You know from day-to-day experience that people can sug-gest many causes for the same outcome. Psychologists face this same problem when they try to make exact claims about cau-sality. To overcome causal ambiguity, researchers use experi-mental methods: They manipulate an independent variable to look for an effect on a dependent variable. The goal of this method is to make strong causal claims about the impact of one variable on the other. In this section, let’s review the problem of alternative explanations and some steps researchers take to counter the problem.
The Challenge to Objectivity When psychologists test a hypothesis, they most often have in mind an explanation for way in which different individuals answer this question—their
beliefs in free will versus determinism—has an impact on how they behave (Vohs & Schooler, 2008). The researchers reasoned that people who are guided by a worldview of determinism would feel less personal responsibility for bad behavior because they’d consider it out of their control. To test this hypothesis, the researchers gave students an opportunity to cheat!
Figure 2.2 presents important aspects of the experiment.
The researchers recruited roughly 120 college undergradu-ates to serve as participants. The independent variable for the study was participants’ beliefs in free will versus determinism.
To manipulate this variable, the researchers presented students with a series of 15 statements and asked them to think about each statement for one minute. As you might expect, those statements were different for the free will and determinism conditions. Figure 2.2 provides examples.
To test their hypothesis, the researchers needed to pro-vide the students with an opportunity to cheat. During the ex-periment, the students attempted to answer 15 problems from Graduate Record Examination (GRE) practice tests. They could earn $1 for each correct answer. Participants scored their own answers in the absence of the experimenter. That provided the context for cheating: The experimenter would never know if a participant paid him- or herself more money than was due.
The dependent variable for the experiment was the amount of money participants paid themselves.
Figure 2.2 provides the results of the experiment. To deter-mine how average students would actually score on the 15 GRE questions, the researchers had an extra condition in which they scored participants’ performance themselves to see how much money the students would earn. The bar labeled “Baseline experimenter-scored” provides that information. As you can see from the other two bars in Figure 2.2, the independent vari-able had the effect on the dependent varivari-able that the research-ers expected. Those students who had been prompted to take the perspective of determinism paid themselves about $4 more than those students who focused on free will. Because of the experimenter-scored baseline—which shows free-will students at the same level as experimenter-scored students—we can infer that the determinism students were cheating. Take a moment to think about other ways in which you might operationalize the experimental variables, to test the same hypothesis by other means. You might, for example, want to measure cheating in some other fashion, to show that the results generalize to other Is violent behavior caused by viewing violence on television?
How could you find out?
The researchers measure the dependentvariable The researchers
manipulate the independent variable
FIGURE 2.2 Elements of an Experiment To test their hypotheses, researchers create operational definitions for the independent and dependent variables.
Data from Kathlee D. Vohs and Jonathan W. Schooler, The value of believing in free will: Encouraging a belief in determinism increases cheating, Psychological Science, January 1, 2008, pages 49–54. © 2008 by the Association for Psychological Science.
experimental method Research methodology that involves the manipulation of independent variables to determine their effects on the dependent variables.
The Process of Research 25 In a psychological research setting, a placebo effect has oc-curred whenever a behavioral response is influenced by a per-son’s expectation of what to do or how to feel rather than by the specific intervention or procedures employed to produce that response. Recall the experiment relating television view-ing to later aggression. Suppose we discovered that experimen-tal participants who hadn’t watched any television at all also showed high levels of aggression. We might conclude that these individuals, by virtue of being put in a situation that allowed them to display aggression, would expect they were supposed to behave aggressively and would go on to do so. Experiment-ers must always be aware that participants change the way they behave simply because they are aware of being observed or tested. For example, participants may feel special about being chosen to take part in a study and thus act differently than they would ordinarily. Such effects can compromise an experi-ment’s results.
The Remedy: Control Procedures Because human and animal behaviors are complex and often have multiple causes, good research design involves anticipating possible confounds and devising strategies for eliminating them.
Similar to defensive strategies in sports, good research de-signs anticipate what the other team might do and make plans to counteract it. Researchers’ strategies are called control procedures—methods that attempt to hold con-stant all variables and conditions other than those related to the hypothesis being tested. In an experiment, instructions, room temperature, tasks, the way the researcher is dressed, time allotted, the way the responses are recorded, and many other details of the situation must be similar for all par-ticipants, to ensure that their experience is the same. The only differences in participants’ experiences should be those introduced by the independent variable. Let’s look at rem-edies for the specific confounding variables, expectancy and placebo effects.
Imagine, for example, that you enriched the aggression ex-periment to include a treatment group that watched comedy programs. You’d want to be careful not to treat your comedy and violence participants in different ways based on your ex-pectations. Thus, in your experiment, we would want the re-search assistant who greeted the participants and later assessed their aggression to be unaware of whether they had watched a violent program or a comedy: We would keep the research assistant blind to the assignment of participants to conditions.
In the best circumstances, bias can be eliminated by keeping both experimental assistants and participants blind to which why change in the independent variable should affect the
de-pendent variable in a particular way. For example, you might predict, and demonstrate experimentally, that the viewing of television violence leads to high levels of aggression. But how can you know that it was precisely the viewing of violence that produced aggression? To make the strongest possible case for their hypotheses, psychologists must be very sensitive to the existence of possible alternative explanations. The more alter-native explanations there might be for a given result, the less confidence there is that the initial hypothesis is accurate. When something other than what an experimenter purposely intro-duces into a research setting changes a participant’s behavior and adds confusion to the interpretation of the data, it is called a confounding variable. When the real cause of some observed behavioral effect is confounded, the experimenter’s interpreta-tion of the data is put at risk. Suppose, for example, that violent television scenes are louder and involve more movement than do most nonviolent scenes. In that case, the superficial and violent aspects of the scenes are confounded. The researcher is unable to specify which factor uniquely produces aggressive behavior.
Although each different experimental method potentially gives rise to a unique set of alternative explanations, there are two types of confounds that apply to almost all experiments, which researchers call expectancy effects and placebo effects.
Unintentional expectancy effects occur when a researcher or observer subtly communicates to the research participants the behaviors he or she expects to find, thereby producing the de-sired reaction. Under these circumstances, the experimenter’s expectations, rather than the independent variable, actually help trigger the observed reactions.
confounding variable A stimulus other than the variable an
experimenter explicitly introduces into a research setting that affects a participant’s behavior.
expectancy effect Result that occurs when a researcher or observer subtly communicates to participants the kind of behavior he or she expects to find, thereby creating that expected reaction.
placebo effect A change in behavior in the absence of an experimental manipulation.
Featured Study
In an experiment, 12 students were given groups of rats that were going to be trained to run a maze (Rosenthal & Fode, 1963). Half of the students were told their rats were from a special maze-bright breed. The other students were told their rats were bred to be maze-dull. As you might guess, their rats were actually all the same. Nonetheless, the stu-dents’ results corresponded with their expectations for their rats. The rats labeled bright were found to be much better learners than those that had been labeled as dull.
control procedure Consistent procedure for giving instructions, scoring responses, and holding all other variables constant except those being systematically varied.
How do you suppose the students communicated their expectations to their rats? Do you see why you should worry even more about expectancy effects when an experiment is carried out within species—with a human experimenter and human participants? Expectation effects distort the content of discovery.
A placebo effect occurs when experimental participants change their behavior in the absence of any kind of experimen-tal manipulation. This concept originated in medicine to ac-count for cases in which a patient’s health improved after he or she had received medication that was chemically inert or a treatment that was nonspecific. The placebo effect refers to an improvement in health or well-being related to the individual’s belief that the treatment will be effective. Some treatments with no genuine medical effects have been shown to produce good or excellent outcomes for patients on whom they were used (Colloca & Miller, 2011).
Chapter 2 Research Methods in Psychology 26
matches the overall characteristics of the population with re-spect, for example, to the distribution of males and females, racial and ethnic groups, and so on. For example, if your study of children’s lying included only boys, we wouldn’t consider that a representative sample of the full population of 4- and 6-year-olds. To achieve a representative sample, researchers of-ten use the procedure of random sampling, which means that every member of a population has an equal likelihood of par-ticipating in the experiment. (In the Statstical Supplement that follows this chapter, we describe the procedures researchers use to determine whether experimental results can be generalized beyond a particular sample. Please read the Supplement in con-junction with this chapter.)
Another type of experimental design—a within-subjects design—uses each participant as his or her own control. For ex-ample, each participant might experience more than one level of the independent variable. Or, the behavior of an experimen-tal participant before getting the treatment might be compared with behavior after. Consider an experiment that examined the accuracy of people’s judgments about future exercise.
participants get which treatment. This technique is called a double-blind control. In our prospective aggression experi-ment, we couldn’t keep participants from knowing whether they had watched comedy or violence. However, we would take great care to ensure that they couldn’t guess that our later anal-yses would focus on their subsequent aggression.
To account for placebo effects, researchers generally in-clude an experimental condition in which the treatment is not administered. This is a placebo control. Placebo controls fall into the general category of controls by which experimenters assure themselves they are making appropriate comparisons.
Suppose you see a late-night TV commercial that celebrates the herbal supplement ginkgo biloba as an answer to all your memory problems. What might you expect if you buy a sup-ply of ginkgo and take it weekly? One study demonstrated that university students who took ginkgo every morning for six weeks did, in fact, show improvements in their performance on cognitive tasks (Elsabagh et al., 2005). On one task, people were asked to view a series of 20 pictures on a computer screen, name them, and later recall those names. The participants were 14 percent better at this task after six weeks of ginkgo. How-ever, participants who took a placebo—a pill with no active ingredients—also improved by 14 percent. The placebo control suggests that improvement on the task was the result of prac-tice from the initial session. The data from control conditions provide an important baseline against which the experimental effect is evaluated.
The Remedy: Research Designs To implement control conditions, researchers make decisions about what type of re-search design best suits their goals. In some rere-search designs, which are referred to as between-subjects designs, different groups of participants are randomly assigned, by chance pro-cedures, to an experimental condition (exposed to one or more experimental treatments) or to a control condition (not ex-posed to an experimental treatment). Random assignment is one of the major steps researchers take to eliminate confound-ing variables that relate to individual differences among poten-tial research participants. This is the procedure you’d want to use for the aggression experiment. The random assignment to experimental and control conditions makes it quite likely that the two groups will be similar in important ways at the start of an experiment because each participant has the same probabil-ity of being in a treatment condition as in a control condition.
We shouldn’t have to worry, for example, that everyone in the experimental group loves violent television and everyone in the control group hates it. Random assignment should mix both types of people together in each group. If outcome differ-ences are found between conditions, we can be more confident that the differences were caused by a treatment or intervention rather than by preexisting differences.
Researchers also try to approximate randomness in the way they bring participants into the laboratory. Suppose you would like to test the hypothesis that 6-year-old children are more likely to lie than 4-year-old children. At the end of your experiment, you’d like your conclusions to apply to the whole population of 4-year-olds and 6-year-olds. However, you can bring only a very small subset—a sample—of the world’s 4- and 6-year-olds into your laboratory. Typically, psychology experiments use from 20 to 100 participants. How should you choose your group of children? Researchers attempt to con-struct a representative sample, which is a sample that closely
double-blind control An experimental technique in which biased expectations of experimenters are eliminated by keeping both participants and experimental assistants unaware of which participants have received which treatment.
placebo control An experimental condition in which treatment is not administered; it is used in cases where a placebo effect might occur.
between-subjects design A research design in which different groups of participants are randomly assigned to experimental conditions or to control conditions.
random assignment A procedure by which participants have an equal likelihood of being assigned to any condition within an experiment.
experimental group A group in an experiment that is exposed to a treatment or experiences a manipulation of the independent variable.
control group A group in an experiment that is not exposed to a treatment or does not experience a manipulation of the independent variable.
populationThe entire set of individuals to which generalizations will be made based on an experimental sample.
sample A subset of a population selected as participants in an experiment.
representative sample A subset of a population that closely matches the overall characteristics of the population with respect to the distribution of males and females, racial and ethnic groups, and so on.
random sampling A procedure that ensures that every member of a population has an equal likelihood of participating in an experiment.
Featured Study
Suppose you are contemplating a visit to the gym. You’re probably more likely to go if you think that you’re going to enjoy the workout. But how accurate are your predictions about your future enjoyment? A team of researchers tested the hypothesis that people habitually underestimate the ex-tent to which they’ll enjoy their exercise (Ruby et al., 2011).
To test their hypothesis, the researchers approached people who attended fitness classes regularly. Before a class be-gan, they asked the participants to predict how much they
within-subjects design A research design that uses each participant as his or her own control; for example, the behavior of an experimental participant before receiving treatment might be compared to his or her behavior after receiving treatment.
The Process of Research 27 children who would not be. The next section turns to a type of research method that often addresses these concerns.