In a review of the literature on policy diffusion in the states, Berry and Berry (1999) conclude that the "gold standard" methodological approach to the empirical modeling of these processes is discrete event history analysis. I concur, but find that the literature largely ignores several important issues in the proper specification of these models, including choice of functional form, modeling different mechanisms of diffusion, modeling duration dependence, and spatial autocorrelation in the cross-sections. I use data from several well-known studies of diffusion to suggest possible improvements to this important research.

In the context of congressional elections, Jacobson and Kernell (1983) were among the first to recognize that strategic challengers take advantage of politically advantageous situations (e.g., an open seat, favorable national conditions) when deciding whether or not to run for elected office. More specifically, students of congressional elections agree that strategic politicians are more likely to emerge when their chances of getting elected are perceived to be high. As some scholars have noted, however, experienced candidates occasionally challenge this convention by running against incumbents. To what degree are experienced candidates' decisions to enter a congressional race based on other strategic calculations? Are experienced challengers attuned to the position taking behavior of incumbent legislators on highly salient or prominent roll call votes? Moreover, to what extent do incumbents decide to opt out of a race as a result of the emergence of an experienced challenger? The answers to these questions have important implications for issues of representation and accountability. Indeed, if representative democracy is working, then we should observe experienced challengers emerge in response to incumbent voting behavior and its reflection of constituency interests.

Prior research suggests that experienced candidates behave strategically in response to anticipated behavior on the part of incumbent legislators. At the same time, the choice by a candidate to challenge an incumbent could impact the likelihood of a legislator to seek reelection (Gates and Humes 1997). Thus, it is important to model this behavior as a sequential process. Given this consideration, it is reasonable to assume that some of the same factors influencing whether an experienced candidate chooses to emerge in a congressional race might also indirectly influence the decision of an incumbent to run again or voluntarily retire. That being the case, it is necessary to consider how factors influencing these processes may be interrelated. If we examine these processes separately, we run the risk of model misspecification resulting from a possible selection effect. Analyzing this relationship jointly with a selection model provides an opportunity to statistically control for sample selection bias, thus avoiding potential threats to statistical inference.

The issue of candidate competition is further complicated by the fact that there is more than one actor involved in this sequential process and each is operating within a strategic context. When two or more actors are engaged in strategic interaction under uncertainty, it is necessary to model such strategic behavior appropriately. If we ignore the strategic dimension to the interaction between the challenger and the incumbent, then we increase the probability of introducing bias into our results. Indeed, Signorino (2002) has demonstrated that parameter estimates from a standard selection model will be biased if the dependent variable we are trying to explain is generated by strategic interaction among the players. To fully capture the strategic nature of the decision calculus involved with challenging an incumbent, this analysis tests hypotheses derived from a game theoretic model and a test of the model using STRAT, a program for analyzing statistical strategic models developed by Curtis Signorino (2001). From a methodological perspective, using an estimation program such as STRAT controls for sample selection bias in the estimation of the model while also incorporating strategic behavior into the actors' utilities.

In examining the issue of candidate competition in congressional elections, I develop a model of strategic interaction between congressional challengers and incumbents and test hypotheses from this model on House and Senate elections from 1976 to 2000 using a strategic probit technique. I control for factors such as incumbents' previous vote margin, partisanship of the district/state, incumbent expenditures, competitiveness of the seat, and institutional factors. In terms of the findings, I show that strategic politicians do emerge in response to the positions incumbents express on key votes, especially when they vote in support of a majority of their party. Moreover, I find evidence that incumbent expenditures discourage the emergence of experienced challengers in House races, but not those in the Senate. Additionally, I show that incumbents make strategic career decisions based on the behavior of experienced challengers in House races, but not those in the Senate. Additionally, I show that incumbents make strategic career decisions based on the behavior of experienced challengers, particularly when they decide to enter a race.

The end of the Cold War brought many changes to the practice of international conflict. Whereas previously the American-Soviet rivalry structured not only international conflicts but also our understanding of them, today the absence of that rivalry brings new issues to the front of international relations. Unfortunately, the study of these new issues has, as is generally the case, lagged behind the changes. Notably, the study of multilateral peacekeeping, largely United Nations peacekeeping, has remained underdeveloped. Journalistic and prescriptive accounts predominate, and where analytical research is done, it has been incomplete and compartmentalized. In this paper, I attempt to improve this situation by simultaneously examining two important components of post-Cold War international relations: United Nations intervention decisions and the subsequent efficacy of those peacekeeping operations.

Researchers know that actions in international relations are often strategic; actors modify their actions based on the expected behavior of their opponents. Consequently, it stands to reason that the UN too is a strategic actor modifying its behavior based upon its expectations of other actors' behavior, in this case the expected behavior of the local factions. Failure to account for these expectations in a researcher's statistical model--misspecifying the strategic functional form of the model--can lead to invalid inferences (Signorino and Yilmaz 2000). However, by basing the statistical model on the theoretical model, here examining the intervention and efficacy questions together in a strategic framework, I can correctly specify the strategic functional form of the intervention process and avoid those misspecification errors.

To examine the United Nations intervention process, I construct and test a strategic statistical model of the UN peacekeeping decision process using a data set of 125 cases of post-WWII intrastate conflict. Additionally, I compare these findings from the strategic statistical model to those from a traditional logit model that fails to account for the proper functional form of the intervention decision process. In this comparison, I illustrate how the properly specified model outperforms the standard model in terms of both how well it explains the observed interventions and by how the strategic model better captures the actual, sometimes nonmonontonic, relationships between the factors affecting the interventions decisions and the intervention decisions themselves.

A key premise in partisan theories of congressional organization of the US House of Representatives is that, to some extent, incumbents of the same party have common electoral fates. Put differently, in addition to candidates' personal votes, there are "electoral tides" that affect all copartisans' vote totals in a given election. Because of this, according to partisan theories, members of the House delegate authority to the party leadership in order to improve their party's electoral fate.

Given the centrality of this premise to contemporary debates about the importance of parties, the empirical work on the extent to which party members share common fates is surprisingly circumscribed. Much of the work that has been done focuses primarily on methodological questions about how to measure copartisans' electoral tides, rather than on substantive questions about over-time changes in these tides. In particular, we know little about the extent to which party members from different regions shared common electoral fates across the 19th century, when we see national parties change from loosely-knit confederations of local party organizations in the 1820's, to powerful and unified legislative forces in the 1890's.

In this paper, I use electoral returns from congressional elections to measure electoral tides across the 19th century, in order to test the hypothesis that the emergence of strong party government in the late 19th century was in part a function of increased internal cohesiveness among members of the two main parties. In particular, I hypothesize that two factors--the elimination of the slavery issue, and 1870's changes in the electoral laws governing congressional elections--led to greater cohesiveness, hence stronger party government.

To do this, I use a technique that combines elements of methods used to measure electoral tides with elements of Cox and Rosenbluth’s (1995) method for estimating the impact of electoral tides on party incumbents ’ probabilities of reelection. For each year and each major party across the time series, I use a probit model to estimate party members ’ probabilities of winning as a function of the 'typical' vote swing to members of the same party in different regions. In this way, I generate a time series showing that party members ’ fates became more closely related to the fates of party members from different regions as the 19th century progressed. I link this increase to institutional and demographic changes that occurred in the wake of the Civil War.

Many previous econometric examinations of the behavior of U.S. tariff rates have ignored crucial time series issues and have employed simple dynamic specifications. Within a framework of testing an access point theory of the institutional determinants of trade policy, this poster will explore the time series aspects of American tariff rate data. The data used in this study are all taken from Historical Statistics of the United States and the Statistical Abstracts of the United States. Tariff rates are operationalized as the trade weighted average tariff, i.e. the value of import duties divided by the total value of imports. The data set consists of annual observations extending from 1890 to 1996, although certain models look at subsets of the data because of limited availability of some of the independent variables.

The access point theory tested here argues that trade policy is a function of the amount of institutional access provided to interest groups. Because of the advantage that protectionists have in counteracting the collective action problem due to their concentrated nature, the more access points there are the more protectionist the final policy will be. Within the American context, this theory hypothesizes that delegation of trade policy-making to the President and unified government should both lead to lower tariff rates by reducing the access points for interest groups. Economic control variables suggested by Magee, Brock, and Young (1989) and Hiscox (1999) are also included in the models. These variables include unemployment, inflation, the terms of trade, the capital-labor ratio, exports as a share of GDP, and imports as a share of GDP.

The time series aspects of the data are explored with a variety of methods. Previous work has explored the issue of unit roots in tariff rates (Gardner and Kimbrough, 1989 Sadorsky, 1994 Lohmann and O'Halloran 1994), but little additional attention has been paid to the dynamic nature of tariff rate data. The Dickey-Fuller analyses conducted here suggest that the presence of a unit root is ambiguous: there probably are no unit roots in the full sample, although this cannot be ruled out entirely, while subsamples may contain unit roots. Many of the economic control variables are non- stationary, though. Further, if tariff rates are non-stationary, then it is found that they are likely cointegrated with the non-stationary economic variables. As such, error correction models (ECM) will be run as well as standard ARMA models. Preliminary analysis also suggests that there may be autoregressive conditional heteroskedasticity in the models. Thus, the analysis will also be conducted using ARCH and GARCH models. All of these different dynamic specifications will be compared to determine which ones best explain the pattern of tariffs in the United States.

This poster serves three purposes. First, from a purely methodological standpoint, it applies ECM and GARCH models to a substantive topic where they have not been used before. Hopefully, this will help demonstrate the usefulness of these techniques. Second, because of the nature of the tariff rate time series, these techniques should enable us to better explain the dynamics of trade policy in the United States, a subject of long-standing debate in both political science and economics. Finally, these techniques are used to test a specific theory of trade policy that seeks to establish how democratic institutions affect trade policy and, in the case of the United States, address the effects of the Reciprocal Trade Agreements Act (RTAA) and divided government on American protectionism, both subjects of debate among political scientists. (See Hiscox (1999) for a discussion of the former and Lohmann and O'Halloran (1994) for a discussion of the latter.) As such, this poster will contribute to our understanding of time series techniques, trade policy in general, and the political economy of American tariff rates.

This paper starts from the assumption that bureaucratic output can be examined in a similar manner to any other system or process, and that equilibrium can and indeed is reached. The process control literature identifies over-controlling processes as one of the biggest sources of error that arises. This is mainly due to an improper understanding of system equilibrium and the normal variation of the system. The same can be true of attempts to model bureaucratic output. The bureaucratic output literature typically uses time series analysis, often ARIMA, to model the output of a bureaucracy. However, this may be an 'over-modeling' of the system since the events that scholars use as impacts, such as budget changes, appointments, etc, are not unusual shocks, but rather they are a standard part of the environment. Given that they are a standard part of the environment, the bureaucratic process finds equilibrium within the context of a constantly changing environment. This results in rather large, but not unexpected, variation around an equilibrium point. Over-modeling of bureaucratic output results in an overstatement of the political control of the bureaucracy.

Wood and Waterman (1991) and Ringquist (1994) attempt to use time series modeling and ARIMA to show political control of the bureaucracy. However, in order to use time-series, some assumptions must be made that certain events cause shocks to the system. These events are then modeled iteratively. This pre-identification of independent variable events inevitably causes researchers to make premature assumptions about the environment and context. These assumptions lead to the researcher giving the model more structure than perhaps is necessary. This leads to the over-modeling discussed earlier. Rather than following the models described above, this paper attempts to establish a baseline equilibrium conditions for bureaucratic output of the EPA’s enforcement actions and policy outcomes. We will define equilibrium output for a bureaucracy by using moving averages to determine when the system shows a naturally fluctuating process. Therefore, we make no a priori assumptions about independent variables which may alter this assumption. Consequently, we provide no structure for the output other than that already available. In this case, the data truly leads the analysis. The first data set this paper uses is from shortly after the establishment of the EPA to define baseline equilibrium for EPA enforcements. Using this data, a ‘control chart' will be developed to define the equilibrium point and variance. Using the control chart methodology to define upper and lower bounds as a function of the moving ranges, we will then be able to chart data from the 1980s forward to reveal any disruptions in equilibrium. Further, we are able to determine natural versus unnatural fluctuations in output. The control chart methodology allows a researcher to define the precise time frame in which a disruption to the system has caused a change in the equilibrium of the system.

After identifying the time frames of equilibrium disruptions, disruption hypotheses can be generated and tested. This paper contributes to the literature on bureaucratic output by assuming that many of the variations that may affect bureaucratic output are within normal variation, so that only the events with the most impact are identified. The most important contribution of this paper is its ability to narrowly identify the time frame of disruptions to the system. It does not use an iterative process to see the impact of previously identified disruptions. Rather, it identifies bureaucratic outputs which show natural or unnatural fluctuations, and then shows that naturally fluctuating systems may have high normal variation. This allows the researcher to identify disruptions to the system without attempting to identify such political or bureaucratic shock events first. The control chart methodology provides a stringent test to identify politically-controlled shocks to the system.

Groseclose (1994) tests five alternative theories of Congressional committee organization by using a Monte Carlo technique that avoids problematic assumptions about the ideological distribution of Members of Congress. He generates 20,000 hypothetical committees for each of the ten committees from the 99th Congress, using a number of interest group scores for members of the House.

On some tests (preference outlier hypothesis, for example), he allows for simulated committees that do not conform to the known and institutionalized majority party-minority party split on committees. This means that some of the simulated committees are never possible in reality (i.e., committees can never be controlled by the minority party). We re-run the analyses with a condition on party composition of committees to test for alternative results. We argue that simulations must account for real world conditions in testing alternative hypotheses.

We also re-run the analysis for later Congresses, when party leadership arguably became more important. That is, we allow for theories of Congressional organization to vary over time, and see if such theories are sensitive to asserted periods of ``conditional party government.'' Thus, this project has both methodological and theoretical importance in that we update Groseclose's Monte Carlo simulations and explore theories of Congressional organization over time.

Both case study and aggregate statistical research show that political institutions have an impact upon the survival of dictators. Authoritarians use legislatures and political parties to coopt potential opposition and broaden the base of the dictatorship. Legislatures provide some policy-making realms, even if limited, to the opposition while parties funnel benefits and privileges. In exchange, the dictator receives acquiescence, if not outright support, since participation in these dictatorial institutions vests the opposition with a stake in the regime.

But if dictators make concessions to groups in civil society in terms of institutions, then they must do so in terms of policy as well. Therefore, there are reasons to believe that institutionalized dictatorships produce policies that are different from those in non-institutionalized ones. A legislature and political party, even if constrained by the discretion of the dictator, can affect the content of policies and the flow of benefits.

I test statistically the effect of political institutions upon policy within dictatorships to determine whether, in fact, there is any relationship and which institutions have an impact. As a proxy for policy, I use time series cross-national data on government expenditures and revenues provided by the World Bank. I am especially interested in total expenditures and revenues as well as educational and military expenditures. I have collected data on a variety of institutions such as legislatures, political parties, elections, and the military or civilian nature of the regime for all dictatorships from 1946 to 2000. I will test the impact of these observable factors, rather than of subjective indices (e.g., rule of law, property rights), upon policy under dictatorship.

Due to the nature of the data, there are a number of methodological issues. Because I am using time series cross-sectional data for all dictatorships, I must take into account country heterogeneity. The standard method, fixed effects models, may not allow for any assessment of the impact of the institutional regressors, however, because typically, institutions change infrequently. Furthermore, expenditure and revenue data are notorious for their serial correlation, requiring either correction for the errors or modeling of the dynamic process generating these data. I explore various statistical techniques to deal with these problems.

Empirical investigation of the variation in the electoral success of extreme right parties in Western Europe has produced inconsistent results. For example, some studies find that immigration matters (Martin, 1996 Anderson, 1996 Knigge, 1998), others that it does not (Mayer and Perrineau, 1989), still others that it only matters in some countries (Givens, 2000). The same inconsistent results can be found for unemployment and electoral institutions. This paper suggests that many of these studies suffer from methodological problems relating to selection bias and the incorrect interpretation of interaction effects. I illustrate these problems by replicating Jackman and Volpert's 1996 British Journal of Political Science article entitled ¡®Conditions Favouring Parties of the Extreme Right in Western Europe¡¯. I show in some detail that the authors’ conclusions rely on an incorrect interpretation of their model’s causal effects.

Given these methodological problems and the inconsistent results that have been generated, I reexamine the effect of electoral institutions, immigration and unemployment on extreme right parties. The focus on how unemployment and immigration influence the extreme right vote relies on arguments about what drives voter preferences, while the focus on electoral institutions emphasizes the way in which institutions constrain voters¡¯ choices given their preferences. I use a time-series-cross-sectional Tobit model with interaction and fixed effects to test three common hypotheses in the literature on extreme right parties. The model draws on a new dataset spanning all 165 national elections in nineteen west European countries between 1970 and 2000.

Although this paper does not employ any new statistical techniques, I would argue that it does make a methodological contribution. This is because it illustrates some of the more common errors associated with the interpretation of interaction effects and suggests how they might be better overcome. The paper also makes several substantive conclusions that contribute to the existing body of literature on extreme right parties. The first is that it is important to distinguish between neofascist and populist parties on the extreme right since the electoral fortunes of these parties clearly depend on different factors. Although the theoretical literature has increasingly made this distinction, no previous statistical analysis has taken this into account. The second is that populist parties do better in countries where the district magnitude is larger and more seats are allocated in upper tiers. This suggests that populist party elites and voters are instrumentally motivated! . The third is that although immigration has a positive effect on populist parties irrespective of the unemployment level, unemployment only matters when immigration is high. Finally, there is evidence that the permissiveness of the electoral system mediates the effect of immigration on populist parties.

Understanding why some individuals vote and others do not is one of the classic questions in political science. Most of this research has relied on static analyses of turnout, and typically concludes that the decision to vote is determined by a variety of relatively immobile social, demographic, and political characteristics. Past analyses, however, have often lacked the appropriate data and methods to adequately explore the dynamics of turnout decision-making. Using the Knowledge Networks panel dataset of Election 2000 --- which includes 29,000 respondents and 133,000 observations on turnout and vote preference --- I find that 30% of the electorate changed their mind at some point during the campaign as to how likely they are to vote. In this paper, I analyze the dynamics of turnout intention, specifically investigating how changes in the decision to vote in Election 2000 are related to presidential campaign activities and the individual’s vote choice decision.

The decision to go to the ballot box and the decision about whom to vote for once there are typically studied separately in political science research, yet they are certainly related processes. I argue that you cannot make inferences about the effect of the campaign (or other covariates) on turnout without taking into account an individual’s pre-campaign intentions, regarding both their turnout and vote choice decisions. For instance, I hypothesize that the campaign will be more likely to mobilize the individual who intends to vote but is undecided as to her vote choice than it is to mobilize the individual who does not plan to vote and does not have a preferred candidate. In making the turnout decision, there are four states (from a 2x2 of turnout intention and vote choice decision) from which an individual can transition:

1) Individual intends to vote and has decided on vote choice
2) Individual intends to vote but is undecided on vote choice
3) Individual does not intend to vote but has a preferred candidate
4) Individual does not intention to vote and has no preferred candidate

I estimate a transition logit model of turnout, an approach that incorporates dynamics into the model by conditioning the covariates on an individual’s previous state. This logit transition model not only better predicts turnout than a static logit model, but it offers several substantive contributions. For one, this analysis will identify the individuals for whom the 2000 presidential campaign had a mobilizing effect. Recent research has demonstrated that campaign events influence voter behavior in the aggregate, but this analysis illustrates how campaigns influence changes in the decision to vote at the individual level. Because I treat an individual’s turnout and vote choice decisions as interrelated processes, this paper will also offer insights regarding the relationship between the opinions and the behavior of the American electorate. For instance, are undecided voters more likely to vote than decided non-voters? What factors explain the turnout decision for these different groups? And is there anything that can induce the undecided non-voters to show up on election day?

Considerable theoretical and empirical investigations have refined the liberal peace arguments (Maoz & Russett 1993 Oneal & Russet 1997, 1999, 2001 Barbieri & Schneider 1999 Beck & Tucker 1997 Beck, Katz, & Tucker 1998 Hegre 2000 Polachek, Robst, & Chang 1999). Testing the pacifying effect of trade on conflict, these researches center on the unit of analysis, data problems, temporal dependence problem, endogeneity of trade or economic growth, and so on. Recently, some studies have found that domestic institutional factors condition or constrain the pacifying effect of economic interdependence on military conflict (Garztke & Li 2001, Gelpi & Grieco 2001). If political leaders' decisions covary systematically with the domestic institutional conditions, political leaders' strategic behavior should be explained. Even though political decisions result from the synthetic interaction between structure and agents, the relationship between agents and structure has not been investigated enough in liberal peace studies. This problem is related to ignorance of heterogeneity existing among countries in terms of domestic institutional context. If the unobservable country specific effects are not constant across countries, biased estimators will be obtained.

This research attempts to explain the heterogeneity by testing a proposition that the pacifying effect of trade on conflict is contingent on the domestic institutional context. Domestic institutional context, whether or not political leaders' positions are stable, can affect decision making, constraining the effect of other influential factors on decisions. For this purpose, this analysis emphasizes the importance of the two-level hierarchical model framework. This way of analysis is remarkable in the studies of international conflict. This is because, by employing a two-level hierarchical structure in a model, we are able to successfully separate domestic political institutional context from international structural factors, clarifying the relationship between the two levels and explaining heterogeneity existing in the pacifying effect of economic interdependence on the decision of conflict onset. As a result, this enhances understanding of political leaders' strategic behavior.

To measure the relationship between international factors and conflict, I use Oneal & Russet's (2001) data set of the Militarized Interstate Disputes (MID) complied by the Correlates of War Project. By accepting Kristian Gleditsch's (2002) recent compilation of data regarding states’ bilateral trade and their gross domestic products, they include observations from within the Soviet bloc during the cold war. They also incorporate Zeev Maoz's (1999) revised data on militarized interstate disputes (MIDs) to obtain more accurate data for dyadic analyses. For measuring domestic political institutional context, I use the concepts of the size of winning coalition and the size of the selectorate (Bueno de Mesquita et al. 1999 Bueno de Mesquita and Root 2000). However, arguing that there are some measurement problems in previous studies, I measure these variables in a different way by using Vanhanen's Polyarchy data set (Vanhnen 2000). In short, substantively, this research deepens the liberal peace arguments by explaining heterogeneity existing across countries. Methodologically, this research shows how a two-level game analysis can be performed in a hierarchical model and why this way of modeling is important. Related methodological problems in the previous studies will be clarified.

As far back as Converse's (1964) seminal chapter, scholars in public opinion and political psychology have explored individual differences in how people think about politics. Work in the voting behavior and participation literature has documented the ways in which variations in the political environment--particularly in terms of the intensity of political campaigns-- can catalyze political action (Rosenstone and Hansen 1993). In this paper, I extend the theory to explore whether campaigns act as catalysts of political cognition, as I examine the intersection of individual differences and political context. The question I raise is this: Can highly intense campaigns bridge individual differences in how citizens think about politics?

Dual-process models in social psychology provide a framework for thinking about the intersection between individual differences and political contexts. Although the details vary, dual-process models suggest that individuals differ in the extent to which they seek and process information in their cognitive environment, and a variety of contextual factors can effectively bridge the gap between those less and more inclined to engage in effortful thinking (Cacioppo and Petty 1982, Eagly and Chaiken 1993).

Based on work in social psychology and political psychology, I explore the hypothesis that highly intense campaign environments can increase both the supply of political information available to ordinary citizens and boost their motivation to engage in sophisticated strategies as they think about politics. The hypothesis to be explored in this paper is similar to Kahn and Kenney's (1999) work on how the intensity of senatorial campaigns affects how experts and novices evaluate senatorial candidates, but it extends it in several ways. First, it poses the largely unexplored question of how the intensity of political campaigns might affect not only the ingredients of candidate evaluations but also the ingredients and properties of policy opinions more generally. It poses the possibility that campaigns, particularly intense campaigns, provide citizens with food for thought and the motivation to digest it slowly. Second, the paper extends the conceptualization of how context interacts with individual differences, by empirically testing the possibility that it's not the experts or the novices who are most affected by variations in campaign intensity, but those in-between.

In the paper, I use two datasets: the 1988-1990-1992 Pooled Senate Election Study and the 1980 Major Panel Study (from the National Election Studies). The Senate Election Study provides an opportunity to compare patterns of political thinking in individuals situated within a wide variety of campaign contexts: from respondents in the midst of highly intense campaigns to respondents in the midst of barely contested campaigns, to respondents who do not even have a senatorial campaign in their state. The 1980 Major Panel Study consists of four interviews with individuals during the course of the 1980 presidential campaign and hence allows for tracking patterns of thought for individuals as the intensity of the campaign increases.

I conceptualize modes of political thinking in several ways, including the amount and type of information that is encoded and used for judgment and the relative weight of different considerations that drive policy opinions. The analyses will provide a way of comparing the thinking strategies of (1) different individuals situated in political contexts that vary widely in intensity and (2) of the same individuals who experience varying political contexts during the course of the presidential campaign. To account for the nested nature of the data (for the Senate Election Study, individuals are nested in states and for the 1980 Major Panel Study, the four waves of observations are nested on individual respondents), I employ hierarchical modeling techniques.

Puzzle
Almost all studies of voter behavior assume the proximity model. However, in most multiparty systems voters do not necessarily vote for parties closest to them but rather for parties more extreme than they are. The main puzzle this project poses is "Why do voters vote for parties more extreme than they are?" I propose and test a formal model of voter behavior that recognizes this empirical regularity and provides a micro-foundation explanation to it.

Answer
The model begins with the observations that parties engage in a constant effort to influence policy outcomes, and that voters, as political actors, wish to use their vote choice to affect policy. In other words, voters care not only about proximity to party platforms, but also about policy outcomes. They therefore reward parties based on two considerations. First, as the proximity model predicts, voters reward parties for representing their ideas. Second, voters reward parties for pulling policy outcomes in their direction. Since in most democracies policy outcomes are a product of a compromise among multiple forces (e.g. parties in the governing coalition), voters understand that their vote will be 'diluted' and so compensate for it by voting for parties more extreme than they are. This means that the institutional environment and the method by which votes are converted to policy determine voter behavior. In particular, the more power sharing allowed by the electoral system, the more voters need to compensate for the dilution of their vote. Similarly, the more majoritarian is the electoral system, the more Downsian is voting.

Contribution
The project emphasizes the importance of policy outcomes for voters and builds on this recognition to reinterpret an ongoing debate between directional and proximity theories of voting. I show that policy-oriented voters vote 'directionally'. Yet unlike the directional framework, my framework does not rely on exogenous imposition of decision rules such as 'region of acceptability' and 'neutral point'. These features are endogenous to my model. I also show that predictions of the proximity and the directional models are special cases of my own model, and I test all three. In addition, at the macro level, this study explains party dispersion in policy space, dispersion that deviates from predictions of current models. The study provides micro-foundational explanation to menu-dependence in individual choice an empirical regularity that previous methodological work addressed via the stochastic component (Alvarez and Nagler, 1995, 1998 Alvarez, Nagler, and Bowler, 2000 Lacy and Burden, 1999 Schofield et al., 1998 Quinn, Martin, and Whitford, 1999). Generally, I show that other things equal, the marginal impact of each party on policy declines with similarity across parties. Therefore, policy-oriented voters will prefer distinct parties to substitutable parties. Furthermore, the more power-sharing allowed by the institutional environment, the more menu- dependent is voter behavior.

Data
Drawing on theoretical predictions, the project consists of comparative empirical analysis of voting behavior from various democracies. The model implies that the mapping process from partisan ballot options to policy outcomes depends on institutional design. Hence, my empirical cases vary along this dimension. Voting behavior in Norway (1989) and The Netherlands (1998) exemplify voting in a fragmented proportional representation systems where policy is a product of extensive political bargaining. Britain (1987) and Canada (1997) represent majoritarian cases where little compromise between the governing party's ideal policy and opposition occurs. Finally, Germany (1998) represents a mixed member proportional system.

In the 1950s, Americans’ trust in government reached its zenith buoyed by the nation’s post-war status. In polls, over sixty-five percent of the country trusted the government "most or all of the time." But in the turbulent times of Vietnam and Watergate, trust in government fell precipitously, dropping to half its previous level by the early 1970s. In the decades since, levels of trust in government have never returned to their 1950s level.

Scholarly interest in trust in government has focused on identifying the individual level factors that led to such massive distrust of the government (Citrin and Green 1986 Citrin and Luks 1998 Miller and Borrelli 1991 Chanley, Rudolph, and Rahn 2000 Craig 1993, 1996 Erber and Lau 1990 Williams 1985 Garment 1991 Orren 1997 Brehm and Rahn 1997 Putnam 2000). But the preoccupation with describing and discussing this one time decline in trust in government has had two consequences.

First, we have no systematic theory of why trust in government moves up and down. We also lack any explanation for what might cause trust to change in the short-term. Second, we have very little idea of how to model trust as a dynamic process. Presently, the literature lacks a model of trust that assesses its dynamic properties. Most analyses of trust over time use biennial data, which is too infrequent to use in a dynamic model building exercise. In my paper, using a quarterly trust in government series from 1920-2000, I undertake a careful dynamic model building exercise for trust in government. I use a variety of dynamic methodologies including transfer function analysis, intervention analysis, distributed lag models, and single-equation error correction to model the dynamics of trust. I assess which methodology allows for precision control of the dynamics, provides for the best fit to the error structure, and allows for the easy incorporation of covariates. The end result is one of the first dynamic models that accounts for the effects of presidential approval, scandals, economic prosperity, social capital, and congressional approval on trust in government.

Background to Research:
Work by Hegre, et. Al. ("Towards a Democratic Civil Peace," APSR 2001) suggests that the likelihood of Civil War decreases in countries which are more definitively autocratic or democratic, as well in countries which are longer lasting and that, more importantly, these effects are both separable and significant. More surprising, those authors fail to recover the oft-repeated claim that increasing democracy makes a country less likely to experience civil war. Our work explores the propriety of the authors' application and interpretation of results from the Cox regression model and explores alternative models (Neural Networks and Support Vector Machines) as more productive predictors of civil war.

Data and analysis:
The original authors calculate the hazard ratio (rate ratio) of civil war by applying a Cox regression model to a detailed data set including, at every instance of Civil War from 1820 to 1992, measurements of every country’s democracy/autocracy score (from Polity IIId), economic development, proximity of previous regime change (in days), proximity of previous civil war (in days), and a number of other variables controlling for the proximity (temporal and spatial) of international war. (In total, 12,500 by 30 entries.) The feature of this civil war data set that separates it from other comparable collection efforts is the day-specific coding of civil wars, which promises a greater potential for analysis than the previous approach where cross sections are tabulated by country-years.

We first make several simple corrections to their methodology. (1) The authors use list-wise deletion to deal with missing values (eliminating approximately 25% of entries). We use methods (most prominently the program Amelia) for imputing these missing values and demonstrate that these methods alter the original findings. (2) In the Cox regression model one ignores the time dependence of the hazard rate in order to gain insight into what are assumed to be fixed causal effects of a set of independent variables. With respect to our particular model, it is not wholly clear why, for instance, one should expect the marginal effect of democratization on civil war to be identical across time periods. In particular, estimation by the original researchers suggests that the time dependent factor of the hazard rate of civil war actually increases violently throughout the 20th century, but their translation of the Cox model requires that they exclude the variables of interest (democratization, development, etc) as potential mechanisms for this increase. By simply disaggregating the original data set into increasing small time slices, we apply the original Cox Model (and then a logit model as a comparison) to determine (a) the variation of the effects of the independent variables throughout the 19th and 20th centuries, and (b) the extent to which the independent variables are successful in accounting for the precipitous increase in the occurrences of civil war.

An additional difficulty with the Cox model is its inability to make substantive predictions about when and where civil war will occur. To demonstrate this point, we divide our data into a testing set and training set. Then using Cox estimates derived from the training set data, we examine the effectiveness of these estimates in predicting civil war in the test set. At this point in our research, we take the opportunity to compare several non-linear models for predicting rare events data. We apply Radford M. Neal’s “Flexible Bayesian Modeling Software,” Langche Zeng's "Hidden Layer Feedforward Neural Networks," and Chih-Chung Chang and Chih-Jen Lin's "LIBSVM"(Library for Support Vector Machines), and then compare the performance of each of these as predictors of Civil War.

Question: are legislative parties equally strong in the U.S. House and Senate?

Relationship to literature: a strong version of ‘conditional party government' would imply that parties are equally strong in the House and Senate, since both chambers are accountable to the same voters and parties. Evidence of comparative strength is mixed, with recent studies (Groseclose and Snyder 2000 Campbell, Cox, and McCubbins 2000) suggesting that parties are equally strong. My rival hypothesis is that parties are stronger, particularly on procedural issues, in the House due to the House’s greater size and workload.

Data: I use measures of party conflict, majority party unity, and majority party success for every vote in either chamber for every Congress (1789-2000) as my dependent variable. I use fixed effects dummy variables to filter out temporal variations, then test for the effects of chamber, majority party size, and vote type on partisanship using ordinary least squares and logistic regression.

Findings: I find that House parties are generally stronger than Senate parties, and particularly so on procedural issues. While this is not a ‘party effects ' paper, the influence of party is clear on these votes. I also find that party size has a significant effect on party unity and effectiveness, while presidents have little systematic influence.

Methodological Interest: this is an ambitious and complex time-series regression project. It is part of a broader project that treats the U.S. House and Senate as most similar cases, then seeks to explain bicameral variation.

This paper describes the application of deterministic sampling techniques to the problem of fitting non-linear time series models.

Deterministic sampling, recently introduced in the machine learning literature as the 'unscented kalman filter' [1], provides an approximation method for inference in nonlinear systems that is significantly more accurate than linearisation schemes such as the extended kalman filter, but does not require monte-carlo methods.

In deterministic sampling a fixed number of control points, determined by the dimension of the model's state are propagated through the true non-linear state (or observation) equations. The resulting points are used to reconstruct the mean and, most importantly, an accurate approximation to the variance of the transformed state variables. This procedure gives more accurate results than the extended kalman filter because control points are propagated through the true system equation, not a linearisation of it. The procedure also maintains fixed computational cost since the number of control points for a d-dimensional state variable is typically a constant 2d+1, and no Jacobians need be computed.

This work extends the range of non-stochastic methods available to time series analysts. It should also make it easier for non-experts to fit non-linear models to their data: the researcher needs only to plug-in the functional form of the relevant nonlinear mapping. No Jacobians need to be derived, and since there is carefully programmed stochastic simulation, the need to monitor simulation properties and judge convergence is also removed.

The paper demonstrates the method by fitting a range of nonlinear time series models to machine-generated event data from the Bosnia conflict, previously studied by Pevehouse and Goldstein [2].
[1] E. A. Wan and R. van der Merwe (2001) "Kalman Filtering and Neural Networks" in S. Haykin (ed) The Unscented Kalman Filter, Wiley Publishing.
[2] J. S. Goldstein and J. C. Pevehouse (1997) "Reciprocity, bullying and international conflict: Time-series analysis of the Bosnia conflict", American Political Science Review 91(3).

Introduction
Between 1848 and 1920, thirty states adopted woman suffrage measures. Early victories came in four Western states in the 1890's. Following these early victories, however, was a dry spell - there were no suffrage victories between 1896 and 1910. The next decade witnessed a steady progression for woman suffrage, as state-by-state the battles were won. Thus the question emerges: What accounts for the difference in time to adoption of woman suffrage in the various states? I argue that to understand where and when state-level woman suffrage measures were adopted, we must begin by addressing the social and political structural constraints within which the battles were fought. I will employ duration analysis techniques to test the hypotheses regarding the importance of structural constraints, such as the competitiveness of the partisan environment, the strength of presence of third parties, the level of industrialization in the state, the presence of racial minorities in the population, and the institutional rules governing the suffrage extension process in the state, in determining the time it took state governments to respond to the demand of the suffrage activists.

Methodological Concerns
The substantive question of interest is what made some American states more ready adopters of woman suffrage than others? In other words, what systematic factors, if any, made the hazard of adoption greater for some states than it was for others? This question implies the modeling of the time elapsed between the start of the woman suffrage movement and the adoption of woman suffrage in the states as a function of states' relevant characteristics. Because I am explicitly interested in how changes in population characteristics and political and economic conditions may change a state's likelihood of adopting woman suffrage, the static specification offered by simple OLS regression cannot serve my analytic interests. I turn, instead, to the set of statistical techniques commonly referred to as "duration analysis." I will be interested specifically in comparing the suitability of various continuous time parametric models (i.e., the appropriateness of various distributional assumptions), the Cox proportional hazards approach, and the binary dependent variable time series cross section (BTSCS) approach using logistic regression.

The difference between the various continuous time parametric models is the distributional form specified for a, the baseline hazard (Cox and Oakes 1984 Box-Steffensmeier & Jones 1997). The inclusion of a baseline hazard, however, means that all of these models employ the assumption of proportional hazards, which implies that the explanatory variables have the same impact on the hazard, regardless of location in time (Box-Steffensmeier & Zorn 2001). This assumption may be substantively unfounded. In the case of the woman suffrage issue, for example, assuming the effect of ethnic composition of the population on the behavior of politicians should be the same across time ignores any social theory of assimilation and/or acculturation. In addition to causing us theoretical discomfort, violating the assumption of proportional hazards will result in bias in the estimated coefficients and a loss of efficiency (Box-Steffensmeier & Zorn 2001). The Cox proportional hazards approach alleviates the need to make a distributional assumption, but does not remedy the concerns about the assumption of proportional hazards. It may also be considered an inefficient use of the data on time to adoption of suffrage measures as it takes the ordering of events and not the elapsed intervals of time between events as its base of information. Treating the data as BTSCS and using logistic regression has the advantage of relative ease of interpretation, especially for audiences unfamiliar with event history techniques.

My task, then, is to compare these techniques in the specific context of the substantive question about state's propensities to act on the woman suffrage issue. I must be especially concerned about taxing a small sample (forty-eight relevant states over a seventy-two year time period). I must also be concerned about the impact of unmeasured heterogeneity between the American states and the clustering of information by region (e.g., the possibility of policy diffusion along geographical patterns). Cognizant of these concerns, I use a dataset of yearly observations on the forty-eight American states over the time period 1848 through 1920, to compare the empirical results on the determinants of time until adoption of woman suffrage offered by these various approaches. I then use simulated data to compare the performance of the three modeling options given known assumption violations in small samples.

The data for this paper were compiled from the Historical U.S. Census, the Congressional Quarterly’s Guide to U.S. Elections, 3rd ed. (1994), and the United States Historical Election Returns, 1824-1968, available from the Inter-University Consortium for Political and Social Research (ICPSR). Information on adoption of woman suffrage measures in the states was taken from a publication of the National Woman Suffrage Association (The National Woman Suffrage Association. 1940. Victory: How Women Won It. New York: The H.W. Wilson Company).

Works Cited Box-Steffensmeier, Janet M. and Bradford S. Jones. 1997. Time is of the Essence: Event History Models in Political Science. American Journal of Political Science 41: 1414-1461.
Box-Steffensmeier, Janet and Christopher J. W. Zorn. 2001. Duration Models and Proportional Hazards in Political Science. American Journal of Political Science 45: 972-988.
Congressional Quarterly. 1994. Guide to U.S. Elections, 3rd Edition. Washington, D.C.: Congressional Quarterly, Inc.
Cox, D.R. and D. Oakes. 1984. Analysis of Survival Data. London: Chapman and Hall.

Research Question:
Why do some newly independent states experience large-scale violence while others do not?

Since 1945, the United Nations recognized the independence of more than 125 new states. Between the 1940s and 1970s, the majority of new states emerged through decolonization, primarily in Asia, Africa, Oceania, the Caribbean and the Middle East. While states such as Kenya, Botswana and Malaysia escaped with only limited violence after independence, other states, including Indonesia, Myanmar (Burma), and the Belgian Congo, experienced devastating collective violence repeatedly.

Some twenty years later, in the early 1990s, collapsing states such as the former Soviet Union and Yugoslavia splintered into new states throughout Europe and Asia. Once again, some states transitioned into new statehood in a relatively peaceful manner, but other states still struggle with internal violence.

The experiences of mature states suggest once state consolidation occurs, whether through decisive internal wars or through internally negotiated agreements, violence does not usually reoccur. Large-scale violence may be understood as a proxy for a state’s failure to consolidate economic and political power against domestic rivals. Studying new states, as opposed to all states, offers a fantastic experimental framework to understand violence, stability and state consolidation given one simplifying assumption: well-established political and economic institutions are absent at the time of independence . Research into violence and state consolidation suggests that these same institutions may cause or inhibit violence. These causal relationships can be tested using data over a forty-year period for a sample in which all cases begin at the same institutional point, i.e. ground zero.

Theory:
My theory explains why some newly independent states avoided large-scale violence, or managed to consolidate peacefully, while other states could not. At the time of independence, some domestic agreement about the future of politically salient groups within the new state must exist. This agreement may be formal and elaborate, or informal and ad hoc. The type of agreement is not necessarily important. However, the agreement, as with any other contract, must have a reliable enforcement mechanism for the agreement to last. Ideally, a domestic contract among politically salient groups over the terms of the new state and government will have a self-enforcing mechanism. With a self-enforcing mechanism, each group has an interest in supporting and sustaining the new contract, i.e. the new state. In such a situation, violence is less likely.

Independence does not necessarily bring trusted friends into a friendly contract. If groups are suspicious of one another, a self-enforcing mechanism may prove impossible if even one group decides to violate or walk away from the contract. In such situations, an external guarantor may provide necessary security guarantees and oversight functions until the new state can develop a new contract with a self- enforcing mechanism. If the new state contains a strong civil society, the civil society may act as the enforcer. If a strong civil society does not exist, an acceptable third party (such as a departing colonial power, another state or international organization) may enter the contract as the enforcer.

When new states have a self-enforcing contract or an external guarantor, violence is less likely. However, if a new state cannot sustain a self-enforcing contract, or if the proposed external guarantor walks away from the contract before the new state is able to create a replacement contract with a self-enforcing mechanism, violence is likely.

Contribution to Literature:
Very few researchers have systematically studied large-scale violence in new states (Cohen et al. 1981, Henderson and Singer 2000, Henderson 2000). The few large-N studies of collective violence in new states that do exist suggest that domestic political variables associated with state making most strongly influence the probability of violence. They do not present detailed theory along with their empirical results, nor do they explore underlying institutional mechanisms that might explain patterns of violence and stability.

There is also the question about state formation and consolidation specifically in new states. Economic enterprises led settlers to colonies, but the influence of private vs. government enterprises over economic and political institutions instituted by the colonial administration differed (Crowder 1968). Settlers did influence colonial policies during colonization (Bates 1983), while indigenous groups had little control over institutional design. Post-independence, did indigenous groups retain colonial institutions?

Several researchers argue that colonial institutions persisted post-independence (Engerman and Sokoloff 1997, La Porta, Lopez-de- Silanes, Shleifer and Vishny 1998, 1999). In specific countries, governments adopted the tactics and institutions of the former colonizers to cement their political power and extract resources from the rest of society (Reno 1995, Boone 1992, Turner and Young 1985). These countries later imploded into massive violence against the government, so it is unclear what affect colonial institutions had on state consolidation and subsequent violence.

Methodology and Data
I will test my explanation of enforcement mechanisms against alternative hypotheses using a large-N data set I am presently constructing. The data set includes all newly independent countries (N = 78), from the time of independence until 2000. While I am coding some variables personally, I am also utilizing existing data sources, including Polity IV, Singer and Small's COW sets, Gleditsch et al's new conflict data set, ACLP data, World Bank data, the State Failure Task Force data, and others. At this point, I plan to employ hazard modeling and regression analysis to test my hypotheses against alternative hypotheses. I also will use the data set to explore five specific patterns of violence and stability in newly independent states, and try to determine if predictions can be made on the duration of peace and stability within new states.

Proponents of the inexorable impact of globalization argue that international financial integration has resulted in an environment in which countries can no longer assert control over their ties to the international financial system. While international financial integration may have changed the attractiveness of the different policy options available to countries, I argue that international financial integration has not obviated the role of domestic politics. I construct a theory of domestic preference formation in which domestic preferences are shaped by sectoral and class characteristics. I demonstrate the role of domestic preferences in the choice of monetary regime first through large-n statistical analyses of monetary regime choice in OECD and non-OECD countries from 1960 to 1998, and then through more detailed case studies of monetary regime choice in Ireland, the United Kingdom, Argentina and Mexico. For this poster, I will focus on the quantitative analysis of monetary regime choice in the OECD.

This poster presentation aims to demonstrate how techniques from quantitative geography can help political scientists better explain the institutional influences upon national economic systems. The project will be undertaken under the supervision of Torben Iverson, Margarita Estevez-Abe, and Gary King, Professors of Government at Harvard University. Work will be completed this spring and summer in conjunction with participation in a larger project supervised by Gary King that focuses on developing new regression techniques for comparative politics.

Project Description:
In their introduction to Varieties of Capitalism: The Institutional Foundations of Comparative Advantage, Peter Hall and David Soskice argue that the key variations in capitalism surface at the national level (Hall, Peter and David Soskice, eds. 2001). Nation- states, after all, regulate the determinative relationships within the economy: the financial system and corporate governance arrangements internal firm structures industrial relations education and training systems and inter-company relations. This focus on national regimes, they argue, improves upon the earlier ‘social systems of production' literature that examines the role played by a variety of tiers of government and non-governmental actors. This literature of the 1980s and 1990s had highlighted the role of trust, cooperation, and learning within economic communities. Much of the theoretical work drew on case study research in regional conglomerations such as Baden Württemberg and Emilia-Romagna.

As a litany of critiques have emphasized, the 'industrial districts' and 'learning regions' literature dismissed by Hall and Soskice did not offer a parsimonious or representative account of the new forms of political economy emerging with globalization (Markusen, A. "'Sticky Places in Slippery Space: A Typology of Industrial Districts." Economic Geography 72: 293-313.) In simply retreating to a focus on national political economies, however, the Varieties of Capitalism literature fails to profit from recent innovations within economic theory. The 'new Economic Geography ' and 'endogenous Growth Theory'--pioneered by scholars such as Masahisa Fujita, Paul Krugman, and Anthony Venables demonstrates that economic growth is ultimately a cumulative, spatial process that usually evolves on a regional as opposed to national basis (Fujita, Masahisa, et al. The Spatial Economy: Cities, Regions, and International Trade. Massachusetts Institute of Technology, 1999). If political scientists want to maintain that institutions are integral to such growth processes, as the Varieties of Capitalism literature argues, they need to identify more clearly how national (and/or regional and local) institutions interact with these regionally-based economic processes.

Academics working within the field of quantitative geography have developed a number of new techniques that could be applied fruitfully to the comparison of political-economic systems (for an excellent review, see Fotheringham, et al, Quantitative Geography. London: Sage Publications, 2000). First of all, geographers have placed an increasing emphasis upon producing local or mappable statistics and estimations of geographical variance, steering away increasingly from aggregate figures. Secondly, geographers have developed geographically-weighted regression techniques. Finally, others have created spatial regression models that deal with auto- correlation amongst spatial variables. Political scientists concerned primarily with voting behavior have been the first within the discipline to adopt some of these techniques. These approaches also presents scholars of political economy with the opportunity to comb their data at a finer degree of resolution and thereby complement the case studies of the earlier ‘industrial districts' and 'learning regions' literature.

In the context of this project, I plan to demonstrate the insights that can be gained through disaggregating geographical data and employing geographically-weighted regression techniques to a substantive problem in political economy. In a classic article on 'economic voting,' Bingham Powell and Guy Whitten assess the influence of national economic performance upon election outcomes across nineteen industrial democracies (Powell, G. Bingham Jr. and Guy Whitten. "A Cross-National Analysis of Economic Voting Taking Account of Political Context." American Journal of Political Science, Vol. 37, No. 2, May 1993, pp.391-414.) They argue that when scholars control for the ideological image of the incumbent government, its electoral bases, and the clarity of its responsibility, patterns of ‘economic voting' are observable. This analysis could be improved, I will argue, through attention to the geographic dispersion of unemployment. During both upturns and downswings, unemployment will occur with varying levels of intensity throughout a country unemployment levels will be closely tied to the health of the particular industries that animate regional economies. We should observe voter backlash in ‘high unemployment' districts and the reverse in less vulnerable regions. I plan to replicate Powell and Whitten’s project at the district level in a representative subset of the original sample to see if the relationships they observe at the national level become yet stronger when regional variation is taken into account.

The majority of U.S. foreign policy studies focus on interactions between the executive and legislative branches of government during the conduct of foreign affairs. They examine the politics of a decision-making process, designed to confront the numerous challenges encountered from the participation of a government in an interdependent, international system. Thus, scholars focus primarily on those actors within the United States who proactively determine foreign relations policy.

Consequently, in an effort to concentrate on the President, Congress, or agencies such as the CIA or Department of State, these examinations neglect the roles played by the judiciary. While the political branches of government most directly determine policy outcomes, the contributions of the judiciary are no less significant. Many foreign policy questions involve constitutional interpretations regarding the authority vested in the executive and legislative branches. Since the courts possess the final opinion on the Constitution, judicial decisions often define the parameters and boundaries within which the political branches must operate. Despite this substantial impact on foreign policy decision making, little scholarship exists on judicial influences in the conduct of foreign affairs.

The terrorist attacks of September 11, 2001, and subsequent actions by the Bush administration, such as his executive order declaring the use of military tribunals to preside over the trials of suspected terrorists, have increased the salience of federal court oversight for U.S. foreign policy. Unfortunately, three significant limitations have hindered our understanding of how the judiciary operates in the foreign relations scheme. First, within the small body of literature examining courts and foreign policy, a majority of these studies utilize qualitative techniques to assess historical relationships between the three branches of the federal government. These studies examine whether the Supreme Court defers to either the President or Congress in the formulation and conduct of U.S. foreign policy. Second, the constitutional authority imposed upon the judiciary extends beyond balancing disputes between the political branches of government. Courts are responsible for protecting the civil liberties of citizens within the United States. Nowhere is this responsibility more important than when judges resolve disputes between the rights of individuals and the authority of the federal government to engage in foreign affairs or protect national security. Finally, most studies focus exclusively on the United States Supreme Court. The Federal Courts of Appeals and District Courts receive virtually no attention. With the Supreme Court gaining more control over its docket, thereby reducing the number of cases it hears, the decisions of the lower federal courts become more significant because the possibility of review is reduced. This paper examines the roles of the federal judiciary in resolving foreign policy disputes since World War II, by asking whether the federal courts are defenders of civil liberties or champions of national security. It contributes to the literature on U.S. foreign policy by focusing on a historically neglected branch, and to the literature on judicial decision making by comparing structural differences among the three levels of the federal court system.

Theories regarding institutional constraints on individual behavior are useful for analyzing judicial resolution of foreign policy disputes. The most prominent theory of judicial behavior argues that judges cast votes according to their personal policy preferences (Segal and Spaeth 1993). When explaining the fundamental tenets of the ‘attitudinal model' Segal and Spaeth comment on certain institutional features which facilitate the application of this theory to the Supreme Court. Specifically, the justices are free to vote their sincere preferences through a combination of three institutional facets: discretionary control over the docket, lack of higher political ambition, and the existence of no higher judicial authority (1993, 70-72). However, the authors do not empirically test these assertions in the lower courts.

Subsequent analyses have provided initial evidence that institutional features constrain individual behavior of judges in the federal district courts (Rowland 1991 Mather 1995) or state supreme courts (Brace and Hall 1990). Yet, questions remain regarding the precise relationship of these constraints to judicial outcomes. For example, if judges are motivated by policy concerns as the attitudinal model suggests, then researchers need to identify the extent to which all levels of the federal judiciary make policy. Jacob (1965, 1991) argues that trial courts are not policy-making institutions because the judges typically confine their decisions to norm enforcing declarations. Contradicting this argument, Mather (1991) and Rowland (1991) contend trial courts can either restrict or expand policy through their decisions and that their ability to frame legal issues extends beyond norm enforcement thereby impacting judicial policy. Given this debate, it is apparent that more research is needed to better understand the extent to which the attitudinal model (and specifically its institutional assumptions) applies all levels of the federal judiciary.

To accurately model the influence of all levels within the federal judicial system, one can conceive of the system as a three-tiered, where each tier selects a specific choice (whether to rule in favor of civil liberties or national security). This is a logical conception since the majority of the Supreme Court’s docket involves reviewing decisions from lower courts.

After each court rules upon a case, the losing litigants have the option of accepting the decision, or appealing to a higher court. If the litigant chooses not to appeal then the sequence is over. However, if the losing party chooses to appeal, then we move to the next subgroup. According to Greene (2000), the nested logit technique provides an adequate model for this multiple-choice problem. Nested logit allows variation in subgroups to differ across those groups, while still maintaining the independence of irrelevant alternatives assumption within the groups. By incorporating a nested logit model, I can test whether structural constraints within the federal court system systematically affect the likelihood of a court to rule in favor of civil liberties or national security. The data consist of a random sample of federal court cases from 1946-2000. Cases were identified using Lexis-Nexis and involve issues pertaining to national defense, foreign affairs, international law, immigration, and war powers.

Sometimes states that could otherwise be independent countries choose instead to voluntarily merge with other states to form federations. Why? States might give up their independence in favor of federating in order to secure trade benefits or to jointly produce public goods. I develop a general model that explains why states would choose federation instead of international cooperation. I test the model with three case studies: the Australian states, which federated peacefully the Argentine states, which unified only after protracted violence and the East African states, which remained independent.

This is an important topic of study. First, the study of federations can tell us about international cooperation in anarchy generally. Second, better models of federation can contribute to predictions about political integration in specific regions, such as the member states of the E.U., APEC, or ASEAN. This poster will highlight some of the methods I use in my case study of the origins of the 1901 Australian federal constitution. Historians disagree about the Australian states’ motives when they federated. Some argue that political identity considerations drove decisions to join the union, while others argue that commercial considerations were crucial. I collected data on divisions (roll-call votes) from 19th century Australian state legislatures. I use several legislative scaling techniques on this data to show the relationships between parliamentary votes on federation and parliamentary votes on other issues. Both the NOMINATE technique (Poole and Rosenthal) and a Bayesian technique using MCMC (Jackman) show that, in the two major Australian state parliaments, divisions over federation were systematically related to divisions over commercial issues like the tariff and not political identity issues like the relationship with Britain and constitutional liberalism or republicanism.

The substantive value-added of this project is that it, for the first time, introduces quantitative evidence into historical debates about the origins of the Australian constitution. It also demonstrates that market considerations can drive international political integration. Methodologically, the project introduces data from several new legislatures. It also provides an example of using scaling data to draw inferences about the political motives behind a particular set of related bills.

One of the best-known models of public opinion is John Zaller’s RAS (receive-accept-sample) model. The model has been used successfully to represent things as diverse as opinion on Vietnam War involvement and candidate choice at the voting booth. One of the model’s key insights is that the probability that individuals both receive and accept messages from a candidate or other source depends on their levels of information.

Zaller’s model is less nuanced, however, in describing how the characteristics of the information environment affect message reception. In most instances, Zaller infers the intensity of a message from the opinions of survey respondents rather than measuring it directly. Indeed, many models of voter learning in campaigns give the information environment less attention that it deserves. Some authors assume that the information environment is limitlessly rich, some infer its characteristics, and others measure it with crude proxies. As a result, political scientists have an incomplete understanding of how the characteristics of a campaign’s information environment influence citizen learning and, in turn, the choices that voters make.

My paper seeks to remedy that situation by focusing not only on the volume of messages that a candidate may send, but on the timing of those messages and the length of the campaign as well. All three of these aspects of the information environment vary considerably across campaigns and candidates, and all three are key to understanding voter learning in campaigns.

The context of my investigation is the 2000 presidential primary campaigns. I use this setting for several reasons. First, there is substantial variation across states in the volume of candidate messages disseminated, ranging from almost zero in some states to hundreds per day in states like New Hampshire and South Carolina. Second, the effective length of primary campaigns also varies considerably across states. While candidates may begin intense campaigning in New Hampshire months before that state’s primary, they first set foot in other states only days before citizens vote there. Third, presidential primary campaigns often provide instances of unbalanced message flows across candidates, the situation most amenable to finding campaign effects. A final reason for studying presidential primaries is that the partisanship of voters does not affect the choices they make. The absence of this anchor in primaries makes what happens during the campaign potentially salient for many more voters.

My model extends the work of Zaller in several significant ways. First, and most important, is the emphasis I place on the information environment, modeling it explicitly in the reception function. I also incorporate several different measures of the information environment, including its density, and the length of the campaign. Third, I allow the information environment to vary depending on the state and media market in which one lives. I also incorporate citizen motivation to learn about the candidates, allowing motivation to vary across individuals and across states.

The special features of a primary race require some additional modifications. For instance, my model generalizes to a multi-candidate setting while Zaller's model was developed for only two candidates. I also incorporate voter expectations about candidate viability into my model to take account of the tactical voting that occurs in situations of more than two candidates.

I advance three hypotheses about the impact of the campaign environment on voter learning. First, and consistent with previous work, increasing message exposure should aid voter learning. Second, longer campaigns, which give voters more time to process candidate messages, should also promote voter learning. Finally, and perhaps counterintuitively, a denser information environment--more messages per unit of time--should decrease learning. This hypothesis draws on theory developed in the marketing literature.

I test these hypotheses statistically, using Stata’s maximum likelihood estimation routine. I use commercial polls conducted during the 2000 presidential primaries to tap learning and advertising tracking data to measure the information environment. The tracking data describe the number and density of candidate messages in the country’s top 75 media markets. They also contain information about each advertisement’s sponsor, and its time and place of airing.

In sum, a better understanding of how the campaign information environment influences voter learning and choice can help scholars evaluate current campaign institutions and practices and can aid in the design of potential reforms as well. But this understanding can be achieved only when statistical models realistically incorporate the characteristics of the information environment.

Political scientists frequently make inferences about how survey respondents and voters are thinking about political issues based on their behavior. In this dissertation, I use functional Magnetic Resonance Imaging (fMRI) to study political cognition and affect during three experiments. The subject population consists of undergraduate political sophisticates and novices. In the first experiment, I show subjects the faces of white and black, famous and not famous, political and nonpolitical persons. In the second study, subjects view images of whites and blacks who are either "value violators" or "value exemplars." The third study asks subjects to answer a series of questions that are either political or nonpolitical, and threatening, non-threatening, or racial. This experimental design enables me to test many theories simultaneously. For instance, behavioral differences in political novices and sophisticates are expected to have neural correlates, "racial politics" may be more about race than politics, emotions are suspected to influence sophisticates more than novices, etc.

Preliminary analysis of the four subject pilot study suggests the intriguing possibilities of fMRI for political science. The white liberal political sophisticate had substantial activations in the amygdala and anterior cingulate (threat and conflict monitoring) while viewing black faces, consistent with the idea of "white liberal guilt." The political sophisticates had higher lymbic activations (emotional regions) while viewing political faces. These differences in activation for each subject are statistically significant at the 0.01 level. By PolMeth, I will have completed the analysis of all three experiments for the pilot study and will be presenting those results for the first time.

Under what conditions should we expect changes in the dominant patterns of interaction in protracted conflicts? This paper addresses this question through the lens of the conflict in Israel and Palestine and events data.

Substantively, this paper seeks to analyze when and how cooperation and conflict breakout between Israel and Palestine. Previous studies have suggested the role of third parties (Goldstein, Pevehouse, Gerner, and Telhami 2001), exogenous shocks and policy entrepreneurs (Rasler 2001), ripeness (Zartman 199?), and a variety of other factors as significantly affecting the level of cooperation or violence between the two parties. Often, however, these studies talk past one and other. We seek to better incorporate the ideas of exogenous shocks and leadership as elements interacting with the actions of Israelis and Palestinians themselves as well as affecting the relationship between possible third parties and Israelis and Palestinians (these latter variables are also measured as event counts through KEDS and are based on cooperative acts other than agreements such as extension of aid, verbal support, etc. see Rasler 2001 for details). This more inclusive model should give us a better idea of if and how parties are more likely to become committed to cycles of cooperation and when those cycles may deteriorate into cycles of violence.

In terms of methodology we evaluate of two different event count models. We use event counts based on the Kansas Event Data System (KEDS) coding of Reuters articles from 1979-2002. We conduct two separate analyses with number of agreements and number of violent events serving as dependent variables (see below for a discussion of coding schemes and aggregation of events). Earlier authors have made use of negative binomial models to analyze interactions in the region (see Rasler 2001). While this is an improvement over using Poisson models (which require the expected value and variance to equal each other), the negative binomial is still an imperfect fit. The negative binomial combines the Poisson and gamma models allowing for over dispersion of the dependent variable (by allowing lambda to vary according to phi and sigma-squared). For the case of cooperation in Israel and Palestine, however, the data are clearly not over dispersed. Rather there is very little before Oslo and quite a bit after.

This can be better modeled with a hurdle model (see King 1998, pg 222 King 1989). One can think of hurdle models as allowing for a Bernoulli process of determining when an event is likely to occur, and then a Poisson process establishing how many of a given event one would observe given that the event occurs at all. King applies this idea to the studying of alliances in war initiation: alliances do not cause wars, but they make them more likely to spread. The assumptions of the hurdle model are a better fit for event counts in our context, particularly given that there were no official contacts between Israelis and Palestinians prior to the Oslo Accords. As a means of comparison we will also evaluate King’s Generalized Event Count model (King and Signorino 1996), which allows for over-, Poisson, and under-dispersion of the dependent variable.

As a further innovation, we make use of a new coding scheme developed by the researchers at KEDS. The Conflict and Mediation Event Observations (CAMEO) scheme is a successor to the World Events/Interaction Survey (WEIS) as the coding scheme of choice for KEDS/Tabari. KEDS researchers report that CAMEO has better sensitivity to low-level agreements and other events related to mediation efforts (Gerner, Schrodt, Abu-Jabr, and Yilmaz 2002). While we are not interested in mediation, per se, our dependent variables of number of agreements and number of violent events should be more realistic under CAMEO than WEIS. Similarly, the independent variables counting reciprocity between Israelis and Palestinians and third party actions by the United States should become more robust under the new scheme. This paper would serve as one of the first substantive applications of this newly coded event data.

Most work on legislators concentrate on what determines their decisions on individual roll calls. This paper instead examines when major policy changes occur. For these changes to occur, the policy coalition that previously rejected the policy must have been altered, and a new coalition forms that passes the legislation. Following Brady and Sinclair (1984), I examine whether or not the changes in policy coalitions (and subsequently, policy) are the result of replacement or conversion. Replacement occurs when a new member is elected to Congress, and her preferences are different from the previous occupant of the seat. Conversion occurs when current legislators change their position on a particular issue. I will examine whether or not the change in the distribution of preferences results primarily from replacement, conversion, or both. The answer to this question indicates whether legislators are responsive to policy demands of their constituents, or major policy change only occurs with substantial electoral turnover.

Brady and Sinclair restrict themselves to cases studies on particular policies this paper tries to generalize the study to examine overall legislative productivity (using measures from Mayhew's Divided We Govern and subsequent studies). Based on Krehbiel's critique of Mayhew, legislative productivity is modeled as a function of changes in the gridlock region. The gridlock region shifts when there are changes in the distribution of legislator preferences, which are the result of either replacement or conversion. The paper examines whether the gridlock region primarily changes due to replacement or conversion, and then subsequently examines whether policy changes substantially with changes in the gridlock region.

The difficulty in testing the conversion-replacement hypotheses at a general level is the inadequacy of most scale measures. The tests requires estimates of the distribution of preferences for Congress and the President. Furthermore, in order to construct estimates of the gridlock region, the estimates of the ideal points must be comparable over time, as well as across institutions (i.e., the House, Senate, and President must all be on the same scale). These estimates also must be able to separate the conversion and replacement effects. Most ideal point estimates over time either assume legislator ideal points do not change or place strong restrictions how they change, implicitly assuming only replacement effects shift the distribution of preferences. This paper, using methods of ideal point estimation based on Jackman, Clinton, and Rivers (2001), Quinn and Martin (2001), and Bailey and Chang (2001), calculates ideal points that are comparable over time and institutions, and allows legislator ideal points to change over time.

The primary challenge to understanding the relationship between interest groups and Congress has been methodological. It is empirically difficult to develop a causal model that explains how interest group behavior affects legislative outcomes. In this paper, I employ three methodological techniques that help to overcome this challenge and begin to fill in the gaps in this ever-expanding literature.

First, scholars who wish to determine how groups affect congressional legislation face the challenge of empirically linking group behavior to congressional outcomes. Prior work in this area has either concentrated on making generalizations about group behavior (see for example Schlozman and Tierney 1986) or is based on the assumption that influence over votes is the same as influence over policymaking (see for example R. Smith 1984 Kau and Rubin 1982 Kollman 1997).1 I overcome this problem by developing a unique survey sample and instrument. I examine four congressional committees, and their subcommittees, from the 106th Congress.2 Using legislative hearings; I identify groups that testify before Congress. Then, I identify the issues the groups were interested in and the bills the hearing references. Finally, I survey the groups about their activities on the specific issues on which they testified. It is necessary to survey groups about their activities on specific policy issues (as represented by one or more bills) rather than generalize about lobbying activities.3 In the fall of 2001 I sent out 1550 unique surveys to interest groups across the country.4

Second, scholars in this field are often faced with the challenge of defining and measuring "influence." What does it mean to exert influence over legislation? Many answers have been offered, but most are either well defined and unmeasured, or ill defined and well measured. I overcome this challenge by developing a unique dependent variable. I use the final status of a piece of legislation as a proxy for group influence. I argue that group influence in Congress can be measured by the progress of legislation in which a group is interested, regardless of whether a group was trying to kill or pass a bill. My sample of bills comes from the four congressional committees mentioned previously. I control for the congressional context of the legislation and interest groups' organizational characteristics. The primary independent variable of interest is the aggregate strength of lobbying efforts on a bill.

Third, I contribute to our understanding of interest group influence over legislation by developing and ordinal probit model to estimate the parameters of the theoretical model. I hypothesize that groups can use their range of strategies to effectively influence legislation in the desired direction. I use state-of-the-art software developed and distributed by Gary King, et al. to impute the missing observations that are so common with survey research.5 Then, I present the results with intuitive graphics after estimating predicted probabilities for the variables of interest.6

This paper fills a substantive and empirical gap in the literature. While several congressional and interest group scholars have noted the need to account for the strategic context in which a bill or issue exists, few studies consider it in their theoretical models. This paper incorporates the needed element of strategic context of legislation into a model of group influence. Empirically, this paper presents a new approach to collecting a sample of groups and uses it to make a connection to legislative outcomes. While not a probability sample, the procedure is useful because it provides the needed link between group behavior and congressional legislation.

Footnotes:
1. Studies that have not used votes as a dependent variable have either gone untested empirically, or they do not provide a direct measure of policymaking. Formal models of this type include Austen-Smith (1993), Ainsworth (1993), Ainsworth and Sened (1993). Examples in other areas include Arnold (1993) Hall and Wayman (1990) Hall (1996) Austen-Smith and Wright (1994) Baumgartner and Jones (1993) Evans (1996) M. Smith (2000).
2. Committees included the House Agriculture Committee, House Education and the Workforce Committee, House Energy and Commerce Committee, and House Ways and Means Committee.
3. Groups tend to care about policies rather than bills, so surveys must ask groups about their activities on specific issue areas. However, Congress make decisions over bills, not policies, so issues must be traced to bills and tracked via legislation in Congress.
4. The survey was conducted nationally, however about half of the groups in the sample were located in the Washington, D.C. area and mailings were compromised by the anthrax scare in the fall of 2001. Response rates were affected by this complication.
5. I use the Amelia program to prevent the case-wise deletion that would have been necessary otherwise in this research (see King, Honaker, Joseph, and Scheve 2001).
6. I use the Clarify program to create the distributions for predicted probabilities (see Tomz, Wittenberg and King 2001).

Economists have long considered the marginal rate of substitution and the analysis of substitutes in goods markets as critical elements of a theory of choice. There are strong reasons to believe that these models yield insights into the study of political choice by unitary actors, e. g. an entire special issue of the Journal of Conflict Resolution develops this question. Unfortunately, the econometric issues accompanying simultaneous equations models for limited dependent variables are substantial and complex. In the goal of developing proper models for the study of policy substitution, this paper first examines the shortcomings of many existing techniques for examining interdependent policy choice, before turning to a discussion of simultaneous equations models with limited dependent variables and the problems they must confront. After exploring the relevant issues and defining the range of permissible problems, the models are applied to controversies in the study of international finance and to United States foreign interventions. The applications illustrate the importance of simultaneous equations models for discrete choice problems.

The assumption that voting that is motivated solely by "preferences" provides the foundation for a wide range of empirical and theoretical models of roll call voting behavior. However, not all scholars adhere to this assumption. For example, models of vote buying, vote options, and party influence all assume that legislators' vote can be influenced by considerations other than immediate policy preferences. Similarly, models of position-taking assume that legislators sometimes face conflicts between how they would like to vote and how their constituents would like them to vote.

These "conflict" models share a common characteristic. How a legislator ultimately votes depends on the probability that she will cast the decisive vote. An efficient leader will buy votes or call options only if those votes are going to change the outcome (Snyder and Groseclose 2000 Zeckhauser and King 2000). A conflicted legislator will vote against constituent only if her vote is going to matter (Bianco 1994 Denzau Riker and Shepsle 1985 Austen-Smith 1991). In other words, these models require that each legislator believe that she will cast the pivotal vote.

Importantly, the evidence offered in support of these models has almost always come from roll call outcomes that are nowhere near minimum winning. For example, Snyder and Groseclose (2000) use 65/35 as their cutoff for identifying the close votes where they expect party influence to occur. It is a common feature of the literature to assume (implicitly or explicitly) that each legislator believes that her vote is pivotal, before studying roll calls that are far from minimum winning.

We propose and test a conflict model that does not involve pivotalness. This model predicts differences in behavior on votes that are not minimum winning, while accounting for patterns that others have attributed to behavior that assumes pivotalness.

Empirical studies indicate that elected officials tend to have more extreme policy preferences than those of their constituencies (Ansolabehere, Snyder and Stewart, 2001 Erikson and Wright, 1999). We use this insight to theorize and test for systematic position- taking effects across congressional roll call votes. We begin with a simple single-dimensional spatial model that demonstrates 1) that the number of legislators facing a position-taking conflict gradually increases as the margin on a roll call decreases and 2) that position-taking effects on a typical roll call are asymmetric and correlated with party affiliation. We next apply monte carlo simulation techniques to generalize these basic formal results. Finally, we provide empirical support for the model by examining NOMINATE errors across all roll calls in a Congress. Using an indicator of electoral vulnerability as our proxy for the incentive to position-take, we show that vulnerability predicts errors on close votes (65/35), and that the rolls calls where these effects occur for Republicans are different than for Democrats.

Thus, the model directly addresses the issue of pivotalness that is ignored in tests of other conflict models. And it accounts for empirical patterns in roll call voting that others attribute to party influence and vote buying.