OVERVIEW

In this chapter we discuss the art of theory building. Unfortunately there is no magical formula or cookbook for developing good theories about politics. But there are strategies for developing theories that will help you to develop good theories. We discuss these strategies in this chapter.

GOOD THEORIES COME FROM GOOD THEORY-BUILDING STRATEGIES

In Chapter 1 we discussed the role of theories in developing scientific knowledge. From that discussion, it is clear that a “good” theory is one that, after going through the rigors of the evaluation process, makes a contribution to scientific knowledge. In other words, a good theory is one that changes the way that we think about some aspect of the political world. We also know from our discussion of the rules of the road that we want our theories to be causal, empirical, nonnormative, general, and parsimonious. This is a tall order, and a logical question to ask at this point is “How do I come up with such a theory?”

Unfortunately, there is neither an easy answer nor a single answer. Instead, what we can offer you is a set of strategies. “Strategies?” you may ask. Imagine that you were given the following assignment: “Go out and get struck by lightning.” There is no cut-and-dried formula that will show you how to get struck by lightning, but certainly there are actions that you can take that will make it more likely.

OVERVIEW

Descriptive statistics and descriptive graphs are what they sound like – they are tools that describe variables. These tools are valuable, because they can summarize a tremendous amount of information in a succinct fashion. In this chapter we discuss some of the most commonly used descriptive statistics and graphs, how we should interpret them, how we should use them, and their limitations.

KNOW YOUR DATA

In Chapter 5 we discussed the measurement of variables. A lot of thought and effort goes into the measurement of individual variables. Once measurement has been conducted, it is important for the researcher to get a good idea of the types of values that the individual variables take on before moving to testing for causal connections between two or more variables. What do “typical” values for a variable look like? How tightly clustered (or widely dispersed) are the these values?

Before proceeding to test for theorized relationships between two or more variables, it is essential understand the properties and characteristics of each variable. To put it differently, we want to learn something about what the values of each variable “look like.” How do we accomplish this? One possibility is to list all of the observed values of a measured variable.

OVERVIEW

Most political science students are interested in the substance of politics and not in its methodology. We begin with a discussion of the goals of this book and why a scientific approach to the study of politics is more interesting and desirable than a “just-the-facts” approach. In this chapter we provide an overview of what it means to study politics scientifically. We begin with an introduction to how we move from causal theories to scientific knowledge, and a key part of this process is thinking about the world in terms of models in which the concepts of interest become variables that are causally linked together by theories. We then introduce the goals and standards of political science research that will be our rules of the road to keep in mind throughout this book. The chapter concludes with a brief overview of the structure of this book.

POLITICAL SCIENCE?

“Which party do you support?” “When are you going to run for office?” These are questions that students often hear after announcing that they are taking courses in political science. Although many political scientists are avid partisans, and some political scientists have even run for elected offices or have advised elected officials, for the most part this is not the focus of modern political science. Instead, political science is about the scientific study of political phenomena.

OVERVIEW

As researchers begin to consider possible tests of their theoretical propositions, they must make a series of important decisions. In this chapter we provide a discussion of choices of population and sample and inferences from samples about populations. We introduce this topic by using examples familiar to political science students – namely, the “plus-or-minus” error figures in presidential horse-race polls, showing where such figures come from and how they illustrate the principles of building bridges between samples we know about with certainty and the underlying population of interest.

POPULATIONS AND SAMPLES

In Chapter 6, we discussed how to use descriptive statistics to summarize large amounts of information about a single variable. In particular, you learned how to characterize a distribution by computing measures of central tendency (like the mean) and measures of dispersion (like the standard deviation). For example, you can implement these formulae to characterize the distribution of income in the United States, or, for that matter, the scores of a midterm examination your professor may have just handed back.

But it is time to draw a critical distinction between two types of data sets that social scientists might use. The first type is data about the population – that is, data for every possible relevant case.

OVERVIEW

Regression models are the workhorses of data analysts in a wide range of fields in the social sciences. We begin this chapter with a discussion of fitting a line to a scatter plot of data, and then we discuss the additional inferences that can be made when we move from a correlation coefficient to a twovariable regression model. We include discussions of measures of goodness-of-fit and on the nature of hypothesis testing and statistical significance in regression models. Throughout this chapter, we present important concepts in text, mathematical formulae, and graphical illustrations. This chapter concludes with a discussion of the assumptions of the regression model and minimal mathematical requirements for estimation.

TWO-VARIABLE REGRESSION

In Chapter 8 we introduced three different bivariate hypothesis tests. In this chapter we add a fourth, two-variable regression. This is an important first step toward the multiple regression model – which is the topic of Chapter 10 – in which we are able to “control for” another variable (Z) as we measure the relationship between our independent variable of interest (X) and our dependent variable (Y). It is crucial to develop an in-depth understanding of two-variable regression before moving to multiple regression. In the sections that follow, we begin with an overview of the twovariable regression model, in which a line is fit to a scatter plot of data. We then discuss the uncertainty associated with the line and how we use various measures of this uncertainty to make inferences about the underlying population.

OVERVIEW

Despite what we have learned in the preceding chapters on hypothesis testing and statistical significance, we have not yet crossed all four of our hurdles for establishing causal relationships. Recall that all of the techniques we have learned in Chapters 8 and 9 are simply bivariate, X-and-Y-type analyses. But, to fully assess whether X causes Y, we need to control for other possible causes of Y, which we have not yet done. In this chapter, we show how multiple regression – which is an extension of the two-variable model we covered in Chapter 9 – does exactly that. We explicitly connect the formulae that we include to the key issues of research design that tie the entire book together. We also discuss some of the problems in multiple regression models when key causes of the dependent variable are omitted, which ties this chapter to the fundamental principles presented in Chapters 3 and 4.

MODELING MULTIVARIATE REALITY

From the very outset of this book, we have emphasized that almost all interesting phenomena in social reality have more than one cause. And yet most of our theories are simply bivariate in nature.

We have shown you (in Chapter 4) that there are distinct methods for dealing with the nature of reality in our designs for social research. If we are fortunate enough to be able to conduct an experiment, then the process of randomly assigning our participants to treatment groups will automatically “control for” those other possible causes that are not a part of our theory.

OVERVIEW

Given our focus on causality, what research strategies do political scientists use to investigate causal relationships? Generally speaking, the controlled experiment is the foundation for scientific research. And some political scientists use experiments in their work. However, owing to the nature of our subject matter, most political scientists adopt one of two types of “observational” research designs that are intended to mimic experiments. The cross-sectional observational study focuses on variation across individual units (like people or countries). The time-series observational study focuses on variation in aggregate quantities (like presidential popularity) over time. What is an “experiment” and why is it so useful? How do observational studies try to mimic experimental designs? Most important, what are the strengths and weaknesses of each of these three research designs in establishing causal relationships between concepts? That is, how does each one help us to get across the four causal hurdles identified in Chapter 3? Relatedly, we introduce issues relating to the selection of samples of cases to study in which we are not able to study the entire population of cases to which our theory applies. This is a subject that will feature prominently in many of the subsequent chapters.

COMPARISON AS THE KEY TO ESTABLISHING CAUSAL RELATIONSHIPS

So far, you have learned that political scientists care about causal relationships. You have learned that most phenomena we are interested in explaining have multiple causes, but our theories typically deal with only one of them while ignoring the others.

OVERVIEW

Although what political scientists care about is discovering whether causal relationships exist between concepts, what we actually examine is statistical associations between variables. Therefore it is critical that we have a clear understanding of the concepts that we care about so we can measure them in a valid and reliable way. In this chapter we focus on several examples from the political science literature, such as the concept of political tolerance. We know that political tolerance and intolerance is a “real” thing – that it exists to varying degrees in the hearts and minds of people. But how do we go about measuring it? What are the implications of poor measurement?

WHY MEASUREMENT MATTERS

We have emphasized the role of theory in political science. That is, we care about causal relationships between concepts that interest us as political scientists. At this point, you are hopefully starting to develop theories of your own about politics. If these original theories are in line with the rules of the road that we laid out in Chapter 1, they will be causal, general, and parsimonious.

OVERVIEW

Once we have set up a hypothesis test and collected data, how do we evaluate what we have found? In this chapter we provide hands-on discussions of the basic building blocks used to make statistical inferences about the relationship between two variables. We deal with the often misunderstood topic of “statistical significance” – focusing both on what it is and what it is not – as well as the nature of statistical uncertainty. We introduce three ways to examine relationships between two variables: tabular analysis (crosstabs), difference of means tests, and correlation coefficients. (We will introduce a fourth technique, bivariate regression analysis, in Chapter 9.)

BIVARIATE HYPOTHESIS TESTS AND ESTABLISHING CAUSAL RELATIONSHIPS

In the preceding chapters we introduced the core concepts of hypothesis testing. In this chapter we discuss the basic mechanics of hypothesis testing with three different examples of bivariate hypothesis testing. It is worth noting that, although this type of analysis was the main form of hypothesis testing in the professional journals up through the 1970s, it is seldom used as the primary means of hypothesis testing in the professional journals today. This is the case because these techniques are good at helping us with only the first principle for establishing causal relationships. Namely, bivariate hypothesis tests help us to answer the question, “Are X and Y related?” By definition – “bivariate” means “two variables” – these tests cannot help us with the important question, “Is there some confounding variable Z that is related to both X and Y and makes the observed association between X and Y spurious?”

OVERVIEW

In this chapter, we show how political scientists across a variety of subfields have used multiple regression to test theories about causal processes involving politics. In particular, we show how clever research design and accompanying statistical analyses uncover interesting patterns of causal dynamics in the domain of American presidential popularity, in economic voting in European democracies, and in patterns of international conflict and cooperation. One of the core principles of this chapter – and of the book – is that solid research design is a prerequisite for insightful data analysis.

WHY CONTROLLING FOR Z MATTERS

Thus far, the discussion of the effects of multiple regression has been mostly abstract, referring to X, Y, and Z instead of to real social phenomena that most political scientists – and most political science students – care about. Some of this abstraction is necessary, of course, but we want you to see examples from actual research on how multiple regression is used, and how including new variables can change our theoretical conclusions. That is the goal of this chapter. What you will see, in these varying examples, is how introducing new controls for variables can change our inferences about the causal structure of the political world, which is what we have emphasized throughout this book.

To highlight the changes that result to parameter estimates when we change our model by adding a new variable, we show only a small portion of the results from the different models that we feature in this chapter.

Let us assume that you have an idea that has led you to identify a topic that you believe to be of sufficient importance and of feasible execution to conduct research on it. It may be a doctoral dissertation, or just a seminar exercise, but regardless of length and complexity no topic can ‘research itself’. You will have to translate it – via a series of strategic choices – into a project. It is this process of translation from something problematic or puzzling into something on which you can gather valid data and about which you can make compelling inferences that constitutes your research design.

Granted, much social scientific research is not self-consciously designed – it is not subject to a deliberate and critical process of choosing its components and defending its overall configuration. In many areas of inquiry, the design is literally given along with the topic. So much research has already been conducted on it that adding yet another case or extending it to yet another time period does not seem to require a novel effort of translation. Indeed, the universal desire of all sciences to produce cumulative knowledge seems to militate against continuously challenging and changing the standard way of doing research. If you do propose a change in design – say, a reconceptualization of the topic, a revised instrument for measuring variation, a different way of selecting relevant cases, or a novel method of testing for association – you will risk confusing your reader-cum-critic.