New and Old Working Papers and Reports and Presentations
All titles are preliminary and comments are welcome. This collection is somewhat disorganized at the moment. Newer works are on the top and older/back-burner work, is on the bottom of each section.
Applied Statistics
Randomization and Design Based Statistical Inference for Causal Inferences
How to increase the precision of causal inferences in experiments using machine learning (but without data snooping). A paper showing how covariates can be used to increase the precision of statistical inferences about causal effects without data snooping (with Mark Fredrickson and Ben Hansen).
Design of Randomized Experiments on Networks When Treatment Propagates: An Exploration A presentation at the end of my participation in the Statistics and Applied Mathematical Sciences Institute year focused on computational social science that was also presented at the Political Networks Conference, May 2014. A collaboration with Bruce Desmarais, Mark Fredrickson, and Nahomi Ichino to learn about the design of randomized experiments to learn about how treatment may propagate across a social or geographic network.
Ethnicity and Electoral Fraud in New Democracies: Modelling Political Party Agents in Ghana A paper using randomization inference with agent-based models to learn about theories of party competition and ethnicity in Ghana (with Nahomi Ichino and Mark Fredrickson).
Regression without Regrets: A modular approach to linear models in (quasi)-experiments. Keywords: Causal inference; Machine Learning; Lasso; Randomization Inference
Do Newspaper Ads Raise Voter Turnout? (with Costas Panagopoulos)
A General Representation of Potential Outcomes for Graphs/Networks (draft)
Fisher's randomization mode of statistical inference, then and now.
How is statistical inference possible when n = 8? How can we infer
without a sample from a population? How should we choose methods for
assessing causal claims when we have low information (like a small
sample, a binary outcome, a multilevel design with few clusters, or a
weak instrument)? R. Fisher answered these questions in 1935 showing
that valid small sample hypothesis tests are possible, inference does
not require a population, and choices about assessing causal effects
can arise from design.
This paper reframes and extends Fisher's method, showing that it is a
practical alternative for political scientists. As an example, we show
how to assess treatment effects using a field experiment of the effect
of newspaper advertising on aggregate turnout with only eight
observations. In the end, we produce confidence intervals using linear
models, but requiring none of the standard assumptions of linear
models to guarantee valid statistical inferences. keywords:
randomization inference; analysis of experimental data; covariance
adjustment; small sample statistical inference
(with Costas Panagopoulos ).
"Probability of What?": A Randomization-based Method for Hypothesis Tests and Confidence Intervals about Treatment Effects This drafty working paper provides my perspective on randomization-based inference for randomized experiments. (with Costas Panagopoulos).
Attributing Effects to A Cluster Randomized Get-Out-The-Vote Campaign: An Application of Randomization Inference Using Full Matching
Statistical analysis requires a probability model: commonly, a model for the
dependence of outcomes Y on confounders X and a potentially causal variable
Z. When the goal of the analysis is to infer Z’s effects on Y, this requirement
introduces an element of circularity: in order to decide how Z affects Y, the
analyst first determines, speculatively, the manner of Y ’s dependence on Z and
other variables. This paper takes a statistical perspective that avoids such cir-
cles, permitting analysis of Z’s effects on Y even as the statistician remains
entirely agnostic about the conditional distribution of Y given X and Z, or
perhaps even denies that such a distribution exists. Our assumptions instead
pertain to the conditional distribution Z|X, and the role of speculation in set-
tling them is reduced by the existence of random assignment of Z in a field
experiment as well as by poststratification, testing for overt bias before accept-
ing a poststratification, and optimal full matching. Such beginnings pave the
way for “randomization inference”, an approach which, despite a long history in
the analysis of designed experiments, is relatively new to political science and
to other fields in which experimental data are rarely available.
The approach applies to both experiments and observational studies. We
illustrate this by applying it to analyze A. Gerber and D. Green’s New Haven
Vote 98 campaign. Conceived as both a get-out-the-vote campaign and a field
experiment in political participation, the study assigned households to treat-
ment and desired to estimate the effect of treatment on the individuals nested
within the households. We estimate the number of voters who would not have
voted had the campaign not prompted them to — that is, the total number of
votes attributable to the interventions of the campaigners — while taking into
account the non-independence of observations within households, non-random
compliance, and missing responses. Both our statistical inferences about these
attributable effects and the stratification and matching that precede them rely
on quite recent developments from statistics; our matching, in particular, has
novel features of potentially wide applicability. Our broad findings resemble
those of the original analysis by Gerber and Green (2000).
(with Ben Hansen ) prepared for presentation at the Political Methodology meetings, July 2005
Fixing Broken Experiments: How to Bolster the Case for Ignorability with Full Matching (with Ben Hansen).
Attributing Effects to a Get-Out-The-Vote Campaign Using Full Matching and Randomization Inference (with Ben Hansen ) prepared for presentation at the MPSA meetings, April 2005
Miscellaneous Fun Stuff
Cycling Involvements: Frequency Domain Time Series Analysis and Political Participation in the USA This paper shows that decomposing a time-series into periodic components can provide po- litically useful information about the shape of aggregate political participation in the United States. Specifically, it provides statistical tests for the periodicity of the aggregate time series of political participation and explains how this decomposition and associated tests work. Between 1973 and 1994 there appears to be an annual cycle in the reporting of political participation by respondents to a series of polls conducted by Gallup 10 times per year. This seasonality has been noted by in one other publication, by Rosenstone and Hansen (1993), but was explained as tied to a summer political cycle. In this article I suggest that this discovery has more to do with annual cycles in the composition of the Gallup sample than politics. I am currently trying to obtain detailed information on the monthly mail volume into and out of Congress. With this information, I will be able to test more directly if, despite the changes in sample composition of the Gallup polls, the political participation of Americans ought to be see as an "output" of Congressional mobilization or an "input" or in what way the flow of participation into Congress is related to the flow of mobilization out of it. Very drafty. If someone has flows of mail to and from Congress, I would love to collaborate. I think this approach would be very useful for relating such nearly continuous political "signals". This paper contains a basic description of some frequency domain time series analysis as applied to a political science topic.
Work Flow and The Work of Social Science
A Proposal for a Political Science Registry (A document drafted as a member of the Society for Political Methodology along with representatives of the Experimental Methods and Qualitative and Mixed Methods subsections of APSA.
Political Behavior
Can television encourage anti-violence norms in Northern Nigeria? This new collaboration with Annette Brown (of the International Initiative for Impact Evaluation) and Graham Couturier (Equal Access International) and Chris Grady (Univ of Illinois) uses a series of experiments to assess theories of social learning for norm change when stakes are high
A Framework for Studying the Dynamics of Political Participation. Political action is driven by events. Although the effects of particularly dramatic events on social movements is well documented, the effects of events, quotidian or exceptional, on the behavior of individuals are significantly less well understood. This paper proposes a framework for understanding how a moment of political action may occur in the life of an ordinary person. It synthesizes past literature and theories that explain variation among people at a single point in time on the basis of largely time-constant attributes of people and elaborates on this literature to suggest when we might expect the poor and disadvantaged to surmount such resource, skills, and status barriers to get involved in politics. Furthermore, this framework suggests a way for future syntheses, theory-building, and empirical studies to coordinate such that all of our disparate findings about political participation cumulate more effectively. An Appendix for this working paper.
Maps In Their Heads/Community Mapping (with Cara Wong , Daniel Rubenson , and Mark Fredrickson ) This project is currently funded by an Insight Development Grant from the SSHRC of Canada (Feb 2011 Competition).
Threat, Mobilization, and Participation: The Impact of Crossburnings on Political Behavior in North Carolina (with Mark Fredrickson )