Association for Public Policy Analysis & Management

Local governments face significant resource constraints, and everyday public service provision demands make it hard for local leaders to prioritize prospective actions taken towards uncertain, irregular events. This paper uses the cases of severe drought and wildfire in the state of California to understand how different localities conceive of chronic and acute risks, and how events both within the jurisdiction and elsewhere induced shifts in stated priorities, policy actions, or both. This study takes advantage of institutional documents such as yearly budget proposals and budget hearing minutes, which present a wealth of untapped information concerning the policy behavior of local government organizations. Spending justification statements and budget hearing minutes provide key insight into the ways that each organization conceives of major risks, and speaks to the deliberation process by which competing policy priorities are rectified. The end goal is to understand how local governments can more effectively prepare for uncertain, potentially high consequence events even in the face of ongoing resource limitations and everyday public service demands.

Every county in the state of California produces a yearly executive budget that includes a justification for expenditures and outlines policy priorities for that year. Further, these proposed spending priorities are scrutinized in a series of public meetings held by county commissioners. This paper applies machine learning text analysis, specifically natural language processing and topic modeling, to these budget justification documents and county commission hearing minutes for each of the 58 counties in California from 2012 to 2017. For example, given a set of budget statements from each county in California in 2017, topic modeling is used to quantify the proportion of content within each document that is devoted to a particular topic. It is this distribution that will be used to construct a metric of stated preferences which will be then linked to actual spending to explore covariation between these over time. Similarly, applying this approach to budget hearing minutes will further allow us to quantify key aspects of deliberation occurring within county board of supervisors meetings and to measure priorities as emphasized by both the executive branch county officials and the members of the board of supervisors.

These textual data are then coupled with budget and spending data over the same period in order to understand how stated priorities relate to subsequent spending and implementation decisions. Finally, we use wildfire outbreak and drought records obtained from the United States Geological Survey in order to quantify the risk that each locality faces and each county’s spatial proximity to past threats. We then trace how these priorities shift over time as a function of water supply and wildfire outbreak risks, and model the connection--or lack-thereof--between planning discourse (as reflected in hearing minutes), stated policy emphases (from budget narratives and messages) and resulting actions (fiscal behavior) using highly detailed local government expenditure data.