Do Wage Cuts Damage Work Morale? Evidence from a Natural Field Experiment Sebastian Kube∗ , Michel André Maréchal∗∗ , Clemens Puppe‡‡ Dezember 20, 2007

Abstract Contractual incompleteness characterizes many employment relations. High work morale is therefore fundamental for sustaining voluntary cooperation within the rm. In the presence of reciprocity motives, wages can inuence work morale. Using a natural eld experiment, we provide clear-cut evidence showing that wage cuts have a signicant and large negative impact on work morale. An equivalent wage increase, however, results in a much weaker gain in productivity. Together, the results highlight a strongly asymmetric response of eort to wage variations.

JEL classication: C93, J30. Keywords: morale, reciprocity, gift exchange, eld experiment, wage rigidity.

∗

University of Karlsruhe, Department of Economics, Kollegium IV am Schloss, 76128 Karlsruhe, Germany. email: kube(at)coll.mpg.de ∗∗ University of St. Gallen, Forschungsinstitut für Empirische Ökonomie und Wirtschaftspolitik (FEW-HSG), Varnbüelstr. 14, CH-9000 St. Gallen, Switzerland. email:michelandre.marechal(at)unisg.ch ‡‡ University of Karlsruhe, Department of Economics, Kollegium IV am Schloss, 76128 Karlsruhe, Germany. email: puppe(at)wior.uni-karlsruhe.de

Dissatisfaction of the workers with their treatment by the management is to be counted among the most important causes of low morale, for it is common knowledge that men tend to hold back and to do little as possible for those against whom they feel a grievance. Sumner H. Slichter (1920, p.40)

1 Introduction Why are rms reluctant to cut wages during economic downturns? A prominent explanation for this puzzling question is based on the psychology of work morale:1 Work morale reects the degree to which workers voluntarily cooperate and contribute to the employer's goals in the absence of reputation or pecuniary incentives. According to this view, work morale is sensitive to the relationship between the workers' actual wage and some reference wage (e.g. see Bewley (1999)). Positive and negative deviations from the reference wage are interpreted as kind or unkind; employees then reciprocate by exerting higher or, respectively, lower eort. While this theoretical argument has a long tradition in economics (see Slichter (1920, 1929), Solow (1979) or Akerlof (1982)), corresponding eld evidence is scarce. In particular, we are not aware of any study providing controlled eld evidence on the impact of wage cuts on work morale. This paper sheds light on the interplay between wages and work morale 1 See

Azariadis (1975), Lindbeck and Snower (1986) and Katz (1986) for alternative theories providing an explanation for the phenomenon of downward wage rigidity.

1

in a naturally occurring labor market. We conduct a controlled eld experiment and test the extent to which workers reciprocate to changes in their hourly wages.2 We hired job applicants to catalog books for a limited time duration (i.e. excluding any possibility of reemployment) at an announced wage of presumably 15 Euro per hour. We actually paid this amount out in our benchmark treatment and it serves as a reference point. In our main treatment, we inform subjects immediately before they begin working that we will only pay them 10 Euros per hour. In a second treatment, we do the opposite and communicate a raise in the hourly wage from 15 to 20 Euros in order to explore the asymmetry between the impact of wage cuts and pay raises on work morale.3 Since we are also interested in the duration of the eects, after six hours of work, we oered each subject the opportunity of working an additional hour for 15 Euros regardless of their initial treatment. The results show that wage cuts have a severe impact on the eort workers provide. Productivity is on average about 30 percent lower if workers experience a wage cut. This negative eect is remarkably persistent in size and signicance, suggesting that workers do not adapt over the observed time span. On the contrary, the detrimental eect is so strong that it cannot be compensated by setting the wages back to their original level. Once work morale is hurt, it cannot be re-established, at least not in the short run. Our results suggest that negative reciprocal behavior plays an important 2 By

reciprocity we refer to the behavioral phenomenon of people responding towards (un)kind treatment likewise, even in the absence of reputational concerns. For formal models of reciprocity, see Rabin (1993), Charness and Rabin (2002), Dufwenberg and Kirchsteiger (2004), Falk and Fischbacher (2006) or Cox et al. (2007). 3 The second treatment is similar to Gneezy and List (2006), where workers had to catalog books and wages were increased from 12 to 20 US Dollars.

2

role in actual labor markets. In contrast, we nd only moderate evidence of a positive impact on work morale from wage increases. The eect of the pay raise on workers' productivity occurs with a time lag and is then about half the magnitude of the impact of an equivalent wage cut. Our eld experiment makes several contributions to the existing literature. First, an impressive amount of laboratory evidence highlights the fact that reciprocal behavior can aect economic outcomes in experimental labor markets (e.g. see Fehr et al. (1993, 1997, 2007), Abeler et al. (2006), Charness (2004) or Hannan et al. (2002)). However, laboratory experiments are generally characterized by a high level of experimenter scrutiny, which creates potential demand eects. Moreover, lab experiments generally do not involve the exertion of actual eort but simply monetary transfers. It is thus not clear to what extent these results can be generalized to real world labor markets (see Levitt and List (2007)). We were able to observe subjects in a more natural - yet still controlled - work environment, because they performed a typical student helper's task and do not know that they are part of an experiment. Second, to the best of our knowledge, this is the only study to date providing clear-cut evidence for negative reciprocal behavior in a natural labor market situation. The few existing eld experiments focus on the economic consequences of positive reciprocity, and their conclusions still remain ambiguous. Falk (2007) for instance shows that charitable donations increase substantially with the size of gifts included in the solicitation letters, rendering gift giving protable. Gneezy and List (2006) on the other hand nd that an increase in hourly wages has only a transient eect which ultimately did 3

not pay o for the employer.4 In addition to testing for reciprocal reactions towards wage cuts, our design allows for a novel direct comparison between the inuence of wage cuts and pay raises within the same framework, and highlights signicant asymmetries in the eld. Third, Bewley (1999)) conducted interviews with compensation executives, exploring the reasons why rms are reluctant to cut wages or avoid hiring underbidders during economic downturns (see also Blinder and Choi (1990), Levine (1993), or Campbell and Kamlani (1997)). The general insight from these interview studies is that the desire to maintain a good work morale seems to be a key rationale employers provide for their policies. Even though this line of research provides a valuable rst indication on the role of work morale in the labor market, several potential drawbacks should be taken into consideration. Social desirability eects are a well-known phenomenon in survey research; they question the extent to which we can take answers from interviews at face value (see Bertrand and Mullainathan (2001) or Krosnick (1999)). More important, while interviews may provide some information with regard as to why rms are reluctant to cut wages, ultimately they only reect beliefs and fail to answer the question of how and the extent to which wages aect work morale. Fourth, identifying the causal impact of wage cuts on work morale poses serious diculties in the eld. Changes in compensation generally reect rms' choices and are therefore potentially endogenous due to unobservable confounds (see Shearer (2003)). Consequently, only a few studies have ana4 For

further eld evidence concerning positive reciprocity, see Maréchal and Thöni (2007), Kube et al. (2007), Cohn et al. (2007), Bellemare and Shearer (2007), Al-Ubaydli et al. (2006), and Hennig-Schmidt et al. (2005).

4

lyzed the impact of wage cuts on productivity using non-experimental eld data (see Mas (2006), Krueger and Mas (2004), and Lee and Rupp (2007)).5 All of these studies are embedded in an ongoing relationship between workers and employers, making it impossible to fully separate work morale from reputational motives. There are at least two alternative pecuniary reasons why workers provide less eort after a wage cut in repeated interactions. First, workers could play a trigger strategy and punish the rm for cutting their wages by exerting lower eort (see Howitt (2002)). Second, lower wages reduce future rents and dampen the disciplining eect of getting red (see Shapiro and Stiglitz (1984) or MacLeod and Malcomson (1989)). We take great care in making clear that we oer a one time job without any possibility of reemployment and are therefore able to rule out reputational motives. Furthermore, while eort often manifests itself in a multitude of dimensions, our simple data entry task allows us to measure eort very accurately. In contrast, Lee and Rupp (2007) for instance have to rely on ight delays as the single proxy for the eort airline pilots provide. However, ight delays can serve only as a very crude proxy for eort and are strongly inuenced by other forces beyond the control of pilots. The remainder of this paper is organized as follows: In the next section, we describe the experimental design. Subsequently, in Sections 3 and 4, the experimental results are presented and discussed. 5 Mas

(2006) does not analyze the impact of actual wage cuts but negative deviations from expected wage increases.

5

2 Experimental Design In August 2006, the library of an economic chair at a German University had to be cataloged. We took this opportunity to run a eld experiment and used posters to recruit students from all over the campus. The announcement read that it was a one-time job opportunity for one day (six hours), and that pay will presumably amount to 15 Euros per hour.6 The announced wage of 15 Euro served as a reference point. About 200 students applied during the two month announcement phase. A research assistant randomly picked 30 persons out of the list of applicants. They were invited via email and asked to conrm the starting date, reminding them that the job would presumably pay 15 Euros per hour. Upon arriving, the subjects were seated in front of a computer terminal with a table of randomly picked books beside them. Their task was to enter the book's author(s), title, publisher, year of publication, and ISBN number into an electronic data base. Participants were allowed to take a break whenever necessary. A research assistant explained the task to them, strictly following a xed protocol. Then, subjects were told their hourly wage and started working. We conducted three dierent treatments. The hourly wage paid in our benchmark treatment was 15 Euros (treatment Neutral ), 20 Euros in the pay raise treatment (treatment Kind ) and 10 Euros in the wage cut (treatment

Unkind ).7 We gave the subjects no reason why they were paid more or less 6 The

announcement said presumably, or prospectively (the exact German wording was voraussichtlich), in order to raise appropriate expectations without cheating. 7 10 Euros still exceed the hourly wages usually paid to a student helper at German universities, which is about 8 Euros. We paid higher wages in the benchmark treatment, because we wanted to focus on the eect of the unexpected change rather than the absolute amount of the wage, which would otherwise have been extremely low in the Unkind

6

than the publicized 15 Euros.8 We had 10 subjects in the benchmark as well as in the wage cut treatment, and 9 subjects in the pay raise treatment, because one subject did not show up for work. The experiment took place over a 10 day period, with 3 subjects per day (one in each treatment). The subjects showed up successively at dierent times and were separated from each other, in dierent rooms at an online computer terminal, without being monitored.9 The computer application in which they entered the details of the books recorded the exact time of each log, allowing us to reconstruct exactly the number of books each person entered over time and the time span between two books entered.10 After 6 hours of work, all subjects had to complete a questionnaire. In order to observe them in a natural working environment, subjects were not told that they were taking part in an experiment. After completing the questionnaire, all subjects were asked if they were willing to work for an additional hour immediately following completion of the questionnaire for 15 Euros (regardless of their initial treatment). When they rejected this oer, they were paid their total wage and left. Otherwise, they worked one additional hour and received their total wage payment afterwards. The data gained from the additional hour enables us to compare the performance of a subject under dierent wages. treatment. 8 None of the subjects actually asked for an explanation. 9 All this was done in order to minimize any possible bias due to peer eects, monitoring, time of start or day of work. Furthermore, all subjects interacted with the same research assistant to eliminate experimenter eects. In order not to induce an experimenter demand eect, the research assistant knew neither the purpose of the study nor the reason for the various wages. 10 See Figure 2 in the Appendix for a screen shot.

7

3 Results Panel (a) in Figure 1 illustrates the average eort (measured by the number of books logged) per 90 minute time interval, or quarter, for each of the three dierent treatments. Table 1 contains the average treatment eects, i.e. the dierence in average number of books logged, and the p-values (two sided) from the corresponding nonparametric Wilcoxon rank-sum tests for the null hypothesis of equal eort between treatments. Figure 1: Work Morale as a Function of Wages (b) Cumulative Distribution Function 1 .8 .6 .4

Cumulative Probability

60 50

.2

40 90

180

270

Unkind

Unkind

Neutral

Neutral

Kind

Kind

0

30

Average # of books logged per 90 minutes

70

(a) Effort over Time

360

Minutes

75

100

125

150

175

200

225

250

275

300

# of books logged during regular working time

We observe a substantial dierence in eort between subjects in the Neu-

tral and Unkind treatments over the entire duration of the experiment. This eect is highly signicant from a statistical and economical point of view (see columns three and four in Table 1). On average, workers in the Unkind treatment logged 56 books (or 27 percent) less than in the benchmark treat8

ment. Moreover, there is no indication of any adaptation over the observed time span. The average treatment eect remains large and signicant in all quarters.11 The cumulative distribution functions in Panel (b) of Figure 1 show that it is not merely one or two workers alone who drive the above mentioned results; instead they reect a broad behavioral phenomenon. For example, while the fraction of workers who logged 177 books or less is only 20 percent in treatment Neutral it is 90 percent in treatment Unkind. Table 1: Average Treatment Eects by Time Intervals: # Books Logged Time interval Minutes 1-90 Minutes 91-180 Minutes 181-270 Minutes 270-360 Total regular time

Kind - Neutral

(1)

(2) p > |z|

Unkind - Neutral

(3)

(4) p > |z|

0.1 6.4 6.7 7.7 20.9

0.623 0.164 0.094 0.347 0.165

-16.3 -13.7 -12.9 -13.3 -56.2

0.002 0.010 0.006 0.017 0.003

Notes: Columns 1 and 3 report average treatment eects for the treatments Kind (20 Euros hourly wage) and Unkind (10 Euros hourly wage) in comparison with the benchmark Neutral (15 Euros hourly wage) for the dierent time intervals. The outcome variable is the number of books logged as a measure for the eort provided. Columns 2 and 4 report the corresponding p-values from a nonparametric (two-sided) Wilcoxon rank-sum test for the null hypothesis of equal eort between treatments.

On the other hand, the average treatment eect for the pay raise is marginally small during the rst quarter. Interestingly, the eect tends to strengthen over the course of time. We nd an eect in the third quarter which is signicant at a 10 percent level (see column two of Table 1). Overall, 11 Alternatively,

one could dene adaptation as the dierence between the number of books entered in the rst and the last quarter. In this sense, treatment Neutral has a roughly 14% increase whereas treatment Unkind has a roughly 30% increase, suggesting that convergence might ultimately occur as in Gneezy and List (2006). We are grateful to John List for pointing this out. However, the null hypothesis of equal dierences between the treatments Unkind and Neutral cannot be rejected using a two-sided Wilcoxon ranksum test (p=0.45).

9

there is moderate support for positive reciprocal behavior. Workers provide on average 10 percent more eort (or logged 21 more books) due to the wage increase. This dierence is statistically insignicant and less than half of the absolute magnitude of the discrepancy between the treatments Unkind and

Neutral. The panel regression results in Table 2 are in line with the preceding nonparametric analysis. In the benchmark model (column 1), the number of books logged is regressed on the treatment indicators Unkind and Kind, the time interval dummies (Timet ) and their interactions. We choose a random eects specication for the error term, accounting for individual dependency over time. The coecient for Unkind is highly signicant, whereas the coefcient for Kind does not reach any conventional signicance level. Moreover, all of the U nkind ∗ T imet interaction eects are relatively small and insignificant, suggesting that there was no adaptation in the observed periods. On the other hand, the estimated Kind ∗ T imet coecients indicate, the eect of treatment Kind is signicantly higher after quarter one. Hence, positive reciprocal reactions tend to strengthen with the elapse of time.12 A further interesting result - which is also clearly visible in Figure 1 - is that the number of books logged increased substantially over time, which we interpret as a learning eect. We performed several robustness checks. First, we control for age, gender, subject of studies, how workers perceived the task (e.g. interesting, strenuous etc.), and the hourly wage earned at the most recent job prior to the 12 Nevertheless,

the null hypothesis that all Kind ∗ T imet interaction eects are equal to zero cannot be rejected (Wald test χ2 = 5.70, (P rob > χ2 ) = 0.1271).

10

experiment (as a proxy for human capital).13 As demonstrated in columns (2) to (4) in Table 2, the inclusion of an extensive set of control variables fails to aect the key results.14 Second, as reported in column (6), the results are robust if higher frequency data - consisting of 15 minute time intervals - are used.15 In addition to the eect on the quantity of eort, we also investigated the impact of our treatments on the quality of eort. Similar to Hennig-Schmidt et al. (2005) we measured the quality of output by the ratio of faultless logs to the total number of books entered.16 Treatment dierences in quality ratios are economically and statistically insignicant: The quality ratios are 0.932 in Neutral, 0.934 in Unkind and 0.921 in Kind.17 The regression results in column (5) of Table 2 show that our main results are robust if we use only error-free logs as our measure of workers' eort. After completion of the regular working time, subjects were oered the opportunity of working one additional hour for 15 Euros. This allows us to test whether the detrimental impact of the wage-cut on eort can be oset 13 Table

3 and 4 contain summary statistics and denitions for all included control variables. 14 Seven subjects did not have any previous work experience and are therefore excluded in our model where we control for previous wages (column 4 of Table 2). 15 Note that we chose an alternative specication for the higher frequency data in order to avoid the large number of coecients that would have resulted if we used dummy variables. Instead of the T imet dummies we constructed the variable T ime which takes values from 1 to 24 indicating the time period. As an alternative to the random eects Generalized Least Squares regression we also ran OLS regressions with clustered standard errors. The coecient estimates for Kind∗T ime3, Kind ∗ T ime4 and - using 15 minute intervals - Kind ∗ T ime become insignicant. All other results remain unchanged in terms of signicance and magnitude. 16 Two research assistants searched for spelling mistakes in the titles of the books by running an automatic spell check program. 17 Using a two-sided Wilcoxon rank-sum test, we cannot reject the null hypothesis of equal output quality between Kind and Neutral (p=0.624) as well as Unkind and Neutral (p=0.7624).

11

Table 2: Panel Regression Results (1)

Kind Unkind Time2 Time3 Time4 Kind ∗ Time2 Kind ∗ Time3 Kind ∗ Time4 Unkind ∗ Time2 Unkind ∗ Time3 Unkind ∗ Time4 Time

(2)

(3)

(4)

(5)

90 min. total logs

90 min. total logs

90 min. total logs

90 min. correct logs

15 min. total logs

0.078 (4.670) -16.300*** (5.014) 1.600 (2.574) 1.700 (2.664) 7.000** (2.968) 6.289** (3.069) 6.633** (3.356) 7.667* (4.491) 2.600 (3.766) 3.400 (3.836) 3.000 (3.877)

-2.239 (5.565) -20.585*** (5.484) 1.600 (2.756) 1.700 (2.885) 7.000** (2.848) 6.289* (3.263) 6.633* (3.569) 7.667* (4.412) 2.600 (4.020) 3.400 (3.998) 3.000 (3.855)

-2.311 (5.869) -20.773*** (5.882) 1.600 (2.633) 1.700 (2.851) 7.000** (2.854) 6.289* (3.263) 6.633* (3.518) 7.667* (4.417) 2.600 (3.945) 3.400 (3.961) 3.000 (3.862)

2.745 (5.352) -23.836*** (5.601) 3.250 (2.625) 5.750** (2.528) 8.625*** (2.955) 4.875 (3.458) 3.000 (3.605) 6.875 (4.839) 0.083 (3.261) 0.417 (4.229) 2.042 (3.678)

1.544 (4.611) -14.500*** (5.082) 0.500 (2.872) 2.200 (2.693) 8.500** (3.349) 4.833 (3.312) 4.356 (3.330) 1.722 (4.636) 2.900 (4.106) 2.600 (3.836) 0.400 (4.135)

0.029 (0.728) -2.662*** (0.764)

49.700*** (3.681)

82.564*** (16.273)

106.061** (42.955)

97.059** (47.968)

45.900*** (3.829)

0.054** (0.021) 0.067** (0.031) 0.026 (0.027) 8.040*** (0.591)

NO NO NO

YES NO NO

YES YES NO

YES YES YES

NO NO NO

NO NO NO

116 0.429 0.000

116 0.573 0.000

116 0.608 0.000

88 0.727 0.000

116 0.387 0.000

696 0.250 0.000

Kind ∗ Time Unkind ∗ Time Constant Controls: Socioeconomic? Task perception? Previous wage? Obs. R2 Prob> χ2

(6)

90 min. total logs

Notes: The table reports GLS coecient estimates (robust standard errors in parentheses) from a linear Random Eects model. The dependent variables is the number of books (correctly [in column 5]) logged per time interval in columns (1)-(4) and (6). The treatment dummies Unkind and Kind are interacted with the time interval dummies T imet (or the variable T ime which takes values from 1 to 24 in the high frequency data). Denitions and summary statistics for the variables in the three dierent sets of controls (socioeconomic, task perception and previous wage) can be found in Tables 4 and 3. Since seven students stated that they had never been employed before, we loose those observations when controlling for previously earned hourly wages in column (4). Signicance levels are denoted as follows: * p<0.1, ** p<0.05, *** p<0.01.

12

by a subsequent wage increase.18 During the last 60 minutes of the regular working time, the workers in the Unkind treatment logged on average 29.6 books. Even though the hourly wage was increased by 50 percent, back to the original level of 15 Euros, the average eort level remained unchanged at 28.3 books logged (Wilcoxon signed-rank test: z =0.170 and p > |z| = 0.865). The ineectiveness of the wage increase is unlikely the result of exhaustion. In the additional hour, subjects in the treatments Kind and Neutral logged in on average 43.6 books, which is signicantly more than in Unkind according to a Wilcoxon rank-sum test (z = 2.847 and p > |z| = 0.004).

4 Discussion and Concluding Remarks Maintaining high work morale is of paramount importance for rms whenever workers' eort is not fully contractible. A longstanding explanation for the phenomenon of downward wage rigidity presumes that wage cuts damage work morale (see Slichter (1920, 1929), Solow (1979) or Akerlof (1982)). However, corresponding evidence from the eld is scarce because compensation schemes usually do not vary exogenously but reect rms' decisions (see Shearer (2003)). Apart from these identication problems, ongoing relations between workers and rms make it hard to disentangle work morale from alternative pecuniary or reputational motives in the eld. This study lls this gap and provides clear-cut evidence on the impact of wage cuts on work morale using a labor market eld experiment. In addition, 18 Only

a few subjects were willing to work for an additional hour in the treatments Kind (N=2) and Neutral (N=3). We therefore only consider the 7 subjects who agreed to work for an additional hour in the Unkind treatment in the following.

13

the paper provides a novel direct comparison between the impact of wage cuts and corresponding wage increases within the same framework. In summary, our results show that wage cuts have a severe impact on the quantity of eort workers provide. Moreover, this negative eect remains large and signicant over the course of the entire working period. The detrimental eect is so strong that it cannot be compensated by setting wages back to their original level, at least not in the short run. While these results are supportive for the notion that wage cuts damage work morale, the evidence for pay raises is less conclusive. We do not nd a signicant positive relationship between higher wages and eort. Response of eort to the pay raise lags and amounts to about half the magnitude of the response to a wage reduction of the same size. Together, our results provide new evidence stressing the importance of work morale and highlight a strongly asymmetric response of eort to wage variations.

14

References Abeler, J., S. Altmann, S. Kube, and M. Wibral (2006): Reciprocity and Payment Schemes: When Equality Is Unfair, IZA Discussion Paper No. 2500. Akerlof, G. A. (1982): Labor Contracts as Partial Gift Exchange, Quarterly

Journal of Economics, Vol. 97, pp. 543569. Al-Ubaydli, O., S. Andersen, U. Gneezy, and J. List (2006): For Love or Money? Testing Non-Pecuniary and Pecuniary Incentive Schemes in a Field Experiment, Working Paper. Azariadis, C. (1975): Implicit Contracts and Unemployment Equilibria,

Journal of Political Economy, Vol. 83, pp. 11831202. Bellemare, C. and B. Shearer (2007): Gift Exchange within a Firm: Evidence from a Field Experiment , IZA Discussion Paper No. 2696. Bertrand, M. and S. Mullainathan (2001): Do People Mean What They Say? Implications for Subjective Survey Data, American Economic Re-

view, Vol. 91, pp. 6772. Bewley, T.F. (1999): Why Wages Don't Fall During a Recession, Cambridge, Massachusetts: Harvard University Press. Blinder, A. S. and D. H. Choi (1990): A Shred of Evidence on Theories of Wage Stickiness, Quarterly Journal of Economics, Vol. 105, pp. 1003 1015. Campbell, C. M. and K. S. Kamlani (1997): The Reasons for Wage Rigidity: Evidence From a Survey of Firms, Quarterly Journal of Economics, Vol. 112, pp. 759789. Charness, G. (2004): Attribution and Reciprocity in an Experimental Labor Market, Journal of Labor Economics, Vol. 22, pp. 665688.

15

Charness, G. and M. Rabin (2002): Understanding Social Preferences with Simple Tests, Quarterly Journal of Economics, Vol. 117, pp. 817869. Cohn, A., E. Fehr, and L. Goette (2007): Do Fair Wages Increase Labor Supply? Evidence from a Large-Scale Field Experiment, Working Paper. Cox, J. C., D. Friedman, and S. Gjerstad (2007): A Tractable Model of Reciprocity and Fairness, Games and Economic Behavior, Vol. 59, pp. 1745. Dufwenberg, M. and G. Kirchsteiger (2004): A Theory of Sequential Reciprocity, Games and Economic Behavior, Vol. 47, pp. 268298. Falk, A. (2007): Gift-Exchange in the Field, Econometrica, Vol. 75, pp. 15011511. Falk, A. and U. Fischbacher (2006): A Theory of Reciprocity, Games and

Economic Behavior, Vol. 54, pp. 293315. Fehr, E., S. Gächter, and G. Kirchsteiger (1997): Reciprocity as a Contract Enforcement Device, Econometrica, Vol. 65, pp. 833860. Fehr, E., G. Kirchsteiger, and A. Riedl (1993): Does Fairness Prevent Market Clearing? An Experimental Investigation, Quarterly Journal of Eco-

nomics, Vol. 108, pp. 437460. Fehr, E., A. Klein, and K. Schmidt (2007): Fairness and Contract Design,

Econometrica, Vol. 75, pp. 121154. Gneezy, U. and J. A. List (2006): Putting Behavioral Economics to Work: Testing for Gift Exchange in Labor Markets Using Field Experiments,

Econometrica, Vol. 74, pp. 13651384. Hannan, R. L., J. Kagel, and D. V. Moser (2002): Partial Gift Exchange in an Experimental Labor Market: Impact of Subject Population Dierences, Productivity Dierences and Eort Requests on Behavior, Journal

of Labor Economics, Vol. 20, pp. 923951. 16

Hennig-Schmidt, H., B. Rockenbach, and A. Sadrieh (2005): In Search of Workers' Real Eort Reciprocity - A Field and a Laboratory Experiment, mimeo. Howitt, P. (2002): Looking Inside the Labor Market: A Review Article,

Journal of Economic Literature, Vol. 40, pp. 125138. Katz, L. F. (1986): Eciency Wage Theories: A Partial Evaluation, NBER

Macroeconomics Annual Review of Psychology, Vol. 1, pp. 235276. Krosnick, J. A. (1999): Survey Research, Annual Review of Psychology, Vol. 50, pp. 53767. Krueger, A. and A. Mas (2004): Strikes, Scabs and Tread Separations: Labor Strife and the Production of Defective Bridgestone/Firestone Tires,

Journal of Political Economy, Vol. 112, pp. 253289. Kube, S., M. A. Maréchal, and C. Puppe (2007): Money Can't Buy Everything - Cash, Perks and Workers' Motivation in a Randomized Field Experiment, Working Paper. Lee, D. and N. G. Rupp (2007): Retracting a Gift: How Does Employee Effort Respond to Wage Reductions?, Journal of Labor Economics, Vol. 25, pp. 725761. Levine, D. I. (1993): Fairness, Markets, and Ability to Pay: Evidence from Compensation Executives, American Economic Review, Vol. 83, pp. 1241 1259. Levitt, S. D. and J. A. List (2007): What do Laboratory Experiments Measuring Social Preferences tell us about the Real World, Journal of Eco-

nomic Perspectives, Vol. 21, pp. 153174. Lindbeck, A. and D. J. Snower (1986): Wage Setting, Unemployment, and Insider-Outsider Relations, American Economic Review, Papers and Pro-

ceedings, Vol. 76, pp. 235239.

17

MacLeod, B. W. and J. M. Malcomson (1989): Implicit Contracts, Incentive Compatibility, and Involuntary Unemployment, Econometrica, Vol. 57, pp. 447480. Maréchal, M. A. and C. Thöni (2007): Do Managers Reciprocate? Field Experimental Evidence From a Competitive Market, University of St. Gallen Economics Discussion Paper No. 2007-09. Mas, A. (2006): Pay, Reference Points, and Police Performance, Quarterly

Journal of Economics, Vol. 121, pp. 783821. Rabin, M. (1993): Incorporating Fairness into Game Theory and Economics, American Economic Review, Vol. 83, pp. 12811302. Shapiro, C. and J. E. Stiglitz (1984): Equilibrium Unemployment as a Worker Discipline Device, American Economic Review, Vol. 74, pp. 433 444. Shearer, B. (2003): Compensation Policy and Worker Performance: Identifying Incentive Eects from Field Experiments, Journal of the European

Economic Association, Vol. 1, pp. 503511. Slichter, S. (1920): Industrial Morale, Quarterly Journal of Economics, Vol. 35, pp. 3660. Slichter, S. H. (1929): The Current Labor Policies of American Industries,

Quarterly Journal of Economics, Vol. 43, pp. 393435. Solow, R. M. (1979): Another Possible Source of Wage Stickiness, Journal

of Macroeconomics, Vol. 1, pp. 7982.

18

5 Appendix Table 3: Summary Statistics: Control Variables Variable

Mean

Std. Dev.

Min.

Max.

Obs.

Age Male Math and Physics Engineering and Computer Science Arts and Social Science Economics Strenuous Interesting Got tired Had problems Worked before Previous wage (Euro per hour)

23.31 0.552 0.138 0.207 0.207 0.448 3.690 3.621 2.862 4.586 0.759 10.94

3.465 0.506 0.351 0.412 0.412 0.506 0.967 0.979 1.156 0.568 0.435 4.921

19 0 0 0 0 0 1 1 1 3 0 7.5

37 1 1 1 1 1 5 5 5 5 1 30

29 29 29 29 29 29 29 29 29 29 29 22

Worked before Previous wage

Previous wage

Strenuous Interesting Got tired Had problems

Task perception

Age Male Math and Physics Engineering and Computer Science Arts and Social Science Economics

Socioeconomic

Variable

1=yes ; 0=no Euro per hour

1=fully agree ... 1=fully agree ... 1=fully agree ... 1=fully agree ...

years 1=yes; 0=no 1=yes ; 0=no 1=yes ; 0=no 1=yes ; 0=no 1=yes ; 0=no

Denition

5=fully disagree 5=fully disagree 5=fully disagree 5=fully disagree

Did you work somewhere before this job? [yes, no] If yes, what was your hourly wage on your last job? [free form]

Please rate the following statements on a scale from 1 to 5 where 1 means fully agree and 5 means fully disagree. I perceived the job as strenuous [1...5] I perceived the job as interesting [1...5] I got tired with the elapse of time [1...5] I had problems categorizing the books [1...5]

Age? [free form] Gender? [free form] Subject of studies? [free form]

Question wording [Possible answers in brackets]

Table 4: Control Variables: Wording and Coding (Translated from German to English

(10 Euros)

Unkind

(20 Euros)

Kind

(15 Euros)

Neutral

Treatment

21 23 38 32 27 32 31 30 38 62 33.4

42 40 48 45 58 53 50 55 57 49.8

3 5 24 20 17 13 11 27 22 Average

25 10 6 1 18 15 29 14 7 23 Average

44 45 38 47 45 57 44 55 49 73 49.7

Minutes 1-90

26 9 4 8 16 19 28 12 2 21 Average

Participant #

22 21 30 38 34 38 35 50 40 68 37.6

46 45 54 56 62 64 65 59 68 57.7

39 45 42 47 52 55 56 47 65 65 51.3

Minutes 91-180

9 29 35 38 44 42 45 37 48 58 38.5

50 43 53 61 51 60 61 69 75 58.1

39 46 49 48 50 46 58 64 59 55 51.4

Minutes 181-270

27 27 31 39 44 50 52 50 50 64 43.4

42 52 56 59 55 73 76 76 91 64.4

42 41 60 53 61 52 55 57 67 79 56.7

Minutes 270-360

79 100 134 147 149 162 163 167 176 252 152.9

180 180 211 221 226 250 252 259 291 230

164 177 189 195 208 210 213 223 240 272 209.1

Total regular time

31 36 34 28.3

23 24 33

17

57 48.0

39

40.7

42 41

39

Overtime 60 minutes

0.937 0.980 0.925 0.986 0.980 0.932 0.804 0.940 0.869 0.984 0.934

0.900 0.978 0.867 0.964 0.978 0.884 0.853 0.931 0.945 0.921

0.939 0.921 0.915 0.990 0.947 0.976 0.934 0.803 0.958 0.938 0.932

Quality ratio regular time

Table 5: Summary Data: Number of Books Logged During 90 Minutes Intervals and 60 Minutes of Overtime

Figure 2: Screenshot: Computer Application