Working Papers

• "What to do about Atheoretic Lags" [Paper] [R package]
  
  Abstract
  The problem of what I term atheoretic lags arises when the researcher's theory cannot specify the amount of time which passes between the observation of the explanatory variable and the realization of its effect on the outcome variable. When such a situation arises, researchers often use the "default" lag of one time period, arbitrarily specify some longer lag, or (worst) try several lags and present the results which best supports their theory. I present methods for the analysis and the presentation of results which encompass the set of reasonable lag lengths for a given explanatory variable based on repeated regression across lags. I examine a series of graphical techniques which are useful for presenting the results of such repeated regressions.



• "Offering the Olive Branch: Friendly and Hostile Strategies for the Demilitarization of Terrorist and Insurgent Groups." [Paper]
  
  Abstract
  This paper presents a formal model of terrorist violence under varying conditions of inter-terrorist competition. Specifically, I examine the levels of terrorist violence expected in competitive (multiple groups compete with each other) and non-competitive (only one terrorist group exists in the environment) environments. I examine these two competitive conditions under different informational conditions. The model yields predictions regarding expected levels of terrorist violence and has implications for counter-terrorist policy.



• "Multiple Hot Deck Imputation: A Non-Parametric Alternative to Multiple Imputation" [with Jeff Gill] [Paper] [Data] [R package]
  
  Abstract
  In this paper, we develop a technique for imputing missing observations which is particularly well suited for discrete data. The technique used is a variant of hot deck imputation which we call multiple hot deck. imputation. Because the imputed value is a draw from the conditional distribution of the variable with the missing observation, the discrete nature of the variable is maintained as its missing values are imputed. We introduce an affinity scoring method to locate the set of donors from which imputed values will be drawn. Several multiply imputed data sets are then imputed with the values from these donors, giving the technique the repetitive property which prevents artificially small standard errors in more traditional multiple imputation; hot decking in the past has been limited by the inability to account for imputation variance, we rectify this.



• "When It's Not All About Me: Altruism, Participation, and Political Context" [with Cindy Kam and James Fowler] [Paper] [Data] [Codebook]
  
  Abstract
  Altruism refers to a willingness to pay a personal cost to make others better off. Past research has established a link between altruism and political participation, primarily among college students. We show that dictator game behavior predicts support for humanitarian norms and donations to Hurricane Katrina victims, suggesting that dictator game allocations are valid measures of altruism. Moreover, we show that this measure of altruism predicts participation in politics, suggesting that past results with students can be generalized to a broader population. Finally, consistent with the argument that altruists only participate when they think doing so will make everyone better off, we show that there is no relationship between altruism and voter turnout in an election where the outcome is distributive and where it is not clear that either political outcome will produce a net societal gain.



• "Beyond Reducing Nonresponse Bias: Modeling Item Nonresponse in Survey Research" [with Lee Walker] [Paper] [Paper] [R package]
  
  Abstract
  This paper discusses the limitations of whitening out nonresponse bias with multiple imputation and discusses the extent to which item nonresponse can be modeled as a response in and of itself. Specifically, I propose three techniques for modeling nonresponse at two different levels. I introduce a technique for modeling nonresponse in a particular variable of interest with a regression model using multiply imputed values of the nonresponses as a key explanatory variable. I then introduce two techniques for modeling nonresponse in a series of variables or over the entirety of the dataset. Finally, I introduce a technique for weighting respondents by their rate of response. The methods described in this paper are implemented in the free software which accompanies this research.



• "Measuring Regime Preference in Latin America: Exposure to and Acceptance of the Democratic Ideal" [with Lee Walker] [Paper] [Data]
  
  Abstract
  Studies of support for democracy that evaluate regime support as an ordinal variable suffer from a democracy as ideal type bias. This means all other regime preferences are inferior preferences. We propose instead that regime preference is a nominal response variable that follows a multinomial distribution. We use a variant of the exposure-acceptance model proposed by Geddes and Zaller (1989). Using the 2003 Latinobarometro, we examine regime preference attitudes of individuals in 17 Latin American countries. We find that highly aware persons are more likely to prefer democracy, and moderately aware persons are more likely to prefer authoritarianism than are lowly aware people.



• "Terrorist Violence and Inter-Terrorist Competition" [Paper]
  
  Abstract
  This paper presents a formal model of terrorist violence under varying conditions of inter-terrorist competition. Specifically, I examine the levels of terrorist violence expected in competitive (multiple groups compete with each other) and non-competitive (only one terrorist group exists in the environment) environments. I examine these two competitive conditions under different informational conditions. The model yields predictions regarding expected levels of terrorist violence and has implications for counter-terrorist policy.