Association for Public Policy Analysis & Management

The difficulties of measuring government performance have been well documented in public management literature (Boyne 2003; Moynihan 2008). Assessing performance in the public sector remains a key question to public management, in which an entire subfield of performance measurement is dedicated to tackle the problem (Meier and O’Toole 2012). Majority of previous studies that focused the performance of federal government have relied on employees’ or managers’ self-assessment of performance of their agencies in surveys like Federal Employee Viewpoint Survey, or Merit Principles Survey. Studies using “objective” measures of government performance often adopt proxy measures or indicators (test scores, crime rates, etc.) that only capture parts of what public organizations are doing. Moreover, the cost of collecting these subjective or objective performance data is usually rather high. In this project, I explore the possibility of using “big data” from social media to measure government performance. Specifically, I examine whether sentiments of performance-related tweets and general sentiments of tweets about a particular federal agency would match 1) self-reported performance data from the agency employees and 2) citizens’ favorability rating of this agency. Performance-related tweets replied to or mentioning 17 department or independent agencies of federal government are collected and sentiment analyzed using a machine learning algorithm called Naïve Bayes classifier. Sentiment score of each agency would be compared with perceptual performance data from 2015 Federal Employee View Point Survey and favorability data from a 2015 political survey conducted by Pew Research Center. Preliminary results show strong correlation between sentiment score of performance-related tweets and citizens’ favorability, but weak correlation between sentiment scores and employees’ self-reported performance. This paper contributes to the field of public management by offering an alternative way of measuring performance of public organizations – using sentiments of performance-related tweets, which is easily accessible, in real-time and less costly. As field of public affairs enters era of “big data”, this paper answers the call of taking advantage of newly emerging data and meaningfully combining them with administratively collected data to have value in improving public programs (Mergel et al. 2016) and furthers our knowledge of whether adopting social media increases citizen participation, collaboration and transparency.