Social Psychology. Daniel W. Barrett

causally related. Without further evidence, we cannot assume that one causes the other. Since there is no meteorological theory that would predict that seasonal changes in one hemisphere cause the seasonal changes in the other, the most we can say is that the two are correlated.

      Similarly, if all that we know is that teens who play more violent video games also tend to be more aggressive (but we can’t say which causes the other), then we call the relationship correlational. It is possible that excessive playing of violent video games causes teens to become aggressive, or it may be that aggressive teens are more likely to play violent video games. As with the weather example, merely knowing that they covary—or change together—does not tell us whether one causes the other. All we can say is that the relationship is correlational. This is an illustration of a scientific mantra that you will often hear as you learn more about psychological science: correlation does not mean causation. Simply because two variables are correlated does not imply that one causes the other (see Figure 1.7).

      Figure 1.7 Aggression and Violent Video Games: Correlation Is Not Causation

      The second type of relationship between two variables is called a causal relationship, and it exists when a change in one of the variables can be shown to produce a change in the other one. The study design used by Bargh et al. (1996) allowed them to infer that a causal connection existed between priming and walking. Another example is research that demonstrated that thinking about death caused people to express more support for then-President George W. Bush than they otherwise would have (Landau et al., 2004).

      The best strategy for discerning whether there is a causal relationship between two variables is to carefully control the context in which we examine them by utilizing the experimental method. An experiment can be defined as research in which one or more variables are systematically varied in order to examine the effects on one or more other variables. The experimenter manipulates or changes the independent variable (IV) to determine whether or not it causes a change in a different one, the dependent variable (DV). The IV is the purported cause, and the DV the predicted effect.

      To learn about the effects of playing violent video games on aggression, we can easily perform an experiment in which we manipulate game playing and then measure resulting aggression. We manipulate the independent variable—type of game played—by giving participants different levels or versions of it; some play a violent game and others play a nonviolent one. The IV is the potential cause, and the DV the expected effect. We could recruit teens from a local high school and randomly assign half to play Wulfenstein (a shoot’em up game) and half to play Tetris (a nonviolent game). After they played their respective games, we could have all the participants play a second game in which they have the opportunity to be aggressive toward an opponent. This is just what Anderson et al. (2004) did, and they found that the Wulfenstein players acted more aggressively in the second game than did the Tetris players. At least within the context of this experiment, playing a violent game (IV) caused participants to be more aggressive (DV). Similarly, the content of the unscrambled sentences differentially affected walking speed.

      One of the key features of experiments is that they have two or more conditions that participants can be assigned to. The manipulation of the IV produces at least two levels of that variable, each representing a different condition. The aggression experiment had two groups or conditions corresponding to the two games. One group is called the treatment group, because the participants assigned to it receive the treatment (in this case they played Wulfenstein). The treatment is the variable being tested and thus is the primary interest of the experimenter. The other group is called the control group, because its participants did not receive the treatment (they played a nonviolent game). A control group serves as a comparison group against which we may measure the effects of the treatment. In the video game study, if we find that there is no difference in aggression between the groups in the second game, then the treatment had no effect. In the priming study, the elderly-related word condition was the treatment condition, and the unrelated word condition was the control. Similarly, in cancer research, for instance, the treatment group receives the drug being tested, whereas the control group is given a placebo. If the cancer treatment and the control groups recover at the same rate, then there was no treatment effect.

      A second key feature of experiments is control: The experimenter needs to be certain that the only variable that could cause the DV to change is the IV. The potential influence of outside variables, called extraneous variables, must be eliminated. Researchers do this by preventing variables other than the IV from changing during the experiment. Let’s say that in the aggression study the experimenters allowed participants to pick which game to play, and the more aggressive ones played Wulfenstein and less aggressive ones chose Tetris. If a difference were found in aggressive behavior between the two groups in the second game, can we say what caused it? Think about it. The answer is no: With this design, we would not know if prior aggressive tendencies or playing the violent game caused the Wulfenstein group to act more aggressively. By allowing the participants to choose their condition, we have introduced a confound or confusion variable. Confound variables are factors that change along with the independent variable and can complicate a clear assessment of the effects of the IV on the DV. Confound variables are extraneous and undesirable. If we can eliminate them, then we can have more confidence in our results. There are many possible sources of confounds, including some based on participant characteristics and others on features of the situation.

      How do we rule out the participant-based confounds? We do this by ensuring, as much as possible, that the participants in the groups are similar in all relevant ways. In the aggression study, we would want the participants at the beginning of the experiment in the treatment group to be no more or less aggressive than those in the control group. In the priming study, the experimenter has to maximize the likelihood that there were as many “slow” walkers as “fast” walkers in each group. To ensure parity between the groups, the experimenter assigns participants to the groups in a random fashion. Random assignment means that each participant has an equal chance of being assigned to any condition. Random assignment can be done by flipping a coin, pulling numbers out of a hat, or in countless other ways. By randomly placing participants in the two game conditions, the number of previously aggressive participants should be about the same in each. Furthermore, the two groups should have about the same proportion of extroverts, artists, fast walkers, and chemistry majors. With random assignment, we can be reasonably confident that differences on the dependent variable between the groups could only have been caused by the independent variable. In all relevant ways, the groups are otherwise essentially the same.

      Although random assignment can minimize the likelihood of participant-based undesirable effects, it may not prevent situational factors from inadvertently influencing experimental outcomes and confounding the research. What if, say, all participants who played Wulfenstein did so in a very hot, humid room, whereas those playing Tetris sat in a cool, dry room? Since research has shown that heat can increase aggression (Anderson, 2001), we would be unable to determine whether increased aggression in the Wulfenstein condition was because of the game or room temperature. Therefore, we must carefully design our experiments to prevent the unwanted influence of such situational variables.

      In summary, all of us create informal or lay theories about the causes of social behavior that are usually based on casual observations and/or anecdotes. Although occasionally accurate, these lay theories must be scientifically tested. Correlational studies can show that two variables are related to one another, but controlled, randomized experiments are necessary to demonstrate cause and effect. Researchers design experiments that manipulate at least one variable, called the IV, and measure its potential influence on at least one other variable, called the DV. In order to help prevent confounds, experimenters randomly assign participants to condition. In addition, researchers control situational features so that all participants are tested in nearly identical circumstances, with the only differences being the level of the IV as determined by the experimenter.

      Theory: Set of interrelated