The first stage involves identifying the basic data. This has proven a considerable challenge in psychology, although behavior analysts long ago decided that behavior would be a sufficient source. The second stage focuses on expressing relations among these data. This effort culminates in well-established laws or summaries of consistent relationships. The science of behavior has made considerable headway on this task. The third stage goes

48

beyond observable relationships to unobservable concepts. For instance, Galileo studied the relationship between the position of a ball on an inclined plane and the elapsed time since its release, and developed the unobservable concept of acceleration. (He did not, however, attribute acceleration to a force inherent in the ball.) Skinner (1961) emphasized the importance of this stage by pointing out that terms like wants, faculties, attitudes, drives, ideas, interests, and capacities should be developed as proper scientific concepts at this level.

In other words, theories are built on concepts that refer to unobservable features of the observable subject matter. These unobservable concepts must not be simply invented, however. In order for a theory to work, it must emerge from a sound database describing functional relations in the subject matter. Ultimately, explanation in the natural sciences is made in terms of such theories (Lee, 1981).

Challenges like these result from the fact that psychology is burdened with cultural preconceptions about human nature. Not surprisingly, a definition of behavior based on sound scientific findings fails to accommodate many of these preconceptions. Some readers may even feel that the definition leaves no place for cherished human qualities. It may help to consider that these omissions are only a matter of the difference between how we have learned to talk about behavior in everyday life and what is really there.

For instance, when we speak of someone as having a particular attitude, we imply that attitudes really exist as physical things. The fallacy in this assumption can be uncovered by considering what prompts us to describe someone as having a certain attitude. In fact, it is observing this person doing something or emitting certain behaviors under certain conditions that encourages us to use the attitudinal terms we have been taught by our verbal community. At best, attitudes are only crude ways of summarizing many instances of behavior. They may be convenient for everyday conversation, but they are not useful for researchers because they lead to scientific dead ends.

The problem, then, is not that the definition of behavior leaves out real aspects of human nature but that the culture has created so many fictional qualities. It is only how we have learned to talk about behavior that implies that these qualities actually exist nonphysically inside us. In fact, when researchers study mental activity, they almost always measure either behavior or physiological events because that is all there is to measure. They must then infer some relationship between their measures of these physical phenomena and the mental events of interest. The resulting subject matter is therefore a collection of these kinds of culturally based assumptions. This sets psychology and the social sciences apart from sciences that investigate phenomena whose existence is the basis for, rather than the result of, experimental inquiry.

Box 2.6

Parsimony

Even though it is not often spoken of, parsimony is one of the most respected attitudes of science. In a word, it means to be stingy. Scientists are parsimonious in their preference for exhausting simple and well-established explanations of phenomena before turning to complex and less-well-understood explanations. Scientists follow this strategy because they have learned over the years that it makes science more efficient. It helps to avoid wild goose chases and blind alleys, which are always a risk in scientific research. Instead of offering fanciful explanations of some event, scientists cautiously insist on trying to explain the event in terms of laws that they already understand fairly well because the odds of successful resolution are usually better than with more far out notions.

This approach urges us to explain the observable facts of behavior with reference to variables in the physical world, which the natural sciences understand pretty well, before inventing a nonphysical world of the psyche, which certainly goes beyond the laws of nature. We need to remember the principle of parsimony when we theorize about human qualities and their explanation.

49

Learning to talk about behavior, especially human behavior, without using familiar everyday terms is difficult at first. The correct terms seem awkward and insufficient, and something often seems left out. The richness with which we talk about behavior in ordinary language is clearly missing. After all, we have had a lifetime of informal training from our verbal community in how to talk about human behavior. It may be some consolation to realize that students new to other sciences such as biology, physics, and chemistry typically face the same challenges. Unfortunately, the colloquial language that works so well for poets contrasts with the need for a precise scientific vocabulary that is devoid of surplus meaning and limited to real events. With a scientific language that follows rather than leads discovery, behavioral researchers are uncovering the kind of richness and complexity that scientists long ago revealed in other natural phenomena.

Methodological Consequences

As the chapters in this book unfold, it will become clear that the primary importance of this definition of behavior lies in how it guides researchers and practitioners to work with this subject matter. Almost every facet of the methods described in this book is partly based on the implications of this definition. In fact, these methods have emerged from what science has already taught us about behavior. We know, for example, that behavior occurs only at the level of individuals, that it has effects on the environment and is in turn affected by the environment, that it is a continuous process, and that what has happened with a particular behavior in the past may continue to influence it in the present. Of course, these are only a few of the most basic facts about behavior. Over the years, the accumulating research literature has revealed a wealth of general principles and complex details (see, for example, Catania, 2007).

When we look to an evolving understanding of behavior for guidance about the best ways to study and manage it, a number of experimental practices seem important. For instance, in order to capture all of the fundamental qualities of behavior, we need to measure the behavior of each participant separately. Furthermore, we need to keep each individuals data separate as we analyze their performance. As we have already suggested, combining observations of different participants into group averages makes it difficult to see any differences in how interventions affected the behavior of each individual.

What we know about behavior also suggests that it is important to measure a target behavior a number of times under each phase of an experiment or a clinical intervention. Any behavior is influenced by its continuously evolving history, and the more recent the history, the greater its impact. This means that we can get a clear picture of any particular environmental condition only if we give each participant enough exposure to that condition to be sure that the effects we see on the target behavior are due primarily to that condition and not to prior experiences.

The following chapters will explain these and many other methodological practices that have proven to be highly effective for behavioral researchers and practitioners. The scientific bottom line is that natures secrets will be revealed only when we ask the right questions in the right way. We cannot force nature to fit our preconceptions and methods. Those who work with behavior must adapt the general principles of scientific method that have worked so well with other sciences and their technologies to suit the features of behavior.

Box 2.7

Pure versus Quasi-Behavioral Research

The ways in which investigators approach behavior as a scientific subject allow a distinction between pure behavioral research and quasi-behavioral research (Johnston & Pennypacker, 1993b). Pure behavioral research uses methodological practices that preserve the fundamental qualities of behavior in undisturbed and uncontaminated form. Quasi-behavioral research is based on data that originated with observations of behavior but whose methods prevent the data from representing its fundamental qualities fully and without distortion or contamination.

Another way of characterizing this distinction is between studies designed to learn about the effects of

50

environmental variables on behavior versus studies that use behavioral data to learn about behavior in a more superficial or informal sense. Pure behavioral research requires identifying relationships at the level of the individual, whereas quasi-behavioral research may permit aggregating the data from different individuals.

For example, in a study analyzing the behavioral effects of a treatment procedure, interest may lie in understanding not just its main effects, but the reasons why those effects occur. Such corollary questions might concern the role of different components of the treatment procedure, whether participant characteristics are relevant to the effects, and the factors that will ensure that the treatment yields consistent effects. These interests would require measuring and analyzing the behavior of individual subjects, which is where the treatments effects actually occur.

On the other hand, an investigators interest might be more superficial, limited to a statistical description of the average outcomes of the treatment. In this case, collating data across participants might be quite permissible. However, this would prevent the investigator from finding a clear answer to more analytical questions that require understanding what is happening at the level of the individual participant. Using behavioral data to answer questions about aggregate outcomes can be quite legitimate. However, the investigator should not suggest that the grouped data represent the effects observed for each individual or argue that the study identified the reasons for the aggregate outcome.

Choosing one kind of behavioral interest over the other should be done at the outset based on the focus of the question. However, this distinction often unintentionally results from the experimenters methodological decisions. A study aimed at discovering behaviorenvironment relationships can become quasi-behavioral by incorporating methodological practices that prevent a clear picture of the fundamental qualities of behavior.

The challenge lies in figuring out when each type of question about behavior is appropriate, as well as being willing to live within the interpretive constraints required by the way the experiment is conducted. This text focuses on pure behavioral research interests. In doing so, it highlights the weaknesses of some tactics (such as aggregating data across individuals) that might otherwise be acceptable in studies aimed at quasi-behavioral interests.

51

Chapter Summary

1. Prescientific conceptions of behavior were invented to explain what people did not understand. They have become part of the culture and tend to discourage scientific investigation and conflict with its results.

2. Social scientific approaches to studying and explaining behavior tend to be based on hypothesized inner processes (prescientific fictions) and to use inferential statistical techniques as a methodological framework.

3. The natural scientific approach to studying behavior involves direct measurement and controlled experimentation and focuses on studying behavior for its own sake, rather than as a way of getting at inner processes.

4. Behavior is a biological phenomenon, but the skin is not a useful dividing line between behavior and biology. The distinction is somewhat arbitrary and not often important.

5. The biological basis for behavior makes it clear that behavior occurs only at the level of individual organisms. This means that there is no such phenomenon as group behavior because groups are not organisms. When people behave in a way influenced by others, each is still behaving individually. Group data cannot reveal behavioral effects.

6. Behavior involves movement, though it can be subtle. The Dead Mans Test is an easy way to determine if movement is involved in a definition of a behavior.

7. Behavior results from interactions between the organism and its environment and is therefore not a part of or possessed by the organism. This means that states of the organism, even real biological conditions, are not behavior, and neither are independent conditions or changes in the environment behavior.

8. Behavior must have some impact on the environment, however small in some cases. This effect is often what is measured in research, although such data must not be the only evidence that the behavior exists.

9. Environment refers to the physical circumstances of the organism, including its own body. 10. Following from these considerations, behavior is defined as that portion of an organisms interaction

with its environment that involves movement of some part of the organism. This definition excludes events that do not involve an organism, events that do not involve the physical environment, events that do not apparently involve movement, movements that do not involve an interactive condition between organism and environment, and events for which the sole evidence is their effect on a measuring instrument.

11. The definition of behavior guides methodological decisions made by researchers and practitioners in ways detailed in subsequent chapters. Research will only be effective to the extent that we adapt our experimental methods to the nature of this subject matter.

52

Text Study Guide

1. The authors pointed out that the natural science approach to behavior involved accepting behavior as a subject matter in its own right. What does this mean?

2. What function does a scientific definition of a phenomenon serve? 3. Why is the skin not a good basis for distinguishing between biology and behavior? 4. What does it mean to say that behavior is an intraorganism phenomenon? 5. What is group behavior? 6. What is the Dead Mans Test? 7. What are some examples of behaviors that do not pass the Dead Mans Test? 8. Behavior is the biological result of the interaction between an organism and its environment. Why does

this suggest that behavior is not a property, attribute, or possession of the organism? 9. Why is it that real or hypothetical states cannot be behavior? 10. Why is waiting for someone not a behavior? 11. Can you think of an example of a behavior that does not affect the environment in some way? If you can,

are you sure that it meets the requirements of the definition of behavior? 12. What would be the problem if the only evidence for the existence of a supposed behavior was from the

measurement process? 13. What does it mean to say that there must be some independent evidence of the existence of a behavior? 14. Why is aggression or being aggressive not usefully referred to as a behavior? 15. Explain how an individual can be part of the environment for his or her own behavior. 16. What are three key implications of the definition of behavior? 17. Why could researchers not develop a comprehensive catalog of all human behaviors? 18. What is a good one-sentence summary of the exclusions implied by the definition of behavior? 19. Describe the influence of cultural language on how we talk about and explain behavior. 20. What are the problems for a science of behavior with hypothesizing mental qualities?

53

Box Study Guide

1. What is wrong with referring to inner causes in attempting to explain behavior? 2. Define and provide examples of neural, psychic, and conceptual inner causes. 3. What is the Dead Mans Test? 4. Explain the reasons why it is useful to conceptualize thinking as a behavior, even though there is no

obvious movement. 5. Describe two reasons why traits are not useful concepts for a science of behavior. 6. What are the three stages of theory building that Skinner described? 7. How does the approach to explaining unobservable concepts taken by the natural sciences differ from that

often found in cognitive psychology? 8. Describe the scientific attitude of parsimony. 9. How is the scientific attitude of parsimony relevant to how behavioral scientists approach a definition of

behavior? 10. Distinguish between pure and quasi-behavioral research. 11. What are the limitations of quasi-behavioral research?

54

Suggested Readings

Baum, W. M. (2005). Understanding behaviorism (2nd ed.). Malden, MS: Blackwell Publishing. Hefferline, R. F., & Keenan, B. (1963). Amplitude-induction gradient of a small-scale (covert) operant. Journal

of the Experimental Analysis of Behavior , 6, 307315. Moore, J. (2008). Conceptual foundations of radical behaviorism. Cornwall-on-Hudson, NY: Sloan Publishing.

[Chapter 4: Behavior as a subject matter in its own right]

55

Discussion Topics

1. Consider phenomena at the boundary between behavior and biology, such as heart beating, burping, etc. Discuss the pros and cons of considering these as behavioral or biological events.

2. Discuss how to study the actual behaviors involved in what might generally be talked about as group phenomena, such as a conversation between two people, three friends loading a sofa on a pickup truck, or a riot.

3. Discuss the list of exclusions resulting from the definition of behavior. Is there anything else the definition seems to exclude? If so, consider why these exclusions might not actually be behavioral events.

56

Exercises

1. Come up with examples of what might usually be referred to as behavior but would not pass the Dead Mans Test.

2. Choose some examples of everyday behaviors that we do not pay much attention to and figure out their specific effects on the environment.

57

Chapter Three Asking Experimental Questions

THE NATURE OF EXPERIMENTAL QUESTIONS

Questions as Verbal Behavior

Sources of Control

Graduate Training

Experimental Literature

Observing Behavior

Existing Resources

Experimental Contingencies

Extra-Experimental Contingencies

Developing and Phrasing Questions

A Strategic Approach

Criteria

Wording

THE FUNCTIONS OF EXPERIMENTAL QUESTIONS

Guiding Experimental Procedures

Strategy

Selecting Participants

Choosing a Response Class

Designing Measurement Procedures

Selecting Independent Variables

Creating Experimental Comparisons

Data Analysis

Guiding Experimental Inferences

Guiding the Science and the Culture

There are doubtless many men whose curiosity about nature is less than their curiosity about the accuracy of their guesses, but it may be noted that science does in fact progress without the aid of this kind of explanatory prophecy.

B. F. Skinner

58

59

The Nature of Experimental Questions

Questions as Verbal Behavior

Before an investigator can actually begin a study, he or she must have some experimental question that can guide methodological decisions and interpretations. Developing such questions is a task that may seem difficult at first, but the more you know about a particular topic of interest, the easier it is to come up with a variety of questions. The real challenge, however, is not just finding a question but figuring out the best question for the topic under investigation.

Asking good questions is a challenge not only for researchers. When practitioners need to change an individuals behavior, there are a number of questions they must answer. These questions come from assessing the nature of behavioral deficits or excesses and designing interventions intended to produce specific behavioral outcomes. For example, assessment procedures usually involve collecting behavioral data under specific conditions in order to determine the nature of the target behavior and the variables influencing it. Sometimes these assessment procedures are explicitly experimental in style (Iwata, Vollmer, & Zarcone, 1990).

Furthermore, treatment programs aimed at changing a target behavior often take the form of informal experiments. Most interventions involve a series of phases in which specific environmental variables are manipulated in order to produce particular changes in a target behavior. Ongoing measurement allows practitioners to evaluate the effects of each adjustment of intervention procedures. (This is illustrated in chapter 1 by the project focusing on increasing walking and decreasing crawling with a girl diagnosed with mental retardation.) Sometimes these adjustments are planned, but when expected outcomes are not obtained, the clinician must figure out what changes may be more effective. At this point, the practitioner is asking the same kind of questions that researchers might entertain (What are the effects of this condition?).

Whether developed by a researcher or a practitioner, perhaps the most important fact about experimental questions is that they are verbal behavior. This means that investigators do not create experimental questions any more than they create other behavior such as eating, driving a car, or watching a baseball game. Furthermore, composing research questions is not the only behavior of interest. This behavior is the result of an extended process involving the reading of experimental and clinical literature, talking with colleagues, and writing and thinking about the topic of interest. Being aware of the behavioral status of experimental questions helps us to consider both the factors that influence question-asking behavior and how these questions affect the way we design, conduct, and interpret experiments.

Sources of Control

Graduate Training. Although it may often be overlooked, graduate training is certainly one of the most powerful influences on question-asking behavior. It is in graduate school that we are metaphorically conceived as scientists or practitioners. Through the contributions of our professors, we gestate as fetal behavioral scientists and practitioners. We are finally thrust out of the ivy-covered womb wearing only a scanty thesis or dissertation, still very much a neonate in a field that is itself still relatively young.

We come to graduate school knowing relatively little about this field and leave with an extensive professional repertoire. Along the way, we learn to restrict much of our culturally learned repertoire about behavior to everyday situations. A specialized, scientific verbal repertoire about behavior gradually replaces this colloquial repertoire in a growing range of professional circumstances.

This academic experience also creates a variety of powerful professional reinforcers that influence our behavior throughout our careers. As a result, for some, laboratory research may be more reinforcing than clinical practice. Working with individuals diagnosed with mental retardation who live in group homes may be more enjoyable than working with children with autism spectrum disorders in school settings. Others may prefer working as supervisors or managers in business environments. Certain issues in the research literature

60

are likely to be much more interesting than others. These preferences often stay with us throughout our careers and lead us in one direction or another as researchers or practitioners.

Experimental Literature. One of the effects of graduate school is to make the research literature of the field a powerful influence over our professional activities, especially on how we develop experimental questions. Of course, any single individual can be familiar with only some areas of the fields literature. Although this unavoidable specialization is a blessed relief to the overachievers among us, it also means that we might not be aware of studies that could be relevant to our interests.

The primary literature concerning a topic should guide the development of research questions. That is, if an investigator is interested in learning more about stimulus equivalence (Sidman, 1994), for example, it is natural that the existing literature on this topic would be the best place to look for guidance about what has already been accomplished and what issues need attention.

However, it is also important to examine broader areas of literature that might suggest useful ways of approaching the issue of interest. Some areas of literature might represent conceptual perspectives and methodological practices that are not well suited to the study of behavioral phenomena. However, behavioral researchers can often find ideas in such literature that merit improved experimental attention. Deguchi (1984), for example, reviewed the literature on observational learning and provided interpretations of this distinctively cognitive literature that suggested interesting opportunities for behavioral researchers.

Finally, familiarity with the fields literature contributes to a researchers overall perspective about its general directions and needs. This sense of where the field is going and where it needs to go are important influences on how a researcher reacts to particular studies in a literature. A newly published experiment does not lead everyone in the same direction, and some directions will probably be more productive than others. The best experimental questions lead to answers that are not only useful for a specific topic but serve the larger interests of the field.

Observing Behavior. Relying too heavily on the archival literature as a source of experimental questions carries its risks. Questions that emerge from published studies naturally tend to pursue the general directions already established in the literature. This is not necessarily bad, but it can lead to the questions gradually becoming more about the literature than about the phenomenon of interest. To the extent that an area of research does not probe all of the interesting possibilities, it is less likely that new questions will move into uncharted areas. This was one of the complaints made by Zeiler (1984) concerning the state of research on schedules of reinforcement.

It is a matter of balance. How researchers develop questions should be influenced not only by the existing literature but by our experiences with the subject matter itself. Behavior is a phenomenon that is always present in accessible and relevant forms. The well-trained behavioral scientist is skilled in observing behavior in any settings. Even observations from everyday life can lead to fruitful research directions. This finding has a long tradition in other sciences, and in his autobiographical volumes B. F. Skinner (1976, 1979, 1983) revealed that his contributions to the study of behavior often had such origins. Many examples of his lifelong observations of behavior in daily life are captured in a sampling of entries from his notebooks (Epstein, 1980).

Existing Resources. It is understandable that researchers tend to develop experimental questions that fit the resources to which they have access. If an investigator is affiliated with an early intervention program for children with autism, it is unlikely that he or she would consider questions that required a laboratory setting and rats as subjects. Budgetary or personnel limitations have a similar constraining effect on scientific curiosities. Of course, this argument can be turned around. Existing resources encourage investigators to search for sound questions that can make good use of participants, settings, and other experimental necessities that are already available.

Whether existing resources facilitate or constrain the kinds of questions asked by each investigator, the real issue is how well the resulting question suits the needs of the literature on a topic. If investigators tend to frame questions based on access to a particular kind of setting or type of participant, could this priority make it less likely that the literature on a topic will grow in needed directions? This possibility suggests that it may be important to balance the influence of existing resources with consideration of the larger needs of the literature in the area of interest. The published studies on a topic can guide the development of new experimental

61

questions by revealing uncertainties that need to be resolved, weaknesses in experiments that should be overcome, issues that have not yet been addressed, and where research interests are going over time. The importance of using the literature as a touchstone in developing experimental questions suggests that questions that can be pursued are not necessarily those that most need to be pursued. The goal is that each research project be as useful as possible, not just for the investigator but for the larger scientific community.

Box 3.1

Research Styles

Scientific research has little in common with the fashion industry, but there are different styles of investigation. Instead of changing seasonally, these styles can be observed year after year in all scientific disciplines. The style of a research project refers to the function that it serves for the experimenter and the effect that it has on the field. A studys style is largely determined by its question and the questions impact on how the researcher designs, conducts, and interprets the experiment.

These research styles sometimes come in complementary pairs. For instance, a thematic research style can be contrasted with an independent research style (see Box 3.2). A thematic study is designed to fit into a predetermined position in a larger research program or area of literature, whereas an independent experiment is not conceived of in the context of any ongoing program or research area. These differences have important consequences for the field (see Box 3.2).

Another contrasting pair of research styles may be labeled demonstration research versus explanatory research. Demonstration research focuses on showing that a certain result is possible, which can be a very important achievement. In contrast, explanatory research strives for an understanding of how and why certain relationships work. Although they can be valuable, demonstration studies leave important questions unanswered because they fail to follow up and show why their results occurred. This is why it is important that any area of literature have many explanatory studies for each demonstration study.

Pilot and in-house research styles often involve methodological short cuts. Pilot research is conducted as a preliminary effort to learn things that will guide a more substantial experimental project. The risk is that the preparatory nature of the effort will justify weak or sloppy methods, which will be an unsound basis for making decisions about how to conduct the planned research project. Research that is only designed for local or in-house use carries the same risk. Sidman (1960) offered a simple rule for avoiding such problems: If the methods are sound, then conclusions are justified; if not, then conclusions may be inappropriate.

A research style that raises a similar concern about what the outcomes mean results from a researcher using experiments to advocate for a particular result. Advocacy research may involve decisions about what to do and how to do it that are biased toward the investigators interests in getting certain results (see Box 3.3). Such studies can often be identified by examining the focus of the question and the related methodological decisions. These choices may facilitate a certain result instead of evaluating the impact of experimental conditions in a neutral manner. The risk is that the findings may differ from those that would have been obtained from a less biased approach.

Finally, although these research styles represent important aspects of how scientists approach their work, these distinctions are not official in any way, and you can look for other styles as you learn more about research literatures. Each of these research styles will be discussed further as they appear in different chapters.

Experimental Contingencies. For many investigators, experimental questions flow from an ongoing program of research. In one way or another, each study may provide the foundation for the next. One study often yields findings that suggest a new question. Sometimes this next question might have been anticipated from the outset, but it is common for the emerging data to point toward the next project. (If researchers could

62

accurately predict their results there would be little reason to do experiments.) It may also be the case that the results of a study are unclear or confusing. For example, it may be that some participants reacted to a condition in one way and others responded in a different way. This kind of result begs for resolution, usually by trying to determine what factors underlie the different outcomes. Finally, conducting a study may reveal problems with measurement procedures or how control and experimental conditions were designed or implemented. These problems may need to be corrected by conducting an improved study.

These outcomes are consequences for the investigators behavior in planning and conducting the experiment. Such contingencies are a valuable influence on asking questions because they encourage thematic research efforts. In contrast to experiments conceived independently of other studies, experiments conducted in a thematic research style fit systematically into the literature on a topic and are designed to serve particular functions for that literature. Although experiments developed more independently of the literature may be quite sound in every respect, studies designed to play a specific role in relation to past and future research projects are likely to be especially valuable for the field (see Box 3.1).

Extra-Experimental Contingencies. Another set of contingencies may have somewhat different effects. These contingencies might be described as extra-experimental because they do not directly concern the details of the experiment or its role in the literature. Instead, these contingencies involve the nonscientific consequences of a studys results. These include professional reputation, grants, patents, consulting contracts, books, tenure and promotion, access to a setting and population, and so forth.

When these issues intrude on the process of developing experimental questions, the resulting findings may not always be in the best interests of the field. They can lead to questions that serve personal interests more than the needs of the literature. More extreme examples of this influence can involve biased interpretations or even fudging the data in some way, although this outcome is rare.

Box 3.2

Thematic versus Independent Research Styles

Researchers who design experiments that are related to other studies in specific ways are using a thematic research style. Any experimental question can lead to thematic research. The key is that the study is part of a coherent program or aims at filling in gaps in the literature. With this approach, the resulting picture is likely to be complete and easily interpretable. The value of thematic research lies in the integration of the resulting data. Thematic findings are likely to have collectively greater utility than the results from the same number of independent experiments addressing the same topic because they were designed to fit together, like pieces of a jigsaw puzzle.

An independent research style results from a researcher conceiving of a study independently of any coordinated research program or even the existing literature. In other words, such studies are not designed to precede or follow other studies in a predetermined way. They often originate from opportunistic situations having to do with the availability of subjects or a setting. Experiments conducted in an independent style may be perfectly worthy studies in every way, and their results may be quite useful. Sometimes their importance may only become clear at a later point in a literatures development. This style becomes a problem, however, when it tends to dominate a literature or field. It can lead to a lack of cohesiveness among studies and contribute to a lack of direction in the literature.

Developing and Phrasing Questions

A Strategic Approach. It is not difficult to come up with experimental questions. Even modest familiarity with an area of literature will suggest possibilities. At times it almost seems as easy as starting a sentence with Which, Why, What, or When and ending it with a question mark. Even such casual questions might be

63

able to generate a workable experiment. The challenge for investigators is more demanding than this. Their objective is to develop a question whose

subsequent experiment will generate data that are more revealing and useful than might be produced by any other question. What this objective implies is that for any area of literature some questions are simply better than others. The best question for a topic results from a process of comparing and refining possible questions in the context of the literature. Eventually, one question will seem to be better than the rest, at least to a particular investigator.

Box 3.3

Advocacy Research

When an experiments question, procedures, and write up suggest that the investigators primary goal was to generate support for a predetermined conclusion, it can be called advocacy research. Because an investigator has preconceptions about the topic of a study and an interest in how the data turn out, does not mean that the experimental procedures and interpretations will be inappropriately biased. However, when the extra-experimental contingencies are powerful, there is naturally a good chance that the way an experiment will be conceived, conducted, and interpreted will be biased in favor of the preferred outcome, even if the researcher is unaware of this impact.

The ways in which a study can be improperly prejudiced are endless, ranging from asking a question that looks only for a desired outcome, to making measurement and design decisions that facilitate preferred results and discourage contrary data. Sometimes the most obvious evidence of an advocacy posture is the tone of the published report.

We all have our prejudices, and it is the rare scientist who can scrupulously ignore all of them. But when the influence of such partialities begins to interfere with conducting a fair experiment, the result can be promoting results that are less than fully true or do not hold up well when others use them. Under these conditions, the researcher has taken on the role of an advocate who defends a cause, not a scientist who searches for understanding.

Along the way, perfectly sound questions may be bypassed in favor of one that is a bit better. This approach certainly does not mean that these good questions should never be dignified by an experiment. The researcher may find some candidate questions worthy of experimental attention at a later point. Different investigators may believe some of these other questions to be the best from their perspective. Each researcher that takes this approach to developing questions will certainly not decide on the same best question.

The objective of figuring out the best question for a particular topic is inevitably a productive and profitable activity. Carefully sifting through existing studies, considering different research issues, and comparing one question to another can only lead to more useful experiments and research literatures. Even though different investigators will arrive at different choices, the field will surely gain from such efforts. Without this kind of investment in developing experimental questions, there is a greater risk that a fields literature will be less useful than it could be. This can be an expensive risk. Any fields scientific resources are finite, and studies that fail to make valuable contributions take the place of those that might have generated more valuable findings.

Criteria. What makes one question better than another? This is a difficult challenge. Great questions tend to be those that go to the heart of an issue and lead to experimental findings that help to resolve the issue. Great questions address gaps in existing knowledge so that their results help to complete a picture. Great questions overcome weaknesses in other studies, correct misunderstandings, or clarify what was previously unclear. Great questions show new possibilities or identify new directions that turn out to be valuable. Perhaps we could summarize these criteria by stating that, above all, great questions lead to results that are importantly useful to other researchers or to practitioners.

What does it take to generate such questions? Certainly the researcher must be thoroughly familiar with the

64

relevant literature. The focus, methods, and findings of published studies should be considered carefully in order to see if there are weaknesses or omissions that might need to be addressed. The relationship among studies should be assessed to determine what is clearly known, what is unclear, and what is only suspected to be true.

Beyond this general advice, it is hard to say what criteria might assure that one question will be better than another. Nevertheless, scientists routinely confront this challenge themselves, as well as make their own assessments of the questions posed by their colleagues. In fact, the soundness of the experimental question is one of the standards that peer reviewers use in deciding whether to fund a research proposal or recommend that an experimental report be accepted for publication in a journal.

Wording. We began this chapter by pointing out that experimental questions are the verbal behavior of investigators. Let us look at this verbal behavior more closely. It turns out that committing a specific question to paper is a revealing and educational task. Sometimes what is revealed is that there really is no question as such. That is, the real interest may lie more in demonstrating an outcome that is already known or at least strongly suspected than in learning something new. Demonstration studies certainly have their place in science, but their focus is often more in the form of a prediction (this will happen when you do that) than a question. For example, in 1949 Paul Fuller, a graduate student at Indiana University, showed that he could systematically increase and decrease a simple motor response by a profoundly retarded individual (Fuller, 1949). This early demonstration of operant conditioning applied to human behavior was important because it showed that profoundly retarded individuals could indeed learn, a conviction not widely held at the time. However, Fullers demonstration did not explain the variables responsible for his results (although they are now well understood).

Another observation is that questions are sometimes not actually asking about behavior. They may instead focus on a theoretical issue or a method of controlling behavior. It is common in the applied literature for studies to assess whether a certain procedure will change behavior in particular ways. Although such questions clearly involve behavior, their primary focus tends to lie with procedures. This emphasis is not improper in any way, but such questions may reveal less about behavior than questions that directly ask about behavior. For example, a study by Fisher, Piazza, Bowman, Hagopian, Owens, and Slevin (1992) compared two approaches for identifying reinforcers for individuals with severe and profound mental retardation. Although their results were clear and valuable, the outcomes revealed more about the procedures being compared than the behavior of individuals with significant disabilities.

Box 3.4

Experimental Questions versus Hypotheses

Although this chapter is about experimental questions, you may be surprised that we hardly mention hypotheses. In a way, a hypothesis is the opposite of an experimental question. Instead of asking a question, a hypothesis states a conjecture or prediction about the outcome of an experiment, which is a fancy way of referring to an educated guess. In this sense, all researchers have hypotheses or make speculations about the outcomes of their experiments. If an investigator had absolutely no suspicions about how a study might turn out, he or she could probably be accused of not knowing the relevant literature very well. Furthermore, even though scientists are supposed to be fairly neutral or at least open to any and all results, it is the rare investigator who does not have some preference for one outcome over another.

Any such speculation can be dressed up and called a hypothesis as long as you understand the risks of doing so. Because even the best investigators do not approach their studies in a completely neutral manner, there is a real possibility that any biases will have an impact on how experiments are designed, conducted, and interpreted. When the investigator replaces the experimental question with a public commitment in the form of a hypothesis, the prediction only increases this risk. Having forecast the experiments outcome, it is understandably tempting to design and conduct the study in a way that

65

confirms the prediction. This certainly does not mean that researchers who state formal hypotheses intentionally bias their experimental decisions, but there are many ways of increasing the chances of getting the right results that have become an accepted part of this tradition. One such tactic is to increase the number of participants in each group to help reach statistical significance.

Actually, researchers do not need to state hypotheses if they are asking questions about nature. When the experimental question simply asks about the relation between independent and dependent variables, there is no scientific reason to make a prediction about what will be learned from the data. Whatever is learned will describe something about that relation that was presumably not known before. Whether it matches anyones expectations has nothing to do with its truth or importance (Sidman, 1960). It is only when the question is about a formal theory or when statistical procedures will be the basis for inferences that a formal hypothesis becomes part of the exercise (for reasons having to do with experimental logic). Otherwise, heed B. F. Skinners warning quoted at the beginning of this chapter. Ask yourself whether you are more interested in learning about nature or the accuracy of your guesses.

These examples suggest that it is important to recognize the subtle themes implied by how questions are phrased. The real interest of the investigator needs to be plainly stated so that no one is likely to be misled about the objectives of a study. For the researcher, this clarity helps to guide the decisions that transform a question into an experiment. It also helps those who read the published study to evaluate the investigators methodological choices in light of the conclusions that the study offers.

Even when the focus of a question is clear, there are still issues about how it is phrased. Consider the following examples:

1. Will a fixed-interval schedule of positive reinforcement generate positively accelerating patterns of responding (scalloping) in verbal human subjects if they are instructed about the schedule contingency?

2. What are the relations between responding by verbal human subjects under a fixed-interval schedule of positive reinforcement and contingency instructions?

At first glance, both questions may appear to pose the same query, and in a general sense they do. However, the differences between how these questions are phrased suggest that their authors may well set up their experiments somewhat differently. If so, there is a good chance that one experiment might be more fruitful than the other. The first question appears to be fairly narrow in its focus. It asks if giving participants instructions about a fixed-interval schedule will produce a certain pattern of responding. The answer will likely be equally narrow: yes or no. When questions are asked in a way that allows them to be answered in such a simple and specific manner, it can be important to see how the investigators limited curiosity affected the construction and interpretation of the experiment. Such questions suggest that the researcher is more interested in one outcome than the other. This is hardly improper, but it could encourage designing the experiment in ways that make the desired results more likely than might be the case had a somewhat different preparation been used. Sometimes this subtle bias is evident in how the investigator approaches data analysis.

In contrast, the second question is open-ended in that it does not imply a specific outcome. It simply asks what will happen to fixed-interval performance when contingency instructions are given, which might suggest that the researcher is interested in whatever results might be obtained. This possibility is further implied by the fact that relations is plural, suggesting that the results could be complex: not simply yes or no. An experimenter that phrased the question in this way might design procedures that probe for the fullest possible description of these relations.

In general, it seems wise to phrase questions in ways that do not anticipate experimental results. What are the relations between is a pretty good representation of this neutral posture. When a questions wording leans toward a particular outcome, this bias, even if innocent, may show up in how the experiment is designed, conducted, and interpreted.

Box 3.5

The Null Hypothesis Game

66

For historical reasons that are beyond the scope of this book, a tradition has evolved in psychology and the social sciences (in marked contrast to the natural sciences) that has a rigid influence on how experimental interests are conceived and phrased. When inferential statistics will be used to interpret experimental data, the investigator must usually identify both a null hypothesis and an alternative hypothesis. The null hypothesis is a routine statement that there is no difference between the performance of the experimental group and the control group. The alternative hypothesis states that there is a difference and that it is due to the only thing that distinguished the two groups: exposure to the independent variable.

If a statistical test of the data shows that there is indeed no substantial difference, then the null hypothesis must be accepted, and the investigator can only argue that there would have been a difference if things had been different. If the test shows that there is a difference, the investigator happily rejects the null hypothesis and turns to the alternative hypothesis as the best explanation of the difference between the groups (given a small likelihood that the difference occurred by chance).

There are a number of problems with this approach to asking experimental questions that will be examined in other chapters. However, in the present context, you should understand that it restricts the type of question asked to the general form, Is there a difference (between experimental and control groups)? The sole permissible answers are inevitably Yes or No, which is both inefficient and a distortion of customary scientific curiosity. The tradition of statistical hypothesis testing requires researchers to play science by the rules of the Twenty Questions Game, in which queries must be phrased in ways that can be answered by yes or no.

In fact, if we look closely enough, there will almost always be a difference. What we really want to know, however, is the nature of the difference and the reasons for it. Instead, this approach encourages the researcher to focus experiments on detecting a particular predicted difference rather than on capturing any features of the relationship between independent and dependent variables. Worse still, this wording encourages researchers to design and conduct the experiment in whatever ways are necessary to get an affirmative conclusion, regardless of its truth in nature.

67

The Functions of Experimental Questions

Guiding Experimental Procedures

Strategy. A good experimental question about behavior is a carefully worded expression of the investigators best judgments about the direction and focus of his or her interests. As such, it should serve as a guide for the many decisions the researcher must make in order to create and manage an experiment that will satisfy those interests. The question may suggest obvious or subtle choices from one topic to another, but almost every methodological decision can be usefully related to the experimental question.

The reason for making sure that the question serves as a touchstone for these decisions is that we expect the experimental procedures to yield data that answer the question. When the design and conduct of an experiment fail to properly reflect the interests of the question, there is a risk that the results will not provide a clear answer. The problem arises when the investigator fails to recognize this risk and proceeds to answer the question based on procedures and data that do not fully suit the question. In this situation, the conclusions may not be entirely correct and therefore may not hold up when others use them.

Consider, for example, an experimental question that asks about the effectiveness of procedures for training students to use good study skills. If the investigator chooses to measure only test performance to assess the possible effects of training in particular study skills, the data may not clearly answer the question. Although we might expect changes in study skills to result in changes in test performance, these test scores may not clearly reveal the effects of the training procedures. One reason for this limitation is that test scores will not provide evidence that the trained study skills were actually used as designed. Test performance is also likely to reflect the influence of factors other than study practices, such as pre-existing knowledge of the material and test-taking skills. Keep in mind that the question was about the effectiveness of procedures for training students to study in particular ways, not the effects of certain study practices on test performance. With this focus, a decision to measure the effects of study skill training procedures using only test performance would not generate data that directly reflected the impact of these training procedures and nothing else. Although measures of test performance might partly represent the effects of the training procedures, they would also reflect the influence of other factors and thereby limit the accuracy and generality of conclusions.

Selecting Participants. One of the first challenges that a researcher faces is selecting participants whose behavior will serve as the dependent variable in the experiment. The criteria for choosing some individuals over others emerge from examining the needs of the experimental question. The question should suggest certain participant characteristics that will help to reveal how their behavior is affected by experimental variables, as well as other features that might interfere with seeing these effects.

For many experimental questions, the need for specific participant characteristics is obvious. Species, gender, age, history, and repertoire are almost always important considerations. The focus of the question, the nature of experimental conditions, and the planned setting will usually dictate some of these criteria. For instance, a study investigating procedures for obtaining compliance with exercise regimens for arthritis patients will presumably require not only individuals with this disorder but even a particular type of arthritis or a certain level of severity. An experiment investigating error correction procedures associated with teaching skills to individuals with mental retardation will probably not only require such individuals but they may also need to share a certain level of functioning and prerequisite skills so that they are suitable for the training procedure that will provide the context for different error correction procedures.

Box 3.6

The Hypothetico-Deductive Method

68

Perhaps the most elaborate way of asking experimental questions is the hypothetico-deductive method. In this approach, the experimenters reasoning is from a set of general postulates or axioms to a series of specific theorems that can then be verified by experiment. Each verification not only confirms the empirical relation suggested by the theorem, but substantiates the validity of the postulate set as well. The rules by which the experimenter deduces relations from the postulates or established theorems are quite formal and generate relatively specific predictions.

To the extent that experimental outcomes verify the predictions, the experimenter may increase confidence in the validity of the postulates. However, if the experiment fails to support a logically valid prediction from a postulate, then one or more of the postulates have been falsified. The appropriate action is then to revise the postulate set until the obtained results can be predicted and to confirm the new prediction by repeating the experiment. In this manner, the entire enterprise is supposedly self- correcting (Salmon, 1966).

One of the problems with this strongly deductive mode of inquiry is that it often leads to a more restricted form of question than when the experimenter approaches the topic inductively. Experimental hypotheses take the general form, When x occurs, y will occur. When y fails to occur, little attention may be paid to what happened instead (see Box 3.5).

Another difficulty of this approach is that it requires precise measurement and a clear correspondence between the formal elements in the theoretical structure and the empirical elements involved in the experimental procedures. The method worked for Newton in the 17th century when he arrived at the basic laws of motion in physics, but it did not work in 20th century psychology. These requirements had not been met when Hull (1940, 1943) launched the first major effort to apply the hypothetico-deductive method to the study of behavior. The coordinating definitions relating abstract concepts in the formal system (habit strength, for instance) to experimental operations in the laboratory were simply too vague, so that the data did not constitute an unequivocal test of the hypotheses (Koch, 1954).

Aside from personal features, each experimental question will require participants who can contact experimental conditions in a certain manner. Even simple availability can be an issue. For instance, a laboratory study designed to measure the effects of psychotropic drugs on the behavior of individuals with diagnoses of schizophrenia might require a decision to use persons residing in a facility of some sort rather than individuals living on their own or with their families. Those living with family might be less likely to consistently attend daily testing sessions.

It is also important to select participants whose repertoires include the behavioral characteristics needed to serve as the dependent variable. By directing selection of treatment and control conditions, the question may suggest how participants will need to behave in order to react meaningfully to these conditions. For example, a study of how the stimuli used in a matching-to-sample procedure affect discrimination learning in young children would probably require that the children share a certain level of development in their verbal repertoire, as well as being able to respond to match-to-sample tasks using a touch screen.

Table 3.1 Considerations in selecting participant characteristics

Species Gender

Age

Accessibility Repertoire

Environmental history

Choosing a Response Class. The experimental question is also important in determining the characteristics of the behavior or response class that the investigator chooses as the dependent variable. (Chapter 4 will address this topic in detail.) In some studies, particularly those guided by applied interests, the target behavior may be largely specified by the focus of the question. If the experiment is to examine procedures for reducing self-injurious behavior, for instance, the researcher will only have to determine whether there are any particular features of such behavior that might be important to include. In other studies, often those conducted under laboratory conditions, the selection of a response class may be fairly open.

The general task is to figure out the features that a response class should, or should not, exhibit. With such a list, the researcher is then ready either to identify a behavior already in the participants repertoire with these

69

characteristics or to determine how to create the desired behavior by arranging pretraining or experimental conditions. There are a number of key features that might be of interest.

For instance, the procedures making up the control and treatment conditions will require participants to behave in particular ways. In a project focusing on procedures for training new skills using individuals with mental retardation, for example, certain prerequisite skills such as making eye contact on command, sitting at a table, and perhaps following basic instructions will likely be necessary. Other behaviors will be required by the nature of the procedures making up the control and experimental conditions. At least one of these behaviors is likely to be used to measure the effects of experimental conditions. If the task in the skills training project involves packaging items, making change, or setting a table, for instance, the target behavior will necessarily involve certain forms of such behaviors.

Box 3.7

Ethical Considerations in Behavioral Research

The days are long past when behavioral researchers could design and conduct a study with concern only for methodological niceties. Today behavioral scientists using any vertebrate species in their studies must comply with a complex array of detailed regulations and guidelines. By design, these policies have a pervasive and often intrusive impact on the kinds of questions that may be asked and how they may be pursued with experimental procedures. In fact, their influence usually touches most features of a study, including the characteristics of participants, how they are acquired or recruited, the circumstances under which their participation may continue throughout the project, the procedures and conditions to which they may be exposed, and how they must be treated when the study is completed.

Research ethics are about cultural values. There are different ways of defining cultures and cultural values, but it is at least clear that there are widely varying views about science and how it should be used and conducted. For example, although some revere science and have high hopes for its accomplishments, others question what science can learn and therefore the value of scientific knowledge. Most people are comfortable with humans using other animal species for food, products, and entertainment, but we may also know people who avoid eating meat, refuse to buy leather clothing, and view the animal acts at the circus as mistreatment. And when it comes to research, a sometimes vocal community believes it is immoral to use animals as subjects, especially when it involves any form of discomfort. Others are willing to use animals as subjects as long as the research can be justified in terms of its potential benefits for society.

The practical challenge is how to resolve these differences among various cultural convictions and interests. The laws and regulations governing behavioral research today are the result not only of governmental commissions, panels, bureaucratic procedures, and legislative actions, but also of public debate, citizen protests, and even terrorist activity against universities and individual scientists. Furthermore, these laws and regulations are not static. Cultures and cultural values change, and practices that were once acceptable may later become objectionable. For example, it was once common to design research procedures that involved deceiving human participants about not only the reasons for the procedures they would experience but what these procedures actually involved. Today the requirements for informing participants about the conditions they will experience and getting their consent greatly limit this kind of deception. The transitional nature of cultural values and regulations is particularly clear in animal research. For instance, societys interest in continuing to reduce or limit the use of animals in research is built into animal welfare policies and regulations.

In sum, compliance with research regulations demands careful study of the rules that apply to human or nonhuman species, consultation with regulatory bodies about the details of a proposed project, submitting required paperwork, obtaining necessary approvals, and conforming to the protocols that have been approved.

70

Perhaps the most important role for any behavior that will serve as the dependent variable is to be sensitive to the independent variable. Sensitivity does not mean that the candidate response class must already be known to be strongly affected by the treatment condition. If this were true, there might be little reason to conduct an experiment. In this context, sensitivity means that, whatever the nature of the relationship between a treatment condition and the behavior, the behavior will change in ways that reflect that relationship. An appropriately sensitive target behavior would require, for example, that it be able to vary over an adequate range of values. This would not be possible if the chosen behavior occurred very infrequently (or very often) because then it might be difficult to detect decreases (or increases).

One of the most common constraints on the sensitivity of a behavior comes from powerful extraneous variables, which are often not under the researchers control. If measures of participant behavior reflect not just the impact of carefully controlled experimental conditions but the effects of factors that are not of interest, it will be difficult to draw conclusions about the effects of treatment conditions alone.

This is a common problem in laboratory research using college students as participants, for example. If the researcher is trying to study the behavioral effects of reinforcement schedules, it is well established that individuals with good verbal repertoires are likely to talk to themselves during sessions about what they think is going on. This self-instructional behavior often influences responding and prevents seeing the effects that particular schedules might have in the absence of this extraneous influence (Catania, 2007).

Selecting response classes that are not overly susceptible to extraneous factors is a frequent challenge in applied studies because nonlaboratory settings typically involve many powerful influences that are not the focus of the experiment. Although one of the reasons for choosing applied research settings is to assess interventions under real-world conditions, if extraneous factors are too powerful then their effects can make it difficult to see the effects of treatment variables. (Later chapters will consider extraneous variables in greater detail.)

Another characteristic of a response class that can have a big impact on what the data reveal is its dimensional quantities. (Chapter 5 will address this topic fully.) The dimensional quantities associated with behavior include the number of responses (count), their duration, and the frequency of responding (how often the behavior occurs), among others. Each quantity reflects a different aspect of a behavior, much as describing an object in terms of its weight, volume, or color reveals different characteristics. The experimental question, together with the nature of the experimental and control conditions, may suggest that some aspects of the behavior will be more useful than others in reflecting the effects of experimental conditions. For instance, if a treatment procedure is likely to produce important variations in the duration of responding, it may be necessary to choose a response class that can vary in this feature, such as watching television. On the other hand, if the experiment is to use a discrete trial procedure, in which the participant must wait for the investigator to start each trial, measuring the participants behavior in terms of frequency of responding is likely to be misleading. In such a procedure, the pace of the participants responding will partly depend on the behavior of the investigator and therefore it will not clearly represent only the impact of the treatment condition.

Finally, because the target behavior will serve as the dependent variable, it is important that it can be accurately measured. This means that the selected feature of each instance must be observed and recorded. This is usually not difficult in more controlled research settings, but it can be challenging in some field settings, especially when human observers are used. Chapter 6 will consider these issues.

Table 3.2 Considerations in choosing a response class

Compatibility with procedures Dimensional quantities Sensitivity to independent variable Measurability Influence by extraneous variables

Designing Measurement Procedures. Many of the considerations involved in setting up measurement procedures are fundamental and do not vary much from one experiment to another. For instance, obtaining accurate measures of responding is always important. However, the issue of how long and how often periods of measurement should occur partly depends on what the question needs to learn about how the treatment condition influences responding. Sometimes relatively brief daily sessions may provide adequate information,

71

but under other circumstances measurement may need to continue for extended periods. Consider an investigation of the effects of diet on hyperactivity in young children. An experiment might

compare the effects of a therapeutic diet versus a typical diet, with each type of diet being followed for a few weeks at a time in turn. In this situation, it may be necessary to obtain samples of behavior throughout each day in varied settings, probably involving parents, teachers, and others as observers. In contrast, a different experiment on this topic might ask about the effects of a particular food by preparing a series of dietary challenges. Under this procedure, the effect of consuming the food in question would last only a few hours, and measurement would therefore need to occur continuously throughout this period following each challenge in order to capture any effects.

Selecting Independent Variables. The independent variable (also called the treatment condition or intervention) is the centerpiece of experimental procedures, and the experimental questions most important role is guiding its selection. The question is only the researchers verbal behavior and is not worth much until it leads to experimental action. The independent variables that are embedded in experimental procedures are the means by which we translate the question into environmental features that the participants will experience. The correspondence between the question and its treatment conditions must therefore be very close in order for the data to clearly answer the question.

Unfortunately, there are no simple rules for how the experimental question should lead to selecting independent variables. One way to approach the problem, however, is to ask what kind of data will be required to answer the question. That is, given the question, imagine the ideal kind of data that would unambiguously answer it. With this guide, then consider what particular features of a treatment condition might generate such data.

For instance, what if a question asks about the effects of quantity of food consumed on the behavior of ruminating after meals in individuals diagnosed as severely or profoundly mentally retarded (Rast, Johnston, Drum, & Conrin, 1981)? If we ask what kind of data would answer this question, we might find that we would need to be able to compare the effects of eating different amounts of food on ruminating behavior. Although we could certainly select three or four different amounts of food for comparison, if there is no evidence from prior studies about whether food quantity has any effects on ruminating at all, it might be sufficient for an initial study just to see if such a relationship exists.

In this case, then, it would be important to be able to compare frequencies of ruminating under a typical diet with frequencies under a diet that showed a clear impact on ruminating. A typical diet might mean single portions of foods that would normally be served to individuals residing in a developmental center. It would not be permissible to serve less food than this, so in order to assess the effects of food quantity the comparison condition would need to involve larger amounts of food. Because we would not know how big a difference in food quantity might be required to see an effect on frequencies of ruminating, we would want to choose a sufficiently large amount of food to capture any effect, if there was such a relationship. Assuming no associated health risks, we could encourage participants to eat until they were satiated (defined as refusing additional food).

In sum, a question focusing on the effects of food quantity on ruminating might be well served with an independent variable defined as satiation quantities of foods that were in all other respects the same as food consumed under control conditions involving single-portion quantities. Developing the full details of both control and experimental conditions would require further decisions about how the food was presented and consumed, how satiation was defined in terms of participant behavior, and so forth.

Selecting an experiments independent variable immediately defines all other variables as extraneous to experimental interests. In a way, the unfortunate function of extraneous variables is to supply alternative explanations for what might otherwise appear to be the effects of the treatment condition. Sometimes these extraneous factors come attached to the independent variable. For example, it would unavoidably take longer to eat a satiation meal than a single-portion meal, and it is possible that the extra time has its own effects on ruminating, unrelated to how much food is consumed. Extraneous factors can also accompany features of control conditions, the general experimental setting, the behavior serving as the dependent variable, or other characteristics of participants. In fact, extraneous variables can intrude on the effects of treatment conditions in endless ways, and they are a major challenge to drawing clear and sound conclusions about the contribution of the independent variable alone. Discussion of extraneous variables will occur throughout this book.