7

A major aim of research is to understand the relations among variables with many different values. It is helpful to remember that variables vary. For example, when studying a discrete nominal variable such as gender, we refer to gender as the variable because it can vary—either male or female. The term level, along with the terms value and condition, all refer to the same idea. Levels are the discrete values or conditions that variables can take on. For example, male is a level of the variable gender. Female is another level of the variable gender. In both cases, gender is the variable. Similarly, when studying a continuous, scale variable, such as how fast a runner completes a marathon, we refer to time as the variable. For example, 3 hours, 42 minutes, 27 seconds is one of an infinite number of possible times it would take to complete a marathon. With this in mind, let’s explore the three types of variables: independent, dependent, and confounding.

The three types of variables that we consider in research are independent, dependent, and confounding. Two of these, independent variables and dependent variables, are necessary for good research. But the third type, a confounding variable, is the enemy of good research. We usually conduct research to determine if one or more independent variables predict a dependent variable. An independent variable has at least two levels that we either manipulate or observe to determine its effects on the dependent variable. For example, if we are studying whether gender predicts one’s attitude about politics, then the independent variable is gender.

The dependent variable is the outcome variable that we hypothesize to be related to or caused by changes in the independent variable. For example, we hypothesize that the dependent variable (attitudes about politics) depends on the independent variable (gender). If in doubt as to which is the independent variable and which is the dependent variable, ask yourself which one depends on the other; that one is the dependent variable.

By contrast, a confounding variable is any variable that systematically varies with the independent variable so that we cannot logically determine which variable is at work. So how do we decide which is the independent variable and which might be a confounding variable (also called a confound)? Well, it all comes down to what you decide to study. Let’s use an example. Suppose you want to lose weight, so you start using a diet drug and begin exercising at the same time. The drug and the exercising are confounded because you cannot logically tell which one is responsible for any weight loss. If we hypothesize that a particular diet drug leads to weight loss, then whether someone uses the diet drug becomes the independent variable, and exercise becomes the potentially confounding variable that we would try to control. On the other hand, if we hypothesize that exercise leads to weight loss, then whether someone exercises or not becomes the independent variable and whether that person uses diet drugs along with it becomes the potentially confounding variable that we would try to control. In both of these cases, the dependent variable would be weight loss. But the researcher has to make some decisions about which variables to treat as independent variables, which variables must be controlled, and which variables to treat as dependent. You, the experimenter, are in control of the experiment.

8

You probably have a lot of experience in assessing variables—at least on the receiving end. You’ve taken standardized tests when applying to your university; you’ve taken short surveys to choose the right product for you, whether jeans or smart phones; and you’ve taken online quizzes—perhaps ones sent through social networking sites like Facebook, such as the Dogster Breed Quiz, which uses a 10-item scale to assess the breed of dog you are most like (http://www.dogster.com/bolz/dog-breed-selector-quiz, 2015).

How good is this quiz? One of the authors took the quiz—answering one of the questions by choosing a light chicken salad over alternative choices of heavier fare—and was declared to be a bulldog: “You may look like the troublemaker of the pack, but it turns out your tough guy mug is worse than its bite.” To determine whether a measure is a good one, we need to know if it is both reliable and valid.

A reliable measure is consistent. If you were to weigh yourself on your bathroom scale now, and then again in an hour, you would expect your weight to be almost exactly the same. If your weight, as shown on the scale, remains the same when you haven’t done anything to change it, then your bathroom scale is reliable. As for the Dogster Breed Quiz, the bulldog author took it twice and was a bulldog the second time as well, one indication of reliability.

But a reliable measure is not necessarily a valid measure. A valid measure is one that measures what it was intended to measure. Your bathroom scale could be incorrect but consistently incorrect—that is, reliable but not valid. A more extreme example is using a ruler when you want to know your weight. You would get a number, and that number might be reliable, but it would not be a valid measure of your weight.

And the Dogster Breed Quiz? It’s probably not an accurate measure of personality. The quiz, for example, lists an unlikely mix of celebrities, with seemingly different personalities, as bulldogs—Ellen DeGeneres, Whoopi Goldberg, Jack Black, and George W. Bush! However, we’re guessing that no one has done the statistical work to determine whether it is valid or not. When you take such online quizzes, our advice is to view the results as entertaining rather than enlightening.

A measure with poor reliability cannot have high validity. It is not possible to measure what we intend to measure when the test itself produces varying results. The well-known Rorschach inkblot test is one example of a test whose reliability is questionable, so the validity of the information it produces is difficult to interpret (Wood, Nezworski, Lilienfeld, & Garb, 2003). For instance, two clinicians might analyze the identical set of responses to a Rorschach test and develop quite different interpretations of those responses—meaning it lacks reliability. Reliability can be increased with scoring guidelines, but that doesn’t mean validity is increased. Just because two clinicians scoring a Rorschach test designate a person as psychotic, it doesn’t necessarily mean the person is psychotic. Reliability is necessary, but not sufficient, to create a valid measure. Nevertheless, the idea that ambiguous images somehow invite revealing information remains attractive to many people; as a result, tests such as the Rorschach are still used frequently, even though there is much controversy about them (Wood et al., 2003).

9