay a fixed wage of 50 ECUs to the selected worker.² The selected worker decides an effort level by choosing a discrete integer in the interval $e=1,2,\ldots ,10$. Effort is costly, with the cost of effort c(e) being stictly increasing in the level of exerted effort (we use the same cost function as in Fehr et al. 1998). All workers experience the same cost of effort, but they have different abilities. Denote $w^k$, with $k=L,H$ the worker with ability $\theta ^k$; we assume $\theta ^{H}=0.5$ and $\theta ^{L}=0$. The monetary payoff of the employer who hires a worker of ability k is $5(e+\theta ^{k})$. It therefore depends positively both on the worker’s ability and effort provided. The chosen worker receives a monetary payoff equal to $m=50-c(e)$. If the selected worker does not accept the employer’s offer, then the game ends and both he and the employer receive 0 ECUs.³ The candidate who is not selected receives an unemployment benefit of 10 ECUs. Effort levels, relative costs and earnings of the employer and the workers are depicted in Table 1.

Table 1 Costs of effort c(e), earnings of the worker m and earnings of the employer in case she hires a high ability, $\pi (w^{H})$ and a low ability worker $\pi (w^{L})$, for each level of effort e

Note that, given the choice of our parameters, an employer finds it profitable to hire a low ability worker only if she expects that he exerts an effort at least one level greater than the high ability worker’s effort.

We run four treatments. In the Random Device Treatment (RDT), a random device selects one worker to be hired. In the No Communication Treatment (NCT) and in the Communication Treatment (CT), the employer selects one worker to be hired. In the CT, when hiring the worker, the employer can send a free form message. The content of the message is not restricted, and it is public information that messages are non-binding and costless in the common sense.⁴ Finally, in the attempt to disentangle the effect of expectation and non-verbal communication, we run as control the Suggestion Treatment (ST). The ST is identical to the CT with the only exception that messages can only suggest a number representing the effort desired by the employer. In all treatments, first the worker learns that he has been selected (and in the CT and ST simultaneously reads the employer’s message / numerical suggestion) and then he chooses an effort.

3 Hypothesis

In this section, we briefly present the three hypotheses we aim to test with our experimental design.

Our first hypothesis predicts that the selection process produces gratitude toward the employer: workers’ reciprocal response toward the employer should be larger when the selection is made by the employer rather than by a random device.

Hypothesis 1

In all treatments the workers’ selection triggers a provision of effort higher than the minimum enforceable one. Effort exerted in the RDT is lower than in the NCT and in the CT.

The first part of this hypothesis is based on the predictions of distributional preferences models (see, e.g., Bolton and Ockenfels 2000; Charness and Rabin 2002; Fehr and Schmidt 1999). In our experiment reciprocity cannot fully display since the wage is fixed. However, Charness (2004), shows that, in a standard gift exchange experiment, outcome based other regarding preferences affect workers’ effort choice. In our setting effort exertion above the minimum enforceable level is compatible also with inequality aversion or efficiency concerns.

The second part predicts that the employer’s intentions matter in explaining the effort choice of the worker; being chosen is a kind of gift which could induce some gratitude by the selected worker. So, if the intentions of the employer are relevant, then we should observe that the average effort exerted in the NCT and CT treatments is higher than the level exerted in the RDT (in a similar vein to Geanakoplos et al. 1989; Rabin 1993; Dufwenberg and Kirchsteiger 2004).

Our second hypothesis refers to the impact of workers’ ability on effort exertion:

Hypothesis 2

In the NCT and the CT low ability workers exert more effort than high ability workers.

This hypothesis is related to the ‘empathy wage’ notion formulated by Baron (2013): given the small but positive differences in abilities, high ability workers may feel more entitled to get the job than low ability candidates. Therefore, in both the NCT and the CT the same gift (i.e., being selected) should arouse more gratitude by the workers who are relatively disadvantaged (i.e., the low ability workers) compared to the high ability workers. Moreover, choosing the low ability worker is a ‘riskier’ action for the employer; actually, this is the only way (given that the wage is fixed) for the employer to signal her kindness and, consequently, to induce higher reciprocal response by the worker (Levine 1998; Ellingsen and Johannesson 2008).

Our last hypothesis refers to the effect of communication:

Hypothesis 3

When receiving a message from the employers, selected workers exert on average a higher effort compared to the NCT, where communication is not allowed.

As in our experiment the kindness of the action of the employer in choosing the respective employee may not be sufficiently salient, we expect that communication helps in clarifying the intentions of the employer and, therefore, it may induce a higher effort exertion by the selected workers.

4 Procedures

The experiment was programmed in z-Tree (Fischbacher 2007) and conducted at the experimental laboratory of the Max Planck Institute of Economics (Jena, Germany). The subjects were undergraduate students from the Friedrich Schiller University, Jena. They were recruited via the ORSEE software (Greiner 2004). We conducted 22 sessions (see Table 2), featuring 210 groups with a total of 630 subjects, from November 2011 to July 2012. The sessions lasted about 50 minutes. Average payment was 9.88 Euros⁵ including the show-up fee.⁶

Table 2 Participants and treatments

Upon arriving at the laboratory, participants were randomly allocated to visually isolated computer terminals and given a paper copy of the instructions that were read aloud to ensure common information.⁷ At the beginning of each session, each participant was randomly assigned a role and matched with two other participants to form a group of three. We referred to each group as a firm and to the group members as employer and low and high ability workers.

5 Results

In this section, we present our experimental results. We will first focus on the effort exertion irrespective of the workers’ ability. Then we will look at the impact of ability. Finally we will focus on the impact of communication.

Our first result refers to the average effort exerted by the hired workers:

Result 1

In all treatments, the average effort exerted by the selected workers is significantly higher than the minimum enforceable level. Average efforts in the NCT and RDT are not significantly different.

Support for Result 1 can be foundin Table 3 and Figure 1. In each treatment the average effort level is significantly higher than one, which is the minimum enforceable level, (Wilcoxon matched-pairs signed-ranks test, RDT: $z=5.00$, $p<0.01$; NCT $z=6.28$, $p<0.01$; CT: z = 6.35, $p<0.01$).

The average effort exerted in the RDT and NCT is not significantly different [Wilcoxon–Mann–Whitney test (WMW henceforth): $z=0.17$, $p=0.87$]. That is, we refuse the second part of hypothesis 1. Intention based reciprocity seems to play a minor role in explaining effort choice in our experiment. One possible explanation for this result is that there is a little room for the employers to display their kindness: they cannot refuse to hire a worker to get an alternative sure payoff, and the wage paid to the hired worker is fixed.⁸

Table 3 Average effort exerted by the workers (SD) and Wilcoxon–Mann–Whitney test comparisons

Our second result summarizes our findings on the relation between effort exertion and workers’ ability.

Result 2

The low ability workers increase their effort in the CT compared to the NCT (and to the RDT). The effort exerted by high ability workers is not significantly different across treatments.

Communication has a significant impact on the behavior of low ability workers. Inspection of Table 3 reveals how the increase in average effort in the CT compared to the NCT is driven by the behavior of low ability workers. Hypothesis 2 is only confirmed in the presence of communication.

Fig. 1

Average effort of workers by treatment

The increase in the effort exerted by low ability workers strongly benefits employers who select them: on average, in the CT the employers who chose a low ability worker gained 41 % more than employers who selected a high ability worker (35.59 ECUs vs. 25.18 ECUs, WMW, $z=2.55$, $p=0.01$). Note, however, that this positive effect is not fully anticipated by employers, who do not increase the number of low ability workers hired in the CT compared to the NCT.⁹

5.1 Communication content

In order to analyze whether the difference in the effort exerted by high and low ability workers in the CT can be due to the messages they received, in this section we focus on the content of the messages. We restrict our analysis to non-empty messages¹⁰ and identify two broad categories:

• 1. Suggestion Messages containing (a) an explicit suggestion, i.e. a numerical suggestion of an effort choice, that can be either a single number or a range; or (b) an implicit suggestion, i.e. a precise description of a behavior, that allows the worker to infer the request of a level/range of effort.

  • An example of a message containing an explicit suggestion is mess. No. 15 in Appendix: “I’m asking you to choose a job performance of 7. Then I’d receive 37.5 ECU and you 40 ECU. We’d both profit.”

  • An example of a message containing an implicit suggestion is mess. No. 30: “Hey congratulations. You got the job I know if you give me only one ecu, you’ll get the highest possible share, but if you give me a 2, it’ll only cost you one ecu and I already get 5 more. Since I gave the job to you, it’d be nice, if you could agree on a value, where we both receive a similar share”.

• 2. Fairness Messages explicitly appealing to fairness, equalization of earnings, gratitude or reciprocal kindness.¹¹

  • An example of a message appealing to fairness is mess. No. 9: “The employer would be happy about a fair decision that will meet both—employer and employee ? :)”

  • An example of a message appealing to equalization of earnings is mess. No. 34: “I ask you to choose a contribution of 7 or 8. Then both of us will have a balanced earning (you have 40/38 and me 35/40). I’m asking you to also decide in my interest.”

To classify messages we followed a procedure frequently used in other economic experiments, see for instance Cooper and Kagel (2005), Cooper and Lightle (2013), and Sutter and Strassmair (2009). Two research assistants, not aware of the hypothesis being tested, independently assigned messages to the categories.¹² A binary coding rule was used: 1 if the message belongs to the category and 0 otherwise. Each message could be assigned to all, one, or none of the categories. Disagreement only happened in 2 % (1 / 48) of the cases for the category “suggestion” and in 10 % (5 / 48) of the cases for the category “fairness”, Cohen’s $\kappa$ for suggestion and fairness is equal to 0.94 and 0.72, respectively. In cases of disagreement a third assistant was asked to classify the message to break the ties.¹³

37 / 48 of the messages are assigned to the category suggestion, 36 / 48 to fairness and 6 / 48 of the messages are not assigned to any of the two categories. According to a set of two samples test of proportions, when comparing the messages sent to low ability and high ability workers, we do not find significant differences in the frequency of the arguments used (an appeal to fairness is present in 76 % of the messages sent to $w^H$ and in 71 % of those sent to $w^L$, $z= 0.37$, $p=0.71$; a suggestion is present in 79 % of the messages sent to $w^H$ and in 71 % of those sent to $w^L$, $z=0.60$, $p=0.55$). Similarly, when looking at the average effort suggested by the employers, we do not find significant differences depending on the worker’s ability: the average suggestion to low ability worker is 7.20 while to high ability workers it is 6.94 (WMW, $z= 1.56$, $p=0.12$).¹⁴

Employers are likely to combine fairness and suggestion: this happens in 65 % (31 / 48) of the cases (Spearman’s rank correlation coefficients, $\rho = 0.37$, $p < 0.01$).¹⁵ Of the remaining 17 messages (35 %) not simultaneously displaying fairness and suggestion, 6 (13 %) do not contain any suggestion and do not appeal to fairness; 6 contain a suggestion without appealing to fairness and 5 (10 %) only appeal to fairness.

Table 4 reports the average effort exerted depending on the content of the messages sent by high and low ability workers. It can be noted that low ability workers exert a higher average effort than high ability workers when receiving a message containing a suggestion (WMW, $z=2.92$, $p<0.01$) despite the fact that the average effort suggested is not significantly different depending on workers’ ability. Low ability workers exert a higher average effort compared to high ability workers also when receiving a message appealing to fairness ($z=2.93$, $p<0.01$). This result also holds when the two categories are combined ($z=3.38$, $p<0.01$).¹⁶

Table 4 Communication treatment: average effort exerted by the workers (SD) depending on the content of the message and Wilcoxon–Mann–Whitney test comparisons

In the CT low ability workers exert significantly higher effort compared to high ability workers; moreover, when receiving a message that contains a suggestion, they also tend to match the average suggested effort (Wilcoxon signed-rank test testing that the average effort exerted by $w^L$ is equal to 7.20, $z = 0.26$, $p = 0.80$). This is not the case for high ability workers, who exert an effort level significantly lower compared to the one suggested (Wilcoxon signed-rank test testing that the average effort exerted by $w^H$ is equal to 6.94, $z =2.6$, $p< 0.01$).¹⁷ So, a possible explanation for the observed difference in effort provision may be that low ability workers, feeling less entitled to the position, react differently than high ability workers once they receive a suggestion by the employer. However, given that the 84 % (31 / 37) of messages containing a suggestion also appeal to fairness, it is difficult to disentangle the pure effect of these two factors on the effort exerted.

To isolate the effect of suggestion on agents’ effort provision we run an additional treatment, identical to the CT, with the only exception that communication is limited to a numerical suggestion. The ‘Suggestion Treatment’ (ST) aims to disentangle the relative importance of the disclosure of expectations (informing the workers about the desired effort level) from the effect of words and in particular, the explicit appealing to fairness (calling for gratitude and/or the indication of a desired final outcome of the interaction, as for example, equality in earnings).¹⁸

In the ST we find that, as in the CT, the average effort levels suggested do not differ depending on workers’ ability (effort suggested to $w^L$=7.60 and to $w^H$ =7.02, WMW test, $z=1.53$, $p=0.13$) and compared to the average effort levels suggested in the CT (WMW test, overall: CT: 7.01, ST: 7.13; $z = 0.58$, $p = 0.56$; for $w^L$: CT: 7.21, ST: 7.60; $z = 1.10$, $p = 0.27$; for $w^H$: CT: 6.94, ST: 7.02; $z = 0.28$, $p = 0.78$).¹⁹ Despite this, in the ST the difference in effort provision by the low ability workers vanishes: the average effort exerted by low ability workers is 4.30, which is significantly lower than the effort exerted by low ability workers receiving a message with a suggestion in the CT (WMW test, $z=2.07,$ $p=0.04$). Similarly, the effort exerted by the low ability workers is not significantly different compared to the effort exerted by high ability workers both in the ST (WMW test, $z=0.54$, $p=0.59$) and in the other treatments, as reported in the Online Appendix (supplementary material). Table 5 compares the average effort exerted by workers who receive a suggestion both in the ST and in the CT. All results are stronger when we restrict the comparisons to workers who in the CT received messages both containing a numerical suggestion and an appeal to fairness.²⁰

Table 5 Average effort exerted by the workers (SD) who receive a suggestion in ST and CT and Wilcoxon-Mann-Whitney test comparisons

Overall, our results are in line with results from the literature, showing that rich communication is more effective than bare messages (see, for example Charness and Dufwenberg 2010; Cooper and Kühn 2014; Janssen et al. 2010).

6 Conclusions

As mentioned by Fiedler et al. (2011), “the choice of one partner over another is in itself a favorable action towards the chosen partner, and may be reciprocated by that partner” (p. 402). In our experiment, we find that this effect is present, but only for low ability workers when employers can explicitly point out to them their favorable action. Intriguing questions remain for future research.

First, our design highlights on the effect of the selection process that precedes a job relation. In real organizations, both the selection and the wage level are dimensions the employers may determine. Therefore, it seems important to understand how the effects induced by competitive selection of workers (who differ in their ability) interacts with reciprocity concerns (related to the decision about the level of the wage).

Second, we analyze a game between one employer and only two workers. The choice of the employer is simply whether to select the worker with the higher or the lower ability. If the employer selects one worker from a larger group of candidates with different abilities, his choice becomes more complex. The employer should, in fact, predict how the effect of the feeling of entitlement varies according to the ranking of the candidates, taking into account factors such as anchoring and counter-factual reasoning (Baron 2013; Medvec and Savitsky 1997). Hence, a relevant question for the employer is whether it is more profitable to select the second-ranked individual, who already feels indebted toward her but is relatively more able with respect to the other candidates, or to choose a much lower-ranked individual who is much less able but even more indebted toward her.

Finally, our analysis suggests that workers who feel less entitled to fill a position may be more grateful and exert a higher effort once employed. This entitlement may also derive from belonging to a discriminated group. In the presence of commonly held negative ex-ante beliefs about some characteristics of the members of a given group (e.g., ethnic or religious group), employers who hire members belonging to discriminated groups could appeal to their moral sentiments to induce a higher effort (O’Reilly III and Pfeffer 2000).

Overall, our findings suggest that the effect of selection might have been underestimated in previous research. In our experimental design differences in abilities are very small. While we replicate the finding of Brandts and Solà (2010) in that the choice of the low ability worker has no effect on the low ability’s behavior, we show that matters change completely once communication is introduced. Since communication is present almost everywhere in the real world, employers might want to keep in mind that selection of a low ability employee might have a strong motivating effect, when accompanied by the right words.

Appendix: Messages

In Table 6 we report all the messages (translated from German) sent in the CT and the pertaining categories (described in section B.2 in the main file) to which they have been assigned. For each message, column 4 ($w^k$) indicates whether it was sent to a low ability (L) or to a high ability worker (H); columns 6 and 7 indicate the final categorization where ‘S’ stands for ‘suggestion’ and ‘F’ for ‘fairness’. For the cases where the two coders disagreed, the final classification is indicated in bold. Disagreement for the category ‘suggestion’ occurred in one out of 48 of the cases (2 %). Specifically, in message No. 42 the first coder assigned 1 while the second coder 0. Disagreement for the category ‘fairness’ occurred in five messages over 48 (10 %). Specifically, in messages No. 12, 13, and 32 the first coder assigned 1 while the second one 0; in messages Nos. 31 and 58 the opposite happened.

Table 6 Messages and categories, part I and part II