September 2011
Sun Mon Tue Wed Thu Fri Sat
        1 2 3
4 5 6 7 8 9 10
11 12 13 14 15 16 17
18 19 20 21 22 23 24
25 26 27 28 29 30  

Authors' Committee

Chair:

Matt Blackwell (Gov)

Members:

Martin Andersen (HealthPol)
Kevin Bartz (Stats)
Deirdre Bloome (Social Policy)
John Graves (HealthPol)
Rich Nielsen (Gov)
Maya Sen (Gov)
Gary King (Gov)

Weekly Research Workshop Sponsors

Alberto Abadie, Lee Fleming, Adam Glynn, Guido Imbens, Gary King, Arthur Spirling, Jamie Robins, Don Rubin, Chris Winship

Weekly Workshop Schedule

Recent Comments

Recent Entries

Categories

Blogroll

SMR Blog
Brad DeLong
Cognitive Daily
Complexity & Social Networks
Developing Intelligence
EconLog
The Education Wonks
Empirical Legal Studies
Free Exchange
Freakonomics
Health Care Economist
Junk Charts
Language Log
Law & Econ Prof Blog
Machine Learning (Theory)
Marginal Revolution
Mixing Memory
Mystery Pollster
New Economist
Political Arithmetik
Political Science Methods
Pure Pedantry
Science & Law Blog
Simon Jackman
Social Science++
Statistical modeling, causal inference, and social science

Archives

Notification

Powered by
Movable Type 4.24-en


« July 2011 | Main | October 2011 »

29 September 2011

Tweeting how you feel

Benedict Carey of the New York Times discusses a paper (gated) by Scott Golder and Michael Macy showing that people's moods -- as expressed within the character limit of twitter -- have remarkably predictable patterns. The authors' interpretation is that our moods are fundamentally linked to our circadian rhythms.

First, I really like this paper and I'm glad to see it come out in Science. An earlier version was presented at one of the conferences put on by Arthur Spirling and the Harvard Program on Text Research, and it caught my eye then.

On one hand, it's obviously innovative research that is making great use of the reams data that are now sitting in the interwebs somewhere, waiting to be analyzed. The possibilities in this new, data rich realm are seemingly endless: the culturomics/ngrams project, work on political blogs (Abe Gong), congressional tweeting (Drew Conway), the news cycle (Leskovec, Backstrom, and Kleinberg), and so on.

But we also should spend more time stepping back and asking hard questions about the data. Are tweets really a great measure of sentiment if the decision to tweet isn't random? Who is online and how do they differ from the offline folks? There is basically no discussion of this in the Golder and Macy article. Perhaps the lack of attention to the limitations of "big data" research is just an inevitable part of the fad cycle, but that doesn't mean we should let our standards slide just because someone has cool data.

Posted by Richard Nielsen at 8:35 PM

27 September 2011

Cross Validated

Cross Validated is a question-and-answer site dedicated to statistics and statistical computing. It is part of the completely awesome Stack Exchange network of Q/A sites which I rely on heavily when coding. The questions range from the fairly straightforward (“How can a regression be significant yet all predictors be non-significant?”) to the fairly complicated (“What’s the difference between principal components analysis and multidimensional scaling?”) to the fairly abstract (“How are we defining ‘reproducible research’?”).

There is a ton of interesting instruction being done there. I have heard others scoff at the idea of giving away your expertise for free on sites like these, but I think that Cross Validated and other sites are crucial and vibrant places for students to learn about statistical methods. And this is a chance to actually help people in a concrete way.

Also, check out their community blog, which promises to have great little tidbits, mostly focused on R.

Posted by Matt Blackwell at 9:17 AM

25 September 2011

App Stats: Spirling on "Partisan Convergence in Executive-Legislative Interactions: Modeling Debates in the House of Commons, 1832-1915"

We hope you can join us this Wednesday, September 28, 2011 for the Applied Statistics Workshop. Arthur Spirling, Assistant Professor at the Department of Government at Harvard University, will present a paper entitled "Partisan Convergence in Executive-Legislative Interactions: Modeling Debates in the House of Commons, 1832-1915". A light lunch will be served at 12 pm and the talk will begin at 12.15.

"Partisan Convergence in Executive-Legislative Interactions: Modeling Debates in the House of Commons, 1832-1915"
Arthur Spirling
Government Department, Harvard University
CGIS K354 (1737 Cambridge St.)
Wednesday, September 28th, 2011 12.00 pm

Abstract:

We consider the interaction between members of the executive and backbenchers in the House of Commons between the Great Reform Act and the Great War, a period of radical internal reform that birthed the Westminster system in its current form. We gather new data of over a million speeches in seventeen thousand debates to model the way in which the cabinet-legislative relationship changed over time. In particular, we conceptualize debates as Markov chains moving between speaker states and focus on estimating transition probabilities of the same. We take a Bayesian mixed model approach, allowing for debate-level and ministry-level variation. We show a remarkable "convergence" in the behavior of ministers from different parties, beginning between the mid-1870s and late-1880s and coinciding with a series of important standing orders relating to the ability to ask questions in the Commons. While Tory ministers generally become more responsive, Liberal ministers are less involved in debate.

Posted by Konstantin Kashin at 9:44 PM

19 September 2011

App Stats: Sen on "Natural Experiments, Judicial Quality, and Racial Bias in Federal Appellate Review"

We hope you can join us this Wednesday, September 21, 2011 for the Applied Statistics Workshop. Maya Sen, a Ph.D. candidate from the Department of Government at Harvard University, will give a practice job talk entitled "Natural Experiments, Judicial Quality, and Racial Bias in Federal Appellate Review". A light lunch will be served at 12 pm and the talk will begin at 12.15.

"Natural Experiments, Judicial Quality, and Racial Bias in Federal Appellate Review"
Maya Sen
Government Department, Harvard University
CGIS K354 (1737 Cambridge St.)
Wednesday, September 21st, 2011 12.00 pm

Abstract:

In this paper, I find that cases decided by black federal lower-court judges are consistently overturned more often than cases authored by similar white judges. I estimate this effect by leveraging the fact that incoming cases to the U.S. courts are randomly assigned to judges, which ensures that black and white judges hear similar sorts of cases. The effect is robust and persists after matching exactly on measures for judicial quality (including quality ratings assigned by the American Bar Association (ABA)), previous professional and judicial experience, and partisanship. Moreover, by looking more closely at the ABA ratings scores awarded to judicial nominees, I demonstrate that this effect is unlikely to be attributable exclusively to differences between black and white judges in terms of quality. This study is the first to explore how higher-court judges evaluate opinions written by judges of color and it has clear normative implications: attempts to make the judiciary more reflective of the general population may have actually resulted in inequality in the aggregate, both for litigants and for judicial actors.

Posted by Konstantin Kashin at 2:20 AM

12 September 2011

App Stats: Blackwell on "A Dynamic Causal Inference Approach for Estimating the Effectiveness of Negative Campaigning"

We hope you can join us this Wednesday, September 14, 2011 for the Applied Statistics Workshop. Matt Blackwell, a Ph.D. candidate from the Department of Government at Harvard University, will give a practice job talk entitled "A Dynamic Causal Inference Approach for Estimating the Effectiveness of Negative Campaigning". A light lunch will be served at 12 pm and the talk will begin at 12.15.

"A Dynamic Causal Inference Approach for Estimating the Effectiveness of Negative Campaigning"
Matt Blackwell
Government Department, Harvard University
CGIS K354 (1737 Cambridge St.)
Wednesday, September 14th, 2011 12.00 pm

Abstract:

Traditional single-shot causal inference models investigate the effect of a single action at a single point in time and are an invaluable tool for political scientists. Often, however, actions unfold over time, with political entities reacting to a shifting environment. Accordingly, single-shot methods leave researchers unable to extract meaningful causal inferences about these dynamic processes. This stems from a fundamental tension: in dynamic settings, regression and matching force a choice between omitted variable bias on the one hand, and post-treatment bias on the other and are unable to simultaneously correct for both. To avoid these problems, I introduce a framework for dynamic causal inference and utilize marginal structural models to estimate dynamic causal effects. The effectiveness of "going negative" serves as a motivating example---an apt illustration since candidates change their strategy as the campaign unfolds. Furthermore, I introduce novel diagnostics and a sensitivity analysis for the model.

Posted by Konstantin Kashin at 2:42 AM

5 September 2011

App Stats: Robins on "Parametrizations, Likelihoods, Semiparametrics, Causal Graphs, Model Selection and Discovery for Complex Causal Models"

We hope you can join us this Wednesday, September 7, 2011 for the first Applied Statistics Workshop this semester. Jamie Robins, the Mitchell L. and Robin LaFoley Dong Professor of Epidemiology at the Harvard School of Public Health, will present his paper entitled "Parametrizations, Likelihoods, Semiparametrics, Causal Graphs, Model Selection and Discovery for Complex Causal Models". A light lunch will be served at 12 pm and the talk will begin at 12.15.

"Parametrizations, Likelihoods, Semiparametrics, Causal Graphs, Model Selection and Discovery for Complex Causal Models"
Jamie Robins
Harvard School of Public Health
CGIS K354 (1737 Cambridge St.)
Wednesday, September 7th, 2011 12.00 pm

Abstract:

I will discuss recent results on novel factorizations of the likelihood for both (i) semiparametric causal models (marginal structural models and structural nested models) and (ii) nonparametric causal graphical models with unmeasured confounders (hidden variables). I will show how the causal question of substantive interest dictates the choice of factorization (eg a causally complete MSM factorization is appropriate for inference on direct effects ) and a R(recursive)-factorization is appropriate for the construction of algorithms to perform model selection and causal discovery in the setting of nonparametric causal graphical models. Associated with each factorization is a parametrization . The parametrization dictates both the form of doubly robust estimators and the likelihood to be maximized in scoring algorithms such as BIC used for model selection. I will derive the relationship (ie mappings or diffeomorphisms ) between these alternate parameterizations. This is joint work with Thomas Richardson, Ilya Shpitser, Robin Evans, Eric Tchetgen and Andrea Rotnitzky.

Posted by Konstantin Kashin at 12:47 AM