• Share
  • Email
  • Embed
  • Like
  • Save
  • Private Content
04.24.2013 - Maitreesh Ghatak
 

04.24.2013 - Maitreesh Ghatak

on

  • 245 views

Motivating Knowledge Agents: Incentive Pay vs. Social Distance

Motivating Knowledge Agents: Incentive Pay vs. Social Distance

Statistics

Views

Total Views
245
Views on SlideShare
245
Embed Views
0

Actions

Likes
0
Downloads
0
Comments
0

0 Embeds 0

No embeds

Accessibility

Categories

Upload Details

Uploaded via as Adobe PDF

Usage Rights

© All Rights Reserved

Report content

Flagged as inappropriate Flag as inappropriate
Flag as inappropriate

Select your reason for flagging this presentation as inappropriate.

Cancel
  • Full Name Full Name Comment goes here.
    Are you sure you want to
    Your message goes here
    Processing…
Post Comment
Edit your comment

    04.24.2013 - Maitreesh Ghatak 04.24.2013 - Maitreesh Ghatak Presentation Transcript

    • Motivating Knowledge Agents:Incentive Pay vs Social DistanceMaitreesh Ghatak (LSE)Erlend Berg (Oxford)R Manjula (ISEC)D Rajasekhar (ISEC)Sanchari Roy (Warwick)IFPRI, 24 April 2013
    • Introduction Theory Empirics Discussion and conclusionMotivationPublic services in developing countries are often dysfunctionalSchools, health care, contract enforcement, social protection. . .A body of work on supply-side constraintsBut little is known about demand-side constraintsWe hypothesise that intended beneficiaries often don’t have enoughinformation to be able to benefit from a serviceInformation costs responsible for low take-up of welfare schemes indeveloped countries (Aizer 2007; Daponte et al 1999)In India, awareness about the National Rural Employment Guarantee(NREG) is very low in some of the poorer statesMaitreesh Ghatak (LSE) Motivating knowledge agents
    • Introduction Theory Empirics Discussion and conclusionWe look at two aspects of information transmissionIncentivesPerformance pay is rare in the public sector (for several reasons)Hence, little is known about the role of incentives in spreadingawareness of government schemesSocial barriersEvidence that public goods are under-provided in fragmented societies(Easterly & Levine 1997; Kimenyi 2006)Possibly because people prefer to interact with ‘their own kind’(Banerjee & Munshi 2004)If so, information may not easily cross social boundariesMay explain heterogeneity in programme awarenessInteraction effects: Do incentives alleviate, or exacerbate, thepotentially negative effects of social barriers?Maitreesh Ghatak (LSE) Motivating knowledge agents
    • Introduction Theory Empirics Discussion and conclusionWhat we doDevelop a simple model of motivated agents (Besley and Ghatak,2005) to generate predictions on impact of incentive pay and socialdistanceRun a randomised experiment in which we hire agents to spreadinformation about a government welfare programmeAim to answer the following questions:1 Do ‘knowledge agents’ actually improve programme knowledge?2 Does incentive pay make a difference?3 Does improved programme knowledge translate into higher programmetake-up?4 Does social distance between agent and target household have aneffect on knowledge transmission?5 Does incentive pay reinforce or weaken any social distance effect?Maitreesh Ghatak (LSE) Motivating knowledge agents
    • Introduction Theory Empirics Discussion and conclusionWhat we findHiring agents to spread information has a positive impact on the levelof knowledge of the scheme in the village populationThe effect is driven entirely by agents on incentive-pay contractsIn turn, improved knowledge increases programme take-up. . . establishing that information costs are an impediment to demandSocial distance between agent and beneficiary has a negative impacton knowledge transmissionIncentive pay seems to cancel out negative effect of social distance. . . but incentive pay has no impact on knowledge transmission forsocially proximate agent-beneficiary pairsNo evidence of ‘crowding out’Maitreesh Ghatak (LSE) Motivating knowledge agents
    • Introduction Theory Empirics Discussion and conclusionTheoretical frameworkAim: A model to look at interactions of intrinsic preferences andincentive pay in determining effortAn agent exerts effort in a task (later, two tasks)Later we will think of each household, or group of households, as a taskSuccess in a task is binary. Probability of success depends on agent’seffortThink of success as having the household’s knowledge about thegovernment scheme exceed a certain thresholdAlternatively, the task is successful if the household signs up for theschemeThe principal values success in the taskObserves task outcome, but not effortMaitreesh Ghatak (LSE) Motivating knowledge agents
    • Introduction Theory Empirics Discussion and conclusionTheoretical frameworkLet e be the unobservable effort exerted by agentThe outcome variable Y is binary and 0 and 1 denote ‘failure’ and‘success’ respectivelyThe probability of success is p(e) = eMaitreesh Ghatak (LSE) Motivating knowledge agents
    • Introduction Theory Empirics Discussion and conclusionTheoretical frameworkThe probability of success is bounded: 0 < e ≤ e ≤ e < 1Both principal and agent are risk neutral but agents are poor, solimited liability (no fines)The agent’s disutility of effort is 12ce2 for constant cIf project succeeds, the agent receives a non-pecuniary pay-off of θ(this is her intrinsic motivation) and the principal receives a pay-off ofπ (which may have a pecuniary as well as a non-pecuniarycomponent)We assume that the principal’s pay-off incorporates the direct pay-offof the beneficiaries as well as how the rest of society values theirwelfareMaitreesh Ghatak (LSE) Motivating knowledge agents
    • Introduction Theory Empirics Discussion and conclusionTheoretical frameworkLet w be the pay that the principal offers to the agent in the case ofsuccess, and w, the pay in the case of failureThen b ≡ w − w can be interpreted as bonus pay with w as the fixedwage componentThe agent’s objective is to maximise:maxe(θ + w)e + w(1 − e) −12ce2This yields the solution:e = max min{b + θc, e}, eMaitreesh Ghatak (LSE) Motivating knowledge agents
    • Introduction Theory Empirics Discussion and conclusionThe one-task solutioneebeMaitreesh Ghatak (LSE) Motivating knowledge agents
    • Introduction Theory Empirics Discussion and conclusionTwo tasksExtend basic model: Two tasksInterpretation: Trying to increase the knowledge among two differenttypes of householdsUnlike the classic multi-tasking model, the outcomes associated withthe tasks are here assumed to be equally measurableInstead, the differences between the two tasks are in terms of:1 the agent’s intrinsic pay-off for success for each task, θ1 and θ22 the cost-of-effort parameters, c1 and c2Assume without loss of generality that task 1 is the ‘easier’ task,c1 < c2Maitreesh Ghatak (LSE) Motivating knowledge agents
    • Introduction Theory Empirics Discussion and conclusionThe bonus cannot vary across tasksWe assume that the principal is constrained to offer the sameconditional payments, w and w, for the two tasksMay be politically, socially, or legally constrained to offer the same payfor all tasksRelevant characteristics of the agent and/or tasks may not beobservable to principal (e.g. favourite students)Maitreesh Ghatak (LSE) Motivating knowledge agents
    • Introduction Theory Empirics Discussion and conclusionA solution without crowding out (relatively lowsubstitutability)eebe1e2Maitreesh Ghatak (LSE) Motivating knowledge agents
    • Introduction Theory Empirics Discussion and conclusionA solution with temporary satiation in e2 (intermediatesubstitutability)eebe1e2Maitreesh Ghatak (LSE) Motivating knowledge agents
    • Introduction Theory Empirics Discussion and conclusionA solution with crowding out (relatively highsubstitutability)eebe1e2Maitreesh Ghatak (LSE) Motivating knowledge agents
    • Introduction Theory Empirics Discussion and conclusionThe intrinsically preferred taskDefine intrinsically preferred taskThe task that receives the most effort from the agent in the absence ofincentive payThe intrinsically preferred task is not necessarily the one for whichintrinsic pay-off is the greatestIntrinsic pay-off could be outweighed by greater cost of effortConceptually classify household into two categories:Households similar to agent in terms of social characteristics (‘own’group)Households socially distant from agent (‘other’ or ‘cross’-group)Map the agent’s ‘own’ group to the intrinsically preferred taskMaitreesh Ghatak (LSE) Motivating knowledge agents
    • Introduction Theory Empirics Discussion and conclusionThe picture emerging from the theoryIn the absence of incentive pay, the agent’s own group is preferredWhen bonus pay is introduced:Total effort (weakly) increasesEffort in the easier task increases at least as much as effort in theharder taskIf the two tasks are relatively complementary: Effort in both tasksincreasesIf the two tasks are relatively substitutable: Effort in the easier taskincreases, effort in the harder task decreases → crowding outEffort can saturate at lower or upper boundsIf so, may not respond to incentivesMaitreesh Ghatak (LSE) Motivating knowledge agents
    • Introduction Theory Empirics Discussion and conclusionExperimental contextThe experiment was conducted in the context of an Indianpublic-private health insurance scheme for the poor called ‘RashtriyaSwasthya Bima Yojana’—henceforth, RSBYSet in two districts in the south Indian state of Karnataka: Shimogaand Bangalore RuralThe scheme was launched in these districts in Feb–March 2010Key features of programme:Eligibility criterion: Below-Poverty-Line (BPL) designationCovers hospitalization expenses for 700 specified medical conditionsand proceduresAnnual expenditure cap of 30,000 rupees (600 USD) per householdPolicy underwritten by insurance company selected in state-wide tenderMaitreesh Ghatak (LSE) Motivating knowledge agents
    • Introduction Theory Empirics Discussion and conclusionExperimental design151 randomly selected villages in Bangalore Rural and ShimogaThree experimental groupsFlat-pay group: local agent recruited and paid fixed amount 400rupees every three months (38 villages)Incentive-pay group: local agent recruited and paid a fixed amount of200 rupees, plus a bonus depending on the level of RSBY knowledgeamongst the eligible households in her village (74 villages)Control group: no agent appointed (39 villages)All agents are female and live locally. Many are members of a localSelf-Help Group (SHG)Agent’s task: spread information about RSBY among eligiblehouseholdsMaitreesh Ghatak (LSE) Motivating knowledge agents
    • Introduction Theory Empirics Discussion and conclusionExperimental designAverage pay designed to equal 400 rupees across the two treatmentgroupsThat is, the average bonus was 200 rupeesThe aim was to isolate the ‘incentive effect’ of the contract structurefrom the ‘income effect’Payment structure revealed to agent after recruitmentThe aim was to isolate the ‘incentive’ effect of contract structure frompotential ‘selection’ effectAttrition could re-introduce selection bias, but no agent quit afterbeing told about the payment structureFour agents quit a few months later, one due to pregnancy and threedue to migrationThose villages excluded from our analysisFinal number of villages in our sample is 147Maitreesh Ghatak (LSE) Motivating knowledge agents
    • Introduction Theory Empirics Discussion and conclusionDataThree waves of surveys conducted post-interventionA random sample of eligible household in our sample villages wereinterviewed in each wave, leading to a partial overlapA few months’ gap between each waveAim of the surveys:Administer knowledge test to eligible households to determine level ofknowledge about RSBY (also used to pay agent)Measure take-up of RSBYCollect limited background information on householdsEach knowledge test consisted of 8 questions relating to RSBY (score0–8)Main outcome variable is the knowledge-test z-scores; we also look atenrolment in the schemeMaitreesh Ghatak (LSE) Motivating knowledge agents
    • Introduction Theory Empirics Discussion and conclusionAgent summary statisticsFlat pay Incentive pay DifferenceAgent age 34.8 34.8 0.018(8.81) (8.08) (1.69)Agent is married 0.81 0.92 0.10(0.40) (0.28) (0.066)Agent is of forward/dominant caste 0.43 0.35 -0.080(0.50) (0.48) (0.099)Agent’s household head has completed 0.62 0.56 -0.058primary school (0.49) (0.50) (0.10)Agent household has ration card 0.89 0.79 -0.10(0.31) (0.41) (0.077)Agent owns her home 0.86 0.87 0.0084(0.35) (0.34) (0.069)Agent is Self-Help Group president 0.30 0.28 -0.016(0.46) (0.45) (0.093)Agent autonomy score (the higher, 5.57 5.68 0.11the more autonomous) (0.93) (0.84) (0.18)Agent pay in round 1 400 507.7 107.7(0) (478.5) (78.8)Agent pay in round 2 400 403.0 3.04(0) (209.1) (34.5)Observations 37 71[para,flushleft] Note: Standarddeviations/errors in parentheses.Maitreesh Ghatak (LSE) Motivating knowledge agents
    • Introduction Theory Empirics Discussion and conclusionHousehold summary statisticsControl Flat Inc’tive Flat Inc’tive Inc’tivepay pay −Control −Control −FlatHousehold is of forward/ 0.25 0.18 0.17 -0.070 -0.084∗-0.015dominant caste (0.43) (0.39) (0.37) (0.054) (0.046) (0.041)Household head has com- 0.30 0.25 0.31 -0.051 0.011 0.062∗pleted primary school (0.46) (0.43) (0.46) (0.042) (0.036) (0.035)Household has ration card 0.94 0.94 0.92 0.00078 -0.019 -0.020(0.25) (0.24) (0.28) (0.020) (0.023) (0.022)Household owns its home 0.67 0.64 0.68 -0.023 0.017 0.040(0.47) (0.48) (0.47) (0.047) (0.035) (0.045)Observations 375 348 625Notes: Standard errors are in parentheses and p-values in bracketsMaitreesh Ghatak (LSE) Motivating knowledge agents
    • Introduction Theory Empirics Discussion and conclusionEmpirical specificationBasic specification: Yhv = α + βTreatv + ehvOutcome is test z-score; β captures overall effect of knowledge agentsAll regressions are weighted least squaresNot all households observed in every wave, but there is overlapWeighted least squares with total weight 1 assigned to each householdStandard errors are robust and clustered at village levelSurvey (wave) and taluk fixed effects includedTaluks are sub-district administrative divisions4 in Bangalore Rural, 7 in ShimogaMaitreesh Ghatak (LSE) Motivating knowledge agents
    • Introduction Theory Empirics Discussion and conclusionImpact of agents on knowledge(1) (2) (3)Knowledge Knowledge KnowledgeAgent in village 0.175*** 0.187***(0.0645) (0.0572)Flat-pay agent in village 0.0722(0.0919)Incentive-pay agent in village 0.246***(0.0569)Survey wave fixed effects No Yes YesTaluk fixed effects No Yes YesObservations 5641 5641 5641t-test: flat=incentivised (p-value) 0.0600Notes: Standard errors, in parentheses, are clustered at the village level. * p<0.10, ** p<0.05, *** p<0.01Maitreesh Ghatak (LSE) Motivating knowledge agents
    • Introduction Theory Empirics Discussion and conclusionPricing out prejudice? Incentives vs social distanceRecent work has shown social distance (identity) to be an importantdeteminant of insurance take-upDo incentives reinforce or weaken the role of social distance?Construct a metric of social distance based on:Forward/dominant caste status (0/1)Whether household head has completed primary school (0/1),Ration card status (0/1)Home ownership (0/1)Define social distance between agent and household as the absolutedifference in the agent’s and the household’s characteristicsConstruct ‘composite’ social distance as the sum of the fourindividual distance measuresMaitreesh Ghatak (LSE) Motivating knowledge agents
    • Introduction Theory Empirics Discussion and conclusionPricing out prejudice? Incentives vs social distanceSpecification: Yhv = α + βDhv + γTv + δDhv ∗ Tv + πX + uhvWithin-treatment analysis, with flat-pay villages as comparison(control villages necessarily dropped)β captures the effect of social distance on knowledge when the agentis not incentivisedγ captures the effect of incentive pay for socially proximate(non-distant) agent-household pairsδ captures the differential effect of incentive pay for socially distantagent-household pairs relative to socially proximate onesMaitreesh Ghatak (LSE) Motivating knowledge agents
    • Introduction Theory Empirics Discussion and conclusionPricing out prejudice? Incentives vs social distance(1) (2) (3) (4) (5) (6)KnowledgeIncentive pay 0.16* -0.13 0.01 0.19 0.09 0.08(0.09) (0.14) (0.11) (0.12) (0.09) (0.11)Social distance -0.66*** -0.38*** 0.09 -0.27** -0.21*(0.21) (0.10) (0.09) (0.13) (0.12)Incentive pay x 0.79*** 0.37*** -0.05 0.38** 0.27**social distance (0.22) (0.13) (0.11) (0.14) (0.12)Agent, village and Yes Yes Yes Yes Yes Yeshousehold characteristicsTime and region Yes Yes Yes Yes Yes Yesfixed effectsObservations 2900 2900 2900 2900 2900 2900Social distance metric N/A Compo- Caste Educ- Ration Homesite only ation card ownershiponly only onlyNotes: Standard errors, in parentheses, are clustered at the village level. * p<0.10, ** p<0.05, *** p<0.01Maitreesh Ghatak (LSE) Motivating knowledge agents
    • Introduction Theory Empirics Discussion and conclusionRelating empirics to theoryes(b) denotes effort of agent when dealing with her own social groupeo(b) denote effort when dealing with the other groupEmpirically observe four points: es(0), es(b ), eo(0) and eo(b )The key empirical findings can be summed up as follows:eo(0) < eo(b ) = es(b ) = es(0)Maitreesh Ghatak (LSE) Motivating knowledge agents
    • Introduction Theory Empirics Discussion and conclusionTheoretical frameworkMaitreesh Ghatak (LSE) Motivating knowledge agents
    • Introduction Theory Empirics Discussion and conclusionRelating empirics to theorySuggests that for own group, agents were already choosing maximumeffort and with bonus pay there were no additional effectsFor other group, agents were choosing the minimum effort level, andwith bonus pay effort goes up to the same level as with own groupWhy is there a maximum effort?We do not observe crowding out, but this could still happen outsidethe observed parameter valuesSpecifically, if we increased or decreased b enough, effort with respectto own group could decreaseFrom the four points we observe, we cannot tell whether we are in acrowding-out worldMaitreesh Ghatak (LSE) Motivating knowledge agents
    • Introduction Theory Empirics Discussion and conclusionConclusionsHiring information-spreading agents has a positive effect onknowledge about the schemeThe effect is driven by agents on incentive payFlat-pay agents not significantly different from no agentIn turn, increased knowledge about the scheme increases take-upShows that information costs can be important even indeveloping-country contextsIncentive pay works by increasing effort with respect to sociallydistant householdsIncentive pay does not change effort with respect to socially proximatehouseholdsIncentive pay may overcome social barriers—in this specific contextMaitreesh Ghatak (LSE) Motivating knowledge agents
    • Introduction Theory Empirics Discussion and conclusionMore on the two-task modelLet Y1 and Y2 be the binary outcomes for the two tasks and e1 ande2 the corresponding effort levels0 < e < e < 1 define bounds for both e1 and e2Let θ1 and θ2 denote the non-pecuniary pay-offs to the agent fromsuccess in task 1 and 2, respectivelyThe principal receives the same pay-off π for both tasksMaitreesh Ghatak (LSE) Motivating knowledge agents
    • Introduction Theory Empirics Discussion and conclusionThe agent’s problemAgent’s cost of effort:c(e1, e2) =12c1e21 +12c2e22 + γe1e2The parameter γ ≥ 0 can is a measure of the substitutability of task 1and 2 in the cost of effortIf c1 = c2 = γ = c and θ1 = θ2 = θ, then the setup collapses to thesingle-task modelWLOG, assume c1 ≤ c2; as before, b = w − wThe agent maximises:maxe1,e2(θ1 + w)e1 + (θ2 + w)e2 + w(1 − e1) + w(1 − e2) − c(e1, e2)Maitreesh Ghatak (LSE) Motivating knowledge agents
    • Introduction Theory Empirics Discussion and conclusionTwo-task solutionSolution when both effort curves are internal:˜e1 (b) =(c2 − γ) b + c2θ1 − γθ2c1c2 − γ2˜e2 (b) =(c1 − γ) b + c1θ2 − γθ1c1c2 − γ2Maitreesh Ghatak (LSE) Motivating knowledge agents
    • Introduction Theory Empirics Discussion and conclusionTwo-task solutionDefineˆe1(b) =θ1+b−γec1if ˜e2(b) ≤ e˜e1(b) if e < ˜e2(b) < eθ1+b−γec1if ˜e2(b) ≥ eˆe2(b) =θ2+b−γec2if ˜e1(b) ≤ e˜e2(b) if e < ˜e1(b) < eθ2+b−γec2if ˜e1(b) ≥ e.The complete second-best solution for the two-task model is given by:e∗1 (b) = max{min{ˆe1(b), e}, e}e∗2 (b) = max{min{ˆe2(b), e}, e}.Maitreesh Ghatak (LSE) Motivating knowledge agents
    • Introduction Theory Empirics Discussion and conclusionCrowding outThe second-order condition for a maximum requiresγ2< c1c2Allows for ‘crowding out’: bonus payment may crowd out intrinsicmotivation (Gneezy & Rustichini; B´enabou & Tirole; Frey)Two main cases:γ < c1 < c2 (relatively low substitutability): no crowding outc1 < γ < c2 (the tasks are relatively substitutable in the cost of effort):crowding out when both curves are internalMaitreesh Ghatak (LSE) Motivating knowledge agents
    • Introduction Theory Empirics Discussion and conclusionThe intrinsically preferred taskThe ‘intrinsically preferred task’: The task in which the agent exerts agreater effort when there is no bonus pay.Task 1 is the intrinsically preferred task iff ˜e1 (0) > ˜e2 (0), orθ1c1 + γ>θ2c2 + γIntuitively, task i is more likely to be intrinsically preferred if θi islarger or ci smaller.Maitreesh Ghatak (LSE) Motivating knowledge agents
    • Introduction Theory Empirics Discussion and conclusionRaw scores by minisurveyMaitreesh Ghatak (LSE) Motivating knowledge agents
    • Introduction Theory Empirics Discussion and conclusionEnrolment by minisurveyMaitreesh Ghatak (LSE) Motivating knowledge agents
    • Introduction Theory Empirics Discussion and conclusionKnowledge test questions, survey wave 11 Does the programme cover the cost of treatment received while admitted to a hospital(hospitalisation)?Yes.2 Does the programme cover the cost of treatment received while not admitted to ahospital (out-patient treatment)?No.3 Who can join this programme?Households designated as being Below the Poverty Line.(Those who said ‘the poor’, ‘lowincome’ or similar were marked as correct.)4 What is the maximal annual expenditure covered by the scheme?30,000 rupees.5 How much money do you have to pay to get enrolled in the scheme?30 rupees per year.6 How many members of a household can be a part of the scheme?Up to five.7 What is the allowance per visit towards transportation to the hospital that you are entitledto under the RSBY scheme?100 rupees. (This was the expected answer, although strictly speaking the transportationallowance is subject to a maximum of 1000 rupees per year, i.e. ten visits.)8 Is there an upper age limit for being covered by the scheme? If yes, what is it?There is no upper age limit.Maitreesh Ghatak (LSE) Motivating knowledge agents
    • Introduction Theory Empirics Discussion and conclusionKnowledge test questions, survey wave 21 What is the maximum insurance cover provided by RSBY per annum?30,000 rupees.2 Does the beneficiary have to bear the cost of hospitalisation under the RSBY scheme upto the maximum limit?No.3 Are pre-existing diseases covered under RSBY?Yes.4 Are out-patient services covered under RSBY?No.5 Are day surgeries covered under RSBY?Yes.6 Does the scheme cover post-hospitalisation charges? If yes, up to how many days?Yes, up to 5 days. (Anyone who answered ‘yes’ was marked as correct.)7 Are maternity benefits covered?Yes.8 If a female RSBY member has given birth to a baby during the policy period, will thebaby be covered under RSBY?Yes.Maitreesh Ghatak (LSE) Motivating knowledge agents
    • Introduction Theory Empirics Discussion and conclusionKnowledge test questions, survey wave 31 Under RSBY, how many family members can be enrolled in the scheme?Five.2 What is the maximum insurance cover provided by RSBY per policy period?30,000 rupees.3 If hospitalised, does an RSBY cardholder have to pay separately for his/her medicines?No.4 If hospitalised, does an RSBY cardholder have to pay separately for his/her diagnostictests?No.5 Is it compulsory for an RSBY cardholder to carry the smart card while visiting the hospitalfor treatment?Yes.6 If an RSBY cardholder is examined by a doctor for a health problem but not admitted tothe hospital, will the treatment cost be covered under RSBY?No.7 What is the duration/tenure of the RSBY policy period?1 year.8 How can an RSBY cardholder check if a particular health condition is covered underRSBY prior to visiting the hospital for treatment?Multiple correct answers, see text.Maitreesh Ghatak (LSE) Motivating knowledge agents
    • Introduction Theory Empirics Discussion and conclusionImpact on enrolment(1) (2) (3) (4)Enrolled Enrolled Knowledge Enrolled(OLS) (Reduced form) (First stage) (IV)Knowledge 0.206*** 0.390***(0.00910) (0.128)Incentive-pay agent in village 0.0816** 0.209***(0.0362) (0.0618)Time fixed effects Yes Yes Yes YesTaluk fixed effects Yes Yes Yes YesObservations 5641 5641 5641 5641[para,flushleft] Notes: Weighted least squares regressions. Each household is given the same weight, divided equallybetween all observations of that household. Standard errors, in parentheses, are clustered at the village level. * p<0.10, **p<0.05, *** p<0.01Maitreesh Ghatak (LSE) Motivating knowledge agents
    • Introduction Theory Empirics Discussion and conclusionImpact of knowledge on take-upIV estimates suggest one SD increase in knowledge score increaseslikelihood of enrolment by 39% pointsIV estimates nearly double that of OLSPotential explanation: LATE (average effect on ‘compliers’)Possible concerns with IV:Persuasion to enrol (but incentives not based on enrolment)Endorsement effect (should not differ b/w incentive-pay and flat-payagents)Strategic behaviourMaitreesh Ghatak (LSE) Motivating knowledge agents
    • Introduction Theory Empirics Discussion and conclusionPhysical distance(1) (2) (3)Knowledge Knowledge KnowledgeIncentive pay 0.160* -0.128 -0.254(0.0923) (0.139) (0.162)Social distance -0.655*** -0.766***(0.214) (0.249)Incentive pay x social distance 0.794*** 0.867***(0.217) (0.260)Castes live apart -0.384**(0.182)Incentive pay x castes live apart 0.392*(0.200)Village size in thousands 0.196 0.205 0.288*(0.154) (0.158) (0.158)Agent and household characteristics Yes Yes YesTime and taluk fixed effects Yes Yes YesSocial distance metric - Composite CompositeObservations 2900 2900 2327[para,flushleft]Maitreesh Ghatak (LSE) Motivating knowledge agents
    • Introduction Theory Empirics Discussion and conclusionMain results, Shimoga district only(1) (2) (3)Knowledge Knowledge KnowledgeAgent in village 0.210** 0.191**(0.0823) (0.0743)Flat-pay agent in village -0.0289(0.121)Incentive-pay agent in village 0.317***(0.0683)Time fixed effects No Yes YesTaluk fixed effects No Yes YesObservations 2885 2885 2885t-test: flat=incentivised (p-value) 0.007[para,flushleft]Notes: Weighted least squares regressions. Each household is given the same weight, divided equally between all observations ofthat household. Standard errors, in parentheses, are clustered at the village level. * p<0.10, ** p<0.05, *** p<0.01Maitreesh Ghatak (LSE) Motivating knowledge agents
    • Introduction Theory Empirics Discussion and conclusionTreatment effect interacted with agent characteristics(1) (2)Knowledge KnowledgeTreatment (agent in village) 0.187*** -0.499(0.0572) (0.331)Treatment x agent is 30+ 0.0453(0.0916)Treatment x agent is 50+ -0.0826(0.0938)Treatment x agent of forward/dominant caste -0.102(0.0894)Treatment x agent household head has completed primary school -0.105(0.0931)Treatment x agent has ration card -0.0642(0.123)Treatment x agent owns her home 0.148(0.108)Treatment x agent is Self-Help Group president 0.00918(0.0865)Treatment x agent autonomy 0.121**(0.0466)Time fixed effects Yes YesTaluk fixed effects Yes YesObservations 5641 5641[para,flushleft] Notes:Weighted least squares regressions. Each household is given the same weight, dividedMaitreesh Ghatak (LSE) Motivating knowledge agents
    • Introduction Theory Empirics Discussion and conclusionIs the effect symmetric (or, does direction matter)?Dominant-caste agentDominant Non-dominant Differencehousehold householdFlat pay 0.22 -0.17 0.39***(0.1) (0.05) (0.11)Incentive pay 0.12 0.11 0.01(0.09) (0.04) (0.09)Difference -0.1 0.28** -0.38***(0.14) (0.06) (0.15)Non-dominant-caste agentDominant Non-dominant Differencehousehold householdFlat pay -0.2 0.09 -0.29**(0.09) (0.05) (0.11)Incentive pay 0.08 0.17 -0.09(0.08) (0.03) (0.08)Difference 0.28** 0.08 0.2(0.12) (0.06) (0.14)[para,flushleft]Maitreesh Ghatak (LSE) Motivating knowledge agents
    • Introduction Theory Empirics Discussion and conclusionDiscussion: The effect sizeOn average, agents were paid a bonus of 8 rupees per ‘successful’householdIncreased raw score by 0.6 and take-up rate by 8 %-pointsIt appears that a modest amount of incentive pay can wipe outknowledge gap between own and cross groups (part-time work)Suggestive evidence that people spend on average 3–4 days full-timeon agent work in each round400 rupees average pay / 4 days of work = 100 rupees per dayCorresponds to typical unskilled wage in the areaAgents may be more sensitive to having an incentive rather than levelof incentive (Filmer and Schady, 2009, Thornton, 2008, Banerjee etal., 2010)Maitreesh Ghatak (LSE) Motivating knowledge agents
    • Introduction Theory Empirics Discussion and conclusionExtra materialMore detail on the two-task modelShimoga onlyImpact on enrolmentThe time dimensionAgent characteristicsPhysical distanceSymmetryDiscussion: The effect sizeKnowledge test questionsMaitreesh Ghatak (LSE) Motivating knowledge agents