As a scientiﬁc enterprise, psychology follows the methods of science, meaning that it uses data to generate testable hypotheses and constructs theories to explain the results of those tests. Research efforts are largely devoted to trying to describe people but also attempt to construct explanations. A theory is an explanatory framework, based on the observations. It suggests certain additional testable questions, known as hypotheses, which are tested by further research. Theories are always open to further reﬁnement, but to become widely accepted in the ﬁrst place, they must be based on replicable evidence. In the popular imagination, however, “theory” and “hypothesis” are often mixed up with each other, with theories dismissed as tentative guesses. One often hears evolution dismissed in this manner by religious fundamentalists, “It’s only a theory!”
Actually, it’s a whole range of theories, as there is much disagreement among biologists as to how exactly the processes involved in natural selection work. What they do not disagree on, however, is the question of whether it occurs. Mountains of data support it, and the general theoretical framework is the basis of all modern biology. Consider a problem in physics: gravity is only a theory, after all. There are actually some very different interpretations of how it operates, with Newtonian physics favoring the idea of a force that draws things towards objects as a function of their mass, while Einsteinian theory favors the idea that large objects actually warp space-time around them, causing objects to move towards them (note to any physicists who may be reading: Yes, I recognize that this is a very highly simpliﬁed explanation). Either way, however, loads of evidence suggests that if I drop an object, it will fall. We may have different theories about why it falls, but whether or not it will isn’t a question anymore.
Psychology is a young science—people have observed selective breeding and the gravity of falling objects for thousands of years, and reaction time to subliminally presented stimuli has been measured for only a few decades. The theories aren’t so well established, and they’re all still looking for supportive evidence.
Psychological scientists seek evidence in several ways, but the ideal method for determining cause-and-effect relationships is the experimental method. In an experiment, a variable (a variable is anything at all that can have more than one value and can be measured—favorites in psychology include gender, age, and intelligence, among many others) is manipulated by the experimenter, and changes in another variable are looked for as a result of the manipulation. For example, if the experimenter wants to see if the Mozart effect actually works, he or she might take a group of people and give them a math test. After the math test, the experimenter will have them listen to some Mozart, and then give them the math test again. If they improve in their performance on the math test, it might be tempting to give Mozart credit, but it isn’t a true experiment yet. How does the experimenter know that they wouldn’t have done better on the second test even without the Mozart? The only way to ﬁnd out is to take a second group of people (the control group) and have them take both tests at the same times as the ﬁrst group (the experimental group), but without having them listen to Mozart. If the two groups perform differently from each other, then it may be assumed that Mozart was the cause, because there was control over that variable. In any experiment, the condition that the experimenter controls and varies (in this case, whether or not the subjects listened to Mozart) is called the independent variable, and the variable that is used to determine whether that manipulation had an effect (in this case, the math test) is the dependent variable.
For this experiment to serve as a deﬁnitive test of the hypothesis that listening to Mozart will improve math skills, however, there are some other variables that must also be dealt with, as they might affect the outcome. What about musical taste, for example? Some students may be big Mozart fans, but others might hate all music other than country or Swedish death metal. Such variables that are not part of the experimental design but that may provide alternative explanations of the data, are called confounding variables. A quick, efﬁcient way of dealing with such things is random assignment to the experimental condition. Assuming that various musical tastes are distributed somewhat randomly among the participants, randomly picking who goes into the experimental group and who goes into the control group is a good way to get those Swedish death metal fans evenly distributed among the two groups, thus ensuring that they don’t have a disproportionate impact on one group or the other.
Sometimes it isn’t possible or ethical to use random assignment to groups, even though the research question is an important one. Consider the question of the effects of cocaine exposure on prenatal development. The ideal way to investigate this question would be to conduct an experiment, in which half of a sample of pregnant women is randomly assigned to smoke crack during their pregnancy . . . the ethical and legal problems in this example are clear. Because people in the world already perform this type of behavior, it is still possible to study the effects, and even to use the same dependent variables, by simply recruiting a sample of women who are already using cocaine and comparing them to a carefully matched control group of women who aren’t. Since the independent variable (cocaine use) wasn’t manipulated by the experimenter, this isn’t a true experiment. This kind of design is a quasi-experiment. It doesn’t allow the same degree of causal inference as an experiment, because of the lack of manipulation, but it allows us to study problems that can’t be looked at experimentally.
Less rigorous than experimental or quasi-experimental research is correlational research, in which multiple variables are measured, without any experimental manipulation, and the degree of relationship between the variables is measured. Most survey research is of this type. From a questionnaire, one might ﬁnd, for example, that men who drive large SUVs are more likely to own basketballs than men who drive compact gas/electric hybrids. Cause-and-effect conclusions can’t be drawn from this sort of data. How the two variables are related is impossible to discern from the data. This sort of research is very good for suggesting relationships between variables that may or not be causal, but which can then be examined more closely in experimental research. Before the experimental research (animal studies, mostly) that showed a causal link between smoking and lung cancer, the two were known to be correlated, but the causal inference could only be drawn after experiments were carried out (inspired by the correlational data).
Some of the questions that are of interest to psychologists involve very rare circumstances: the effects of having parents killed by terrorists on a child’s school adjustment, for example. In these cases, traditional group research, be it experimental, quasi-experimental, or correlational, isn’t appropriate. If there is only one person with the problem under study, or only a handful, it may be better to just observe that person in great depth to see what can be found out.
Such an in-depth examination of a single person is a case study, and while generalization of ﬁndings may be problematic (after all, one person can’t be representative of the whole population), such research is invaluable in discovering more about rare clinical syndromes.
- Mook, D. G. Psychological Research: The Ideas behind the Methods. New York: W. W. Norton, 2001.