Public water utilities have increasingly turned to increasing block rate price structures to reduce water consumption and signal the high environmental costs of water supply. This paper evaluates the impact of a new and substantially higher price block added to the existing block rate price structure in Seattle -- often referred to as a `shock rate' -- which only affected those who consume very high quantities of water.  The public water utility in Seattle added such a rate to its existing price structure in 2001, and has subsequently seen significant decreases in per capita water demand.

Rigorous evaluation of policies such as the shock rate is often constrained by the limited availability of appropriate data that describes individual consumption decisions.  Much of the previous literature of water demand relies upon aggregated data, which often results in theoretically implausible results such as price elasticities with the wrong signs.  As a result of these empirical data limitations, there has also been limited application of theoretically appropriate models such as discrete-continuous choice (DCC) within the literature of demand for water and other resources.

This paper applies a DCC model to a new, rich source of observational micro-data to evaluate changes in water consumption in Seattle as a result of the new pricing structure.  A comprehensive billing database of water consumption for individual households in the period from 1991 to 2007 was obtained from Seattle Public Utilities.  The DCC model was applied in order to accomplish three complementary goals: first, to use a theoretically appropriate model to describe realistically the effect of inclined block rate price structures on water consumption; second, to identify group-level social, geographic, and climatic effects on water consumption; and third, to establish causal inference for, and quantify the effect of, the `shock rate' on aggregate water consumption. 

The DCC model is estimated using Bayesian Monte Carlo Markov Chain (MCMC) methods implemented in the Python language and using the PyMC module.  Finally, the paper concludes by considering possible extensions of this model, including time series and spatial components, and analysis of the distributional effects of new proposed pricing policies.