Thoughts:

1) I love seeing other people's first sketches. I sketch first too, and I find that the quality of any graphic can mostly be determined by how good the idea was when I first sketched it.

2) This reminded me that rather than using R to make my final figures, I really need run them through Illustrator. Nathan Yau's book Visualize This gives some awesome worked examples of how to clean up R graphics in Illustrator. (And for Harvard folks, the book is available online through Widener library!).

"Endogenous Selection"
Felix Elwert
Department of Sociology, University of Wisconsin-Madison
CGIS K354 (1737 Cambridge St.)
Wednesday, April 25th, 2012 12.00 pm

Abstract:

Selection bias is a central problem for causal inference in the social sciences. Quite how central a problem it is, however, is often obscured by ambiguous terminology, needlessly technical presentations, and narrow rules of thumb. This paper uses directed acyclic graphs (DAGs) to advance a precise yet intuitive global definition of endogenous selection bias and argue its theoretical and practical centrality for causal inference. The paper clarifies the fundamental structural difference between confounding and endogenous selection, shows that nearly all non-parametric identification problems relate to either confounding or endogenous selection, and argues that the problem of endogenous selection is indifferent to timing. Perhaps most importantly, we illustrate the importance of endogenous selection bias with numerous and varied examples from empirical social research.

This is joint work with Chris Winship.

"Violence and Voting: Did the 1960s Urban Riots Reshape American Politics?"
Omar Wasow
Government Department, Harvard University
CGIS K354 (1737 Cambridge St.)
Wednesday, April 18th, 2012 12.00 pm

Abstract:

Between 1964 and 1971, more than 750 riots flared up in black neighborhoods across the United States. Scholarship on how the American polity respond to these violent protests is contested. Some scholars argue that urban riots produced a conservative ``backlash'' among white voters, while other scholars find little or no effect. Using a measure that incorporates the location, timing and severity of urban riots between 1964 and 1971, I examine whether increased exposure to urban riots is associated with decreased support for the Democratic party. In the 1964, 1968 and 1972 presidential elections, I find a strong negative relationship between exposure to civil unrest and the county-level Democratic vote share. I find a similar negative relationship between exposure to riots and Democratic vote share in congressional elections between 1968 and 1972. Finally, I find that in counterfactual scenarios of fewer riots the Democratic presidential nominee, Hubert Humphrey, would have beaten the Republican nominee, Richard Nixon, in the 1968 election. As African Americans were strongly identified with the Democratic party in this time period, my results suggest that, in at least some contexts, political violence by a minority group may contribute to a backlash among segments of the mass electorate and encourage outcomes directly at odds with the preferences of the protestors.

"Using Post-Treatment Variables to Establish Upper Bounds on Causal Effects: Assessing Executive Selection Procedures in New Democracies"
Adam Glynn
Government Department, Harvard University
CGIS K354 (1737 Cambridge St.)
Wednesday, April 11th, 2012 12.00 pm

Abstract:

In this paper we propose an adjustment based on post-treatment variables for some standard estimators of the average treatment effect on the treated. Under relatively weak conditions, this adjusted estimator will provide an upper bound for the effect and in some cases lower bounds on p-values. Additionally, this approach does not place a restriction on the outcome variable and allows for multiple mechanisms by which the treatment has an effect on the outcome. We also demonstrate that this adjustment will reduce the estimated effect in a wide variety of circumstances, and therefore, when the assumptions for the adjusted estimator are preferable to the assumptions for the unadjusted estimator, the adjustment can be used as a robustness check. This method is illustrated with an assessment of the effects of using plurality rules for the first multi-party presidential elections in third wave of democracy in sub-Saharan Africa.

This is joint work with Nahomi Ichino.

"International Knowledge Diffusion and the Comparative Advantage of Nations"
Dany Bahar
Harvard Kennedy School
CGIS K354 (1737 Cambridge St.)
Wednesday, April 4th, 2012 12.00 pm

Abstract:

In this paper we document that the probability that a product is added to a country's export basket is, on average, 65% larger if a neighboring country is a successful exporter of that same product. We interpret our result as evidence of international intra-industry knowledge diffusion. Our results are consistent with the overall consensus in the literature on technology spillovers: diffusion is stronger at shorter distances; is weaker for more knowledge-intensive products; and has become faster over time.

This is joint work with Ricardo Hausmann and Cesar Hidalgo.

"A Multinomial Response Model for Varying Choice Sets, with Application to Partially Contested Multiparty Elections"
Teppei Yamamoto
Department of Political Science, MIT
CGIS K354 (1737 Cambridge St.)
Wednesday, March 28th, 2012 12.00 pm

Abstract:

This paper proposes a new multinomial choice model which explicitly takes into account variation in choice sets across observations. The proposed varying choice set logit model relaxes the independence of irrelevant alternatives assumption by allowing the individual random utility function to directly depend on choice set types, and can be applied to a variety of data in which some individuals can only choose from a subset of the theoretically possible responses. Both frequentist and Bayesian simulation-based estimation procedures are developed using the Monte Carlo expectation-maximization algorithm and Markov chain Monte Carlo, respectively. The proposed model can be used to analyze survey data in partially contested multiparty elections in which some political parties do not run their candidates in every district. For illustration, I apply the proposed method to the 1996 Japanese general election, where none of the districts was contested by all of the six major parties.

"Detecting Novel Bivariate Associations in Large Data Sets"
David Reshef
Harvard-MIT Division of Health Sciences and Technology
CGIS K354 (1737 Cambridge St.)
Wednesday, March 21st, 2012 12.00 pm

Abstract:

Identifying interesting relationships between pairs of variables in large data sets is increasingly important. One way of doing so is to search such data sets for pairs of variables that are closely associated. This can be done by calculating some measure of dependence for each pair, ranking the pairs by their scores, and examining the top-scoring pairs. We outline two heuristic properties--generality and equitability--that the statistic we use to measure dependence should have in order for such a strategy to be effective. We present a measure of dependence for two-variable relationships, the maximal information coefficient (MIC), that has these properties. MIC captures a wide range of associations both functional and not (generality), and assigns similar scores to relationships with similar noise levels, regardless of relationship type (equitability). Finally, we show that MIC belongs to a larger class of maximal information-based nonparametric exploration (MINE) statistics for identifying and classifying relationships.

Starting with Miguel, Satyanath, and Sergenti (2004), a large literature has used rainfall variation as an instrument to study the impacts of income shocks on civil war and conflict. These studies argue that in agriculturally-dependent regions, negative rain shocks lower income levels, which in turn incites violence. This identi…cation strategy relies on the assumption that rainfall shocks affect conflict only through their impacts on income. I evaluate this exclusion restriction by identifying districts that are downstream from dams in India. In downstream districts, income is much less sensitive to rainfall fluctuations. However, rain shocks remain equally strong predictors of riot incidence in these districts. These results suggest that rainfall affects rioting through a channel other than income and cast doubt on the conclusion that income shocks incite riots.

It's a short, readable paper -- worth checking out if you're into this kind of thing.

"Flaking Out: Snowfall, Disruptions of Instructional Time, and Student Achievement"
Joshua Goodman
Harvard Kennedy School
CGIS K354 (1737 Cambridge St.)
Wednesday, March 7th, 2012 12.00 pm

Abstract:

Recent research on charter schools, summer learning loss, and international achievement suggests that instructional time is a critical input to the education production function. Using student and school-grade fixed effects models with data from Massachusetts, I find no relation between school closures and achievement but a strong relation between student absences and achievement. I then confirm these results using temporal and spatial variation in snowfall to provide better identification. Extreme snowfall induces school closures but does not affect achievement. Moderate snowfall induces student absences and does reduce achievement. Instrumental variables estimates suggest that each absence induced by bad weather reduces math achievement by 0.05 standard deviations. These results are consistent with a model of instruction in which coordination of students is the central challenge. Teachers deal well with coordinated disruptions of instructional time like school closures, but deal poorly with absences that affect different students and different times. These estimates suggest that absences are responsible for up to 20% of the achievement gap between poor and nonpoor students. They also suggest that policies designed solely to increase instructional time may not be effective.

"Visual Computing in Biology"
Hanspeter Pfister
School of Engineering and Applied Sciences, Harvard University
CGIS K354 (1737 Cambridge St.)
Wednesday, February 29th, 2012 12.00 pm

Abstract:

Many areas in science are experiencing a flood of data arising in part from the development of instruments that acquire information on an unprecedented scale. This is particularly true in biology, where huge amounts of heterogeneous data are acquired from microarrays, scanners, microscopes, and various other instruments. Visual computing tools are essential to gain insights into this data by combining computational analysis with the power of the human perceptual and cognitive system and enabling data exploration through interactive visualizations. In this talk I will present some of my group's work in visual computing and give an overview of several successful visualization projects in the areas of genomics and systems biology. I then will focus on our work on visual computing in Connectomics, a new field in neuroscience that aims to apply biology and computer science to the grand challenge of determining the detailed neural circuitry of the brain.

"Bayesian Effect Estimation Accounting for Adjustment Uncertainty"
Francesca Dominici
Department of Biostatistics, Harvard School of Public Health
CGIS K354 (1737 Cambridge St.)
Wednesday, February 22nd, 2012 12.00 pm

Abstract:

Model-based estimation of the effect of an exposure on an outcome is generally sensitive to the choice of which confounding factors are included in the model. We propose a new approach, which we call Bayesian Adjustment for Confounding (BAC), to estimate the effect on the outcome associated with an exposure of interest while accounting for the uncertainty in the confounding adjustment. Our approach is based on specifying two models: 1) the outcome as a function of the exposure and the potential confounders (the outcome model); and 2) the exposure as a function of the potential confounders (the exposure model). We consider Bayesian variable selection on both models and link the two by introducing a dependence parameter ω denoting the prior odds of including a predictor in the outcome model, given that the same predictor is in the exposure model. In the absence of dependence (ω = 1), BAC reduces to traditional Bayesian Model Averaging (BMA). In simulation studies we show that BAC with ω > 1 estimates the exposure effect with smaller bias than traditional BMA, and improved coverage. We then compare BAC, a recent approach of Crainiceanu et al. (2008), and traditional BMA in a time series data set of hospital admissions, air pollution levels and weather variables in Nassau, NY for the period 1999-2005. Using each approach, we estimate the short-term effects of PM2.5 on emergency admissions for cardiovascular diseases, accounting for confounding. This application illustrates the potentially significant pitfalls of misusing variable selection methods in the context of adjustment uncertainty.

"Sparse Joint Estimation of Covariates-Dependent Covariance Matrices"
Tamar Sofer
Department of Biostatistics, Harvard University
CGIS K354 (1737 Cambridge St.)
Wednesday, February 15th, 2012 12.00 pm

Abstract:

We propose an estimation method for the principal components/covariance structures of a set of outcomes, while modeling the effect of covariates. We assume a linear mixed model formulation on the outcomes as response to covariates, a model corresponding to spiked covariance matrices. Since the subject-specific covariance matrices and the effects of covariates are believed to be sparse, we penalize coefficients using an oracle penalty function. Under some assumptions on the parameters and the likelihood, we show that the maximum likelihood estimator of the parameters is asymptotically consistent and is uniformly sparse ("sparsistent"), even when the number of parameters is small. We propose using the Bayesian Information Criterion (BIC) for tuning parameter selection and show that it is consistent for model selection. Using a simple iterated least squares procedure we are able to recover the model parameters with high accuracy. The method is implemented to study the effect of smoking on the covariances of gene methylations in the asthma pathway in smokers and non-smokers US veterans from the Normative Aging Study (NAS).

"Using Regression Discontinuity to Uncover the Personal Incumbency Advantage"
Rocio Titiunik
Department of Political Science, University of Michigan
CGIS K354 (1737 Cambridge St.)
Wednesday, February 8th, 2012 12.00 pm

Abstract:

We study the conditions under which estimating the incumbency advantage using a regression discontinuity (RD) design recovers the personal incumbency advantage in a two-party system. Lee (2008) has introduced RD as a method for estimating what is generally considered the "partisan" incumbency advantage. We present a causal model with some simple but plausible assumptions that allows RD to be used to estimate the "personal" incumbency advantage, as an alternative to sophomore surge, retirement slump, and other commonly used measures. We estimate the incumbency advantage using our model with data from U.S. House elections between 1952 and 2008. Using the assumptions of our model, we also explore the estimation of the incumbency advantage beyond the limited RD conditions where knife-edge electoral shifts create the leverage for causal inference.

"Moving Beyond the Mean: The Role of Variation in Determining Phenotype"
John Quackenbush
Dana-Farber Cancer Institute and Harvard School of Public Health
CGIS K354 (1737 Cambridge St.)
Wednesday, February 1st, 2012 12.00 pm

Two trends are driving innovation and discovery in biological sciences: technologies that allow holistic surveys of genes, proteins, and metabolites and a realization that biological processes are driven by complex networks of interacting biological molecules. However, there is a gap between the gene lists emerging from genome sequencing projects and the network diagrams that are essential if we are to understand the link between genotype and phenotype. 'Omic technologies were once heralded as providing a window into those networks, but so far their success has been limited, in large part because the high-dimensional they produce cannot be fully constrained by the limited number of measurements and in part because the data themselves represent only a small part of the complete story. To circumvent these limitations, we have developed methods that combine 'omic data with other sources of information in an effort to leverage, more completely, the compendium of information that we have been able to amass. Here we will present a number of approaches we have developed, with an emphasis on the how those methods have provided into the role that particular cellular pathways play in driving differentiation, and the role that variation in gene expression patterns influences the development of disease states. In particular, we will challenge the basic analytical that have been used in biomedical research and argue that one should move beyond a simple comparison of the means relative to variance (the t-test) but instead also consider how variance itself changes between phenotypes. Looking forward, we will examine more abstract state-space models that may have potential to lead us to a more general predictive, theoretical biology.

"The Consumer Assessments of Healthcare Providers and Systems (CAHPS) Survey for Medicare: A Review and New Findings from a Mode Experiment"
Alan Zaslavsky
Department of Health Care Policy, Harvard Medical School
CGIS K354 (1737 Cambridge St.)
Wednesday, January 25th, 2012 12.00 pm

Abstract:

We assess health care quality and access in the Medicare system to inform consumer choice, foster quality improvement, monitor health plan quality, and reward high-performing plans. The CAHPS survey has since 1997 been one of the main tools for this assessment. In this talk, I will review some of the more interesting analyses of system quality made possible by using the CAHPS survey and some of the challenging issues in system monitoring overcoming years. I will then describe our analyses of an experiment on the effects of survey mode on CAHPS responses, using a principal stratification framework.