Wage Differentials, Fairness and Social Comparison: An experimental study of the Co-Employment of Permanent and Temporary Agency Workers

Size: px
Start display at page:

Download "Wage Differentials, Fairness and Social Comparison: An experimental study of the Co-Employment of Permanent and Temporary Agency Workers"

Transcription

1 Pepints of the Max Planck Institute fo Reseach on Collective Goods Bonn 2006/8 Wage Diffeentials, Fainess and Social Compaison: An expeimental study of the Co-Employment of Pemanent and Tempoay Agency Wokes Doothea Alewell / Andeas Nicklisch M A X P L A N C K S O C I E T Y

2 Pepints of the Max Planck Institute fo Reseach on Collective Goods Bonn 2006/8 Wage Diffeentials, Fainess and Social Compaison: An expeimental study of the Co-Employment of Pemanent and Tempoay Agency Wokes Doothea Alewell / Andeas Nicklisch Mach 2006 Max Planck Institute fo Reseach on Collective Goods, Kut-Schumache-St. 10, D Bonn

3 Wage Diffeentials, Fainess, and Social Compaison: An expeimental study of the Co-Employment of Pemanent and Tempoay Agency Wokes * Doothea Alewell, Fiedich Schille Univesity Jena, Chai fo Business Administation, Human Resouce Management and Oganization; Cal-Zeiss-Stasse 3, D Jena, Gemany; d.alewell@wiwi.uni-jena.de Andeas Nicklisch, Max Planck Institute fo Reseach on Collective Goods; Kut- Schumache-Stasse 10, D Bonn, Gemany; nicklisch@coll.mpg.de Abstact Recent expeimental liteatue in labo economics shows that fainess concens make a substantial diffeence fo woking decisions. Ou study systematically exploes how the existence of multiple fainess foci influences wage setting and acceptance thesholds. Paticulaly, we focus on the effect of hoizontal fainess concens, i.e., the wage compaison among employees. Fo ou expeiment, we use an institutional design of wage negotiations among employes, employees and tempoay agency wokes. Woking agencies hie these wokes and ent them out to fims. Theeby, we ceate a heteogeneous backgound of the labou foce. Although tempoay agency wokes do the same wok, typically, they eceive lowe wages due to the intemediate agency. The esults of ou laboatoy expeiments indicate that the availability of infomation concening co-employee s wage offes stongly influences the wage set and paticipants acceptance of contacts. Wheeas the elation of aveage wages is not influenced by the ode of the decisions, the absolute level of wages is dependent on the decisions. We find that tempoay agency wokes who decide on a wage offe afte pemanent employees eceive a pemium in addition to thei wages, while pemanent employees take a cut in wages if they get thei wage offe afte tempoay wokes have decided on thei offes. These esults ae moe influenced by self-egading social compaison pefeences than by othe-egading hoizontal fainess concens. Keywods Expeimental economics, hoizontal fainess noms, labou economics, social pefeences, vetical fainess noms JEL C92, J33, M12, M52 We dedicate this aticle to ou colleague, Wiebke Kuklys, who died while we wee planning this study. * We ae vey gateful fo comments and suggestions of Chistoph Engel, Matin Hellwig and Stefan Magen.

4 2 1. Intoduction Fainess noms ae highly elevant fo wok-elated decisions in oganizations and on the labou maket. Although thee has been much in-depth eseach on fainess issues, 1 many aspects of fainess behaviou ae still not vey well undestood. 2 This is especially tue fo the elevance of fainess judgements in complex, ich settings whee multiple efeence points could guide behaviou. In the following, as one example of the wide class of situations with multiple fainess foci, we expeimentally analyse the elation between two efeence points fo social pefeences, vetical fainess consideations and hoizontal fainess consideations in a complex ultimatum game setting with thee inteelated ultimatum games. Heeby, we use the tem vetical fainess consideations fo the elation between the outcome distibutions among poposes and espondes in ultimatum games and espondes acceptance thesholds. By contast, we use the tem hoizontal fainess concens to chaacteize the elation between the outcome distibutions among seveal espondes and thei acceptance thesholds. As one example, we conside the co-employment of egula employees and tempoay agency wokes. This situation is especially inteesting since thee ae vetical foci, e.g., the payoffs of the espective contact patnes (employee and employe, tempoay wok agency and tempoay woke) and hoizontal foci fo fainess judgements, e.g., the absolute o elative wage of the othe woke, with o without efeence to othe diffeences between the two types of wokes. Moeove, this envionment allows fo diffeent intepetations of fainess consideations. We model a situation whee employees have to do the same wok. Theefoe, one may ague that only equal wages would teat tempoay and egula employees faily. On the othe hand, in this institutional setting, thee ae obvious easons that the wages of tempoay and egula employees diffe, since thee is an intemediating agency between employes and tempoay employees. Employes have to pay the agency as well, so that one can ague that wages fo temps have to deviate fom wages of egula employees. Ou teatments systematically vay the sequential ode of decisions, and, theeby, the available infomation about some of these possible fainess foci. We can, thus, deive infomation about the elevance of these elements in the fomation of fainess judgements. 1 To name only a few moe ecent papes, fo example, Bolten & Ockenfels, 2000, Dickinson & Tiefenthale, 2002, Feh & Schmidt, 1999, Gächte & Feh, 2002, Kahnemann et al., 1986, Konow, 1996, 2000, 2001, 2003, Rabin 1993, and Scott, Fo an ealy pape fom the pespective of psychology, see Leventhal (1980); he identifies the multidimensionality of fainess concepts as one of the thee majo poblems with equity theoy.

5 3 The co-employment of hied and ented hands is only one example fom a wide class of situations with multiple fainess foci. Paticulaly, we choose it as ou example since it is cuently of consideable political elevance in the Euopean Union. Seveal institutional changes have been ecently intoduced that aim at deceasing unemployment by futheing tempoay agency wok. As an example, in ode to incease the acceptance of tempoay employment, Gemany intoduced the compulsoy equal teatment of pemanent and tempoay employees if no collective ageement applies (Alewell et al., 2004). The fainess and justice liteatue identifies diffeent fainess noms, efeence points o justice pinciples that can potentially guide the judgement on whethe allocations ae fai; fo example, equality and needs, equity and deset, efficiency o accountability (Konow, 2003), absolute diffeences in pay-off vesus elative pay-off (Bolten & Ockenfels, 2000, Feh & Schmidt, 1999), changes in one s own pay-off vesus pay-off elative to othes (Scott, 2003), effects of entitlements (Kahnemann et al., 1986), total suplus o net suplus ove esevation utility, hoizontal fainess between diffeent wokes o vetical fainess between the employe and woke (Knez & Camee, 1995). Despite the multiple insights on the effect of the context on fainess evaluations (fo an in-depth oveview, see Konow, 2003), specific hypotheses about the elevance of diffeing fainess noms ae still to some extent speculative, since thee ae many situational factos that influence fainess evaluations fo example, the scope of the compaisons, the weighting of justice pinciples, competing foces such as fainess and self-inteest and the elative impotance of pocedual and distibutive justice (Konow, 2003). 3 Theefoe, at this point in the eseach, much can still be leant fom expeimental studies with multiple fainess foci that systematically vay aspects of the situation. Appaently, within oganizations and on labou makets, multiple foci of fainess ae elevant in many contexts, e.g., egading wage decisions concening employees with diffeing sets of human capital investment and heteogeneous needs doing the same wok, layoff decisions concening wokes with diffeing senioity, family esponsibilities, outside options and pefomance, taining investment decisions fo employees with diffeing pefomance, outside options and futue employment pespectives. Fo ou analysis, we chose the issue of the co-employment of egula employees and tempoay agency wokes unde

6 4 heteogeneous contacts, since this allows us to investigate not only two employees, but also two employes with diffeing chaacteistics. Thus, hoizontal as well as vetical social compaisons o foci fo fainess judgements ae available. The game theoetic solution does not hold fo the standad esults of standad ultimatum games with two playes and outside options of zeo. Rathe some kind of fainess consideation will be elevant (see, fo example Güth et al., 1982, Güth & Tietz, 1990, Güth et al., 1998, Camee, 2003). Geneally, in simple two-playe ultimatum games with outside options of zeo, espondes accept splits of between twenty and fifty pecent, while lowe offes ae fequently ejected (Camee, 2003). In ultimatum games with moe than two pesons, multiple foci fo fainess judgements may esult, as hoizontal o vetical compaisons could be elevant. Positive, but diffeent, outside options of espondes futhe add possible foci. Knez and Camee (1995) conduct a thee-playe ultimatum game with one popose offeing to two espondes. These have positive, but diffeing, outside options. Obsevations ae based on the stategy method. Both ultimatum games ae independent fom each othe because neithe esponde s decision has any diect impact on the othe ultimatum game. Howeve, the design offes oppotunities fo social compaison not only vetically, between popose and esponde as in standad ultimatum games, but also hoizontally, between the two espondes, as well. The expeimental data indicate that appoximately half the espondes obviously conducted a social compaison between espondes, while the othe half of the espondes, as well as poposes, did not. Consequently, this setting led to ejection ates of appoximately fifty pecent of all the offes, much highe ates than in othe studies. The authos ague that the intoduction of multiple possible foci fo fainess judgements esulted in a self-seving bias (Babcock et al., 1996) o egocentic selections between these diffeent foci of fainess judgements fo each playe: 4 While poposes select a fainess standad which leads to elatively low offes, espondes focus on efeence points which lead to high offes, and theefoe eject the lowe offes moe fequently than in the simple ultimatum game. Moeove, even fo those individuals who apply social compaisons between espondes, the stength of this motive seems to hinge, in a citical and complex way, on the size of the payoffs eceived by the popose and the othe esponde. Thus, the elevance of hoizontal fainess judgements as opposed to vetical fainess standads, which ae vey well 3 Besides situational factos, diffeent cultual efeence points fo fainess (fo an oveview, see Camee, 2003, chapte 2) and gende diffeences in fainess behavio (e.g. Dickinson & Tiefenthale, 2002) ae found in expeiments. 4 Konow (2000) elaboates on these issues with espect to the accountability pinciple of fainess.

7 5 documented in the expeimental liteatue is quite ambiguous even fo simple situations whee the single ultimatum games ae not diectly inteelated. 5 Extending this insight, we concentate on the analysis of multiple foci fo fainess judgements and constuct thee inteelated ultimatum games with fou playes, whee the esult of the fist ultimatum game influences the thid game. Yet many diffeent contextual elements might be elevant fo the selection of fainess foci. The available infomation will influence the fainess judgement, as non-available infomation might be neglected in the decision o be substituted by ceteis paibus assumptions (see Konow, 2003). Of couse, the impotance of the sequential ode of decisions and the availability of pecise infomation fo decisions has been analysed extensively. The theoetical analysis (e.g., Bagwell, 1995, Huck & Mülle, 2000, Schelling, 1960) pedicts a stategic advantage fo those paties who fist decide about thei shae of a common pie, such that the fist moves eceive the entie suplus fom the bagaining. Howeve, since the wok of Güth et al. (1982), it has been obvious in the expeimental liteatue about ultimatum bagaining that thee is no clea-cut fist move advantage because of the effect of vetical fainess concens. Studies dealing with hoizontal concens attempt to diffeentiate among second moves, i.e., espondes in sequential but independent ultimatum games. Again, theoy pedicts that thee is neithe a fist esponde advantage no a second esponde advantage. Yet, as shown by Chaness & Kuhn (2004), the obsevation of cowokes wages is quite impotant fo wokes' effot decisions. The main esult is that fims eact to the fact that wages become public by suppessing wage dispesion, obviously anticipating negative eactions by wokes to lage wage diffeences. Thus, wage sececy and wage compession policies have elevant implications fo fims pefomance. Consequently, thee is a second esponde advantage; specifically, espondes who obseve othe espondes decisions eceive a pemium in ode to avoid negative wage discimination, which violates thei hoizontal fainess needs. With these esults in mind selected out of the lage numbe of possible elements on which futhe eseach is needed ou pape focuses on the effect of the sequential ode of decisions and the esulting availability of infomation that can be used to fom efeence points. We extend the setting of Chaness & Kuhn (2004) by intoducing an intemediating 5 Knez and Camee (1995, p. 67, fn 2) conclude that to undestand how fainess issues affect oganizations, iche games that ae moe like complicated oganizations would be equied, with thee playe ultimatum

8 6 agency fo tempoay employees. Thus, the fainess consideations fo the elation between employe, agency and tempoay employee ae moe complex. We avoid the salience of equivalent wage offes among employees. Rathe, we will test whethe thee is some kind of substitution effect between hoizontal and vetical fainess consideations. Paticulaly, we analyse the diection of hoizontal fainess consideations. One may ague that playes have othe-egading needs. If so, a atio of offes substantially disciminating against any of the espondes will be ejected, not only offes disciminating against themselves. On the othe hand, one can think of hoizontal fainess consideations as social compaison needs so that offes that would have been below the acceptance theshold when no infomation was available become acceptable due to a favouable hoizontal compaison. Ou esults clealy indicate that, indeed, the public infomation on wages functions as a social compaison nom. Yet, thee is a two-sided effect. Thee ae both second esponde advantages, as well as second esponde disadvantages, due to hoizontal fainess concens. Those espondes who obseve decisions of othe espondes with highe outside options eceive a pemium. Appaently, poposes anticipate the hoizontal fainess concens of the obseving esponde and incease thei offes. Howeve, those espondes who obseve decisions of othe espondes with lowe outside options eceive lowe wage offes, which ae nevetheless acceptable due to thei hoizontal compaison. Thus, poposes benefit fom the espondes expected needs fo social compaison in this setting. This pape is oganized as follows: Section 2 intoduces the model of the intedependent ultimatum game and develops the theoetical pedictions. Section 3 epots the expeimental design. Section 4 elaboates on hypotheses fo playes' behaviou and discusses counteaguments. Section 5 shows the esults of the laboatoy expeiments, while Section 6 concludes the pape with a discussion. 2. The model Thee ae fou playes in ou model: an employe o pincipal P, a tempoay wok agency A, an employee a (hied hand) and a tempoay agency woke (ented hand). Fo both wokes, a and, we assume that effot and effot costs ae exogenously given, and can theefoe be excluded fom ou analysis. Wage payments ae thus the only elevant decision paametes fo the two employees. Both wokes i = {, a} have positive, but diffeing, esevation utilities games being only one step in the ight diection.

9 7 U i fom unemployment benefits. Employee a has a lage esevation utility than tempoay woke, U a > U. If employed, they both do the same wok. Theefoe, each woke has the same level of poductivity, denoted as. We assume that all esevation utilities and poductivity ae common knowledge. Agency A can employ tempoay woke and offe w. If accepts this offe, he is unde contact with A and eans w ; if not, he eans unemployment benefit U. Employe P has to decide on the wage offe w a made to employee a. If a accepts, employee a woks fo P and eans w a ; if he does not agee, he emains unemployed and eans unemployment benefit U a. Additionally, P can contact with tempoay wok agency A to tempoaily hie woke. Fo this, he offes the agency a hiing fee w A. If A accepts (conditional on s acceptance of the contact with agency A), then woke is hied out to P. Othewise, A hies out fo an outside option, yielding U A fo the agency. Defining (1) δ i = 1 if i accepts 0 othewise fo i = {a,, A}, we can compute the pofit of P as (2) π = δ δ + δ ) δ w δ w δ. The pofit of agency A is given by P ( A a a a A A (3) π = δ δ w + (1 δ ) U w ), while ented hand eans A ( A A a A (4) π = δ w + ( 1 δ ) U, and hied hand a eceives (5) π a = δ aw a + ( 1 δ a ) U a. Assuming 2 > U + U + A a 2ε, with ε denoting the smallest feasible incement, and futhe assuming U A > U + ε, the unique subgame pefect Nash equilibium of this game is chaacteized by (6) w w w δ i a A = = = U U U a A + ε + ε + ε = 1 fo i = { a,, A}. Note that the optimal wage offes ae independent of the wage offes to othe playes, but depend solely on the esevation utilities.

10 8 As can easily be seen by this model desciption, the model contains thee inteelated ultimatum games. While the two ultimatum games between the two employes P and A and thei espective employees a and ae independent fom each othe, but could be inteelated indiectly by a hoizontal social compaison if paticipants eceived infomation about the wage offes to the othe playes, the esult of the ultimatum game between the two employes P and A has a diect effect on the game between A and, as the enting out fee w A, if accepted, defines the size of the pie that can be divided between A and. 3. Expeimental designs and teatments The expeiment was conducted at the EconLab at the Univesity of Bonn, Gemany in Octobe, Novembe and Decembe Paticipants wee students fom Bonn Univesity. As the expeiment aims at testing fo the influence of the diffeing fainess foci of the paticipants, we designed two teatments which vay the infomation that is available fo the two employees by changing the sequential ode of the decisions. In both teatments the ultimatum game between P and A is conducted fist. Additionally, A chooses the wage offe w, and all paticipants ae infomed about offe w A and A's decision δ A. The ode of the othe two ultimatum games, howeve, diffes in the two teatments: i. In the teatment staff, employe P offes w a fist, and employee a makes his decision ii. δ a. Employee is infomed about w, w a and δ a, and makes his decision. In teatment temp, the game between agency A and woke is conducted fist. Employee a is infomed about w a, w and δ, and makes his decision. The infomation sets can descibed as follows, with supescipts R and S denoting the teatments staff and temp : 7 (7) I I I I s a s t a t = = = = { w, δ, w } A A a { w, δ, w, δ, w } A A a a { w, δ, w, δ, w } A { w, δ, w }. A A A a 6 Expeiments wee computeized using ztee (Fischbache, 1999). Fo the ecuitment of subjects, we used Osee (Geine, 2004). 7 Paticipants knew about esevation utilities and poductivity, which we have not explicitly mentioned hee.

11 9 Thus in the teatment staff, playes had infomation about an additional hoizontal fainess focus, while in the teatment temp, this additional infomation was available fo playes a. 8 Poductivity of an individual woke was set as 19 expeimental cuency units, while the esevation pay-offs wee defined by U a = 3, U = 1, and U A = 10 expeimental cuency units. Befoe stating the expeiment, paticipants had sufficient time to study the instuctions and ask questions pivately. In the instuctions we efeed explicitly to wage negotiations between fims, agencies, employees and tempoay employees, which made it easy fo paticipants to undestand the entie stuctue of the expeiment. We then applied a questionnaie, which tested paticipants undestanding of the game and of the pay-off stuctue. 9 Only paticipants with a sufficient undestanding and good test esults wee allowed to paticipate in the expeiment. In total, 168 subjects paticipated in the 14 sessions, 7 fo the staff teatment, and 7 fo the temp teatment. Within each session, thee wee 3 subjects fo each ole. Each paticipant played 15 peiods of the intedependent ultimatum game. We used a stange design and constant oles, such that paticipants emained in the same ole thoughout the entie expeiment, but wee anonymously gouped togethe anew by chance in evey ound. The aveage length of the expeiment was 60 minutes. In addition to a fixed show-up fee of 4, playes eaned vaiable pay-offs accoding to thei decisions in the game. All expeimental cuency units wee conveted at a ate of 24 units fo 1.00 at the end of the expeiment. Aveage pay-offs wee 7.26 (standad deviation, heeafte abbeviated sd, 1.36) fo the P ole, 3.16 (sd 0.75) fo the A ole, 5.05 (sd 0.98) fo the a ole, and 3.08 (sd 0.67) fo the ole. 4. Hypotheses Stating with the esults of standad ultimatum games, ou fist hypothesis tagets vetical fainess issues between agency A and tempoay woke, on the one hand, and between employe P and employee a, on the othe hand. H 1 (Standad vetical fainess concens between employes and employees): Against the backgound of the coopeative Nash solution, the size of the pie that can be divided between the playes of the ultimatum games has to be calculated as a suplus ove the esevation pay-offs. Standad esults of ultimatum games with zeo esevation utilities show 8 The instuctions fo teatment staff can be found in appendix A. 9 See appendix B.

12 10 that offes of between 20 and 50 pecent of the total size of the pie ae fequently made and accepted (e.g., Camee, 2003). Combing both esults, fo the standad ultimatum games, i.e., between P and a, and between A and, we expect poposes offes to shae the espective size of the suplus ove esevation utilities appoximately equally. Fo the uppe limit, we expect 0.5, while fo the lowe limit we expect 40 pecent of the uppe limit, i.e., 0.2 of the suplus, o (8) w = U wa = U a + α 1 + α 2 ( w U ) A ( U ) a with 0.2 < α 0.5 and 0.2 < α Offes below the lowe theshold will seldom be made and will fequently be ejected; offes lage than the uppe theshold will seldom be made. Howeve, fo the intedependent ultimatum game between P and A, the evenue has to be divided between thee paties in the tempoay wok agency elationship. Theefoe, we expect the following effect. H 2 (Adaptation of vetical fainess concens to a highe numbe of espondes): A and P eact to the diffeing numbe of pesons that shae the pie by adapting the fainess standads known fom two-playe ultimatum games to the thee-playe ultimatum games: They divide the suplus poduction evenue not by half, but such that P etains a shae of about one-thid fo himself and A eceives appoximately two-thids to shae with woke. Theefoe, each playe eans oughly one-thid of the suplus. As a consequence, the uppe limit is two-thids, i.e., the atio that divides the suplus of the fist ultimatum game by the elation of the numbe of the playes paticipating in the subodinated ultimatum game and the numbe of the paticipating playes in both games. Fo the lowe limit, we expect similaly to the two-peson case appoximately 0.4 of the uppe limit, yielding Theefoe, we pedict (9) w = U + β ( U ) with β A A Smalle offes will seldom be made, and if made, they will fequently be ejected. A Hypotheses H 1 and H 2 do not conside effects that ae caused by hoizontal fainess noms. At the othe exteme, one could ague that vetical noms ae dominated by hoizontal fainess noms, i.e., equal payment fo equal wok. Equal payment fo equal wok could only be offeed at the expense of the tempoay wok agency o at the expense of the egula employee. Thus, one could pedict that vetical fainess noms will be neglected using the following hypothesis.

13 11 H 3 (Dominance of hoizontal fainess noms): The wage offes to employee a and tempoay woke do not diffe since they do the same wok. Consequently, w (10) = 1. w a Howeve, given the stong expeimental esults on the elevance of vetical fainess concens, it seems less plausible to assume that playes exclusively pay attention to hoizontal fainess noms. Theefoe, the main pupose of ou study is to show to which extent hoizontal fainess noms intefee with vetical fainess concens. In the teatment staff, the ultimatum game between employe P and employee a takes place befoe the ultimatum game between agency A and tempoay woke. Thus, a has to make his decision, while ignoing the diect hoizontal wage compaisons between the two wokes. In the teatment temp, the infomation about agency A s offe to hie employee and s acceptance decision is available to a. Thus, a compaison of the behaviou fo a between the two teatments eveals the additional effect of a social compaison between a and. Fo, the same effect could be elevant: While he decides without any infomation about P s offe to a and a s eaction in the teatment temp, this infomation is added in the teatment staff : specifically, thee exists an additional possible focus fo hoizontal fainess concens. Theefoe, we pedict H 4 (Effects of adding hoizontal wage infomation): Adding a efeence point fo the hoizontal fainess concens of wokes changes acceptance ates in compaison to the case in which thee is no hoizontal infomation in two possible ways: On the one hand, due to a favouable hoizontal compaison, e.g., w /w a 1 fo in the staff teatment, one could ague that social compaison (o self-egading hoizontal fainess concens) makes offes acceptable that would have been below the acceptance theshold wee no infomation available. Similaly, due to an inauspicious hoizontal compaison, e.g., w a /w < 1 fo a in the temp teatment, offes ae not acceptable that would have been above the acceptance theshold wee no infomation available. On the othe hand, one could ague that playes may cae about a decent atio on offes that do not disciminate against eithe of the two playes. We efe to this idea as the othe-egading needs. Hence, a atio w /w a (w a /w, espectively) which diffes substantially fom one leads to ejections, although the undelying offe would be acceptable if obseved without consideation of the social compaison.

14 12 Finally, in thei social-compaison, thee-peson ultimatum game, Knez and Camee (1995) obseve high ejection ates unde multiple fainess foci and intepet this as a esult of egocentic selection pocesses. While poposes select fainess standads which lead to elatively low offes, espondes focus on efeence points which lead to high offes, and theefoe eject the lowe offes moe fequently than in the simple ultimatum game. We have to stess that in this expeiment offes wee made simultaneously. In contast, we ague that P and A playes in ou setting conside the influence of the hoizontal fainess concens of playes a and since offes ae made sequentially. H 5 (Anticipation of fainess foci): P and A playes in the expeiment anticipate the effects of the social compaisons of employees in the temp teatment and of tempoay employees in the staff teatment. Playes, who decide about the wage in the staff teatment, eceive a pemium in addition to thei wages, while playe a suffes a wage eduction in the temp teatment. 5. Results We fist look at the aveage offes fo both teatments, which ae shown in Table 1. Obviously, offes do not follow the game-theoetic pediction. Additionally, aveage offes to a ae lage than wage offes to. We find that both accepted and oveall offes w A, w a and w ae significantly highe in the staff teatment than in the temp teatment. 10 Since P eans the esiduals of the ultimatum games, the aveage pay-off fo P is smalle in the staff teatment than in the temp teatment. Aveages staff tempoay w A (sd) (2.25) (3.16) w A δ A = 1 (sd) (1.94) (2.74) w a (sd) 8.94 (2.70) 8.38 (2.84) w a δ a = 1 (sd) 9.44 (2.59) 8.70 (2.91) w (sd) 6.09 (1.94) 5.44 (2.01) w δ = 1 (sd) 6.52 (1.84) 5.79 (2.03) Table 1: Aveage and aveage accepted offes The analysis fo H 1 ties to find the elation between ou expeimental obsevation and the standad esults of ultimatum games. Heeby, we focus on vetical fainess aspects. Fo this pupose, we have to detemine the numbe of obsevations that fall within the paametes

15 13 of the acceptance theshold α. Indeed, fo employees, we find appoximately 86 pecent in the staff teatment and 83 pecent in the temp teatment, espectively, of all w a that fall only within this inteval. Moeove, consideing the ejection ates fo offes, as shown in Figue 1(a), we have to confim H 1 fo employees in the staff teatment since we find a damatic incease in the fequency of ejection fo α < 0.2. Thus, fo the staff teatment, it seems that thee is an acceptance theshold of appoximately α = 0.2, which coesponds to offes of 7. Howeve, the ejection ate in the temp teatment looks athe diffeent. Hee, even fo offes smalle than 7, thee is no damatic incease in the fequency of ejection. Fo tempoay employees, we have to diffeentiate between the cases in which agencies eject the offe of P and the cases in which agencies accept thei offes. 11 Again, we find the majoity of stated offes ae within the limits suggested if agencies eject (94 pecent fo the staff and 91 pecent fo the temp cases). When the agencies accept w A, howeve, in the staff teatment (the temp teatment), they choose w in 73 pecent (70 pecent) of the cases fom the inteval defined in equation (8). Figue 1: Rejection ates of (a) employees and (b) tempoay employees As shown in Figue 1(b), the ejection ates again suggest a ejection theshold of appoximately α = 0.2, 12 although the stuctue is not as clea as fo employees. Quite supisingly, in the temp teatment, the ejection by tempoay wokes follows the same 10 Fo all diffeences, we find p < 0.01, using two-sided Wilcoxon signed ank test fo offes and using two-sided Mann-Whitney test fo accepted offes. 11 In the fome case, the limits ae 6 > w > 3, while in the latte case, we have w A > w > w A. 12 Hee, the data of cases that agencies eject and that agencies accept is pooled.

16 14 unusual stuctue as fo employees. Ou intepetation is that the moe complicated ultimatum game between agencies and tempoay employees makes it moe difficult to see a clea stuctue egading what is consideed to be acceptable and what is not. 13 Then social compaison tansfes this ambiguity to employees. The opposite effect applies fo the staff teatment. Hee, the clea-cut stuctue of ejection thesholds fo employees is tansfeed ove to tempoay employees. Oveall, the data do not eject H 1 fo employees and tempoay employees in the staff teatment. Yet, ejection behaviou in the temp teatment diffes substantially. With espect to H 2, we find a diffeent patten. Only 57 pecent of all obsevations fo w A in the staff teatment, and even less, only 29 pecent in the temp teatment, ae within the pedicted ange. This esult may indicate that behaviou in the independent ultimatum game does diffe quite substantially fom obseved behaviou in standad ultimatum games. Note that the median β fo the temp teatment is 0.222, while it is fo the staff teatment. Theefoe, offes ae much lowe than obseved in standad games. With espect to the ejection ate of agencies fo w A as shown in Figue 2, we do not obseve a clea-cut acceptance theshold. Thee is no theshold to be obseved since thee is no step-wise incease in the ejection ate as offes decease. Thee is little diffeence between ejections in the staff teatment and ejections in the temp teatment. Thus, thee is no diffeence fo agencies acoss the two teatments. In geneal, we find little expeimental evidence suppoting H 2, i.e., that offes in the ultimatum game between P and A ae only adapted to the highe numbe of espondes. Behaviou diffes quite substantially fom what could be pedicted on the basis of standad ultimatum games. 13 When agencies accept thei offes, one may ague that tempoay employees eject offes since they conside the faction that agencies ean fom the entie poduction, i.e., (w A -w )/19, as being unfai. Howeve, since the coelation between the acceptance on the pat of tempoay employees and this atio fo the cases in which agencies accept thei offes does not diffe significantly fom zeo fo both teatments (-0.17 fo the staff teatment, and fo the temp teatment; two-sided Peason coelation tests cannot eject the hypothesis that coelations equal zeo on an α= 0.05 level), we will not focus on this issue in the following analysis.

17 15 Figue 2: Rejection ates of agencies Fo the analysis of H 3, we have to conside the effects of hoizontal fainess concens. One theoetical possibility is that hoizontal fainess noms between wokes could completely offset vetical fainess issues. As mentioned ealie, we do not expect to find this dominance. The aveage development of w /w a thoughout the entie expeiment is shown in Figue 3. Figue 3: The aveage elation w /w a acoss peiods (a) of all offes and (b) of offes that wee accepted by a and only. Neithe the atio based on all offes (shown in Figue 3(a)) no the atio based on accepted offes only (shown in Figue 3(b)) suppots H 3. Howeve, we find a stable atio w /w a in both teatments. Even if we only conside offes that wee accepted by employee a

18 16 and tempoay woke, thee is no significant diffeence between the atios in the staff teatment and in the temp teatment. 14 Of couse, in geneal, w will be smalle than w a, as the suplus that can be divided between agency A and woke is much smalle than the suplus that can be divided between employe P and employee a. Thus, the atio does not suppot the claim of equal payment fo equal wok; but we find that consideing the esults fo accepted offes only a wage atio between 0.7 and 0.8 is acceptable. So fa, we have seen that thee is a quite stable atio between wage offes fo a and. Theefoe, it seems impotant to claify the extent to which hoizontal fainess concens can substitute fo vetical fainess concens. H 4 aims at this question and states that adding a efeence point fo hoizontal fainess concens esults in wage offes that ae acceptable (inacceptable) in the no-hoizontal-infomation teatment, but inacceptable (acceptable) in the infomation teatment. In ode to test the expeimental data fo this question, we un a seies of logit estimations fo the ejection decisions δ a and δ. Of couse, we have to conside that thee ae impotant intepesonal diffeences. Theefoe, we estimate two individual fixed effects models with dependent vaiables δ a, and δ, espectively. Let us define the dummy vaiable as σ = 0 if obsevations ae fom the staff teatment and as σ = 1 if obsevations ae fom the temp teatment. As independent vaiables indicating vetical fainess concens, we use the atio α i fo i = 1,2, as defined in equation (8). We can pedict a positive coefficient fo α i in both models, i.e., the positive influence of α i on both vaiables δ a, and δ. Howeve, we expect vetical fainess consideations to have less influence on δ a in the temp teatment than in the staff teatment, i.e., α i to have a lowe influence on δ a if σ = 1, wheeas the influence of vetical fainess consideations on δ in the temp teatment is stonge than in the staff teatment, i.e., α i to have a highe influence on δ if σ = 1. Results fo this estimation ae epoted as models v in Table 2. Additionally, we test the atio w /w a, indicating hoizontal fainess concens in the infomation teatments, i.e., in staff fo vaiable δ and in temp fo vaiable δ a. 15 Yet, the diection of influence of hoizontal fainess concens diffes consideably acoss the values of w /w a and playes. One may think of two diffeent lines of agumentation. We call one line social compaison needs. Hee, hoizontal fainess concens impove the acceptability of offes if the atio favous one s own payoffs. Fo this line we can 14 On an α= 0.05 level, a two-sided Wilcoxon signed ank test cannot eject the hypothesis that atios fo all offes of both teatments ae the same; on an α= 0.05 level, a two-sided Mann-Whitney test cannot eject the hypothesis that atios fo the accepted offes of both teatments ae the same.

19 17 pedict that the moe the atio favous tempoay employees, the lage the positive influence of w /w a on δ. Thus, we can pedict a positive coefficient fo w /w a if σ = 0. Likewise, we can ague that the moe the invese atio w a /w favous employees, the lage the influence of w a /w on δ a. Thus, we expect a positive coefficient fo w a /w if σ = 1. The estimation esults fo social compaison needs ae epoted as models h soc in Table 2. Yet, the othe line of agumentation suggests that playes cae about a decent atio of offes, which does not disciminate against eithe of the two playes. We call this line the othe-egading needs. Thus, the moe the atio w /w a (the invese atio w a /w, espectively) deviates fom one, the less acceptable the offes ae. Hence, we define a vaiable that measues the quadatic distance fom equal offes, i.e., τ 1 =(w /w a -1) 2 and τ 2 =(w a /w -1) 2. This measue is expected to show a negative coefficient. Results of the estimations ae summaized as models h o in Table 2; 16 goodness of fit is epoted by the Akaïke infomation citeion (AIC). δ a σ = 0 δ a σ = 1 δ σ = 1 δ σ = 0 Model v v h soc h o v V h soc h o constant (2.831) -2.03** (0.816) -2.45*** (0.777) -2.02** (0.79) -2.28*** (0.691) -2.03*** (0.767) -2.53*** (0.864) -2.37** (1.01) α i 22.88*** (0.45) 18.68*** (3.362) 17.81*** (3.333) 18.3*** (3.355) 10.77*** (1.906) 9.55*** (1.71) 8.35*** (1.768) 9.45*** (2.078) w a /w 0.74** (0.334) w /w a 1.38*** (0.534) τ 1 (0.168) 0.11 τ 2 (1.396) AIC Table 2: Logit egession esults fo coefficients 15 We set the atios equal to zeo fo all obsevations whee espondes who had to decide fist ejected the offe, i.e., δ a =0 in the staff teatment and δ =0 in the temp teatment. 16 Standad eos in paenthesis;*** significant on α = 0.01 level, ** significant on α = 0.05 level, * significant on α = 0.1 level.

20 18 As expected, contact acceptance by employee a is stongly and significantly influenced in a positive way by vetical fainess concens (α i ) in both teatments. Confiming ou pedictions, vetical fainess concens ae highe when no infomation is available, i.e., σ = 0. On the othe hand, hoizontal fainess concens in the sense of social compaison, i.e., w a /w, show a significant positive influence on a s contact acceptance if playe decides fist. Howeve, the insignificant vaiable τ 1 indicate that hoizontal fainess concens follow the idea of social compaison athe than othe-egading needs. Hence, we find a substitution effect fo some w a offes. Fo an offe w a, which is acceptable if no hoizontal infomation is povided, the pobability that a will accept deceases in the infomation teatment as the atio w a /w deceases. Fo the offe acceptance by tempoay employee, vetical fainess concens stongly and significantly influence the pobability that a tempoay employee will accept a contact. The highe α i is, the highe the pobability of acceptance. Again, this effect is less ponounced in the infomation teatment, i.e., σ = 0. Hee, hoizontal fainess concens matte quite a lot. The highe the atio w /w a, the highe the pobability that will accept the offe. Fo an offe w which is inacceptable if no hoizontal infomation is povided, the pobability that will accept inceases in the infomation teatment as the atio w /w a inceases. Again, social compaison mattes, but we do not find othe egading needs to have a significant effect. Summaising the esults of the estimation models, we can suppot H 4, i.e., that vetical and hoizontal fainess consideations ae substituted fo one anothe. This effect efes to some social compaison; that is, an offe is acceptable due to a favouable compaison to anothe playe s offe, although the offe may not be acceptable if no compaative infomation is available. With espect to H 5, we claim that ejection ates do not incease unde multiple fainess foci because poposes adapt thei offes, anticipating the hoizontal fainess concens of espondes. Theefoe, we conside the development of efficiency and payoffs acoss expeimental peiods. Please note that the efficiency povides a mio image of ejections. Wheneve offes ae ejected, efficiency is deceased. Figue 4(a) shows aveage oveall efficiency, while we nomalized efficiency such that 0 efficiency chaacteizes the case in which playe a and playe eject thei offes. Figue 4(b) epots the development of efficiency within the ultimatum game between P and a. Finally, Figue 4(c) shows the development of efficiency within the ultimatum game between A and. As one can see, thee is no diffeence between the teatments fo efficiency in geneal and fo the sepaate

21 19 ultimatum games. A non-paametic test suppots this hypothesis. 17 Moeove, we do not obseve a clea time effect. Efficiency emains at appoximately 80 pecent thoughout the entie expeiment. Figue 4: Aveage efficiency pe peiod (a) oveall, (b) within ultimatum game between P and a, and (c) within the ultimatum game between A and thoughout the expeiment Howeve, we find a teatment effect with espect to the distibution of eanings. Figue 5(a) compaes the eanings of playe P and playe A acoss teatments and peiods. As one can 17 Two-sided Wilcoxon signed ank tests cannot eject the hypothesis of equal efficiency on an α = 0.05 level.

22 20 see, on aveage playe P eceives lowe eanings in the staff teatment than in the temp teatment. Indeed, a non-paametic test confims this esult. 18 Although we found ealie that the aveage offe w A is significantly highe in the staff teatment than in the temp teatment, Figue 5(a) illustates that playe A does not ean moe in the staff teatment than in the temp condition. 19 Thus, the pemium is tansfeed to playe. In paticula, as one can see in Figue 5(b), the eanings fo both playe a and playe ae highe in the staff teatment. Again, a non-paametic test confims this esult. 20 Thus, one could say that tempoay employees eceive a pemium payment in the staff teatment in ode to satisfy thei hoizontal fainess concens. To the contay, employees suffe a significantly lowe income in the temp teatment than in the staff teatment. 21 Aveage eanings ae significantly lowe if tempoay agents decide fist about thei wage offe; they ae highe if the decision of the employees comes fist. Figue 5: Aveage eanings of (a) of playes P and A and (b) of playes a and pe peiod thoughout the expeiment Evidently, these esults confim H 5. Poposes adapt thei offes to the diffeent fainess foci. Moeove, one could indeed speak of some kind of fist esponde advantage due to hoizontal fainess concens, e.g., self-egading social compaison needs. The data show that poposes ty to maintain stable atios between offes acoss teatments. Thus, 18 A one-sided Wilcoxon signed ank test ejects the hypothesis of equal efficiency on an α = 0.05 level. 19 A two-sided Wilcoxon signed ank test confims this claim since it cannot eject the hypothesis of equal efficiency on an α = 0.05 level. 20 A one-sided Wilcoxon signed ank test ejects the hypothesis of equal efficiency on an α = 0.05 level. 21 As ealie, a one-sided Wilcoxon signed ank test ejects the hypothesis of equal efficiency on an α = 0.05 level.

23 21 espondes eceive a pemium if thei focus is on the decisions of othe espondes with highe outside options. Howeve, those espondes who focus on decisions of othe espondes with lowe outside options suffe a loss in thei incomes since they accept lowe wage offes due to thei hoizontal compaison. 6. Conclusion Appaently, fainess concens matte fo economics. And they have indeed been studied extensively by economists ove the last decades. Howeve, the ovewhelming majoity of those studies have focussed exclusively on the issue of vetical fainess consideations. Yet, social compaison and the esulting hoizontal fainess concens ae impotant in most aeas of ou lives. Ou expeimental esults demonstate the combined elevance of vetical and hoizontal fainess foci in inteelated ultimatum games. Howeve, the sequential ode of decisions and bagaining steps, which detemines the available infomation fo hoizontal compaison, is cucial fo detemining the elevance of hoizontal fainess concens. Theefoe, thee ae effects that favou the payoffs of the second esponde, but also effects that decease the payoffs of second espondes. If the woke with lowe outside options bagains on the smalle pie, i.e., if tempoay employee and agency A bagain, and wokes with highe outside option obseve the outcome of the bagaining, i.e., employee a, the effect of this sequential ode is disadvantageous fo both wokes. Those playes who have lowe outside options accept offes due to vetical fainess concens. Yet, playes with highe outside options accept wages due to thei hoizontal compaison. By contast, if wokes with highe outside options choose fist, the opposite occus fo both wokes. Yet it hams poposes, i.e., playe P. Hee, espondes with lowe outside options ask fo highe wage offes based on thei hoizontal fainess concens. As we could show in this expeiment, behaviou is influenced in impotant ways by hoizontal fainess noms. In paticula, in ou study we focussed on the inteplay between vetical and hoizontal fainess consideations elated to wage decisions. The high elevance of combined vetical and hoizontal fainess foci fo labou economics is obvious. Since the topic of tempoay agency wok, and, theefoe, commonly, heteogeneous vesus homogenous payment fo simila wok, is at the cente of extensive political discussion in the

24 22 Euopean Union, the significance of effects of multiple fainess foci is inceasing athe than disappeaing. Yet, the co-employment of pemanent and tempoay wokes is only one example fom a wide class of situations. Fims and goups offe multiple souces fo fainess foci. Refeences Alewell, D., C. Fiedich & S. Matin (2004), Gleichbehandlungsgundsatz: Ende de Zeitabeit? Ausgewählte betiebliche Wikungen de Zeitabeit aus ökonomische Pespektive. In: Stuck, O. & C. Köhle (eds.), Beschäftigungsstabilität im Wandel; empiische und theoetische Befunde, München, Meing, p Babcock, L., X. Wang & G. Loewenstein (1996), Choosing the wong pond: Social compaisons in negotiations that eflect a self-seving bias, The Quately Jounal of Economics, 111(1), Bagwell, K. (1995), Commitment and obsevability in games, Games and Economic Behavio 8, Bolton, G.E. & A. Ockenfels (2000), ERC: A theoy of equity, ecipocity, and competition, Ameican Economic Review 90(1), Camee, C.F.(2003), Behavioal game theoy; Pinceton: Pinceton Univesity Pess. Chaness, G. & P. Kuhn (2004), Do co-wokes' wages matte? Theoy and evidence on wage sececy, wage compession and effot, IZA Discussion Pape 1417 Dickinson, D.L. & J. Tiefenthale (2002), What is fai? Expeimental evidence, Southen Ecoomic Jounal. 69(2), Feh, E. & K.M. Schmidt (1999), A theoy of fainess, competition, and coopeation, The Quately Jounal of Economics, 114(3), Fischbache, U. (1999), z-tee Zuich toolbox fo eadymade economic expeiments Expeimente s manual, Woking Pape 21, Institute fo Empiical Reseach in Economics, Univesity of Zuich. Gächte, S. & E. Feh (2002), Fainess in the labou maket: A suvey of expeimental esults. In: Bolle, F. & M. Lehmann-Waffenschmidt (eds.), Suveys in expeimental economics: Bagaining, coopeation and election stock makets, Heidelbeg: Spinge, Geine, B. (2004), An online ecuitment system fo economic expeiments. In: Keme, K. & V. Macho (eds.), Foschung und wissenschaftliches Rechnen 2003, Beicht de Gesellschaft fü wissenschaftlichen Datenveabeitung Göttingen 63, Güth, W., S. Huck & W. Mülle (1998), The elevance of equal splits: On a behaivoal discontinuity in ultimatum games, Discussion Pape Sondefoschungsbeeich 373, Humboldt- Univesity Belin. Güth, W., R. Schmittbege & B. Schwaze (1982), An expeimental analysis of ultimatum bagaining, Jounal of Economic Behavio and Oganization 3, Güth, W. & T. Tietz (1990), Ultimatum bagaining behavio: A suvey and compaison of expeimental esults, Jounal of Economic Psychology 11, Huck, S. & W. Mülle (2000), Pefect vesus impefect obsevability: An expeimental test of Bagwell s esult, Games and Economic Behavio 31, Kahnemann, D., J.L. Knetsch & R. Thale (1986), Fainess as a constaint on pofit seeking: Entitlements in the maket, Ameican Economic Review 76(4), Knez, M.J. & C.F. Camee (1995), Outside options and social compaison in thee-playe ultimatum game expeiments, Games and Economic Behavio 10, Konow, J. (1996), A positive theoy of economic fainess, Jounal of Economic Behaviou and Oganization 31, Konow, J. (2000): Fai shaes: Accountability and cognitive dissonance in allocation decisions, Ameican Economic Review 90(4),

25 23 Konow, J. (2001), Fai and squae: The fou sides of distibutive justice, Jounal of Economic Behavio and Oganization 46, Konow, J. (2003), Which is the faiest one of all? A positive analysis of justice theoies, Jounal of Economic Liteatue 41, Leventhal, G.S. (1980), What should be done with equity theoy? New appoaches to the study of fainess in social elationships. In: Gegen, G.S., M.S. Geenbeg & R.H. Willis (eds.), Social Exchange, Advances in Theoy and Reseach, New Yok: Plenum Pess, p Rabin, M. (1993), Incopoationg fainess into game theoy and economics, Ameican Economic Review 83(5), Schelling, T. (1960), The stategy of conflict, Cambidge, M.A.: Havad Univesity Pess. Scott, P. (2003), Fainess as a souce of hysteesis in employment and elative wages, Discussion Pape, Univesity of Massachusttes.

26 24 Appendix A Tanslation of the instuctions fo tempoay teatment 22 Thank you vey much fo paticipating in this expeiment. Please efain fom talking to o communicating with othe paticipants in any way duing the expeiment. Please note that we have to exclude you fom futhe paticipation if you beak this ule. If you have any questions, please aise you hand. One of the expeimentatos will then pivately answe you questions. Please ead these instuctions vey caefully. In this expeiment, you will make decisions by which you can ean money. How much you will ean depends on you decisions and the decisions of othe paticipants. Duing the expeiment, you can ean points, which will be exchanged fo Euos at the end of the expeiment. The exchange ate is 24 points fo 1 Euo. Additionally, each of you will eceive 4 Euo independent of the numbe of points you ean in the expeiment. At the beginning of the expeiment you will be assigned a ole, e.g., a task, by chance. This ole emains unchanged fo the complete duation of the expeiment. In total, you will inteact with the othe paticipants fo 15 ounds. In each ound, thee othe paticipants will be anonymously assigned to you goup by chance. The oles of the fou diffeent types of paticipants ae labeled fim, agency, employee and tempoay woke. The fim can poduce poducts with he employee and/o with the tempoay woke and gain pofits. The agency can hie the tempoay woke and hie him out to the fim o othe fims. Specifically, each paticipant has the following tasks: The fim has to offe to the agency the fee w A to hie the tempoay woke. Additionally, it has to make an offe w a to the employee. The agency has to accept o eject the fim s fee offe w A. Additionally, it has to make an offe w to the tempoay woke. The agency can cancel tempoay contacts with the fim if it has not succeeded in contacting with the tempoay woke. The employee has to accept o eject the fim s offe w a. The tempoay woke has to accept o eject the agency s wage offe w. The payoffs of the paticipants ae calculated as follows: If the employee accepts wage offe w a, he eans w a points; othewise he eans 3 points fom othe souces. 22 This is a tanslation of the oiginal Geman instuctions. Changes in the staff teatment ae indicated by footnotes.