Module 2 part 3.docx

Module 2 part 3 III. Causal Criteria: What Makes a Cause a Cause? What does it take to convince someone that what you claim is a causal factor really is a cause? Many social researchers focus on the four basic causal criteria discussed below. Time Order Perhaps the most fundamental trait of a cause is time order. The independent variable must be present prior to the hypothesized behavior in the dependent variable. It is easy to see why the time order principle is crucial. No one would want to explain voter turnout in the 1980 presidential elections by referring to things that happened in, say, the year 1984. Covariation The independent variable must vary with the dependent variable in an observable and fairly consistent manner. If we hypothesize that people with a college education are more likely to vote than people without a college education, then we would expect to see education tend to covary with education in the following manner: people without a college education would often fail to vote while those with a college education would usually vote. Notice that this conceptualization of causal covariation is decicdedly probabilistic. We are not claiming that people with a college education will always vote nor are we claiming that people without a college education will never vote. We are simply looking for a probabilistic trend of covariation ? we do not expect to find an iron law. No Spuriousness Time order and covariation are not enough to convince us that something is a cause. The independent variable?s observed correlation with the dependent variable must not be caused by some other potential independent variable that we failed to account for. Consider this remarkable covariation discovered by a startled insurance adjuster: the more firefighters that are sent to fight a fire, the more property damage tends to occur. Perhaps if all fires were battled by the Three Stooges, we might be willing to accept this as a causal relationship. However, since firefighters are trained to put out fires in an effort to save lives and to reduce property damage, we need to consider the possibility that this is a spurious relationship. A spurious relationship is an observed covariation of two variables that is actually caused by some third variable that influences the behavior of the other two variables. What was the true cause of the property damage observed by the insurance adjuster? The size of the fire was the unexamined variable: large fires tend to generate calls for more firefighters and large fires tend to cause more property damage. This story of the confused insurance adjuster hopefully will serve to drive home the cliche often used to describe the call for no spuriousness: "Correlation does not equal causation." The effort to meet the criterion of no spuriousness is what motivates researchers to control for competing potential independent variables. The use of the term "control" stems from the experimental method: the researchers attempt to control other potential causes by holding them constant ? permitting only a single independent variable to vary in the study. Further discussion of the experimental method is beyond the scope of this module; for additional discussion of experimental control as well as other issues in experimental research, click here. Many social research questions are difficult to study via the experimental method. To make a long story short, either it proves difficult to control all major potential influences or the effort to provide such control makes the experiment so artificial that we begin to question whether or not the behavior observed is reflective of real world conditions. [It would be decidedly difficult to study voter turnout via the experimental method because any controlled environment would lead us to question how well it simulates the real-life decision to vote in the United States: what makes people drop what they are doing to vote? In an experiment, people have already dropped what they were doing.] In response to this frequent problem, social researchers often turn to the use of "statistical control." In the statistical approach to control, researchers observe the real-world variation of competing potential causes (and of the dependent variable) and use various mathematical techniques to sort out the separate effect of each independent variable under examination. Further consideration of the details of statistical control is beyond the scope of this workshop. For additional on-line discussion, click here. A Theoretical Justification Whether we are using the experimental method or the statistical approach, how do we determine which factors to control for in our causal model? It is here that the fourth and final causal criterion becomes crucial. For each independent variable that we wish to include in our model, we must be able to put forward some sort of theoretical justification. In other words, we must provide a plausible potential explanation of how each independent variable influences the dependent variable. This need for a theoretical justification is another reason why people argue that correlation does not equal causation. This theorizing is the causal explanation of the observable covariation. In the example below, we will try to think through several potential causes of voting.