Sticks and Carrots in Procurement: An Experimental Exploration

See e.g. Laffont and Tirole (1993), Bajari and Tadelis (2001), Albano et al. (2006). The USA government procurement uses both fixed-price and cost-plus contracts. Recently, President Obama has signed a memorandum addressed to Federal Executive Departments and Agencies where he explicitly declares that “there shall be a preference for fixed-price type contracts” to eliminate wasteful and inefficient government contracts (Obama 2009).

In Anglo-Saxon countries liquidated damages are only enforceable up to the courts’ estimates of the harm produced by the contractual violation; and in most other countries, contractual penalties are only enforceable up to a fraction of the value of the exchange.

Kahneman, Knetsch, and Thaler (1986) pointed out that, in violation of normative standards, people tend to be more sensitive to losses than to foregone gains, which renders their preferences vulnerable to framing effects (see also Tversky and Kahneman 1986).

Hannan, Hoffman, and Moser (2005) confirm Luft (1994)’s result that agents prefer bonus contracts and perceive them to be fairer than penalty contracts. They also find that incentives framed as bonuses foster more trust and reciprocity, although this effect is dominated by a substantially larger increase in effort induced by penalties because of loss aversion. In our set-up, bonus and penalty contracts are such to induce even fully rational agents to exert the maximum level of effort in the verifiable task, hence in principle loss aversion should not induce differences in agents’ behavior.

In the words of Holmstrom and Milgrom (1991), “Incentive pay serves not only to allocate risks and motivate hard work, it also serves to direct the allocation of the agents’ attention among their various duties.”

See MacLeod (2007) for an excellent recent survey.

Bonuses are typically seen as “positive” incentives, favoring trust among the parties and possibly facilitating cooperation (Luft 1994; Hallman, Hartzell, and Parsons 2005). Fehr and Gächter (2002) observe that if, for a given incentive contract, multiple effort levels are consistent with equilibrium behavior, then the framing may affect the selection of equilibria, thus leading to a different outcome for the game. They also point out that, in cases where concerns of fairness and reciprocity are important, the framing of incentives may affect perceptions of the kindness or hostility of the principals’ actions, which are crucial for reciprocal responses.

Hossain and List (2012) also mention that one of the reasons why their framing manipulation is much more powerful among groups of workers than among individuals, is that in the former case workers face a riskier setting because of team production externalities.

If our result is confirmed in future studies, it may imply that social preferences are not only complementary to reputation, as suggested by Brown, Falk, and Fehr (2004), but may actually be essential to its effectiveness.

In our setting, explicit incentives are deterministic, and the punishment in the event of a seller’s default is automatic and cost-free for the buyer. This is consistent with procurement practice but contrasts with the typical set-up adopted in previous experiments (Andreoni, Harbaugh, and Vesterlund 2003; Fehr and Schmidt 2007; Sutter, Haigner, and Kocher 2010).

We opted for an uncertain time horizon to minimize backward induction reasoning, but also to mimic real-life procurement interactions more closely. This marks a difference from previous experiments where subjects knew they were playing for a fixed number of rounds: in Brown, Falk, and Fehr (2004) the number of rounds was 15, in Falk, Huffman, and Macleod (2008), in Fehr and Gächter (2002) and in Gächter, Kessler, and Königstein (2010) it was 18, 12 and 10, respectively.

In line with the instructions of Brown, Falk, and Fehr (2004) – our instructions explicitly state that “The seller is not compelled to provide the quality requested by the buyer, nor to deliver the good by the deadline requested.”

Detailed analytical solutions are presented in Bigoni, Spagnolo, and Valbonesi (2009).

Our theoretical predictions and experimental results are specific to this production function. Whether a higher degree of complementarity between the two efforts would have generated different results is an interesting research question, which is outside the scope of this study.

Even when all traders are rational, Kreps et al. (1982) show that cooperation can be sustained in a finite horizon game by players who are aware of its finite duration as long as they believe that, with some positive probability, they are facing a certain “commitment type”.

As Brown, Falk, and Fehr (2004) write “In the presence of fair-minded workers, who respond to generous offers with a generous effort level, employers have an incentive to offer a rent in the final period. This rent can be used to discipline the selfish workers in the non-final periods by the threat of termination of the relationship.”

We had one session with 14 subjects, and one with 12 because some subjects failed to show up.

The instructions explicitly informed the subjects that, in the event their total number of points at the end of the experiment was lower than zero, they would not have to pay any money to the experimenters, and that they would receive the show-up fee regardless. This circumstance, however, never took place.

In all the analyses contained in this section, we consider only the first 15 periods of each session, so to even out the end game effect. If not otherwise specified, the unit of observation is the average, per session, per period; this allows us to equalize to 45 the number of observations per treatment. In all regressions, we obtain qualitatively similar results by using individual-level data, and adopting a model with random effects at the subject level and standard errors robust for clustering at the session level. Regression results are available from the authors upon request.

Frank, Gilovich, and Regan (1993) report several empirical results suggesting that economics and business students appear less cooperative than others. Yezer, Goldfarb, and Poppen (1996) provide further evidence that undergraduate students of economics display uncooperative behavior in specialized games or surveys, although they disagree with Frank, Gilovich, and Regan (1993) on the additional conclusion that economists’ “real-world” behavior is also less cooperative. More recently, Gaechter et al. (2012) find that economists cooperate less – i.e. contribute less – than non-economists in a leader–follower experimental set-up.

Evidence of the framing effects of instruction in this and similar settings is also somewhat mixed. Charness, Frechette, and Kagel (2004) found that providing complete and detailed information about the players’ feasible payoffs – in their case by means of a profit table – significantly reduces the impact of intrinsic motivation and fairness in gift exchange laboratory experiments. By contrast, Gürerk and Selten (2012) find that granting subjects access to a payoff table induces a more cooperative behavior in oligopoly experiments. Engelmann and Ortmann (2009) recently found that the framing of instructions exerts only limited effects on subjects’ behavior in gift-exchange games. Hoffman et al. (1994) did find that ultimatum game offers are lower in a buyer–seller frame than in a standard bargaining frame, but in a following study, Hoffman, McCabe, and Smith (2000) report treatments where offers in a buyer–seller frame were similar to those from bargaining frames.

This difference does not affect our main results, which stem from the comparison across treatments, because this feature of our design equally affects all treatments.

We drop from the following analyses 3 out of 1,290 observations, in which e1<e1∗ and e2>0.

We set c(e1∗)=0 when no explicit incentives are present.

The difference is not significant when we compare Treatment T with either B or P.

A bonus contract with wage w is equivalent to a penalty contract with wage w+20. Hence, to capture only the framing difference between bonuses and penalties, when the contract includes a bonus the fixed payment is increased by 20 points.

In Treatment T, e1∗=0 and the cost of voluntary effort coincides with the total cost of effort exerted.

These additional controls include: the fixed compensation offered to the seller, a dummy indicating whether the offer is public or private, the number of private offers received by a seller in the period, and the desired levels of non-verifiable and verifiable effort.

We thank Dirk Engelmann for pointing this out.

To avoid such an asymmetry across treatments, we should have set different restrictions on the set of possible fixed wages w, depending on the type of incentive adopted. More specifically, we should have allowed buyers to choose w from the set [16,146] in contracts with a penalty, and from the set [−4,126] in contracts with a bonus, so that the minimum possible earning for the seller is zero, both in trust contracts and in contracts with explicit incentives. Yet, we believe this would have seriously distorted subjects’ perception of the alternatives.

25.9% of the total offers are unacceptable, and 11.2% of the contracts originate from unacceptable offers.

21.4% vs 7.3%. The Wilcoxon signed-rank test with N = 8 confirms that the difference is significant at the 5% level. In 4 out of 12 sessions, all buyers made at least two mistakes, so no matched observations are available.

11.2% vs 6.4%. The difference is significant at the 5% level, according to a Wilcoxon signed-rank test with N=11. In one out of 12 sessions, all sellers made at least two mistakes, so no matched observations are available.

28.0 vs. 20.5. The difference is significant at the 5% level, according to a Wilcoxon signed-rank test with N=9 (the comparison is not possible for 3 sessions).

11.2 vs. 14.8. The difference is significant at the 5% level according to a Wilcoxon signed-rank test with N=10 (the comparison is not possible for 2 sessions).

Filiz-Ozbay and Ozbay (2007) also report the presence of overbidding in experimental auctions and explain it by arguing that bidders may overbid if they anticipate regret for losing the auction.

According to a GLS panel regression, however, the difference is significant only when we compare contracts with bonuses and with penalties in Treatment I, while the estimated coefficients from Treatments B and P are not significantly different at the 10% level according to a Wald test.

In our game, this is always equal to w−(e∗)2.

Table 17 in Appendix B.