Household (Under) Adoption of Sanitation: Importance of Externalities and Borrowing Constraints
Updated draft forthcoming
This paper analyses the problem of under-adoption of sanitation and addresses the current policy debate on the choice between loans and price subsidy policies to increase sanitation coverage in the developing world. While sanitation is a expensive investment for a poor potentially liquidity constrained household, adoption also generates positive health externalities for others within the village. Both factors may result in under-adoption but are driven by different sources of market failure. To investigate impact of these two distinct policies on sanitation coverage I estimate a dynamic model of household sanitation demand with interdependent adoption choice, using a unique dataset from rural India. I use the model to compute equilibrium adoption levels under both loans and subsidy policies to study the optimal design of interventions in an equilibrium setting. Counterfactual analysis reveals existing sanitation level to be below the social planner solution, implying under-adoption. I find price subsidies to be more cost effective at increasing sanitation coverage. But the policy effects are heterogeneous with coverage levels, where loans are found to be equally, if not marginally more, effective in villages with no sanitation coverage. A price subsidy has a high social rate of return where the presence of externalities accounts for a substantial fraction of its impact. While a sanitation loan policy generates smaller social returns it is also cost efficient under targeted delivery.
This paper analyses the impact of externalities on household demand for a preventive healthcare good and the subsequent welfare effects generated from a subsidy program towards its provision. An interesting feature of such goods is that the take-up generates externalities where the privately chosen adoption level differs from what is socially optimal. Using a unique dataset on sanitation take-up from rural India, I estimate a static demand model keeping into account the interdependence of household decision making within the village. A two-step estimating method is implemented to circumvent the computational burden associated with estimating a model under multiple equilibria. To evaluate the impact of subsidy interventions, I formulate analytical expressions to quantify substitution and income effects in the presence of externalities. Using this tool to further separate out the direct and indirect effects under different policy simulations, I find that substitution effects are significantly larger than income effects, and a substantial amount of this price effect is propagated through the indirect channel. The presence of externalities implies a larger welfare gain, while the Deadweight loss generated from the subsidy intervention is realised as a Net-Gain, as the society shifts towards a socially optimal level of adoption.
Work in progress
Two-step Conditional Choice Probability Estimators with Measurement Error
This paper develops a correction method for existing Two-step CCP methods to estimate static and dynamic discrete choice models of incomplete information. Under the assumption that the observed data is generated by one of the possible equilibria, two-step estimators avoid the computational burden associated with repeatedly solving the fixed point for each candidate vector of parameters. However, to obtain consistent choice probability estimates in the first stage, two-step estimators rely on being able to observe the entire vector of states and actions in the data. This data limitation can be treated as a contamination of the variable of interest with measurement error. Using insights from small variance approximation to probability distributions, I extend the error correction method proposed by Chesher (1991) to the estimation of simple static interaction models. The method is applied to estimate Brock & Durlauf's (2001) interaction model which has been the cornerstone in the study of peer effects in recent literature. Monte Carlo simulations are conducted to approximate the magnitude of the impact of error in data and the resulting bias in parameter estimates.
Keeping Money under the Udder: Buffer-Stock Livestock Ownership