14
When you were a child, did it seem like you were always getting into trouble, being a nuisance to your parents and teachers? Or was there someone else in your school or neighborhood who fits this description? When this type of behavior is extreme, it is given a clinical diagnosis of attention deficit hyperactivity disorder (ADHD), or hyperactivity. One topic discussed in this chapter is the outcome of research on experimental approaches to the treatment of ADHD, a condition, incidentally, that has been frequently found predictive of later criminality.
An important type of research has not yet been carefully examined. It is called experimental research, and can be distinguished from other forms of research in that it involves directly manipulating one or more of the variables being studied. In all other research designs, researchers merely observe variations in variables and record what they see. (In nonexperimental research, any manipulation of variables is carried out in the analysis phase of a study, not during the time that the data are being collected.) This chapter demonstrates that experimentation, while often difficult, can accomplish scientific objectives that other types of research can never fully achieve.
Recall that in chapter 4, the adage ‘‘correlation does not equal causation’’ was discussed. What this adage means is that simply observing a relationship between two variables does not warrant the conclusion that one variable caused the other. This is true even when one variable clearly precedes the other in time. For instance, even if Variables A and B are strongly correlated and Variable A occurs in childhood while Variable B occurs in adolescence, one still cannot confidently conclude that Variable A causes Variable B. Why? Perhaps, some third variable occurring in infancy caused both Variable A and Variable B.
In experimental research, a scientist systematically manipulates a suspected ‘‘causal variable,’’ and then monitors the suspected ‘‘effect variable’’ to determine if it changes as expected (Stroebe & Diehl, 1991). These type of experiments are conducted more in the discipline of psychology than criminology or criminal justice simply because crime doesn’t happen in a lab and so it is often difficult to study it there. Studies of aggression, self control, and other concepts that are related to crime and criminality, however, can sometimes be researched using controlled experimentation. Before describing specific types of experimental designs, some terminology needs to be presented.
BASIC EXPERIMENTAL TERMINOLOGY
Among the terms that are fundamental to experimental research are independent variable and dependent variable. An independent variable is one that is under the control of a researcher in the sense that the researcher can manipulate the level at which subjects are exposed to the variable. An independent variable is also called a treatment variable, especially in experiments conducted in clinical settings.
A dependent variable is one that may be altered as a result of changes made in the independent variable (Howell, 1989, 10). A helpful way to remember the distinction between these two terms is to state that the independent variable is considered a possible cause, and the dependent variable is considered a possible effect. Thus the independent variable must occur first.
In most (but not all) experiments, there are at least two groups of subjects. One group, called the experimental group (or experimentals), is exposed to the independent variable to an unusual degree. The other group, called the control group (or controls), is exposed to the independent variable to a ‘‘normal’’ degree (or, in some cases, not at all).
Distinguishing between a control group and an experimental group is sometimes arbitrary, but usually it is obvious. As an example, several experimental studies have been undertaken to determine if exposing males to various forms of sexually explicit mass media causes them to be aggressive toward women (including the commission of sexual assault) (for reviews see Check & Malamuth, 1986; Donnerstein et al., 1987; Ferguson & Hartley, 2009; Zillmann, 1984). In most of these experiments, the independent variable is the exposure to sexually explicit material, and the dependent variable is aggression toward women. The experimental subjects are those shown the sexually explicit material, and the control subjects are those exposed to some ‘‘neutral’’ material such as a nature film, or perhaps those who saw no film at all.
Measuring dependent variables can be difficult. In the case of these experiments, the researchers often simply measure aggression toward women with questionnaires containing questions about how subjects think they would respond under various hypothetical situations (Malamuth et al., 1980). In other studies, the dependent variable has involved asking male subjects about how severely men should be punished for committing various aggressive acts toward women (e.g., Donnerstein et al., 1984). The dependent variable in other studies involved confronting male subjects with a female lab assistant who hurled insults at the subjects after they had viewed either a sexually explicit video or some control video to see if the sexual material increased the likelihood that the males would respond aggressively to the affront (Malamuth & Ceniti, 1986).
In these types of experiments, researchers obviously walk a thin line between making their studies as realistic as possible, while causing no significant harm to subjects or to anyone having contact with the subjects. Issues regarding the protection of human subjects are discussed in chapter 16.
The concept of experimental control is important in most scientific experiments. It refers to a process of assigning subjects to the experimental and control groups at random. For example, if one hundred subjects are considered eligible for exposure to an independent variable, experimental control can be achieved by randomly picking fifty of them to be in the experimental group, leaving the remaining fifty for the control group. Experimental control helps to ensure that the groups to be compared are equivalent with respect to all variables except the one being manipulated. Of course, the more subjects in each randomly selected group, the more confidence one can have that the groups are equivalent to one another.
Another term that is important in experimental research is time frame. This refers to the length of time that the dependent variable is monitored for evidence of any independent variable effects. In most of the experiments with sexually explicit videos, for example, the time frame is the length of time between exposure to the sexually explicit material (or some control material), and when aggressive acts toward women are expected to occur.
The length of the time frame must be relevant to the dependent variable being investigated. Thus, if a particular experiment involves the use of a fast-acting medication on some aspect of alertness, the time frame might be in minutes. But if the independent variable is a prison rehabilitation program designed to reduce recidivism, the time frame would often be years following prison release.
A final term that is frequently used in experimental research is random assignment. Do not confuse random assignment with random selection, the latter being a concept discussed in chapter 7 in connection with surveying. Random assignment means that once a pool of eligible subjects for an experiment have been identified, each subject from the pool is chosen at random to be either an experimental subject or a control subject (Levin, 1993, 379). In most types of experimental designs, the ability to identify causes comes from the ability to take a large sample of subjects and randomly assign them to two (or sometimes more) groups. By doing so, a researcher can be confident that whatever differences appear between two groups in terms of the dependent variable can be attributed to the differential exposure to the independent variable and not differences in the groups themselves.
MAIN TYPES OF EXPERIMENTAL DESIGNS
Although there are a wide variety of experimental designs, most can be subsumed under one of the following six categories: the classical design, the after-only design, the before-after no control group design, the cross-over design, the Solomon four-group design, and the factorial design. Simple diagrams are used to illustrate the essential features of each design. In describing each design, the focus is on their simplest form, although more elaborate versions of each design are often utilized.
Classical Experimental Design
In its simplest form, the classical design involves subjects being randomly assigned to one of two groups, the experimental group and the control group. To ensure that subjects are randomly assigned, a coin flip might be used for each pair of subjects. For an experiment to fit the minimum requirements of a classical design, there must also be at least two time frames.
Figure 14.1 offers a graphic representation of a classical experimental design. For the moment, pay attention only to the four squares outlined with the solid lines. Along the top of figure 14.1, two time frames are represented: T1 and T2. Along the side, an experimental group (GE) and a control group (GC) are represented. Inside each of the four cells, observations on the dependent variable (DV) can be represented. For the experimental group, the imposition of the independent variable is represented by the extra dark line. (Sometimes the independent variable is imposed throughout T2, rather than simply between T1 and T2).
The dashed lines at the bottom and right side of figure 14.1 represent the fact that the classical design can be made more elaborate by adding time frames or groups of subjects beyond those that are minimally required. A second experimental group could be added to a classical experimental design by having two degrees of experimental exposure to the independent variable. For example, if the independent variable were a particular drug, one experimental group might get the drug once a day, whereas the second experimental group might receive it twice a day; the control group, of course, would not receive the drug at all.
Figure 14.2 illustrates a classical experimental design that involved one more time frame than the two that are minimally required. This study comes from a program designed to reduce the number of women who smoke during pregnancy (Burling et al., 1991). The time frames in this study represent the three times that a group of pregnant women who smoked visited a prenatal health clinic. Between the first and the second visit, half the mothers (chosen at random) were sent a one-page letter that briefly informed them of the possible health risks of smoking, particularly to the fetus they were carrying (e.g., low birth weight).
As one can see from viewing figure 14.2, receipt of this letter slightly lowered the number of women who continued to smoke during pregnancy. However, it is also worth noting that some of the expectant mothers who were not sent the letter stopped smoking later on in pregnancy. Statistically, the difference between the two groups of expectant mothers during the second time frame (88.4 percent versus 98.6 percent) were significantly different.
After-Only Experimental Design
As in a classical design, an after-only experimental design must have at least two groups of subjects. However, in the after-only design, no observations occur prior to the time the independent variable is imposed on the experimental group. This is illustrated in figure 14.3. Again, as in a classical design, subjects are almost always assigned to the experimental and control groups at random. There are various reasons why researchers choose not to include pretest observations of the dependent variable in an experiment. One is that repeated measurement of the dependent variable sometimes alters this variable’s subsequent levels in some unacceptable ways. Another reason might pertain to the urgency with which the researcher wants the results; an after-only design often requires half the time to conduct as a classical design.
Examples of after-only experimental designs come from clinical studies of the effects of methylphenidate (Ritalin) and similar drugs on children with ADHD. According to one survey in the United States, these drugs were being given to about 6 percent of the nation’s elementary school children (predominantly boys) in the 1980s (Safer & Krager, 1988).
For children with clinically significant ADHD symptoms, several studies indicate that methylphenidate improves classroom behavior (reviewed by Carlson & Bunner, 1993; Gadow et al., 1992; Whalen et al., 1987) as well as relationships with family members and peers (Barkley et al., 1985; Cunningham et al., 1985). In such experiments, researchers often randomly assigned a pool of ADHD children into an experimental group and a control group and used ratings by teachers or parents of the children’s subsequent behavior. (To make sure teachers and parents are not biased in their assessments by knowing which children were receiving methylphenidate, the control children are typically given a placebo in what is known as a double-blind experiment, as is discussed later in this chapter.) Experiments involving the treatment of ADHD children with methylphenidate have consistently found that about 80 percent of the experimentals exhibit significantly diminished ADHD symptoms compared with controls (Carlson et al., 1992; Whalen et al., 1987).
Before-After No Control Group Design
Although before-after no control group design is a cumbersome term, it precisely characterizes the nature of a widely used experimental design in social science. It is the only experimental design in which there is only one group of subjects, all of whom will be exposed to the independent variable. This design is also unique in that it can even be used with a single subject, although it is certainly preferable from a scientific standpoint to use more than one subject (Barlow & Hersen, 1984; Hersen & Barlow, 1976).
The basic structure of a before-after no control group design is shown in figure 14.4. Note that it consists of a minimum of two periods of time during which a dependent variable is measured. These two time frames are typically separated by the imposition of an independent variable, although in some studies utilizing this design, the researcher imposes the independent variable throughout either one (but never both) of the time frames.
To illustrate the before-after no control group design, let us return to the use of methylphenidate for the treatment of ADHD symptoms. A Norwegian study assessed the effects of withholding treatment for ADHD children who had been taking methylphenidate for several months and substituting a placebo (an inert substance) for three weeks instead (Zeiner, 1999). Twenty-one boys were assessed by both their teachers and parents using rating scales of hyperactive and defiant behavior.
The main results are presented in figure 14.5. The figure shows that at least in school, the hyperactive and defiant behavior of the boys while they were regularly taking methylphenidate was significantly less than during the three weeks they were taking placebos.
Because there is no control group in a before-after no control group design, it may seem poorly equipped to provide convincing evidence of the effect of an independent variable on a dependent variable. This is certainly true regarding the most basic form of the design—as shown in figure 14.4 and illustrated by figure 14.5. However, there are more elaborate forms of before-after no control group experimental design that can provide very convincing evidence of experimental effects.
Consider an experiment undertaken to increase the use of seat belts among government employees in Florida (Thyer & Geller, 1987). In vehicles being used for highway travel, the following sign was either absent or prominently displayed on the dashboard: Safety Belt Use Required in this Vehicle. As you can see in figure 14.6, during the two two-week periods when the sign was present (the ‘‘imposition’’ weeks), seat-belt use nearly doubled that of the two two-week periods when it was absent.
The experiment summarized in figure 14.6 is an example of a type of before-after no control group design sometimes referred to as a reversal design, since it shifts back and forth between the control and the experimental conditions. Another name for the same type of design is an ABAB design (Thyer et al., 1987), which is the single most widely used reversal design (Eichelman, 1992, 491).
Had the experiment illustrated in figure 14.6 stopped after the second week, the researcher could not have been very confident that the increase in seat-belt usage from the first to the second week was really attributable to the sign that was displayed. Along these lines is a well-known phenomenon in clinical research called spontaneous recovery, which refers to the tendency for people to seek treatment when the symptoms are worst. Because symptoms for most ailments fluctuate in severity, there is a reasonable probability that after patients come to a clinician, the symptoms will be less severe than at the point of first appearance. While spontaneous recovery is not of relevance to the seat-belt usage experiment, the concept reinforces the idea that one should be cautious in interpreting before-after no control group experiments with no reversals. Before-after no control group experimental designs are especially common in clinical research. These designs are unique in even being applicable to experiments with just one subject (Benjamin et al., 1983; Monette et al., 1986, 265).
Cross-Over Design
Another type of experimental design that is fairly often used in social science research is called a cross-over design. Its key feature is that, at some point in time, all participants serve as both experimental subjects and control subjects.
One study that used a cross-over design was undertaken to determine if social interactions, especially with an attractive, flirtatious female, would result in elevated testosterone levels among male subjects (Dabbs, Ruback, & Besch, 1987). Testosterone levels were measured in small samples of saliva, which the subjects ‘‘donated’’ by spitting into a vile. Three measures of the dependent variable were taken. The first was a baseline measure obtained at the time subjects entered the laboratory. The second measure was taken after half the subjects had been isolated in a room with another male subject while the other half of the subjects had been isolated in a room with an attractive and friendly female confederate. The third measure of the dependent variable was obtained about fifteen minutes after the males switched rooms, with the controls becoming the experimentals and vice versa. In this study, as in most cross-over designs, each subject served under both experimental and control conditions. The experiment revealed that testosterone levels rose significantly after both forms of social interaction, but especially after interacting with the female confederates.
Another study using a cross-over design was conducted by one of the authors along with a colleague (Ellis & Mathis, 1985). We were interested in retesting earlier conclusions that students can learn as well from watching lectures on television as they can watching the same lectures presented live in a classroom (e.g., Alexander, 1962; Berger, 1962; Thorman, 1975; Thorman & Amb, 1974). In our study, students were randomly selected to watch half the lectures in person and the other half of the lectures live on a television monitor in an adjoining room. All students took four exams separated by four sets of six lectures. After each set of lectures, the students who had watched the lectures in person were switched to the room containing the television monitor, and those who had watched the lectures on television were switched to the in-person lectures.
Did the students learn as well from being in the presence of the instructor as when they watched the instructor in an adjoining room on a television monitor? Results from the experiment are shown in figure 14.7. It revealed that there were no significant differences in overall test scores, with the exception of the first test when the students watching the lectures in person did better. After discussing the issue with the students in the television room, the reason for this initial difference was traced to poor sound quality in the video room. After making adjustments to the sound system, test performance for the two groups was virtually identical.
Notice that in figure 14.7, the groups are simply identified as ‘‘Group 1’’ and ‘‘Group 2’’ rather than ‘‘experimental’’ and ‘‘controls.’’ The reason is that in this and other cross-over experimental designs, all subjects are both experimental and control subjects at varying times throughout the study.
Solomon Four-Group Design
The Solomon four-group design is a specialized and rarely-used design. It is discussed in courses on social and behavioral science research methods primarily to illustrate the extreme lengths to which a researcher can go to achieve experimental control over all causal contingencies.
As shown in figure 14.8, the Solomon four-group design consists of an after-only experimental design attached to a classical design. This design requires assigning subjects to a minimum of four groups, two experimental groups and two control groups. Both experimental groups receive the same level of exposure to the independent variable, and both control groups are denied anything beyond the normal level of exposure to the independent variable. The difference, however, is that one experimental group and one control group are pretested, while the other two groups are not. The purpose of this elaborate set of procedures is to see if repeated measurement of the dependent variable interacts in some way with exposure to the independent variable.
Experimental Factorial Design
Factorial designs are common, especially in the fields of psychology and education. They can be either experimental or nonexperimental, but the focus here is on the former. The purpose of experimental factorial designs is to assess the possibility of two or more independent variables on a dependent variable. Note that all of the other experimental designs have focused on just one independent variable. Factorial designs not only allow researchers to look for the effects of two or more independent variables acting alone, but also interactively. Interactive effects refer to the effects that two or more independent variables may have on a dependent variable that neither independent variable has by itself. Interactive effects are often important to establish because in the ‘‘real world,’’ it is common for many variables to be acting in concert rather than one at a time.
Interactive effects can be of two types. One type is an augmenting effect, where the effects of one variable enhance the effects of the other. The other type is an inhibiting effect, where increases in one variable tend to neutralize or at least dampen the effects of the other variable.
Most experimental factorial designs involve only one time frame, so in this respect, they are a simple design. However, as shown in figure 14.9, factorial designs have a minimum of four groups of subjects. This set of minimum conditions constitutes what is called a two-by-two factorial design. Figure 14.9 illustrates this point with two hypothetical variables (Variables A and B), each of which is presented in a control level and an experimental level.
There are many ways to elaborate on the basic two-by-two factorial design. A factorial experiment in which subjects are exposed to two levels of one variable and three levels of another variable is called a two-by-three factorial design. Some factorial designs have two levels of three different variables, making it a two-bytwo-by-two factorial design. To visualize this design, imagine figure 14.9 having a depth dimension as well as height and length dimensions. The number of cells (and therefore groups) for a two-by-two-by-two design is eight.
Since most research about crime or the criminal justice system assumes that there are multiple variables involved in outcomes—in other words, that there are interactive effects—these factorial designs have been undertaken in criminal justice research as well. An example of a factorial design study about the criminal justice system is one undertaken to determine if two variables affected judgments about the appropriate sentence for persons found guilty in mock trials (Sigall & Ostrove, 1975). Each subject in the study was given one of six portfolios containing a description of two different crimes and a photograph of four different defendants. Three independent variables were manipulated in the portfolios: the physical attractiveness of the defendant (attractive or unattractive), the type of crime (burglary or swindling), and the gender of the defendant (male or female). Results were assessed separately according to the sex of the subjects, effectively adding a fourth independent variable. Thus, the design can be thought of as a two-by-two-by-two-by-two factorial design.
The main findings from the study was that regardless of the sex of the subject, attractive female defendants were sentenced less harshly than other defendants in burglary cases, but not in swindling cases. A relatively recent replication of this study found that for both types of offenses, attractive female defendants were less harshly sentenced (Wuensch et al., 1991).
One last note: Factorial experimental designs can be combined with other basic designs already discussed. No attempt is made here to describe these elaborate combined designs, but be aware that they are possible.
PITFALLS WITH HUMAN EXPERIMENTATION
Experimentation, especially with human subjects, entails some special problems that need to be addressed before concluding this chapter. These problems have both procedural and ethical ramifications.
Expectancy (Placebo) Effect
Any person who agrees to take part in an experiment is bound to be curious as to what will happen. Subtle and often unintentional clues may affect those expectations, which in turn may have major effects on the results of the experiment.
Some of the best evidence for the expectancy effect has come from experiments involving drugs that are actually inert substances (called placebos) like sugar pills. For this reason, the expectancy effect is also referred to as the placebo effect, especially in experiments with drugs.
A special set of experimental procedures that are often used with human subjects to control for any possible placebo effect is called a double-blind experiment. Using this procedure, subjects who agree to take part in a double-blind experiment are randomly assigned to the experimental and control groups and neither they, nor persons administering the treatment to them, know whether they are receiving the real treatment or a placebo. Only a ‘‘third party’’ connected with the experiment has this knowledge, and extra safeguards are taken to ensure that this third party does not divulge this information before the experiment has been completed.
Double-blind procedures are used most often in studies involving the testing of drugs, especially drugs that may influence behavior, mood, and cognition, such as drugs used in the treatment of various forms of mental illness and behavior problems (e.g., Marini et al., 1976; Taylor et al., 1990). For example, a double-blind, placebo-controlled study was used to assess the effectiveness of lithium (a very light metal consumed in capsule form) for treating children with conduct disorder (e.g., Silva et al., 1992). In this study, neither the children being treated nor their attending physicians knew whether the child was receiving lithium or a placebo.
To give another example, an experiment was carried out to determine whether giving young children vitamin and mineral supplements might improve their scores on intelligence tests (Benton & Cook, 1991). Forty-four subjects were randomly assigned to experimental and control groups. Neither they, their parents, nor their teachers knew which students were given capsules containing the vita-mins and minerals and which were being given placebos. After six weeks, the children who had taken the capsules containing the vitamin and mineral supplements registered a 7.6-point gain on the test, compared with a 1.7-point loss by those who had taken the placebos. Although this difference was a statistically significant effect, a similar experiment reported in the same year found no significant effect (Todman et al., 1991).
Before leaving this topic, it should be mentioned that in some clinical experiments, double-blind procedures fail. For instance, in a study of patients with persistent panic attacks, nearly all of the patients and attending physicians were able to correctly guess which patients were receiving a new experimental drug rather than a placebo because of the drug’s obvious effectiveness (Margraf et al., 1991).
The Hawthorne Effect
In the late 1920s, industrial consultants were asked to recommend ways of improving work productivity in an electronics assembly plant in a district of Chicago known as Hawthorne (Diaper, 1990). The consultants explored various possibilities, and, through a series of experiments (using a before-after no control group design), they eventually concluded that improved lighting in the plant promoted worker productivity. Months later, the consultants recommended further increases in factory luminescence and found additional productivity improvements. However, they eventually lowered the luminescence and found that productivity further improved.
What could explain these findings? Operators of the plant finally came to the conclusion that the extra attention and concern given the workers during the experiment were more crucial to increasing worker productivity than was the lighting. As a result, whenever extraneous factors such as extra attention and reinforcement in an experiment with people have effects on behavior, it is referred to as the Hawthorne effect (Jones, 1990).
Writers have embellished the Hawthorne experiments over the years, in part because careful analyses of the findings were never published (Diaper, 1990). Because it was poorly documented, several researchers have seriously questioned the reality of a Hawthorne effect (e.g., Adair, 1984; Jones, 1992; Parsons, 1978), with one report going so far as to calling it a ‘‘phantom phenomenon’’ (Granberg & Holmberg, 1992, 241). Whether real or mythical, the concept of a Hawthorne effect is a useful vehicle from which to emphasize the importance of always being on guard against subtle, unintended factors inadvertently confounding experiments involving human subjects (Gillespie, 1991).
SHORTCOMINGS OF EXPERIMENTAL RESEARCH
Experiments are conducted with one overriding objective: to identify cause-effect relationships between variables. The need for identifying such relationships in science is of major importance in theory development. As mentioned in chapter 4, when nonexperimental research is all that one has for understanding how variables are causally related, one is walking on thin scientific ice, even with the aid of multivariate statistics. With well-designed experimental evidence, however, one can have a great deal of confidence that causation is responsible for differences that are found.
Despite the strengths of experimental research for identifying causes, it is important to be aware of some of the shortcomings associated with experimental research. Most problems fall into the following four interrelated categories: ethics, time and expense, lack of realism, and the fact that many questions are simply beyond the reach of experimentation.
Ethics
Ethical considerations are not confined to experimental research, but there are some unique and perplexing ethical dilemmas that must be confronted before conducting experiments, especially involving human subjects. These are spelled out in more detail in chapter 16, but they include issues surrounding the deception and informed consent of experimental subjects, and legal and moral responsibility for harm suffered by subjects. Another ethical dilemma often accompanying human experimentation involves the fact that subjects must be randomly denied treatment in order to demonstrate the treatment’s effectiveness on other subjects.
Time and Expense
Experimental research is more time-consuming and expensive than nonexperimental research. Imagine how much time and money would be required to follow subjects over the course of most experiments when compared with the time and expense associated with administering a questionnaire.
Realism
Many experiments lack essential elements of realism. This has been one criticism of many criminal justice experiments, and most critics question the conclusions of these experiments. They argue that experiments are happening in a controlled setting and not in the real world where variables often cannot be controlled or manipulated to a great degree. For example, researchers have investigated whether time spent playing violent video games fostered real-world violence, and most have come to an affirmative answer (Sherry, 2001). However, the experimental designs used have largely limited measurement of ‘‘violence’’ to short-term, contrived effects that can be observed in laboratory settings, and then debrief subjects to avoid the risk of any long-term effects. Critics of these experiments have expressed doubts that these short-term laboratory effects have any real bearing on any long-term effects that mass media may have in regard to ‘‘real-life’’ violence. The question is, should researchers ever conduct experiments designed to look for real-world effects, such as violence of an actual criminal nature?
Feasibility
Finally, many questions are simply beyond the reach of experimental research quite apart from considerations of ethics. Examples can be found in questions pertaining to the collective behavior of millions of people. Although it is sometimes possible to ‘‘simulate’’ such behavioral processes in a laboratory on a small scale, researchers are often left wondering how permanent these small-scale simulations really are to the collective behavior in naturalistic settings.
SUMMARY
This chapter has focused attention on the concept of experimentation in social science research. The overriding distinguishing feature of experimental research is that it involves direct manipulation of an independent variable in order to assess the effects of such manipulation on a dependent variable.
Six types of experimental designs were identified and discussed: the classical design, the after-only design, the before-after no control group design, the crossover design, the Solomon four-group design, and the factorial design. A brief description of each design follows.
In the classical design, subjects are divided into at least two groups, an experimental group and a control group. There is also a minimum of two time frames in a classical experiment. One time frame occurs prior to the experimental group receiving exposure to an unusual level of the independent variable, and the other occurs after it has received this exposure. To construct an after-only design, the classical design is cut in half by removing the initial time frame. Provided a researcher has a large group of subjects randomly assigned to each of the experimental and control conditions, this removal makes the after-only design only slightly weaker than its classical counterpart. The only purpose served by the initial time frame in the classical design is to give assurance that the experimental and control groups are indeed statistically equivalent prior to exposing the experimental group to an unusual degree of exposure to the independent variable. In an after-only design, no such assurance is obtained.
The before-after no control group design can also be considered a truncated version of the classical design; in this case, the control group has been removed. With this design, a researcher can strengthen the experiment by adding more than the minimum two time frames. Within these additional time frames, the independent variable can be imposed, withheld, and then reimposed several times to confirm that it is the independent variable that is responsible for significant increases or decreases on the dependent variable.
The cross-over experimental design is unique in that two or more groups of subjects are used, with each group receiving unusual exposure to the independent variable. The sequence of this exposure, however, is different for each group.
The Solomon four-group design essentially affixes an after-only design onto a classical design. This is a rarely used design, but it illustrates the length to which a researcher can go to verify the effects an independent variable has on a dependent variable apart from any effects caused by repeated measurement of the independent variable.
Finally, in the factorial experiment design two or more independent variables are manipulated at a time. In this way it is possible to discover not only their separate effects on one or more dependent variables, but also whether there are interactive effects.
The values and the drawbacks of scientific experimentation were discussed and illustrated near the end of this chapter. Also discussed were the concept of the expectancy effect and the Hawthorne effect. The expectancy (or placebo) effect refers to the tendency for subjects to anticipate results from experimentation, often based on subtle clues that the researcher may neither intend nor be aware of. If these clues are in any way associated with the independent variables being manipulated, they can confound an experiment in ways that may be very misleading to a researcher. A set of experimental procedures specially designed to reduce the expectancy effect is called double-blind experimentation. The Hawthorne effect refers to the unintended effects that researchers may have on the results of an experiment because of the extra attention and reinforcement given to subjects.
Although there are a number of pitfalls that scientists need to watch for, the potential benefits from experimental research are difficult to exaggerate. Whenever a researcher’s primary interest is causal relationships between variables, he or she should give serious thought to experimental research. In the next chapter, research designs that simulate experimental research are explored.
SUGGESTED READINGS
Hersen, M., & Barlow, D. H. (1976). Single case experimental designs: Strategies for studying behavior change. New York: Pergamon. (An easy-to-read book on how to conduct clinical experiments on various aspects of behavior.)
Leavitt, F. (1991). Researcher method for behavioral scientists. Dubuque, IA: Wm. C. Brown. (This text provides a useful overview of social and behavioral science research, with particularly clear coverage devoted to experimental designs).
Solso, R. L., & Johnson, H. H. (1989). An introduction to experimental design in psychology: A case approach (4th ed.). New York: Harper & Row. (This text contains numerous examples of different experimental research designs using both humans and nonhuman animals.)