Thursday, February 16, 2017, 12:30-2pm @ The GovLab, 2 Metrotech Center, 9th Floor, Brooklyn, NY 11201
Getting outside expertise is essential to improving the quality of decisionmaking in government. While universities are a storehouse of knowledge and experience, public officials often turn to lobbyists, think tanks and interest groups, instead, because of the difficulty of quickly ascertaining who has expertise on a given topic and the nature of that expertise. The problem is exacerbated by the absence of (i) well-formatted records of knowledge, when (ii) it is difficult ex ante to specify relevant types of expertise in a survey, and (iii) individuals with overlapping interests may use related, but not identical, language to describe similar issues. This talk explores new computational strategies—in particular, a growing class of “word embedding” models—for creating measures of individual-level expertise that might offer insight into how, in the future, we can match public problems to the supply of experts and help public institutions obtain diverse knowledge more quickly. The talk concludes with a discussion of how model estimates may be used on data outside of the training set (e.g., as merged with resumes, or ‘LinkedIn’ type data), when the quality of observed data varies across individuals, or in tandem with existing user meta data derived from a social network platform.
Michael Gill is a Moore-Sloan Data Science Fellow at the Center for Data Science at NYU, and a Research Fellow at the GovLab. Michael’s substantive interests include the study of special interest groups, the causes and effects of government transparency, and U.S. foreign policy. Methodologically, his research focuses on applications of machine learning methods for causal inference problems in the social sciences, experimental methods, and the analysis of text-as-data. He received his Ph.D. in Government from Harvard University, where he was an affiliate of the Institute for Quantitative Social Sciences.