Tweeting to Conserve: A Natural Language Analysis of Drought Messaging

• In Event: Water Governance Issues

Abstract

Public service providers, like water districts, rely on effective public messaging to persuade their users to adopt preferred behaviors. Known as citizen coproduction governments increasingly rely on the cooperation of their citizens to provide necessary public goods. To do so they rely on effective messaging. While a growing literature, has begun to examine techniques for maximizing the effectiveness of this public outreach, little work has explored the extent to which public agency messaging actually aligns with best practices from behavioral science. Accordingly, this paper examines the case of social media outreach of California water districts in their efforts to promote voluntary water conservation in times of drought.

Specifically, it uses supervised machine learning techniques to categorize the approximately 400k tweets from California water district accounts between 2012 and 2022 aimed at promoting conservation. Using natural language processing(BERT) each tweet was then categorized into one or more nonexclusive labels. This begins with if the tweet is drought-focused or not. If so, it was then categorized if focused on using financial incentives, regulations, or voluntary behavior change to increase conservation. An additional focus is on tweets promoting voluntary behavior change. These behavioral interventions are categorized using the information-motivation-behavioral(IMB) skills framework used previously in experimental research(Ehret et al, 2019). Using machine learning allows for the categorization of the full population of tweets allowing for in-depth analysis of temporal and spatial variation as well as correlations with drought levels and conservation rates. These data are supplemented with interviews with district social media managers. Initial results show that water districts rarely message around incentives or regulation, relying instead on voluntary conservation. However, a large majority of tweets, approximately two-thirds, focus on providing simple information about the drought. By not including adequate motivation and conservation skills in their outreach districts are likely failing to maximize voluntary conservation. The broader implications of this disconnect for the effectiveness of citizen coproduction and the value of government messaging are discussed.

Authors

• Robert Durbin

• Colin Kuehl, Northern Illinois University