Presentation on theme: "Health information, health treatment, and worker productivity: Evidence from Nigeria Andrew Dillon Michigan State University Jed Friedman The World Bank."— Presentation transcript:
Health information, health treatment, and worker productivity: Evidence from Nigeria Andrew Dillon Michigan State University Jed Friedman The World Bank Pieter Serneels University of East Anglia October 2013
Motivation Health largely seen as critical input to economic development … but economic gains from health investment difficult to measure Yet economic gains from health investment important for policy decisions Agricultural productivity is key for growth in low income countries, but productivity increasing investments (irrigation) may increase exposure to vector borne illnesses. Does investment in health increase productivity and income, and by how much?
Motivation Malaria remains a significant illness worldwide, especially for sub-Saharan Africa Estimated 210 million infections and ~1 million deaths per year 20% of estimated malaria cases worldwide are found in Nigeria 51% of households in Nigeria reported at least one episode of malaria (NLSS 2003/4) – in endemic settings, “malaria” often becomes a general term for illness/fever Estimated cost of malaria from previous work: direct cost of treatment absenteeism from work (‘indirect cost’) is estimated between 1-5 days per episode
Motivation Existing micro-economic evidence suggests substantial effects of health on income and labor outcomes in low income countries Disease affects labor outcomes Reduction in malaria exposure at an early age increases income later in life (Cutler et al. 2010 and Bleakley 2010) HIV/AIDS reduces labour supply in Kenya, Botswana (Fox et al 2004;Thiruthmurti et al 2006;Habyarimana et al 2010) Schistomosiasis reduces labor supply in Mali (Audibert and Etard, 1998) Pollution reduces worker output in US (Zivin 2010)
Some micro-estimates of cost per malarial episode Study Country Direct cost (% of total cost) Indirect cost (% of total cost) Morel et al.VietnamUS$ 0.7 (6%)US$ 11.09 (94%) Akazili et al.GhanaUS$ 1.87 (29%)US$ 4.52 (71%) Deressa et al.EthiopiaUS$ 1.60 (28%)US$ 4.08 (72%) Cropper et al.EthiopiaUS$ 1.60 (7%)US$ 22 (93%) Sauerborn et al.Burkina FasoUS$ 1.85 (31%)US$ 4.11 (69%) Ettling & ShepardRwandaUS$ 0.63 (28%)US$ 2.25 (72%) Asenso-Okyere & Dzator GhanaUS$1.81 (21%)US$ 6.87 (79%) Indirect costs (absenteeism) is highest, but estimates derive from self-reports, valued at average wages, not including on-the-job costs. Furthermore, all based on observational data.
Motivation for current study design Existing empirical evidence suffers from problems with: Omitted or unobserved variables E.g. endowment effects: worker’s initial health condition, physical work capacity Reverse causality Are healthier workers more productive or more productive workers healthier? Or both? Measurement of productivity Need a measure at the individual worker level Need uniform measure, so consider homogenous activities
Empirical questions Does access to workplace based malaria testing and treatment affect income, labor supply, and productivity? If malaria treatment leads to higher income, is this because of its effects on labor supply, productivity, or both? Are there other benefits from workplace testing and treatment for malaria?
Study site One large 5,700 hectare sugarcane plantation in rural Nigeria Employs ~800 sugarcane cutters who work throughout the season Workers are organized into work groups, managed by a supervisor and headmen Workers are paid (piece rate) 2.04 naira per “measured rod” of sugarcane they cut A “measured rod” is a physical standard carried by every gang leader
Study site The plantation records for each worker the daily amount cut, days worked, and earnings Workers are paid monthly, and keep careful track of their earnings to verify payments At the start of every work day, workers can select into cane-cutting or scrabbling: Scrabbling pays fixed wage (500 Naira/day) and is less arduous, involves collecting and bundling cut cane, most workers cut cane but many scrabble for a small proportion of days Harvest work is seasonal: December – March and relatively lucrative
Worker’s decision theory Worker’s output is a function of ability and effort. Effort may be influenced by wage, health self-perception, individual motivation, etc. Worker’s earnings are determined by both labor supply and their productivity when working. Worker’s choose not only how much to work but which occupation, cutting for a piece rate or scrabbling for a lower average fixed rate. This choice depends on the worker’s willingness to supply effort.
Experimental design A mobile health clinic visits the workforce during working hours Every sugarcane cutter is tested for malaria and, if positive, treated with ACT over the course of an 8 week period But the temporal order of worker testing and treatment determined on a random basis … … potentially inducing exogenous variation in worker health over study period Conducted at peak of harvest season (Jan, Feb) in 2010
Experimental design (cont.) Each week, 2-3 randomly selected gangs visited by study team, and random subset of workers assessed Study design: blocked by work group, assessment temporally randomized across workers within work group Small survey team administers work and health questionnaire Mobile health clinic tests for malaria (RDT and blood slides) All parasitemic positive workers treated with ACT
Experimental design: validation In principle randomization guarantees valid counterfactual but in reality: 8 work groups Over 1 harvest season ~800 workers Need to explore balance of worker characteristics within work group over time
Worker characteristics and importance of work group strata
Within work group balancing test across survey week, by round
Temporal variation in earnings and labor supply, even across work groups Daily earnings and days worked, by selected week and work group
Measurement and diagnosis of malaria Diagnosis follows the WHO standard of microscopy from thick blood smear, independent validation Results in 3-day lag between specimen collection and diagnosis: Metric is number of parasites observed in 5 randomly selected blood field, MOH cut-off is 3 parasites
Impact estimation strategy Workers grouped by week of interview (and work group) to identify windows of potential impact
Econometric strategy overview We start with an intent to treat effect (ITT) ITT: compares outcomes for workers who were offered access to testing and treatment with outcomes for those who had not yet been offered access to testing and treatment Stratification by work group, and temporal variation within work group, indicates importance of gang-week FE Where W t- is set of all workers assessed before observation period t
ITT results 11-13% gain in earnings from treatment over 2-3 weeks, due in roughly equal measure to increased labor supply and increased productivity
Robustness of ITT Results likely unaffected by externalities from treatment: most malaria transmission occurs in evening or night when workers are off plantation symptoms usually begin 12-25 days after infection Inclusion of worker observables (BMI, hemoglobin, education, etc.) does not affect results
Summary so far A temporally randomized work site regimen of malaria testing and treatment indicates significant increase in earnings (~10%) Increase is due in roughly equal proportions to: Increased labor supply Increased earnings per day
Treatment on the treated We can also attempt to estimate the ToT with an additional assumption ToT: compares outcomes for parasitemic positive workers who received ACT with outcomes for workers who were sick during the same period, using information from future positive tests
Treatment on the treated An earnings effect of roughly same magnitude as ITT, mostly through labor supply response. Here the estimated 3-week earnings benefit ~$9
Compliance protocol Validity of ToT estimates is predicated (in part) on compliance with treatment Workers who tested positive were visited by the health supervisor on: Day 2 of their ACT prescription The day after their prescribed treatment ended Workers were given a small incentive (50 Naira) to return the medicine container and asked follow-up questions to confirm proper compliance with treatment and to identify any adverse effects
Treatment on the treated (cont.) What may account for similar magnitudes of ITT and ToT estimates, given 36% positivity rate? Our counterfactual for the “ill” is noisy – TOT estimates are comparing recovering workers with some proportion of healthy ones (and not all sick ones as necessary). Very difficult to correct for this with existing data The intervention is a combination of two “treatments”: Health information on malaria status Malaria treatment if sick Might workers also be responding to the first component?
‘TOUT’ on malaria negative workers Similar in earnings magnitude to ITT estimates, increases largely coming through increases in daily wage
Reductions in scrabbling greater for malaria negative workers Furthermore it appears that gains for negatives at least in part due to switch into piece-rate from scrabbling
Gains to negative workers not solely due to decreases in scrabbling Can restrict estimates to worker-weeks with no scrabbling
So what is happening here? Does good health news lead to higher worker participation and productivity? New health information has been known to affect subsequent behavior, at least in the case of HIV, where learning positivity status leads to safer practices (Thornton (2008, 2012)) But little literature concerning health status learning effects of communicable diseases Theory may tell us (i.e. Gong 2013) that only surprises from new information should affect behavior – i.e. a change in expectations TOT for malaria negative workers
Suggestive findings consistent with good health news leading to change in expectations of productivity Are there other explanations/possible channels? A “meaning response” a la Moerman (2002) who examines the placebo effect and argues the “meaning” given to placebo is what induces benefits. Can a healthy diagnosis in endemic setting by medical professional convey sufficient meaning to alter subjective perception of well-being and actually increase productivity? Gift exchange a la Akerlof (1982) – gratitude for health test leads to higher morale and greater effort. Unlikely in piece-rate setting and this plantation context. TOT for malaria negative workers
Conclusions Clear that the malaria testing and treatment increases the labor supply and productivity of the average worker Earnings responses for both the malaria positive, who receive medication, and malaria negatives Gains among the sick largely due to increased labor supply, and greatest gains observed among the sickest. At prevailing wages, estimated $9 benefit from treatment Gains among the healthy apparently in part due to change in earnings expectations (shift out of scrabbling). Estimated $11 benefit from receiving good news. Example of workplace based health program that likely pays for itself
Are these gains the “productivity costs” of malaria infection? Only a fraction of workers have malaria, what is the expected gain from treatment for them? 35.9% of workers tested positive and given treatment Wald estimator [ITT estimate over proportion treated] is one take at a TOT Substantial gains from treating malaria! 3 week gain approximately $30, while cost of ACT = $7
Treatment on the treated So we estimate where P t- is set of all workers found parasitemic positive and given ACT before observation period t
Designing Evaluations of Workplace Based Programs Time varying access to intervention facilitates the construction of treatment and control groups Not all types of interventions can use this strategy (treatments with potential spillovers) Resolves ethical dilemmas Measurement of productivity Across firms worker productivity may be highly variable Piece rate wage systems are ideal Recording system of worker output is essential
If parasite presence, even at low levels, affects subjective health and expectations, then perhaps those with parasites more subject to surprise: Evidence on upside surprise
We can also stratify by subjective perception of work capacity: whether worker feels tired at end of day Evidence on upside surprise