Political Selection and Bureaucratic Productivity

Political Selection and Bureaucratic Productivity James Habyarimana 1 Stuti Khemani 2 Thiago Scot 3 June 25, 2018 1 Georgetown 2 World Bank 3 UC Berkeley 1

Motivation: understanding local state capacity Impact of anti-poverty & pro-growth policies depends upon how well they are implemented by government agencies Quintessential delivery unit in poor countries: local governments helmed by centrally appointed bureaucrats and locally elected politicians Corresponds to special features of principal-agent problems in the public sector: multiple principals and tasks; and complexity of objectives (Tirole, 1994; Dewatripont, Jewett and Tirole, 1999; Dixit, 2002; Besley and Ghatak, 2005) 2

Motivation: understanding local state capacity Theory says high-powered incentives might have limited role in fostering effort provision in such organizations; selection of agents with intrinsic motivation might be very relevant. But we have little (albeit growing) empirical evidence on selection (Finan, Olken, Pande, 2015), and particularly in political selection (Gulzar and Khan, 2017; Banerjee et al, 2017). 3

Overview: what we do Rich survey data collected in 2015 on district-level bureaucrats and politicians in Uganda (institutional context typical of delivery units in poor countries) Link survey measures of selection traits to administrative data on the productivity of the local state (Districts) We cannot causally estimate the effect of politicians integrity on service delivery, but provide first robust evidence of which individual traits predict better service delivery. 4

Main findings Politicians and bureaucrats differ in systematic ways, and differences in selection traits persist when controlling for demographics Integrity of politicians is a robust correlate of productivity of the local state: 1 s.d. increase in average politician integrity correlates with 0.2-0.4 s.d. increase in health score index Political integrity is in short supply, however: politicians score significantly lower than bureaucrats, and those who win elections tend to score less than contenders who lost Presence of local media and support to incumbent national party (NRM) are strong predictors of health outcomes, but local political competition is not. 5

Data and Institutional Environment

Survey description Original survey undertaken between September - December 2015 in 75 of Uganda s 112 districts to measure the selection traits of district bureaucrats and politicians. Surveyed politicians include the directly elected District Chairperson and Councillors, as well as contenders for Chairperson position who lost in the last elections (2011). Bureaucrats: leaders such as the Chief Administrative Officer (CAO), well as a roster of other senior technical officers (eg. District Health Officer) In total, over 2,000 individuals were interviewed (770 politicians and 1,357 bureaucrats) 6

How do Politicians and Bureaucrats Differ? Our rich survey data allows us to compare politicians and bureaucrats for a large set of individual characteristics. Some differences might be unsurprising: bureaucrats are appointed from a national cadre of qualified public servants, while politicians are local leaders. Bureaucrats are much more educated, wealthier and more likely to be male. 7

Bureaucrats are much more educated than politicians Fraction 0.1.2.3.4.5 Less than complete primary Less than complete secondary Less than complete college Complete college Coarser measure of individual's education Master's/PhD Bureaucrat Politician 8

How do Politicians and Bureaucrats Differ? Differences in personality traits are less intuitive: are civil servants more or less intrinsically motivated than politicians? What about differences in integrity? We measure a broad range of traits using modules similar to related literature (Dal Bo, Finan and Rossi, 2013; Callen et al., 2014; World Bank STEP survey): Integrity (Moral Disengagement Measure) Public Sector Motivation (PSM) Non-Cognitive ability (Big Five, Grit, Decision-Making and Hostility Bias) Cognitive ability (Memory and Risk understanding) Risk preferences 10

Integrity measure 11

1.8 Cumulative Probability.6.4.2 0-4 -2 0 2 Integrity Index Politician Bureaucrat 12

1 KS test p-value = 0.00.8 Cumulative Probability.6.4.2 0-4 -2 0 2 Integrity Index Politician Bureaucrat 13

Findings Overall findings Bureaucrats score higher in Integrity measures, Cognitive ability and Non-Cognitive traits such as grit and decision-making But politicians perform better in Public Sector Motivation and non-cognitive traits such as Extraversion and Agreeableness Given stark differences in education and wealth, natural question is whether individual traits differences persist once we condition on demographics 14

Gaps in individual traits remain Other Non-Cognitive Integrity index Asset Index Cognitive Choose dominated risk Altruism Risk Aversion Non-Cognitive index PSM index Big Five index -.4 -.3 -.2 -.1 0.1.2.3.4 Point estimate and 95% CI 15

Results Previous results suggest that politicians and bureaucrats differ in relevant dimensions Differences in easily observable demographics do not fully explain gaps in individual traits: politicians and bureaucrats are systematically distinct. To what extent do these individual traits explain variation in service delivery at the District level? 17

Productivity of district bureaucracies

Productivity of district bureaucracies We match survey data on politicians and bureaucrats to productivity of Districts in implementing national health programs Main outcome: FY 2014/2015 District League Tables (DLT) scores DLT summarizes performance in health delivery, compiling data reported by local facilities. It comprises both service delivery measures (antenatal care, vaccination, outpatient visits) and management outcomes (completeness of reports and medicine orders) 18

Coverage and quality care Management Weight Explanation DPT3 Coverage (%) 15 Diphtheria-tetanus-pertussis (DTP3) immunization coverage Share of infants delivered in Deliveries in gov t and PNFP facilities (%) 15 government or Private Non for Profit facilities OPD Per Capita 10 Number of outpatients department (OPD) visits per capita HIV testing in children born to HIV+ women (%) 10 - Latrine coverage in households (%) 10 - IPT2 (%) 5 Share of pregnant women who completed Intermittent preventive therapy (IPT) against Malaria ANC 4 (%) 5 Proportion of pregnant women receiving 4 or more antenatal care visits TB TSR (%) 5 Tuberculosis Treatment Success Rate Approved posts filled (%) 10 - % Monthly reports sent on time 3 - % Completeness monthly reports 2 - % Completeness facility reporting 3 - Completeness of the annual report (%) 2 - Medicine orders submitted timely (%) 5-19

Productivity of district bureaucracies We first present partial correlations of (residualized) health scores and average individual traits Health scores and politicians /bureaucrats quality, however, are correlated with a host of other factors Then investigate whether partial correlation results hold when we condition on districts socio-economic and political characteristics Population, poverty rates, urbanization shares, education performance (UCE scores) Support for President Museveni and competition between local councillors Number of Radio HQs 20

Higher Politician Integrity is correlated with better health outcomes -3-2 -1 0 1 2 Slope = 0.29 [s.d. = 0.11] -3-2 -1 0 1 2 Integrity Index - Average Politicians Figure 1: Integrity Index 21

Politicians traits vs. health outcomes I 22

Bureaucrats traits vs. health outcomes I 23

Conditional on Districts characteristics, politicians integrity matters... Integrity index PSM Index Big Five Index Non-Cognitive Index Altruism Dominated Risk Option Cognitive Ability Risk Aversion -.4 -.2 0.2.4 Point estimate and 95% CI Figure 2: Politicians Traits 24

...but no bureaucrats traits do. Integrity index PSM Index Big Five Index Non-Cognitive Index Altruism Dominated Risk Option Cognitive Ability Risk Aversion -.4 -.2 0.2.4.6 Point estimate and 95% CI Figure 3: Bureaucrats Traits 25

Model Selection - LASSO

Dealing with multiple covariates - LASSO estimates Given large number of individual traits and limited number of observations (districts), we are faced with the question of what model to estimate Strategy: reduce "degrees of freedom" in model selection and rely on regularization techniques of machine learning (Athey, 2018; Abadie and Kasy, 2017) We perform a LASSO estimation to define a subset of variables that explain most of the variation in health outcomes - trade-off minimizing prediction error and number of covariates (Tbishirani, 1995). 26

Dealing with multiple covariates - LASSO estimates The LASSO model chooses 9 covariates with non-zero coefficients: Two politicians traits: Integrity and PSM Three bureaucrats traits: Altruism, Risk-aversion and share of less than college educated Four District socio-economic and political characteristics: Vote share of President Museveni, Poverty Rate, Number of Radio HQs and UCE score We then estimate an OLS model restricted to these explanatory variables Caveat: LASSO doesn t add anything in terms of identification; it is a model selection tool. 27

LASSO model Number Radios (HQ) Average Integrity - Politician Altruism - Bureaucrat Risk Aversion - Bureaucrat Poverty Rate 2013 UCE score 2011 Less than College education - Bur Average PSM - Politician Voter Attachment to National Party -.6 -.4 -.2 0.2.4.6 Point estimate and 95% CI 28

LASSO model 29

Robustness tests Model: OLS estimates with different set of covariates Dependent variable: individual health outcomes Dependent variable: PC of health outcomes Independent variable: breakdown of integrity by politicians type (Councillor, Chairperson or Contender) Further controls: theory-informed controls (management practices, ethnic diversity) 30

Robustness I - Individual Health Outcomes 1 s.d. increase in average politicians integrity is correlated 0.4 s.d. increase in anti-malaria coverage for pregnant women (IPT2 coverage); 0.4 s.d. increase in the share of households having a latrine; 0.3 s.d. increase in the share of infants delivered in government or Private Non- Profit facilities. Completeness reports IPT2 coverage Reports sent on time Latrine Coverage Completeness facilities Antenatal care coverage Deliveries in facilities Total Score Outpatients visit DPT3 coverage Completeness annual report Medicine Order timely TB treatment success HIV testing children Approved posts filled -.5 0.5 1 Point estimate and 95% CI 31

Discussion of mechanisms and conclusion

Alternative Explanations? Causal interpretation of correlations: effort towards public goods by high-integrity politicians However, local economic and political conditions might cause both politicians traits and service delivery; we control for a number of observables, but harder to account for unobservable citizen demand for better services or social norms Results on political integrity survive the inclusion of local media environment, voter attachment to the national ruling party, and electoral competitiveness We argue that action by politicians is the proximate mechanism, even when things like citizen demand/social norms may be the underlying driver (supporting evidence from the field, and in other studies, eg. Grossman, Platas and Rodden, 2018) 32

Conclusion The research agenda has shifted away from comparing centralized versus decentralized provision of public goods towards examining principal-agent relationships within complex organizations of government. Some key theoretical predictions have been empirically confirmed (e.g. higher wages attract better candidates, Dal Bo, Finan and Rossi, 2013) We provide the first evidence that political selection matters for the productivity of bureaucratic organizations Policy efforts to build capacity of the local state should take seriously the role of political selection when crafting the principal-agent relationships in local bureaucracies. 33

Robustness I - alternative OLS models 34

Robustness III - OLS with PCA of health outcomes Back 35

Robustness IV - Integrity by Politician Type Back 36

Robustness V - other theoretical channels Back 37

What correlates with politicians integrity?

Are all politicians the same? 38

Are all politicians the same? 39

Are all politicians the same? 40

What predicts politicians integrity - District level 41

Results - Integrity by Politician Type 42

Bureaucrats also have more assets Fraction 0.1.2.3.4 1 2 3 4 5 Asset Index quantiles Bureaucrat Politician 43

Bureaucrats also have more assets Fraction 0.1.2.3.4 1 2 3 4 5 Asset Index quantiles Bureaucrat Politician 44

Results - Integrity by Politician Type 45

Results - Integrity by Politician Type 46

Results - Integrity by Politician Type 47

Results - Integrity by Politician Type 48

Non-Cognitive traits measure We use two instruments to measure Non-Cognitive traits: Big Five Index and an extended instrument including Grit, Decision-Making and Hostility Bias dimensions. 49