Download Understanding Social Science: Theories, Explanations, and Demographics and more Study notes Social Sciences in PDF only on Docsity!
Theory Development
1. Learning Objectives
After reviewing this chapter readers should be able to:
- Understand basic concepts of causality in terms of 2x2 tables.
- Understand the role of background and theoretical knowledge in social and behavioral research.
- Understand the relation of basic demographic information and other kinds of explanation in social and behavioral research.
- Understand the concept of confounding.
- Understand the role of mechanisms in social and behavioral explanation.
- Understand the problem of the underdetermination of theory by data in social science.
2. Introduction
Theory construction in the social sciences faces a series of difficulties, or different circumstances, from those faced in the physical sciences, with the result that theories in the social sciences, despite some surface similarities, have significant differences from the natural sciences. Among the differences are the massive causal complexity of the subject matter; the fact that the topics of interest to social scientists and the users of social science are generated from folk, normative, or common sense concerns, and cannot be completely separated from ordinary language; the fact that even the most successful models work only in limited contexts whose boundaries are poorly understood; and the fact that successful prediction often, if not always, results from simplifications known to be false or inadequate as explanations.
This means that social science theories are better understood as models which work, either to predict or explain, in limited settings, rather than laws of science which hold and apply universally.
But the identification and discovery of predictive laws faces the same problem: the actual causal facts or relationships which appear empirically are already compounded of a long list of possible causes, from which laws must be extracted and discovered. In a very simple case, one might be able to hypothesize both the laws and the mathematical nature of the additive relationship and find that one set of laws and one rule for combination of causes actually predicted the outcomes. But such simple cases are never found.
3. Causal Complexity
Experiment vs. Nature
It might seem that the solution to the problem of complexity is to simplify by constructing experiments in which the effects of other causes can be neutralized by random assignments of treatments for levels of the causes in question, so that fundamental relationships can be identified. But this strategy has not proved successful. Not only are there few usable results of this kind even in experimental psychology, once the relationships are taken out of the laboratory and applied to causally complex actual situations, they fail to predict successfully as a result of interferences from other causes.
The primary alternative to this method is the identification of patterns of relationships between variables. Normally this is a matter of identifying a correlation or statistical relationship in the data, though usually with a significant degree of error or unaccounted-for variation. In many science contexts, such as engineering, the same kind of empirically-based modeling of predictive relationships is standard practice, and often for the same reason, there is an absence of theories which allow for prediction. Ordinarily in these cases, which involve physical magnitudes, the relevant casual relationships are reasonably well understood and the relationships are estimated from data collected from experiments designed to isolate the relationships in question.
4. Prediction vs. Explanation
Not only the data but the interests of social science theorizing and much of its language are generated from normative, practical, policy, or common sense concerns. Topics such as adolescent pregnancy, for example, are both policy and normative interests that require explanations– explanatory theories or models– in order to intervene in the causal process to alter outcomes. This topic is an example of the problem of the difference between prediction and explanation.
There are some good predictors of risk for adolescent pregnancy, such as smoking. Although knowing that smoking is a predictor might be useful as a means of identifying the adolescents who might be made the subject of an intervention, smoking is not a cause of pregnancy, so intervening by preventing smoking is not going to be an effective method of reducing adolescent pregnancy. This requires causal knowledge.
But the underlying causes that produce both smoking and the behavior that leads to adolescent pregnancy are far more complex, heterogeneous, and difficult to either identify or work with than the simple fact of smoking.
Moreover, this particular relationship only holds in those social contexts in which smoking has a particular meaning for the smoker and for others. In a society in which smoking was universal, or uncommon, the relationship would not hold. But it would likely also not hold or work in the same way in a context in which the social meaning of smoking– the message sent and received by the act of smoking– was different. This problem– that the underlying causal mechanisms themselves vary according to context– limits the generalizability or robustness of models and at the same time reminds us of the importance of the complex but unknown underlying causal mechanisms. This adds a complication of a different character.
Exercise 2: Explanatory or Predictive
6. Common Sense
In the face of these daunting difficulties, social scientists have devised a number of strategies. The simplest and most fundamental is to understand behavioral phenomena in terms of “folk psychology” or common sense, as actions with reasons. The problem of complexity overwhelms such explanations: the kinds of decisions and reasoning that go into an event such as an adolescent pregnancy are complex, and even to turn such an event into an action or a series of actions involves a reconstruction. Even if we think of these events as choices, they are difficult to construct as reasoned decisions. Like most actions, there are many considerations, some spur of the moment, some long term, and disentangling them is not easy, even in such simple market decisions as the purchase of a pair of shoes.
The diagram below demonstrates the point that statistically linked data may not make rational sense when ignoring other factors in the process. In social science all the factors and their relationship to one another must be taken into account.
Figure 1: Goals and Plans in Decision Making
Adapted from Krantz, D and Kunreuther H. "Goals and plans in decision making."
Judgment and Decision Making, Vol. 2, No. 3, June 2007.
6. Common Sense
More complex social behaviors, such as a decision to commit suicide, or the background to and events leading to an adolescent pregnancy, can be made to conform to the model of decisions based on preferences only by reconstructing them as an abstraction and attributing the reasons, preferences, and “decisions” to this abstract model. The model of the decision-maker in turn is constructed to conform to the statistical data by varying the reasons or preferences. One might conclude, for example, that girls with more limited opportunities are more likely to become pregnant because their losses in future earnings would be less than those with greater opportunities, and one would indeed find statistical patterns that confirmed that poor families are more likely to produce such pregnancies. In this case, no claim is made that teenagers in the heat of passion calculate future income probabilities. The claim is that they behave as if they did so, and that this “as if” is what explains their conduct.
This is the strategy of economic theory and rational choice approaches to theory construction. In practice, these models rely on generic knowledge about what sorts of preferences in general drive human action. The construction of such models employs a large set of known corrections, such as discounting future returns, which are used to enable these models to fit the data. At each step, of course, the model becomes farther removed from the kinds of facts that folk interpretations and common sense descriptions of these events rely on. But this kind of abstraction does provide a kind of solution to the problem of complexity.
7. Who Believes What?
If we think of our problem of understanding the phenomena as one of seeking the real causes of the outcome in question, and perhaps we also would like a means of predicting behavior or even intervening so as to change the outcomes, this kind of abstraction is potentially valuable, but only if one can manipulate the situation in such a way that behavior changes. Changing a teenager’s future earnings prospects is not feasible. Nor is this a very good predictor: in the case of adolescent pregnancy, such non-explanatory facts as whether the adolescent smokes turns out to predict better.
One of the standard methods solves the problem in a different way, which compromises between fidelity to the actual thinking and beliefs of the people who are acting and the larger picture of social differences within a society. The method also begins, as the rational choice model does, with some typical, well-proven starting points, but the starting points are “standard demographics” rather than an abstract economic model.
Rates, for example, of smoking, political affiliations, suicide, or adolescent pregnancy, vary between demographic groupings, often dramatically.
At the same time, demographic (and geographic) groupings correspond, loosely, to different social worlds. The analyst’s knowledge of the specifics of the social life of these social worlds may supply other informative categorizations, for example between informal social groupings that can be identified on the basis of members’ knowledge of these social worlds, such as membership in cliques. But there is also a mass of additional concepts, such as “network,” that also enable the analyst to search for categorical distinctions that are possibly relevant to the outcomes of interest, and to test this relevance by comparing rates or degrees of the outcome in question.
This kind of problem, in which prediction, rational decisions, and the kind of information available in the thick description (Geertz, 1973) of actual events pull us in different directions, is normal for behavioral science subjects. And they lead to different approaches to the problem of complexity.
7. Who Believes What?
The method of dividing the social world into smaller and smaller categories to identify differential outcomes does not itself produce an explanation or theory.
But it is highly relevant to the construction of theory, and to understanding the less-theoretical statistical approach to the same kinds of questions. One abstract possibility is this: dividing the population into smaller and smaller categories results in a set of rates of outcomes, such as presidential voting preferences, in which the relation between the outcome and the categorization is more or less self-explanatory, or can be explained on the basis of background knowledge that is widely shared, such as the fact that a particular candidate advocates policies favorable to the group in question. In these cases, little in the way of “theory” would be needed. The puzzle is the overall outcome, in this case likely votes for president.
The bulk of the explanation of the outcome is statistical: counting and adding up the size of the categories determines the outcome. In this case, the candidates’ policies are an intervention which is targeted to specific categories in order to influence the total vote. “Theory” plays little role, though some generic background knowledge about what makes people vote is necessary. As we will see in the final section, this is also how other statistical approaches to causal model building proceed.
Exercise 4: Demographic Categories
8. Understanding and Interpretation
In many of the cases of interest to social and behavioral science, however, background knowledge does not suffice.
A standard approach to these problems of explanation is to use the categories to point to the different communities and social networks that the individuals in the categories are part of, because different beliefs, values, and experiences are sustained and transmitted in groups and networks. The focus of this kind of analysis shifts from the individual to the social world in which the attitudes, interests, experiences, and beliefs are sustained and developed, and often leads to explanations that terminate in conceptual constructions such as “culture” and “world view.” These are themselves abstractions, but they are developed on a different basis, for example by the analysis of open-ended interview material or through ethnographies that supply the material for attributing attitudes, beliefs, and motivations, reasons for acting, different perceptions of the meanings of choices and outcomes, and thus different behavior to composite or idealized members of the group or category in question.
In these cases the attitudes and beliefs themselves may require interpretation, in the sense of making the background knowledge and beliefs of the agents– often contained in uncodified practices– intelligible to the outsider. But even with interpretation and reconstruction into intelligible world views, the behavior may still be puzzling. In the case of the abortion dispute, for example, it is evident that there are social categories, such as working women and mothers, that are more strongly represented on opposite sides of the controversy, and that there are differences in world view and membership in social groups that sustain these views. But these considerations do not seem to explain the passion with which the sides engage in the struggle.
The differences that appear when populations are categorized point to beliefs, interests, values, and so forth that are themselves puzzling and in need of explanation.
8. Understanding and Interpretation
An influential interpretation of this conflict explains the passion in terms of identity: women against abortion are often stay-at-home moms who react to the implicit devaluation of babies as threats to their own value (Luker, 1984). This is a theoretical explanation, in two senses.
- It appeals implicitly to a more general idea, namely that people will react to threats to their identify or ideas that devalue this identity.
- It is also a mechanism, that is to say a model of a causal process that links two apparently unconnected facts: the demographic category of the women opposed to abortion, and their attitudes.
Exercise 5: Hypothesized Mechanisms
10. Mechanisms
The mechanisms approach can be understood by taking an epidemiological case as a point of comparison, consider cholera. There were, in the 1850's, strong correlations in London between altitude of residence and incidence of disease. This suggested "miasma" as a mechanism. The real mechanism, however, was water contaminated with the cholera bacillus. Establishing this required a different kind of study, which eventually showed that the correlation was an artifact of pumping methods, not miasma. Applying this kind of reasoning in the social sciences is more difficult, as the mechanisms in question typically involve the mind. But hypotheses about the motivations of individuals may be supported with various kinds of additional evidence, and tests may be devised of some of these hypotheses.
Table 1: Elevation and Cholera Deaths
Elevation of
Districts in
Feet
Number of
Terraces from
Boston
Deaths from Cholera
in 10,000 Inhabitants
Calculated
Series
Under 20 1 102 102/1 = 102 20-40 2 65 102/2 = 51 40-60 3 54 102/3 = 34 60-80 4 27 102/4 = 26 80-100 5 22 102/5 = 20 100-120 6 17 102/6 = 17 140-60 18 7 102/18 = 6
(Humphreys, N.A. (Editor): Vital Statistics: A Memorial Volume of Selections from the Reports and Writings of William Farr. London , Sanitary Institute, 1885, p. 254-5.)
10. Mechanisms
The notion of mechanism, however, is not well-defined. Some economists, for example, consider that they have a mechanism when they have an equation. In many cases, mechanism accounts rely on rational choice models as discussed earlier (Elster, 1998).
