WPS7345 Policy Research Working Paper 7345 High-Powered Incentives and Communication Failure Ajit Mishra Sudipta Sarangi Development Economics Vice Presidency Office of the Chief Economist June 2015 Policy Research Working Paper 7345 Abstract This paper uses a donor-provider-agent framework to study lead to breakdown of communication between providers and the role of provider incentives for the delivery of develop- agents, leading to undesirable outcomes. The paper studies mental goods like aid, credit, or technology transfer to the the interplay between incentives and communication in the poor. The paper considers a situation where credible com- presence of typical and motivated providers and finds that munication by the provider is the key to successful delivery. in certain situations incentivization leads to worse outcomes. The study shows that the use of high-powered incentives can This paper is a product of the Office of the Chief Economist, Development Economics Vice Presidency. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://econ.worldbank.org. The authors may be contacted at a.mishra@bath.ac.uk. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team High-Powered Incentives and Communication Failure Ajit Mishray Sudipta Sarangiz ds to worse outcomes. Key Words: Incentives; Communication; Motivated Provider. JEL Classification: D8; J3; O1; O3. We are grateful to Kaushik Basu, James Copestake and Debraj Ray for comments and suggestions. Large part of the work was done during Ajit Mishra’ s visit to the Development Economics unit of the World Bank. He would like to acknowkedge help and support from this unit. y University of Bath, Bath, UK Email : a.mishra@bath.ac.uk z Virginia Polytechnic Institute and State University Professor X discourages me from enrolling for a Ph.D. with him because, according to him, my chances of completion are low. But the truth is that he wants someone better than me so that he bene…ts. I know I am good enough to do a PhD. A prospective student. 1 Introduction Consider an agency situation where a donor (Principal) relies on a provider (Intermediary) to deliver certain goods and services to a group of clients (Agent). Various aspects of the delivery system have come under scrutiny in recent times. A key recurring theme that has been emphasized in the literature is the need to incentivize the providers or intermediaries respon- sible for delivery of goods and services.1 For example, in the context of foreign aid, it has been pointed out that the intermediaries may not have the right kind of incentives to see that aid is spent e¤ectively (Easterly and Pfutze (2008)). Likewise, in the context of micro…nance, a major cause for concern is the issue of appropriate incentives for loan o¢ cers to achieve the organizational goals of the micro…nance institutions (Armendariz de Aghion and Murdoch (2004)). The role of incentives in the context of gov- ernment bureaucracy and delivery of various social services has also been a subject of investigation (Dixit (2002), Tirole (1994) and Wilson (1989)). In many such agency settings, successful delivery and realization of bene…ts by the agents requires the provider to communicate relevant infor- mation, which the provider must acquire at some cost before it can com- municate. The objective of our paper is to show that while it is possible 1 Even though we use a three tiered sturcture, our focus is on the interaction between the provider and the agents. Makris (2009) studies a similar problem of incentives for intermediaries providing non-marketable goods. Unlike the agency structure used here, he focuses on a principal-agent framework. 1 to design an incentive scheme to induce the provider to acquire costly in- formation, the incentive scheme can render the process of communication ine¤ective. Using a simple example of targeted technology transfer, we show how the use of high-powered incentives becomes counter-productive. Consider for instance a farmer who is currently earning a …xed, deter- ministic income using traditional technology and is considering the adop- tion of modern technology with stochastic outcomes. Relative to the tra- ditional technology, modern technology can lead to higher as well as lower incomes. The chances of success depend on the characteristics (skill level) of the farmer, the nature of the technology, and the environment in which the farmer will operate it (state of nature).2 It is possible that the farmer, though informed about his own characteristics, is unable to calculate the success probabilities because of lack of information about the state of na- ture. In such an event, even rational farmers may switch to the modern technology with lower expected incomes because they are uninformed about success probabilities.3 The question then is how to provide them with this information in a credible manner. In our setting, such information can be made available to the recipients of the modern technology by the provider, to whom the donor provides the funds for disbursement. It turns out that while providers can success- fully communicate to the relevant agents in the absence of any incentives, the communication process breaks down in the presence of high powered incentives for providers. For a large class of incentive schemes, where the s compensation depends on the total number of successful projects, provider’ her announcement regarding the non-suitability of the transfer for cer- 2 An example in agriculture can be found in the adoption of High Yielding Variety (HYV) seeds. While HYVs are certainly more productive, they are also more sensitive to know-how and resource base of the recipient farmers. 3 Note that in any such modernization process it is not possible to rule out lower income ex post. But in the present context, for certain farmers or in certain cases, modern technology may fail to dominate current practice in an expected sense. 2 tain types (low success probability for modern technology) is non-credible. Hence even though the relevant information is available, the agents do not bene…t from it and we can obtain highly ine¢ cient outcomes. Since the provider must incur some cost to acquire the relevant information before it can communicate, we have a Catch-22 situation when this cost is non- veri…able and cannot be contracted. We need to have an incentive system to induce the provider to acquire information, but by the creation of this incentive we render the process of communication ine¤ective. The situation is improved when we have motivated providers who would acquire and communicate this agent-relevant information.4 These moti- vated providers are driven by the mission to help the disadvantaged (low- skilled in our context) and derive some private bene…ts from doing so. However, we also have non-motivated or typical providers who respond only to pecuniary incentives. The agents have no way of knowing whether they face a motivated or a typical provider. In the absence of any high- powered incentives, the presence of these typical providers does not a¤ect communication between motivated providers and the agents, but with the introduction of incentives communication breaks down due to the presence of the typical providers. Hence, in the presence of incentives, even the mo- tivated providers are of little help. However, the negative implications of incentivization can be avoided when the donor is able to use a richer set of incentive schemes. Using state-dependent contracts, the typical providers can be incentivized to acquire and communicate information truthfully. Our paper is related to several strands in the literature and we draw on many of these sources. Earlier papers by Dur and Swank (2005), and Gerardi and Yariv (2008) have emphasized the interplay between the acqui- sition and transmission of information by interested experts. In the case 4 See Besley and Ghatak (2005) on the signi…cance of these motivated agents. Fran- cois and Vlassopoulos (2008) o¤ers an exhaustive survey on the nature of prosocial motivation and delivery of social services. 3 of Dur and Swank (2005), unbiased experts exert maximum e¤ort, in a moral hazard setting, to acquire information. Hence the principal is better o¤ hiring an expert whose preference is less extreme than her own. Ger- ardi and Yariv (2008) also look at costly information gathering but in their solution the principal employs multiple experts with opposite preferences. We study an entirely di¤erent agency setting where the motivated provider (expert) is more likely to incur costly e¤ort to acquire information. Second, the presence of the typical provider alongside the motivated provider does not help the donor (principal) in addressing the problem of communication failure. The role and signi…cance of various types of motivations has received attention from several economists recently. Besley and Ghatak (2005) point out that it might be cheaper to address the moral hazard problem of induc- ing e¤ort by careful matching of motivated agents rather than the use of high-powered incentives. In our case, reliance on motivated agents may be the only way of solving the problem of information acquisition and commu- nication. While the claim that introduction of incentives can be counter- productive because of demanding informational requirement is not new, a more recent literature shows that even when incentives are appropriately designed, we cannot be certain of e¢ cient outcomes. This can happen since extrinsic motivations lead to crowding out of intrinsic motivations.5 In several principal-agent experimental settings, it has been noted that stronger incentives and control induce weaker performance by the agent. Benabou and Tirole (2006) and Ellingsen and Johannesson (2008) show that when agents care for esteem, material incentives may undermine es- teem incentives. In our case, stronger material incentives do not crowd out motivational incentives of these providers, but material incentives enable 5 See Andreoni (1990), Benabou and Tirole (2003, 2006), Sliwka (2007), Ellingsen and Johannesson (2008) among others. 4 the non-motivated providers to add noise to the communication process. Signalling plays a key methodological role in many of these models of motivations. Individuals have private information regarding own char- acteristics and they try to signal these through generosity, superior per- formances or esteem enhancing acts.6 Our model also involves signalling by the provider (and not the agents) but it is costless. In that sense it is closer to the literature on strategic information transmission and cheap talk (Crawford and Sobel (1992), Farrell (1995), Krishna and Morgan (2001)). It is well known that divergence of interests between the sender (provider) and receiver (recipient) can lead to communication failure. Our paper uses this intuition in a simple setting but with the added features that the sender has to acquire information before communicating and that the nature of incentive schemes for the provider has the potential to a¤ect the degree of divergence in interests. Finally, we do not make any general claims about the usefulness or otherwise of incentive schemes. Ours is an extremely stylized model with two-sided asymmetric information, which we elaborate in the text.7 How- ever, the interplay of incentives and communication failure is the novel feature of our analysis. Section 2 sets up the model and Section 3 has the results. The …nal section summarizes. 6 It is not the case that only agents engage in signalling. There are cases where the principal also signals (through its choice of control, trust, incentive provision) about the private information held by the principal. In Sliwka (2007), the principal chooses the level of control over the agents to signal about the average level of trustworthy agents in the population. In Ellingsen and Johannesson (2008), the principal signals its altruistic characteristics. 7 In our model the provider does not know the skill level of the recipients but is aware of the success probability of each type of recipient. The recipients on the other hand know their types but do not know the success probability since they do not know the state of the world. 5 2 The Model We consider a simple variant of the standard principal-agent framework where there is a Donor who provides a …xed amount of funds denoted by M to a Provider, who then disburses the funds to the Recipients ( or s role is limited to providing …nancing and setting up Agents). The donor’ a compensation scheme for the provider. Most of the paper is about the interaction between the provider and the agents. Money from the donor is used to fund projects that are undertaken M by the agents. Each project costs an amount T ; hence a maximum of T projects can be …nanced. Note that the funding need not only be in the form of cash transfers, it can also take the form of transfer of production technology. We discuss the details of this technology transfer, payo¤s, and strategies of the provider and agents below. 2.1 Agents We assume that there exist two types of agents - high skill (h) and low skill (l). The total population (of agents) is denoted by N = Nh + Nl where Nh and Nl are the number of high-skilled and low-skilled agents respectively. We denote by the fraction of h-type agents in the population. Each agent can supply 1 unit of labor in an inelastic manner and is assumed to be risk neutral. In the absence of donor provided funds (which can be interpreted as the subsistence sector using traditional technology) output does not depend on skill type and is given by Xi = i = h,l (1) where > 0. The transfer T enables the agents to pursue a project with varying 6 returns. For simplicity we consider only two outcomes: the project results in output Yi > 0 when it is successful, and zero otherwise. The probability of success for the h-type is given by ph > 0. For the low-skilled agents, on the other hand, the project success probability depends on one or more factors which are summarized by the state of nature . For simplicity, we consider two possible states 2 fG; B g where G denotes the good state and B the bad state. The commonly held prior belief about the probability that = G is given by > 0. The success probabilities of the l-type are given by plG and plB with ph > plG > plB > 0. Output in the successful state and expected outputs for both types are given below. Yi = i = h,l (2) E (Yh ) = ph (3) E (Yl j = G) = plG and E (Yl j = B ) = plB (4) Note that > . We assume that plG > > plB : (A1) Thus when = G, both types are better o¤ (in an expected sense) by undertaking the project but the h-type is more likely to succeed. However when = B; the l-type is better o¤ using traditional technology and not undertaking the project. To highlight the role of communication, we only consider the case where the prior is such that the low types will choose to undertake the project, i.e., f plG + (1 )plB g > (A2) 7 2.2 Provider In our model the provider has two unique roles: (i) only she can distribute funds to the agents, while being unable to identify their skill types; (ii) only she can learn the true state by incurring costly e¤ort e. We assume that e¤ort is binary with e 2 f0; 1g and the disutility of e¤ort d(e) is given by: d(0) = 0 and d(1) = E > 0. We assume that this e¤ort is observable but not contractible.8 Regarding the provider’ s compensation, we focus on a class of performance based schemes where monetary payments to the provider depend on the number of successful projects. Let the number successful projects be denoted by m; which is observable and veri…able. To begin with, we consider a typical risk neutral provider whose payo¤ is given by U = Z (m) d(e), Z = (m) 0: This also includes …xed compensation scheme with Z (m) = Z: Later in Section 3.5 we consider more general compensation schemes where Z depends on other observable variables too. The provider’ s reservation utility is denoted by U 0:9 Clearly, the donor has to design a suitable incentive scheme for the provider so that the latter undertakes the desired e¤ort to learn the realized . 2.3 Information and Time Line We assume that the output parameters ; , the success probabilities ph ; plG and plB and the compensation scheme chosen by the donor are commonly known. We assume that the agents (for whom this may be thought of as new technology) and the donor (who is removed from the …eld) do not know the realization of . In the model the low-skilled agents know the 8 Assuming observable e¤ort is reasonable since the donor can observe the documen- tation and other material evidence that the provider gathers to …nd out the true state. However, the donor might consider writing contracts (based on e¤ort) to be prohibitively expensive, and/or even when they can be written, contracts may not be enforceable. 9 s compensation and therefore It is assumed that the donor sets aside the provider’ it does not a¤ect M . 8 success probabilities associated with the good and bad state, but do not know . The provider learns the true state through costly e¤ort and can communicate this by sending a signal S 2 fG; B g to the agents. Finally, once all projects are undertaken and outcomes realized, the donor can verify m. The sequence of moves in this game is as follows. 1. Donor provides M to …nance ( M T s ) projects and speci…es the provider’ compensation scheme Z (m). 2. Provider chooses e;and if e = 1;makes announcement S 2 fG; B g.10 3. Agents update their beliefs about and choose whether to apply (A) or not apply (N A) for the project. 4. Provider randomly selects a subset of all applicants and transfers amount T to each of the selected agents. Let n be the total number projects funded. Then nh and nl denote the number of high skill and low skill agents selected to undertake the project respectively. 5. Outputs are realized and the donor learns the number of successful projects (m). Our goal is to study the impact of various incentive schemes on the interaction between the provider and agent. Hence our equilibrium de…n- ition essentially captures the interaction in stages 2-3. An equilibrium is s choice of e¤ort and given by fe ; S ; a g where e denotes the provider’ s choice is denoted by ai : fG; B g ! S is her signal if e = 1. Agent i’ fA; N Ag; i = l; h: Agents choose whether to apply or not apply based on their posterior belief (S; ) : fG; B g [0; 1] ! [0; 1]. In the absence s choice is determined by the prior of any communications, e = 0, agent’ belief : We now study the Perfect Bayesian Equilibria (PBE) of this game 10 There is no announcement when e = 0: Hence the observability of e¤ort decision is integral to our analysis. 9 (stage 2-3) assuming e = 1. For a given equilibrium, the corresponding project allocations will be denoted by nh ; nl : 3 Results and Analysis We …rst illustrate the role of communication in our setup. Then we examine e¢ ciency implications of incentive schemes for the providers and introduce the motivated provider. Initially we focus on a single class of incentive s compensation depends on the total number of schemes where the provider’ successful projects. Later in the section, we consider more general incentive schemes. 3.1 Communication In order to demonstrate the importance of communication, we begin with the payo¤ matrix shown below. These payo¤s are for illustration purposes only and are not derived from the payo¤ speci…cations discussed earlier. Hence we have suppressed the e¤ort dimension here. The provider is des- ignated as the row player and the low-skilled agent is the column player. s payo¤ and the second The …rst element in each box refers to the provider’ s payo¤.11 The provider makes the announcement S refers to the agent’ about and the agent chooses whether to apply (A) or not (N A) in the two di¤erent states. The payo¤s capture the idea that the agent is better o¤ choosing N A in the bad state, and prefers A in the good state. More- over, in the bad state, the provider also prefers the agent to choose NA. Of s preference over the agent’ course, in the good state the provider’ s choice depends on the relationship between x and . 11 The high-skilled agent is missing from the analysis because her choice is not a¤ected s announcement. by the provider’ 10 A NA =G x, X ,0 = B 0, Y 3, 0 Game 1 Suppose, x > , it is clear that communication is informative. It is easy to verify that we have a PBE where S (G) = G, S (B ) = B , a (B ) = N A, a (G) = A, (5) (G; ) = 1, (B; ) = 0: babbling’ equi- It is of course true that we also have the uninformative ‘ librium where (S; ) = ; 8S . The agent learns nothing from the an- s equilibrium announcement nouncement by the provider and the provider’ S (G) = S (B ). We do not go in to equilibrium selection issues here and assume that whenever the fully informative equilibrium exists, player will choose to play according it. Next suppose that x < . In the good state s payo¤ = G, the provider’ is higher whenever the (low-skilled) agent chooses N A. This makes the announcement of S (B ) = B non-credible because the agent realizes that the provider would like the agent to believe the state to be B even when = G: Hence the equilibrium described in (5) cannot be sustained. In fact, the only PBE in this case is the uninformative babbling equilibrium where the agent chooses A irrespective of the announcement. 3.2 E¢ ciency Suppose that for a given M , the donor is interested in maximizing to- tal (expected) output V resulting from project allocations. This amounts to maximizing total success probabilities. Let ni denote the number of 11 projects allocated to type-i agent in state :12 Hence the donor maximizes V = [ph nh + f plG nlG + (1 )plB nlB g] : (6) We consider two benchmark cases where the agents know the realization of the state . This implies that communication is irrelevant, we assume that the donor does not employ any provider in these two cases. First consider the case where, in addition to agents knowing ; information about skill types are known by the donor (as well as the agents). Since the agents know ; the outcome must satisfy the interim participation constraints, E (Yi j ) Xi : Given (A1), it is clear that V is maximized by the following allocation, M M nh = , for Nh (7) T T M M nh = Nh , nlG = Nh , and nlB = 0, for Nh < T T M Remark 1 We refer to this as the …rst-best outcome. When Nh T , only the high-skilled agents get the project in both states. In the complementary M case, Nh < T ; low-skilled agents are allocated the remaining projects in the good state but no projects in the bad state. Observe that this does not involve welfare loss for the low-skilled agents in the bad state. This outcome is clearly interim e¢ cient but ex post ine¢ ciency can- not be ruled out because of the non-deterministic nature of the output. Moreover, interim e¢ ciency requires that some amount of funds will re- M main unused when = B , and Nh < T : For the remainder of the paper M we will focus on the case when Nh T as this is su¢ cient to illustrate the trade-o¤ between costly communication and incentives of the provider, 12 When the allocation of a type is same in both states, we will drop the subscript for convenient reading. 12 when we depart from the benchmark case. Next consider the case where agents know but the donor has no in- formation about skill types. Hence projects are allocated randomly among the applicants. In the context of our simple example, we have a …nite set of outcomes depending on how the di¤erent types apply in the two states. When the low-skilled cannot be prevented from applying, it is clear that expected output V is maximized when the low-skilled apply in the good state but not in the bad state. The maximizing allocation is given below, M M M nhG = ; nlG = (1 ) ; nhB = ;n = 0 (8) T T T lB Remark 2 We refer to it as the second-best outcome. This outcome is also interim e¢ cient. There are several other interim e¢ cient allocations with nl > 0, satisfying E (Yi j ) Xi ; but this allocation yields the highest expected output in this class. Finally, we could also consider a third case where the agents are un- informed but the provider has information regarding the skill types. But M with Nh T , this case is equivalent to the …rst-best outcome. 3.3 Incentives and Communication Failure We now return to our model setting where the donor is uninformed about as well as skill types. The question we want to answer is whether the donor can achieve the e¢ cient outcomes described in (7) and (8) by hiring a provider and providing suitable incentives. M Given that Nh T ; the …rst-best outcome can only be achieved by preventing the l-types from applying in the bad as well the good state, and this is impossible to achieve. Recall that in the absence of any communi- cation about the realized state, assumption (A2) implies that both types will apply to undertake the project. For the low-skilled agents to revise 13 their prior belief we need (i) the provider to engage in costly e¤ort and acquire information about the realized , and (ii) credibly communicate this information. There is a basic tension between these two. Since e¤ort is not contractible, the provider can only be incentivized by making their compensation depend on the outcomes. Since the compensation scheme is assumed to be common knowledge, the communication game between the provider and l-type agents will have a payo¤ matrix that is similar to the one speci…ed in Game 1 with x < . We know that the only equi- librium in this case is the uninformative babbling equilibrium.13 Hence the bene…ts of communication are non-existent and the provider is better o¤ not acquiring any information. Thus for communication to be e¤ective we need Z 0 (m) = 0, but for the provider to acquire information we need Z 0 (m) > 0. Clearly it is not possible to have both simultaneously, render- ing high-powered incentives completely ine¤ective. We summarize this in our …rst proposition. M Proposition 1 Let Nh T . For any compensation scheme Z (m), both types apply in all states and nh < nh ; nl > 0; for = G; B . It is clear that neither of the e¢ cient outcomes can be achieved. It is M easy to verify that in this case nh = T ; nl = (1 )M T ; = G; B: The total expected output V will be given by M M M V = ph + f plG + (1 )plB gf1 g : (9) T T T Note that the donor is not able to do any better by conditioning provider compensation on the number of unsuccessful projects. 13 In a related context, Macchiavello (2008) studies the role of public sector wage pre- mium in screening and ensuring worker honesty. His focus is on the impact of such incentive schemes on corruption, while we examine their impact on credible communi- cation. 14 3.4 Motivated Providers Now suppose that we have some motivated providers who are mission ori- ented (see Besley and Ghatak (2005)). These providers derive additional private bene…ts which are Rawlsian in nature: they seek to maximize the expected bene…t to the most disadvantaged group the low-skilled agents. s utility depends on these private bene…ts, com- Hence a motivated provider’ pensation from the donor and possible disutility of e¤ort. The population of providers consists of both the typical providers (denoted by ) and moti- vated providers (denoted by ). We assume that the fraction of motivated providers is .14 Recall that the typical providers simply maximize Z d(e); which we relabel as U : We now turn to the utility of the motivated providers. In state = G; the (marginal) expected bene…t to the l-type from the new technology will be (plG s private bene…t is maximized when nl ) > 0 and the provider’ is maximized. On the other hand, in state = B the (marginal) expected bene…t of the new technology to the low-skilled type will be (plB )<0 s private bene…t is maximized when nl is minimized. Thus and the provider’ s private bene…ts are state dependent. We can the motivated provider’ s payo¤ as rewrite the motivated provider’ U =Z d(e) + I ( )nl k J ( )nl k 0 , k 0 ; k > 0 (10) where I ( ) = 1 when = G, and zero otherwise. Similarly, J ( ) = 1 when = B and zero otherwise. The constants k; k 0 re‡ect the weights placed s bene…t in di¤erent states. by the provider on agent’ From the above payo¤ function it also follows that, in the absence of any incentives, for = G the motivated provider prefers the l-types to un- 14 This fraction of motivated providers might depend on the nature of incentive schemes due to self selection, but we do not address this issue. See Delfgaauw and Dur (2007) for an analysis of incentive wages and workers’self selection in …rms. 15 dertake the project. This implies that the motivated provider would like to screen out the l-types in the bad state since they are better o¤ using tradi- tional technology. An example of such motivated providers would be loan o¢ cers working for a MFI who would not advance loans to someone that s repayment is most likely to be severely indebted; not because the MFI’ rates are going to be adversely a¤ected, but because the client is strictly worse o¤. 3.4.1 Intrinsic Motivations Only Suppose there are no extrinsic incentives for the providers, i.e. Z is …xed and not performance based. Then the motivated provider will choose e = 1 and communicate the realized state to the agents. The typical provider chooses e = 0 and does not observe the realized state as e¤ort is costly. We show that there is an equilibrium where the motivated provider truthfully conveys information regarding the state and the l-types do not apply in the bad state. Note that given the objective function of the mo- tivated provider, the communication game resembles Game 1 with x . This means the provider would like the l-types to apply in the good state but not in the bad state, making their announcement credible. Conse- quently, the low skilled agent chooses its strategy as follows: a(G) = A, and a(B ) = N A: It is easy to verify that the typical provider does not have any incentive to deviate and acquire information to take advantage of the credibility of communication. Since compensation Z does not depend on the outcome, doing so would simply lead a reduction in equilibrium payo¤ by d(1) = E . So in this setting, with probability we get the outcome where only the high types apply in the bad state and with probability (1 ); we get the ine¢ cient outcome where all types apply in both states. 16 Equilibrium strategies are given by e = 1, S (G) = G, S (B ) = B; ah = A; al (G) = A and al (B ) = N A: e = 0; ah = al = A (11) The corresponding allocation is given by M M nhG = , nlG = (1 ) , T T M M M nhB = + (1 ) , nlB = (1 )(1 ) : (12) T T T When = 1; this reduces to the second-best outcome (8) and expected output is given by M M M V = f ph + plG (1 ) g + (1 )ph : (13) T T T It can be seen that expected output in (13) is higher than the expected output under typical providers with incentives (9). Even when < 1; the above allocation with motivated providers and …xed compensation domi- nates the previous allocation listed in proposition 1. However, the …rst-best can never be achieved with the motivated providers, because the motivated providers would always prefer the low skilled agents in the good state. We summarize this in the following proposition. M Proposition 2 Let Nh > T . (a) When all providers are motivated, = 1; the second-best outcome shown in (8) can be achieved. (b) When some providers are motivated, > 0; the donor achieves higher expected output compared to the case with only typical providers. 17 3.4.2 Intrinsic Motivations and Incentives In the previous analysis, the presence of typical providers has no e¤ect on the credible communication between the motivated providers and the agents. But this is not necessarily true when the provider is incentivized with Z 0 (m) > 0. For the motivated provider this does not change any of the equilibrium strategies for su¢ ciently large values of k and k 0 . Consider the equilibrium strategies given in (11) and the corresponding outcome. The s e¤ort choice is still given by e = 1. It is clear that motivated provider’ they will choose to communicate truthfully in the bad state. But will they choose S = G when the realized state is G? There may exist situations depending on the proportion of high-skilled agents ( ), where Z (m) can be lower according to the equilibrium strategy in (11). However it is easy to show that there exists k k such that the motivated provider will not deviate to S = B , where k is given by the solution to the following15 M M M M Z (ph ) = Z ph + plG (1 ) + (1 ) k : (14) T T T T s But the incentive scheme has a signi…cant impact on the typical provider’ strategies. Given the strategies of the motivated provider and the agents, the typical provider will bene…t from deviating to e = 1 and s(G) = B if the following is satis…ed, M M M M Z (ph ) E Z ph + plG (1 ) + (1 )Z p h : (15) T T T T The right hand side expression in (15) is the payo¤ to the typical provider in such an equilibrium, but with Z 0 (m) > 0: For given e¤ort level and ; this condition depends on the slope of the compensation function Z . The 15 In the context of crowding out, as discussed earlier, this can be interpreted as intrinsic motivation being su¢ ciently strong. 18 slope can be interpreted as the power of the incentive scheme, with a higher value of the slope implying high-powered incentives. If compensation is highly responsive to the outcome (in this case m) then the typical provider will deviate. Suppose, Z (m) = z:m. Then the equilibrium outcome given by (15) cannot be sustained if z > z 0 ; where z 0 is given by M z 0 (ph plG ) (1 ) = E: (16) T Once the typical provider also makes announcements, the agents have s announcement from that of the no way of separating the typical provider’ motivated provider. The signal B could come from a typical provider in state G, or it could come from both types of providers in state B . It can be checked that the posterior belief that the state is good, when e = e = 1 ( ) and the signal is B; is (G j B ) = 1 : Note that for a given prior s posterior belief , the agent’ is determined by the fraction of motivated providers : A high value of induces the l-type agents to apply. Hence, when there are large number of typical providers and agents’belief about the underlying state being good is high, all agents will apply even when the state is bad. Since, according to assumption (A2), plG + (1 )plB > 0 and ! as ! 0; there exists such that 0 plG + (1 )plB for all : (17) Hence the introduction of high-powered schemes leads to a communica- tion failure even in the presence of motivated providers if these incentives are powerful enough to induce costly information gathering by the typical providers. This is summarized below. Proposition 3 Consider linear compensation schemes Z (m) = z:m with high-powered incentives z > z 0 (ref 16). When the fraction of motivated 19 0 providers is small, < (ref 17), in any equilibrium both types of agents apply in all the states. This suggests that when faced with a mixed population of motivated and typical providers, the donor is better o¤ not using any incentive schemes. Information acquisition by the typical providers adds noise to the commu- nication by the motivated providers and it leads to lower expected output. 3.5 State-Dependent Contracts: An Illustration The previous analysis shows how intrinsic motivations could be e¤ective where extrinsic motivations through incentives failed to do so. However, in our previous example, the donor was restricted to a small class of contracts based on total number of successes or failures. Suppose the donor can observe and verify the state ex post and condition compensation contracts on the realization of : Using this expanded set of feasible contracts we can show that extrinsic motivations can be made to work. However, it may be hard to verify in many situations and motivated providers remain the best answer in such scenarios. Since motivated providers do not need any incentivization and our re- sult concerning ine¤ectiveness of incentivization (Proposition 1) had only typical providers, here also we con…ne attention to a world with typical providers only and drop the subscript . As is obvious from Proposition 1 and analysis in Section 3.3, high-powered incentives lead to communication failure because in state = G, the provider has an incentive to dissuade s prefer- the l-types from applying by announcing B: Hence the provider’ ence over types in the good state is responsible for undermining credibility. We can design an incentive scheme where the provider has an incentive to acquire information but at the same time, it is not a¤ected by which 20 types apply in the good state. Since states are ex post veri…able and hence contractible we can consider the following contract: Z = Z if = G, Z = Z (nl + nh m)f if = B; (18) where Z is some …xed payment and f is the penalty for each unsuccessful project. We can …nd suitable Z and f such that the following participation and incentive constraints are satis…ed. M Z (1 )(1 ph ) f E U; (19) T M (1 ) f (1 )(ph pl ) E 0: (20) T It is easy to verify that such a compensation scheme is feasible and the provider will choose e = 1. Moreover, in state = G, the provider has no incentive to lie and choose S = B: Likewise, in state = B , an announcement of S = G will lead to a lower payo¤ for the provider. Hence, we have an equilibrium where e = 1, S ( ) = ; for = G; B , (21) ah = A, al (G) = A, al (B ) = N A: M Clearly, this will lead to an allocation given by nhG = T ; nlG = (1 )M T ; nhB = M T ; nlB = 0: Hence the second-best outcome can be achieved by this contract. Note that this is the same outcome which can be achieved s perspective will with motivated providers. Expected output from donor’ be M M M V = ph + plG (1 ) + (1 ) ph : (22) T T T In our simple example, the provider can use other contracts too.16 Since 16 These need not be state dependent but serve the same purpose as the state depen- 21 we have two states and two types, the number of successful projects can be ordered according to states. Let m1 be the number of successes when only high-skilled types undertake projects, and the number of successful projects with both types applying in good and bad states will be given by m2 and m3 respectively. Clearly, m1 > m2 > m3 : Consider the following contract: Z = Z; if m m3 ; and (23) Z = 0; otherwise, for 0 < < m3 m2 : (24) This is identical to the state-dependent contract discussed earlier. Observe that the provider is not a¤ected by the presence of low-skilled agents in the good state and hence communication is credible. 3.6 s Objectives Donor’ s objective into the picture. The above discussion has brought out the donor’ Throughout we have assumed that the donor is interested in maximizing total expected output. However, it is not the case that donors have to be interested in maximizing returns on every dollar spent. Suppose the donor is interested in only avoiding the worst case, i.e. preventing low-skilled agents from undertaking the project in the bad state but has no preference over types in the good state.17 Hence the donor does not care about nlG or nh : Such a donor can use a contract similar to (18) where failed projects in state B are penalized. If the donor can identify the di¤erent failed types then we could even target only failures by l-types in the bad state. With such a contract the interests of the providers, both motivated as well as dent contract. 17 This would of course include the case where the donor would like the low-skilled agents to get the project in the good state. If wealth and skill level are positively correlated, one can justify such objectives. 22 typical, are aligned with the interests of the agents. Hence the donor can achieve an allocation with nlB = 0: When the donor cares only about successful projects the total number of success is bounded above by what is implied by the second-best outcome. The best an output maximizing donor can do is identical to what a donor, who is interested in avoiding the worst case scenario, would do to prevent low-skilled agents from undertaking the project in the bad state. Thus the output maximizing donor is observationally equivalent to the one who wants to avoid the worst case outcome. 4 Conclusion E¤ective delivery is a critical component of development e¤orts. Using a simple and stylized setting, we have shown that introduction of high- powered incentives can lead to communication failure and undermine the very reasons for the introduction of incentives. Despite the context-speci…c nature of our example, our analysis is relevant to the general case of pro- visioning of goods and services identi…ed by two distinctive features: non- commercial intent and reliance on non-price allocation mechanisms. Ex- amples of these are transfer of modern technology, technological know-how, loans and grants as well as aid to the poor. This framework can also be used to study programs like the provisioning of health services, education and many other public goods. Whenever agents’relevant characteristics are not commonly observed, communication is important and informational prob- lems arise in the absence of e¤ective communication. The severity of the problem can be gauged by the fact that in these situations, even though some types of agents are likely to be worse o¤ than their current status, they end up receiving the transfers. An instance of such communication failure and ine¢ cient uptake can 23 be found in the recent micro-lending programs of several micro…nance in- stitutions. A major crisis broke out in March 2006 when around 50 MFI branches in Andhra Pradesh (a state in India) were closed by the govern- ment because of complaints against practices of these organizations. Some authors, while analyzing this incident, commented on how indiscriminate making a debt trap’for the poor.18 It is argued that several lending was ‘ individuals who (ex ante ) had a very small chance of repaying the loans also entered into debt contracts. This was possible because of a breakdown in credible communication between the loan o¢ cers and the individuals. Excessive incentivization of the loan o¢ cers to maximize the number of clients can be listed as a major cause of this counter-productive outcome.19 While our result is related to the recent literature on intrinsic and ex- trinsic motivation, the emphasis on information ‡ows and communication is a novel feature. We show that while the introduction of extrinsic mo- tivations or incentives does not destroy the intrinsic motivations of the motivated providers, it makes the typical provider act in such a manner that communication between the motivated providers and agents breaks down. Based on our stylized model, we believe that there are two issues which need to be noted. First, we have assumed (in most of the paper) that the number of high skilled agents exceeds the number of projects that can be …nanced. If this is not true then in some states the entire amount of funds supplied by the donor will not be spent. Donors who prefer full utilization (or disbursement) of funds will consider this outcome ine¢ cient. But on the other hand, in the bad state where the low-skilled agents are better o¤ not undertaking the project, it is better to have undisbursed funds. We 18 See Shylendra (2006) and Kumar (2006) for detailed accounts and analysis of this incident. 19 In more general contexts, the recent literature on participatory development can also be viewed as attempts to adopt development practices where there is better information ‡ ow (about local preferences). See for instance Platteau (2009). 24 only make a partial reference to this issue since it is not the main focus of our paper and does not generate additional insights. Second, the provider relies on a random allocation when the number of applications exceeds the number of projects to be …nanced. Since our focus was on communication, in our model the provider can a¤ect the …nal allocation only by commu- nicating the state-related information to in‡uence the agents’decision to apply for projects. In practice however, the provider might undertake costly screening of the applications, an issue that has been left for future research. References Andreoni, James.1990. "Impure Altruism and Donations to Public Goods: A Theory of Warm-Glow Giving?" Economic Journal, 100: 464-477. Armendariz de Aghion, Beatriz and Jonathan Morduch. The Economics of Micro…nance, MIT Press. Benabou, Roland, and Jean Tirole. 2003. "Intrinsic and Extrinsic Moti- vation." Review of Economic Studies, 70(30): 495-520. Benabou, Roland, and Jean Tirole. 2006. " Incentives and Prosocial Behavior." American Economic Review, 96(5):1652-78. Besley, Timothy and Maitreesh Ghatak. 2005. "Competition and Incen- tives with Motivated Agents." American Economic Review, 95(3):616- 36. Copestake, James. 2007. "Mainstreaming Micro…nance: Social Perfor- mance Management or Mission Drift? World Development, 35(10): 1721-1738. Crawford, Vincent and Joel Sobel. 1982. "Strategic Information Trans- mission." Econometrica, 50:1431-1452. 25 Delfgaauw, Josse and Robert Dur. 2007. "Signaling and Screening of Workers’Motivation." Journal of Economic Behavior and Organiza- tion, 62:605-624. Dixit, Avinash. 2002. "Incentives and Organizations in the Public Sec- tor." Journal of Human Resources, 37(4): 696-727. Dur, Robert and Otto Swank. 2005. "Producing and Manipulating Infor- mation." Economic Journal, 115: 185-199. Ellingsen, Tore and Magnus Johannesson. 2008. "Pride and Prejudice: The Human Side of Incentive Theory." American Economic Review, 98(3): 990-1008. Easterly, William. 2003. "Can Foreign Aid Buy Growth?" Journal of Economic Perspectives, 17(3): 23-48. Easterly, William and Tobias Pfutze. 2008. "Where Does the Money Go? Best and Worst Practices in Foreign Aid.". Journal of Economic Perspectives, 22(2): 29-52. Farrell, Joseph, 1995, Talk is Cheap, American Economic Review, 85:186- 90. Francois, Patrick and Michael Vlassopoulos. 2008. "Pro-social Motivation and the Delivery of Social Services." CESifo Economic Studies, 54:22- 54. Gerardi, Dino and Leeat Yariv. 2008. "Costly Expertise." American Economic Review,98:187-93 Ghosh, Suman and Eric Van Tassel. 2008. "A Model of Mission Drift in Micro…nance Institutions." Mimeo, Florida Atlantic University. 26 Kumar, Nagesh. 2006. "The Making of Debt Trap in Andhra Pradesh." The Hindu, April 20. Krishna, Vijay and John Morgan. 2001. "A Model of Expertise." Quar- terly Journal of Economics, 23: 747- 775. Macchiavello, Rocco. 2008. "Public Sector Motivation and Development Failures." Journal of Development Economics, 86: 201-213. Makris, Miltiadis. 2009. "Incentives for Motivated Agents under an Ad- ministrative Constraint." Journal of Economic Behavior and Orga- nization, 71:428440. Platteau, Jean-Philippe. 2009. "Information Distortion, Elite Capture, and Task Complexity in Decentralized Development." in Does De- centralization Enhance Service Delivery and Poverty Reduction? Ed- ward Elgar, Cheltenham. Shylendra, H.S. 2006. "Micro…nance Institutions in Andhra Pradesh." Economic and Political Weekly, May 20: 1959-1963. Sliwka, Dirk. 2007. "Trust as a Signal of a Social Norm and the Hid- den Costs of Incentive Schemes." American Economic Review, 97(3): 999-1011. Tirole, Jean. 1994. "The Internal Organization of Government." Oxford Economic Papers, 46(1): 1-29. Titmuss, Richard. 1970. The Gift Relationship. London: Allen and Unwin. Wilson, James Q. 1989. Bureaucracy: What Government Agencies Do and Why They Do It. New York: Basic Books. 27