Assignment: MEASURING THE DEPENDENT VARIABLE
In previous chapters, we have discussed various aspects of measuring variables, including reliability, validity, and reactivity of measures; observational methods; and the development of self-report measures for questionnaires and interviews. In this section, we will focus on measurement considerations that are particularly relevant to experimental research.
Types of Measures
The dependent variable in most experiments is one of three general types: self-report, behavioral, or physiological.
Self-report measures Self-reports can be used to measure attitudes, liking for someone, judgments about someone’s personality characteristics, intended behaviors, emotional states, attributions about why someone performed well or poorly on a task, confidence in one’s judgments, and many other aspects of human thought and behavior. Rating scales with descriptive anchors Page 187(endpoints) are most commonly used. For example, Funk and Todorov (2013) studied the impact of a facial tattoo on impressions of a man accused of assault. The man, Jack, had punched another man in a bar following a dispute over a spilled drink. A description of the incident included a photo of Jack with or without a facial tattoo. After viewing the photo and reading the description, subjects responded to several questions on a 7-point scale that included the following:
How likely is it that Jack is guilty?
Behavioral measures Behavioral measures are direct observations of behaviors. As with self-reports, measurements of an almost endless number of behaviors are possible. Sometimes, the researcher may record whether a given behavior occurs—for example, whether an individual responds to a request for help, makes an error on a test, or chooses to engage in one activity rather than another. Often, the researcher must decide whether to record the number of times a behavior occurs in a given time period—the rate of a behavior; how quickly a response occurs after a stimulus—a reaction time; or how long a behavior lasts—a measure of duration. The decision about which aspect of behavior to measure depends on which is most theoretically relevant for the study of a particular problem or which measure logically follows from the independent variable manipulation. MEASURING THE DEPENDENT VARIABLE.
As an example, consider a study on eating behavior while viewing a food-related or nature television program (Bodenlos & Wormuth, 2013). Participants had access to chocolate-covered candies, cheese curls, and carrots that were weighed before and after the session. More candy was consumed during the food-related program; there were no differences for the other two foods.
Sometimes the behavioral measure is not an actual behavior but a behavioral intention or choice. Recall the study described in Chapter 3 in which subjects decided how much hot sauce another subject would have to consume later in the study (Vasquez, Pederson, Bushman, Kelley, Demeestere, & Miller, 2013). They did not actually pour the hot sauce but they did commit to an action rather than simply indicate their feelings about the other subject. MEASURING THE DEPENDENT VARIABLE.
Physiological measures Physiological measures are recordings of responses of the body. Many such measurements are available; examples include the galvanic skin response (GSR), electromyogram (EMG), and electroencephalogram (EEG). The GSR is a measure of general emotional arousal and anxiety; it measures the electrical conductance of the skin, which changes when sweating occurs. The EMG measures muscle tension and is frequently used as a measure of tension or stress. The EEG is a measure of electrical activity of brain cells. It can be used to record general brain arousal as a response to different situations, such as activity in certain parts of the brain as learning occurs or brain activity during different stages of sleep.
Page 188The GSR, EMG, and EEG have long been used as physiological indicators of important psychological variables. Many other physiological measures are available, including temperature, heart rate, and analysis of blood or urine (see Cacioppo & Tassinary, 1990). In recent years, magnetic resonance imaging (MRI) has become an increasingly important tool for researchers in behavioral neuroscience. An MRI provides an image of an individual’s brain structure. It allows scientists to compare the brain structure of individuals with a particular condition (e.g., a cognitive impairment, schizophrenia, or attention deficit hyperactivity disorder) with the brain structure of people without the condition. In addition, a functional MRI (fMRI) allows researchers to scan areas of the brain while a research participant performs a physical or cognitive task. The data provide evidence for what brain processes are involved in these tasks. For example, a researcher can see which areas of the brain are most active when performing different memory tasks. In one study using fMRI, elderly adults with higher levels of education not only performed better on memory tasks than their less educated peers, but they also used areas of their frontal cortex that were not used by other elderly and younger individuals (Springer, McIntosh, Winocur, & Grady, 2005).
Although it is convenient to describe single dependent variables, most studies include more than one dependent measure. One reason to use multiple measures stems from the fact that a variable can be measured in a variety of concrete ways (recall the discussion of operational definitions in Chapter 4). In a study on the effects of an employee wellness program on health, the researchers might measure self-reported fatigue, stress, physical activity, and eating habits along with physical measures of blood pressure, blood sugar, cholesterol, and weight (cf. Clark et al, 2011). If the independent variable has the same effect on several measures of the same dependent variable, our confidence in the results is increased. It is also useful to know whether the same independent variable affects some measures but not others. For example, an independent variable designed to affect liking might have an effect on some measures of liking (e.g., desirability as a person to work with) but not others (e.g., desirability as a dating partner). Researchers may also be interested in studying the effects of an independent variable on several different behaviors. For example, an experiment on the effects of a new classroom management technique might examine academic performance, interaction rates among classmates, and teacher satisfaction. MEASURING THE DEPENDENT VARIABLE.
When you have more than one dependent measure, the question of order arises. Does it matter which measures are made first? Is it possible that the results for a particular measure will be different if the measure comes earlier rather than later? The issue is similar to the order effects that were discussed in Chapter 8 in the context of repeated measures designs. Perhaps responding to the first measures will somehow affect responses on the later measures, Page 189or perhaps the participants attend more closely to first measures than to later measures. There are two possible ways of responding to this issue. If it appears that the problem is serious, the order of presenting the measures can be counterbalanced using the techniques described in Chapter 8. Often there are no indications from previous research that order is a serious problem. In this case, the prudent response is to present the most important measures first and the less important ones later. With this approach, order will not be a problem in interpreting the results on the most important dependent variables. Even though order may be a potential problem for some of the measures, the overall impact on the study is minimized.
Making multiple measurements in a single experiment is valuable when it is feasible to do so. However, it may be necessary to conduct a separate series of experiments to explore the effects of an independent variable on various behaviors.
Sensitivity of the Dependent Variable
The dependent variable should be sensitive enough to detect differences between groups. A measure of liking that asks, “Do you like this person?” with only a simple “yes” or “no” response alternative is less sensitive than one that asks, “How much do you like this person?” on a 5- or 7-point scale. With the first measure, people may tend to be nice and say yes even if they have some negative feelings about the person. The second measure allows for a gradation of liking; such a scale would make it easier to detect differences in amount of liking. MEASURING THE DEPENDENT VARIABLE.
The issue of sensitivity is particularly important when measuring human performance. Memory can be measured using recall, recognition, or reaction time; cognitive task performance might be measured by examining speed or number of errors during a proofreading task; physical performance can be measured through various motor tasks. Such tasks vary in their difficulty. Sometimes a task is so easy that everyone does well regardless of the conditions that are manipulated by the independent variable. This results in what is called a ceiling effect—the independent variable appears to have no effect on the dependent measure only because participants quickly reach the maximum performance level. The opposite problem occurs when a task is so difficult that hardly anyone can perform well; this is called a floor effect.
The need to consider sensitivity of measures is nicely illustrated in the Freedman et al. (1971) study of crowding mentioned in Chapter 4. The study examined the effect of crowding on various measures of cognitive task performance and found that crowding did not impair performance. You could conclude that crowding has no effect on performance; however, it is also possible that the measures were either too easy or too difficult to detect an effect of crowding. In fact, subsequent research showed that the tasks may have been too easy; when subjects perform complex cognitive tasks in laboratory or natural settings, crowding does result in lower performance (Bruins & Barber, 2000; Paulus, Annis, Seta, Schkade, & Matthews, 1976). MEASURING THE DEPENDENT VARIABLE.
Cost of Measures
Another consideration is cost—some measures may be more costly than others. Paper-and-pencil self-report measures are generally inexpensive; measures that require trained observers or elaborate equipment can become quite costly. A researcher studying nonverbal behavior, for example, might have to use a video camera to record each participant’s behaviors in a situation. Two or more observers would then have to view the tapes to code behaviors such as eye contact, smiling, or self-touching (two observers are needed to ensure that the observations are reliable). Thus, there would be expenses for both equipment and personnel. Physiological recording devices are also expensive. Researchers need resources from the university or outside agencies to carry out such research.
The basic experimental design has two groups: in the simplest case, an experimental group that receives the treatment and a control group that does not. Use of a control group makes it possible to eliminate a variety of alternative explanations for the results, thus improving internal validity. Sometimes additional control procedures may be necessary to address other types of alternative explanations. Two general control issues concern expectancies on the part of both the participants in the experiment and the experimenters. MEASURING THE DEPENDENT VARIABLE.
Controlling for Participant Expectations
Demand characteristics We noted previously that experimenters generally do not wish to inform participants about the specific hypotheses being studied or the exact purpose of the research. The reason for this lies in the problem of demand characteristics (Orne, 1962), which is any feature of an experiment that might inform participants of the purpose of the study. The concern is that when participants form expectations about the hypothesis of the study, they will then do whatever is necessary to confirm the hypothesis. For example, if you were studying the relationship between political orientation and homophobia, participants might figure out the hypothesis and behave according to what they think you want, rather than according to their true selves.
One way to control for demand characteristics is to use deception—to make participants think that the experiment is studying one thing when actually it is studying something else. The experimenter may devise elaborate cover stories to explain the purpose of the study and to disguise what is really being studied. The researcher may also attempt to disguise the dependent variable by using an unobtrusive measure or by placing the measure among a set of unrelated filler items on a questionnaire. Another approach is simply to assess whether demand characteristics are a problem by asking participants about their perceptions of the purpose of the research. It may be that participants do Page 191not have an accurate view of the purpose of the study; or if some individuals do guess the hypotheses of the study, their data may be analyzed separately.
Demand characteristics may be eliminated when people are not aware that an experiment is taking place or that their behavior is being observed. Thus, experiments conducted in field settings and observational research in which the observer is concealed or unobtrusive measures are used minimize the problem of demand characteristics. MEASURING THE DEPENDENT VARIABLE.
Placebo groups A special kind of participant expectation arises in research on the effects of drugs. Consider an experiment that is investigating whether a drug such as Prozac reduces depression. One group of people diagnosed as depressive receives the drug and the other group receives nothing. Now suppose that the drug group shows an improvement. We do not know whether the improvement was caused by the properties of the drug or by the participants’ expectations about the effect of the drug—what is called a placebo effect. In other words, just administering a pill or an injection may be sufficient to cause an observed improvement in behavior. To control for this possibility, a placebo group can be added. Participants in the placebo group receive a pill or injection containing an inert, harmless substance; they do not receive the drug given to members of the experimental group. If the improvement results from the active properties of the drug, the participants in the experimental group should show greater improvement than those in the placebo group. If the placebo group improves as much as the experimental group, all improvement could be caused by a placebo effect.
Sometimes, participants’ expectations are the primary focus of an investigation. For example, Marlatt and Rohsenow (1980) conducted research to determine which behavioral effects of alcohol are due to alcohol itself as opposed to the psychological impact of believing one is drinking alcohol. The experimental design to examine these effects had four groups: (1) expect no alcohol–receive no alcohol, (2) expect no alcohol–receive alcohol, (3) expect alcohol–receive no alcohol, and (4) expect alcohol–receive alcohol. This design is called a balanced placebo design. Marlatt and Rohsenow’s research suggests that the belief that one has consumed alcohol is a more important determinant of behavior than the alcohol itself. That is, people who believed they had consumed alcohol (Groups 3 and 4) behaved very similarly, although those in Group 3 were not actually given any alcohol. MEASURING THE DEPENDENT VARIABLE.
In some areas of research, the use of placebo control groups has ethical implications. Suppose you are studying a treatment that does have a positive effect on people (for example, by reducing migraine headaches or alleviating symptoms of depression). It is important to use careful experimental procedures to make sure that the treatment does have an impact and that alternative explanations for the effect, including a placebo effect, are eliminated. However, it is also important to help those people who are in the control conditions; this aligns with the concept of beneficence that was covered in Chapter 3. Thus, participants in the control conditions must be given the treatment as soon as Page 192they have completed their part in the study in order to maximize the benefits of participation.
Placebo effects are real and must receive serious study in many areas of research. A great deal of current research and debate focuses on the extent to which any beneficial effects of antidepressant medications such as Prozac are due to placebo effects (e.g., Kirsch, 2010; Wampold, Minami, Tierney, Baskin, & Bhati, 2005).
Controlling for Experimenter Expectations
Experimenters are usually aware of the purpose of the study and thus may develop expectations about how participants should respond. These expectations can in turn bias the results. This general problem is called experimenter bias or expectancy effects (Rosenthal, 1966, 1967, 1969).
Expectancy effects may occur whenever the experimenter knows which condition the participants are in. There are two potential sources of experimenter bias. First, the experimenter might unintentionally treat participants differently in the various conditions of the study. For example, certain words might be emphasized when reading instructions to one group but not the other, or the experimenter might smile more when interacting with people in one of the conditions. The second source of bias can occur when experimenters record the behaviors of the participants; there may be subtle differences in the way the experimenter interprets and records the behaviors. MEASURING THE DEPENDENT VARIABLE.
Research on expectancy effects Expectancy effects have been studied in a variety of ways. Perhaps the earliest demonstration of the problem is the case of Clever Hans, a horse with alleged mathematical and other abilities that attracted the attention of Europeans in the early 20th century (Rosenthal, 1967). The owner of the horse posed questions to Hans who in turn would provide answers by tapping his hoof (e.g., a question of “what is two times five” would be followed by ten taps). Pfungst (1911) later showed that Hans was actually responding to barely detectable cues provided by the person asking the question. The person would look at the hoof as Hans started to tap and then changed to look at Hans as the correct answer was about to be given. Hans was responding to these head and eye movements that went undetected by observers. MEASURING THE DEPENDENT VARIABLE.
If a clever horse can respond to subtle cues, it is reasonable to suppose that clever humans can too. In fact, research has shown that experimenter expectancies can be communicated to humans by both verbal and nonverbal means (Duncan, Rosenberg, & Finklestein, 1969; Jones & Cooper, 1971). An example of more systematic research on expectancy effects is a study by Rosenthal (1966). In this experiment, graduate students trained rats that were described as coming from either “maze bright” or “maze dull” genetic strains. The animals actually came from the same strain and had been randomly assigned to the bright and dull categories; however, the “bright” rats did perform better than the “dull” rats. Subtle differences in the ways the students treated the rats Page 193or recorded their behavior must have caused this result. A generalization of this particular finding is called “teacher expectancy.” Research has shown that telling a teacher that a pupil will bloom intellectually over the next year results in an increase in the pupil’s IQ score (Rosenthal & Jacobson, 1968). In short, teachers’ expectations can influence students’ performance.
The problem of expectations influencing ratings of behavior is nicely illustrated in an experiment by Langer and Abelson (1974). Clinical psychologists were shown a videotape of an interview in which the person interviewed was described as either an applicant for a job or a patient; in reality, all saw the same tape. The psychologists later rated the person as more “disturbed” when they thought the person was a patient than when the person was described as a job applicant. MEASURING THE DEPENDENT VARIABLE.
Solutions to the expectancy problem Clearly, experimenter expectations can influence the outcomes of research investigations. How can this problem be solved? Fortunately, there are a number of ways to minimize expectancy effects. First, experimenters should be well trained and should practice behaving consistently with all participants. The benefit of training was illustrated in the Langer and Abelson study with clinical psychologists. The bias of rating the “patient” as disturbed was much less among behavior-oriented therapists than among traditional ones. Presumably, the training of the behavior-oriented therapists led them to focus more on the actual behavior of the person, so they were less influenced by expectations stemming from the label of “patient.”
Another solution is to run all conditions simultaneously so that the experimenter’s behavior is the same for all participants. This solution is feasible only under certain circumstances, however, such as when the study can be carried out with the use of printed materials or the experimenter’s instructions to participants are the same for everyone.
Expectancy effects are also minimized when the procedures are automated. As noted previously, it may be possible to manipulate independent variables and record responses using computers; with automated procedures, the experimenter’s expectations are less likely to influence the results.
A final solution is to use experimenters who are unaware of the hypothesis being investigated. In these cases, the person conducting the study or making observations is blind regarding what is being studied or which condition the participant is in. This procedure originated in drug research using placebo groups. In a single-blind experiment, the participant is unaware of whether a placebo or the actual drug is being administered; in a double-blind experiment, neither the participant nor the experimenter knows whether the placebo or actual treatment is being given. To use a procedure in which the experimenter or observer is unaware of either the hypothesis or the group the participant is in, you must hire other people to conduct the experiment and make observations. MEASURING THE DEPENDENT VARIABLE.
Because researchers are aware of the problem of expectancy effects, solutions such as the ones just described are usually incorporated into the procedures of the study. The procedures used in scientific research must be precisely Page 194defined so they can be replicated by others. This allows other researchers to build on previous research. If a study does have a potential problem of expectancy effects, researchers are bound to notice and will attempt to replicate the experiment with procedures that control for them. It is also a self-correcting mechanism that ensures that methodological flaws will be discovered. The importance of replication will be discussed further in Chapter 14.
So far, we have discussed several of the factors that a researcher considers when planning a study. Actually conducting the study and analyzing the results is a time-consuming process. Before beginning the research, the investigator wants to be as sure as possible that everything will be done right. And once the study has been designed, there are some additional procedures that will improve it.
After putting considerable thought into planning the study, the researcher writes a research proposal. The proposal will include a literature review that provides a background for the study. The intent is to clearly explain why the research is being done—what questions the research is designed to answer. The details of the procedures that will be used to test the idea are then given. The plans for analysis of the data are also provided. A research proposal is very similar to the introduction and method sections of a journal article. Such proposals must be included in applications for research grants; ethics review committees require some type of proposal as well (see Chapter 3 for more information on Institutional Review Boards). MEASURING THE DEPENDENT VARIABLE.
Preparing a proposal is a good idea in planning any research project because simply putting your thoughts on paper helps organize and systematize ideas. In addition, you can show the proposal to friends, colleagues, professors, and other interested parties who can provide useful feedback about the adequacy of your procedures. They may see problems that you did not recognize, or they may offer ways of improving the study.
When the researcher has finally decided on all the specific aspects of the procedure, it is possible to conduct a pilot study in which the researcher does a trial run with a small number of participants. The pilot study will reveal whether participants understand the instructions, whether the total experimental setting seems plausible, whether any confusing questions are being asked, and so on.
Sometimes participants in the pilot study are questioned in detail about the experience following the experiment. Another method is to use the think aloud protocol (described in Chapter 7) in which the participants in the pilot study Page 195are instructed to verbalize their thoughts about everything that is happening during the study. Such procedures provide the researcher with an opportunity to make any necessary changes in the procedure before doing the entire study. Also, a pilot study allows the experimenters who are collecting the data to become comfortable with their roles and to standardize their procedures. MEASURING THE DEPENDENT VARIABLE.
A manipulation check is an attempt to directly measure whether the independent variable manipulation has the intended effect on the participants. Manipulation checks provide evidence for the construct validity of the manipulation (construct validity was discussed in Chapter 4). If you are manipulating anxiety, for example, a manipulation check will tell you whether participants in the high-anxiety group really were more anxious than those in the low-anxiety condition. The manipulation check might involve a self-report of anxiety, a behavioral measure (such as number of arm and hand movements), or a physiological measure. All manipulation checks, then, ask whether the independent variable manipulation was in fact a successful operationalization of the conceptual variable being studied. Consider, for example, a manipulation of physical attractiveness as an independent variable. In an experiment, participants respond to someone who is supposed to be perceived as attractive or unattractive. The manipulation check in this case would determine whether participants do rate the highly attractive person as more physically attractive.
Manipulation checks are particularly useful in the pilot study to decide whether the independent variable manipulation is in fact having the intended effect. If the independent variable is not effective, the procedures can be changed. However, it is also important to conduct a manipulation check in the actual experiment. Because a manipulation check in the actual experiment might distract participants or inform them about the purpose of the experiment, it is usually wise to position the administration of the manipulation check measure near the end of the experiment; in most cases, this would be after measuring the dependent variables and prior to the debriefing session. MEASURING THE DEPENDENT VARIABLE.
A manipulation check has two advantages. First, if the check shows that your manipulation was not effective, you have saved the expense of running the actual experiment. You can turn your attention to changing the manipulation to make it more effective. For instance, if the manipulation check shows that neither the low- nor the high-anxiety group was very anxious, you could change your procedures to increase the anxiety in the high-anxiety condition.
Second, a manipulation check is advantageous if you get nonsignificant results—that is, if the results indicate that no relationship exists between the independent and dependent variables. A manipulation check can identify whether the nonsignificant results are due to a problem in manipulating the independent variable. If your manipulation is not successful, it is only reasonable that you will obtain nonsignificant results. If both groups are equally anxious after you manipulate anxiety, anxiety cannot have any effect on the dependent measure. Page 196What if the check shows that the manipulation was successful, but you still get nonsignificant results? Then you know at least that the results were not due to a problem with the manipulation; the reason for not finding a relationship lies elsewhere. Perhaps you had a poor dependent measure, or perhaps there really is no relationship between the variables.
The importance of debriefing was discussed in Chapter 3 in the context of ethical considerations. After all the data are collected, a debriefing session is usually held. This is an opportunity for the researcher to interact with the participants to discuss the ethical and educational implications of the study. MEASURING THE DEPENDENT VARIABLE.
The debriefing session can also provide an opportunity to learn more about what participants were thinking during the experiment. Researchers can ask participants what they believed to be the purpose of the experiment, how they interpreted the independent variable manipulation, and what they were thinking when they responded to the dependent measures. Such information can prove useful in interpreting the results and planning future studies.
Finally, researchers may ask the participants to refrain from discussing the study with others. Such requests are typically made when more people will be participating and they may talk with one another in classes or residence halls. People who have already participated are aware of the general purposes and procedures; it is important that these individuals not provide expectancies about the study to potential future participants.
ANALYZING AND INTERPRETING RESULTS