10PART!FUNDAMENTALSOFRESEARCHTHEORYInductionDeductionDATAFIGURE1.1A Theory Organizes and Predicts Data.By means of deduction,particular observations(data) may be predicted. By means of indluction, the data suggest organizing principles (theo-ries). This circular relationship indicates that theories are tentative pictures of how data areorganized.(p.).In thecaseof social loafing,theargumentwould bethatthefactsof social loafingderivedfromexperimentationproducedthetheoryof diffusionofresponsibilityOne problem with a purely inductive approach has to do with the finality ofempirical observations. Scientific observations are tied to the circumstances underwhich they are made, which means that the laws or theories that are induced fromthem must also be limited in scope.Subsequent experiments in different contextsmay suggest another theory or modifications to an existing one, so ourtheories thatare induced on the basis of particular observations can (and usually do) changewhen other observations are made.This, of course, is a problem only if one takes anauthoritarian view of ideas and believes in clinging tenaciously to a particular theory.Thus, theories induced from observations are tentative ideas, not final truths, and thetheoretical changesthatoccurasaresultof continuedempiricalwork exemplifytheself-correcting nature of science.According to the deductive view, which emphasizes the primacy of theory, theimportant scientific aspect of the social loafing research is the empirical guidance pro-vided by the formal theory of social loafing.Furthermore, the more general theory,diffusion of responsibility, provides an understanding of social loafing. The deductiveapproach holds well-developed theories in high regard. Casual observations, informaltheories, and data take second place to broad theories that describe and predict asubstantial numberofobservationsFrom the standpoint of the deductive approach, scientific understanding means, inpart, that a theory will predict that certain kinds of empirical observations should oc-cur. In the case of social loafing, the theory of diffusion of responsibility suggests thatmonitoring individual performance in a group should reduce the diffusion of respon-sibility,which in turn will reduce the amount of social loafing thatis observed.Thisprediction, as wehave seen,provestobecorrect.But what do correct predictions reveal? If a theory is verified by the results ofexperiments, a deductive scientist might have increased confidence in the veracity of thetheory. However, since empirical observations are not final and can change, somethingotherthanverificationmaybeessentialforacceptanceorrejectionof atheory.Popper
10 PART 1 FUNDAMENTALS OF RESEARCH (p. 6). In the case of social loafi ng, the argument would be that the facts of social loafi ng derived from experimentation produced the theory of diffusion of responsibility. One problem with a purely inductive approach has to do with the fi nality of empirical observations. Scientifi c observations are tied to the circumstances under which they are made, which means that the laws or theories that are induced from them must also be limited in scope. Subsequent experiments in different contexts may suggest another theory or modifi cations to an existing one, so our theories that are induced on the basis of particular observations can (and usually do) change when other observations are made. This, of course, is a problem only if one takes an authoritarian view of ideas and believes in clinging tenaciously to a particular theory. Thus, theories induced from observations are tentative ideas, not fi nal truths, and the theoretical changes that occur as a result of continued empirical work exemplify the self-correcting nature of science. According to the deductive view, which emphasizes the primacy of theory, the important scientifi c aspect of the social loafi ng research is the empirical guidance provided by the formal theory of social loafi ng. Furthermore, the more general theory, diffusion of responsibility, provides an understanding of social loafi ng. The deductive approach holds well-developed theories in high regard. Casual observations, informal theories, and data take second place to broad theories that describe and predict a substantial number of observations. From the standpoint of the deductive approach, scientifi c understanding means, in part, that a theory will predict that certain kinds of empirical observations should occur. In the case of social loafi ng, the theory of diffusion of responsibility suggests that monitoring individual performance in a group should reduce the diffusion of responsibility, which in turn will reduce the amount of social loafi ng that is observed. This prediction, as we have seen, proves to be correct. But what do correct predictions reveal? If a theory is verifi ed by the results of experiments, a deductive scientist might have increased confi dence in the veracity of the theory. However, since empirical observations are not fi nal and can change, something other than verifi cation may be essential for acceptance or rejection of a theory. Popper THEORY Deduction DATA Induction ▼ FIGURE 1.1 A Theory Organizes and Predicts Data. By means of deduction, particular observations (data) may be predicted. By means of induction, the data suggest organizing principles (theories). This circular relationship indicates that theories are tentative pictures of how data are organized. 59533_02_ch01_p001-023.indd 10 9533_02_ch01_p001-023.indd 10 3/4/08 11:47:14 PM /4/08 11:47:14 PM
CHAPTERI11EXPLANATIONINSCIENTIFICPSYCHOLOGY(1961),a philosopher of science, has suggested that good theories must befallible; thatis, the empirical predictions must be capable of tests that could show them to be false.This suggestion of Popper's has been called the falsifiabilityview.According to thefalsifiability view, thetemporarynatureof induction makes negativeevidencemore im-portant than positive support, If a prediction is supported by data, one cannot say thatthe theory is true. However, if a theory leads to a prediction that is not supported by thedata,then Popper would argue thatthetheorymust be false, and it should berejected.According to Popper, a theory can never beproven; it can only be disproven.Popper's view about the difficulty of proving a theory can be illustrated by think-ingabout a specifictheory;for example,does a bag of marbles contain onlyblackmarbles? One good way to test this theory would be to reach into thebag and drawoutamarble.Themarbleisblack.Whatcanyouconcludeaboutthetheorythat all themarbles are black? While the datum (one black marble) is consistent with the theory,it does not prove it. There might still be a white marble inside the bag. So pull outanother marble; indeed, pull out ten more marbles. All ten are black. Is the theory nowproved? No, there still might be a single white marble lurking in the bag.You wouldhaveto remove every marbleto ensure that there were no white marbles.It is easy toprove the theory wrong if a white marble gets drawn. Proving the theory to be correctdepends on the size of the bag. If the bag is infinitely large, the theory can never beproven because the next marble you examine might be white.Proctor and Capaldi (2001)have noted two kinds of objections to Popper'sap-proach. First, there is a logical problem (Salmon, 1988). Since a theory potentially canalways bedisconfirmed by the next experiment, the number ofaccomplished experi-mentsconsistentwiththetheoryisirrelevant.Sologicallyawell-collaboratedtheoryisnot more valuable and does not necessarily make better predictions than a theory thathasneverbeentested.Thislogical viewconflictswiththepractical viewthat scientiststendtobemorecomfortablewiththeoriesthathavepassed several experimental tests.Thispracticalview(Kuhn,1970)iswhatProctorand Capaldi(2001)offerasthesec-ond, empirical, objection to falsification:Theories tend to beaccepted, at least initially,onthebasisoftheirabilitytoexplain(organize)existingphenomenamorethanontheirabilitytopredictnewresultsOne problem with the deductive approach has to do with the theories themselves.Most theories include many assumptions about the world that are difficult to test and thatmay be wrong.In Latane's work, one assumption underlying the general theory is thatmeasuring a person's behaviorin an experimental context does not change the behav-iorin question.Although this often is a reasonableassumption,we will show later thatpeople can react to being observed in unusual ways, which means that this assumptionissometimeswrong.Iftheuntestedassumptionsarewrong,thenaparticularexperimentthat falsifies a theory may have falsified it for the wrong reasons.That is, the test of thetheory may not have been fair or appropriate. It can be concluded, therefore, that thedeductive approach by itself cannot lead to scientific understanding.At this point, you may be wondering whether scientific understanding is possible ifboth induction and deduction are not infallible. Do not despair. Science is self-correcting,anditcanprovideanswerstoproblems,howevertemporarythoseanswersmaybe.Sci-entific understanding changes as scientists ply their trade. We have a better understand-ing of social loafing now than wedid before Latane and his coworkers undertook theirresearch. Through a combination of induction and deduction (see Figure 1.1), scienceprogresses toward a more thorough understanding of its problems
CHAPTER 1 EXPLANATION IN SCIENTIFIC PSYCHOLOGY 11 (1961), a philosopher of science, has suggested that good theories must be fallible; that is, the empirical predictions must be capable of tests that could show them to be false. This suggestion of Popper’s has been called the falsifi ability view. According to the falsifi ability view, the temporary nature of induction makes negative evidence more important than positive support. If a prediction is supported by data, one cannot say that the theory is true. However, if a theory leads to a prediction that is not supported by the data, then Popper would argue that the theory must be false, and it should be rejected. According to Popper, a theory can never be proven; it can only be disproven. Popper’s view about the diffi culty of proving a theory can be illustrated by thinking about a specifi c theory; for example, does a bag of marbles contain only black marbles? One good way to test this theory would be to reach into the bag and draw out a marble. The marble is black. What can you conclude about the theory that all the marbles are black? While the datum (one black marble) is consistent with the theory, it does not prove it. There might still be a white marble inside the bag. So pull out another marble; indeed, pull out ten more marbles. All ten are black. Is the theory now proved? No, there still might be a single white marble lurking in the bag. You would have to remove every marble to ensure that there were no white marbles. It is easy to prove the theory wrong if a white marble gets drawn. Proving the theory to be correct depends on the size of the bag. If the bag is infi nitely large, the theory can never be proven because the next marble you examine might be white. Proctor and Capaldi (2001) have noted two kinds of objections to Popper’s approach. First, there is a logical problem (Salmon, 1988). Since a theory potentially can always be disconfi rmed by the next experiment, the number of accomplished experiments consistent with the theory is irrelevant. So logically a well-collaborated theory is not more valuable and does not necessarily make better predictions than a theory that has never been tested. This logical view confl icts with the practical view that scientists tend to be more comfortable with theories that have passed several experimental tests. This practical view (Kuhn, 1970) is what Proctor and Capaldi (2001) offer as the second, empirical, objection to falsifi cation: Theories tend to be accepted, at least initially, on the basis of their ability to explain (organize) existing phenomena more than on their ability to predict new results. One problem with the deductive approach has to do with the theories themselves. Most theories include many assumptions about the world that are diffi cult to test and that may be wrong. In Latané’s work, one assumption underlying the general theory is that measuring a person’s behavior in an experimental context does not change the behavior in question. Although this often is a reasonable assumption, we will show later that people can react to being observed in unusual ways, which means that this assumption is sometimes wrong. If the untested assumptions are wrong, then a particular experiment that falsifi es a theory may have falsifi ed it for the wrong reasons. That is, the test of the theory may not have been fair or appropriate. It can be concluded, therefore, that the deductive approach by itself cannot lead to scientifi c understanding. At this point, you may be wondering whether scientifi c understanding is possible if both induction and deduction are not infallible. Do not despair. Science is self- correcting, and it can provide answers to problems, however temporary those answers may be. Scientifi c understanding changes as scientists ply their trade. We have a better understanding of social loafi ng now than we did before Latané and his coworkers undertook their research. Through a combination of induction and deduction (see Figure 1.1), science progresses toward a more thorough understanding of its problems. 59533_02_ch01_p001-023.indd 11 9533_02_ch01_p001-023.indd 11 3/4/08 11:47:14 PM /4/08 11:47:14 PM
12PARTIFUNDAMENTALSOFRESEARCHBy way of concluding this section, we reexamine social loafing.Initially,positiveexperimental results bolstered our confidence in the general notion of social loafing.These results, in turn, suggested hypotheses about the nature of social loafing. Is it ageneral phenomenon thatwould influence even group-oriented individuals? Does itoccur in the workplace as well as the laboratory?Positive answers to these questionsare consistent with a diffusion-of-responsibility interpretation of social loafing.In the next phase of the research, Latane and his colleagues attempted to eliminateother explanations of social loafing by falsifying predictions made by these alternativetheories. In their earlier work, Latane and his colleagues tested a particular person's effortboth when alone and when in a group. They subsequently reasoned that under these con-ditions, a person might rest during the grouptest so that greater effort could be allocatedto the task when he or she was tested alone.To eliminate the possibility that allocation ofeffort rather than diffusion of responsibility accounted for social loafing, they conductedadditional experiments in which a person was tested either alone or in a group-but notin both situations. Contrary to the allocation-of-effort hypothesis, the results indicated thatsocial loafing occurred when a person wastested in justthat one condition ofbeing inagroup (Harkins, Latane, & Williams, 1980). Therefore, it was concluded that diffusion of re-sponsibility was a more appropriate account of social loafing than was allocation of effort.Note the course of events here.Successive experiments pitted two possible out-comes against each other with the hope that one possibility would be eliminated andone supported by the outcome of the research.Of course, subsequent tests of thediffusion-of-responsibility theory probably will contradict it or add to it in some way.Thus, the theory might be revised or, with enough contradictions, rejected for an al-ternativeexplanation,itselfsupportedbyempirical observations.Inanyevent,wherewe stand now is that we have constructed a reasonable view of what social loafingentails and what seems to cause it. It is the mixture of hypotheses induced from dataand experimental tests deduced from theory that resulted in the theory that diffusion ofresponsibility leads to social loafing.FromTheorytoHypothesisTheories cannot be tested directly.There is no single magical experiment that willprove a theorytobe correct or incorrect.Instead,scientists perform experiments totesthypotheses that are derived from a theory. But exactly what are scientific hypothesesand where do theycomefrom?It is important to distinguish between hypotheses and generalizations (Kluger &Tikochinsky, 2001). A hypothesis is a very specific testable statement that can beevaluated from observable data.For example,wemight hypothesize that drivers olderthan sixty-five years would have a higher frequency of accidents involving left turnsacross oncoming traffic when driving at night than do younger drivers.By looking atpolice records of accident data, we could determine, with the help of some statistics(seeAppendixB),ifthishypothesisisincorrect.Ageneralizationisabroaderstate-ment that cannot be tested directly.For example, we might generalize that older driversare unsafe at any speed and should have restrictions, such as not being able to drive atnight, on their driver's license. Since "unsafe at any speed" is not clearly defined, this isnot a testable statement. Similarly, the generalization does not define an age range forolder drivers. However, it can be used to derive several testable hypotheses
12 PART 1 FUNDAMENTALS OF RESEARCH By way of concluding this section, we reexamine social loafi ng. Initially, positive experimental results bolstered our confi dence in the general notion of social loafi ng. These results, in turn, suggested hypotheses about the nature of social loafi ng. Is it a general phenomenon that would infl uence even group-oriented individuals? Does it occur in the workplace as well as the laboratory? Positive answers to these questions are consistent with a diffusion-of-responsibility interpretation of social loafi ng. In the next phase of the research, Latané and his colleagues attempted to eliminate other explanations of social loafi ng by falsifying predictions made by these alternative theories. In their earlier work, Latané and his colleagues tested a particular person’s effort both when alone and when in a group. They subsequently reasoned that under these conditions, a person might rest during the group test so that greater effort could be allocated to the task when he or she was tested alone. To eliminate the possibility that allocation of effort rather than diffusion of responsibility accounted for social loafi ng, they conducted additional experiments in which a person was tested either alone or in a group—but not in both situations. Contrary to the allocation-of-effort hypothesis, the results indicated that social loafi ng occurred when a person was tested in just that one condition of being in a group (Harkins, Latané, & Williams, 1980). Therefore, it was concluded that diffusion of responsibility was a more appropriate account of social loafi ng than was allocation of effort. Note the course of events here. Successive experiments pitted two possible outcomes against each other with the hope that one possibility would be eliminated and one supported by the outcome of the research. Of course, subsequent tests of the diffusion-of-responsibility theory probably will contradict it or add to it in some way. Thus, the theory might be revised or, with enough contradictions, rejected for an alternative explanation, itself supported by empirical observations. In any event, where we stand now is that we have constructed a reasonable view of what social loafi ng entails and what seems to cause it. It is the mixture of hypotheses induced from data and experimental tests deduced from theory that resulted in the theory that diffusion of responsibility leads to social loafi ng. From Theory to Hypothesis Theories cannot be tested directly. There is no single magical experiment that will prove a theory to be correct or incorrect. Instead, scientists perform experiments to test hypotheses that are derived from a theory. But exactly what are scientifi c hypotheses and where do they come from? It is important to distinguish between hypotheses and generalizations (Kluger & Tikochinsky, 2001). A hypothesis is a very specifi c testable statement that can be evaluated from observable data. For example, we might hypothesize that drivers older than sixty-fi ve years would have a higher frequency of accidents involving left turns across oncoming traffi c when driving at night than do younger drivers. By looking at police records of accident data, we could determine, with the help of some statistics (see Appendix B), if this hypothesis is incorrect. A generalization is a broader statement that cannot be tested directly. For example, we might generalize that older drivers are unsafe at any speed and should have restrictions, such as not being able to drive at night, on their driver’s license. Since “unsafe at any speed” is not clearly defi ned, this is not a testable statement. Similarly, the generalization does not defi ne an age range for older drivers. However, it can be used to derive several testable hypotheses. 59533_02_ch01_p001-023.indd 12 9533_02_ch01_p001-023.indd 12 3/4/08 11:47:14 PM /4/08 11:47:14 PM
CHAPTERIEXPLANATIONINSCIENTIFICPSYCHOLOGY13Figure1.2 illustrates this process.Each generalization can produce more than onehypothesis. Only two are illustrated in the figure to keep it simple, but a good generali-zation can produce a horde of hypotheses.For example,the older-driver generalizationcould produce manyhypotheses about differentkinds of accidents and behaviors thatbefall aging drivers: crashing into stopped vehicles, failing to signal for turns, drivingon the sidewalk, backing up into objects, not keeping within their lane, and so on.These hypotheses could be tested by making observations in traffic, on closed testtracks (safer for the driving public if the generalization is true), or in driving simulators(safestforthedrivingpublic).Now that we have explained that hypotheses come from generalizations, we can goon to the next question:Where do generalizations comefrom?Figure 1.2 shows thereare two sources forgeneralizations.They can come from theory orfrom experienceWhile only three generalizations are shown in Figure 1.2,a good theory will producea gaggle of generalizations.You may think that the aging-driver generalization comesfrom experience rather than from theory. You may have firsthand experience being apassenger in a car driven by a grandparent, and thatexperience may have caused you toagree with the generalization. This is an inductive process (see Figure 1.1) based upondata,namely casual observation of the driving behavior of elderly citizens.Hypothesesderived from this inductive process are called common-sense hypotheses.While testingcommon-sense hypotheses was once frowned upon in experimental psychology as be-ing inferior to testinghypotheses derived from theory,there is currently a new apprecia-tion of the value of common-sense hypotheses (Kluger & Tikochinsky, 2oo1).Nevertheless, most psychologists prefertesting hypotheses based upon theory.Inthis case, the generalization is formed deductively (see Figure 1.1) from the theory. Theaging-driver generalization could also be derived from theories of attention, perception,and decision making (Kantowitz, 2001).As we age, our ability to attend to multiple tasksdecreases and our decision making becomes more conservative, often requiring moretime to accomplish. So an elderlydriver might (a) have trouble seeing oncoming traffic atnight, (b)havetrouble attendingto oncoming traffic whilepaying attention to a radioora passenger, and (c) take a long time to decide if a left-hand turn across traffic is safe, soTheoryGeneralizationGeneralizationGeneralizationHypothesisHypothesisHypothesisHypothesisHypothesisHypothesisEverydayExperienceFIGURE1.2Gaggles of Generalizations ProduceHordes of Hypotheses
CHAPTER 1 EXPLANATION IN SCIENTIFIC PSYCHOLOGY 13 Figure 1.2 illustrates this process. Each generalization can produce more than one hypothesis. Only two are illustrated in the fi gure to keep it simple, but a good generalization can produce a horde of hypotheses. For example, the older-driver generalization could produce many hypotheses about different kinds of accidents and behaviors that befall aging drivers: crashing into stopped vehicles, failing to signal for turns, driving on the sidewalk, backing up into objects, not keeping within their lane, and so on. These hypotheses could be tested by making observations in traffi c, on closed test tracks (safer for the driving public if the generalization is true), or in driving simulators (safest for the driving public). Now that we have explained that hypotheses come from generalizations, we can go on to the next question: Where do generalizations come from? Figure 1.2 shows there are two sources for generalizations. They can come from theory or from experience. While only three generalizations are shown in Figure 1.2, a good theory will produce a gaggle of generalizations. You may think that the aging-driver generalization comes from experience rather than from theory. You may have fi rsthand experience being a passenger in a car driven by a grandparent, and that experience may have caused you to agree with the generalization. This is an inductive process (see Figure 1.1) based upon data, namely casual observation of the driving behavior of elderly citizens. Hypotheses derived from this inductive process are called common-sense hypotheses. While testing common-sense hypotheses was once frowned upon in experimental psychology as being inferior to testing hypotheses derived from theory, there is currently a new appreciation of the value of common-sense hypotheses (Kluger & Tikochinsky, 2001). Nevertheless, most psychologists prefer testing hypotheses based upon theory. In this case, the generalization is formed deductively (see Figure 1.1) from the theory. The aging-driver generalization could also be derived from theories of attention, perception, and decision making (Kantowitz, 2001). As we age, our ability to attend to multiple tasks decreases and our decision making becomes more conservative, often requiring more time to accomplish. So an elderly driver might (a) have trouble seeing oncoming traffi c at night, (b) have trouble attending to oncoming traffi c while paying attention to a radio or a passenger, and (c) take a long time to decide if a left-hand turn across traffi c is safe, so Generalization Hypothesis Hypothesis Hypothesis Hypothesis Hypothesis Hypothesis Generalization Generalization Theory Everyday Experience ▼ FIGURE 1.2 Gaggles of Generalizations Produce Hordes of Hypotheses. 59533_02_ch01_p001-023.indd 13 9533_02_ch01_p001-023.indd 13 3/4/08 11:47:14 PM /4/08 11:47:14 PM
14PARTIFUNDAMENTALSOFRESEARCHthat when he or she finally makes the turn it is too late and oncoming traffic cannot avoidan accident. The advantage of a good theory is that it produces many generalizations.Theories of attention not only deal with agingdrivers butmakegeneralizations aboutmany otherpractical situationssuchas operating airplanes and nuclearpowerplants,tosay nothing of more abstract predictions to be tested in laboratories. For example, manytheories of attention would predict that talking on your cell phone while you are drivingwould be dangerous, and indeed laboratory research suggests that it is (Steayer & Drew,2007).However, common-sensegeneralizations are notproductivebecause, even if theyare correct, they do not create new generalizations.So theories are more efficient inadvancing scientific inquiry.While hypothesis testing is the dominant methodology used in experimentalpsychology, there are other points of view. Most theories in psychology are verbaland qualitative so that mathematical predictions are hard to come by.However, ifa formal model can be generated either mathematically orby computer simulation,then it becomes possible to estimate parameters of the model. Parameter estimationis superiorto hypothesis testing and curve fitting (Kantowitz&Fujita,1990),and aspsychology evolves as a science, estimation will supplement, and perhaps eventuallyreplace,hypothesistesting.Indeed,thereisanewmovementinthephilosophyofscience, called naturalism, that criticizes current methodologies such as hypothesistesting, and its tentacles have reached the shores of psychological science (Proctor& Capaldi, 2001).Naturalism suggests that methodological criteria are not fixed foreternity based on logical premises, but can change and evolve (just like theories) onpragmaticgrounds.EvaluatingTheoriesThe sophisticated scientist does not try to determine if a particular theory is true or falsein an absolute sense.There is no black-and-white approach to theory evaluation.Atheorymaybeknown to be incorrect in someportion and yet continueto be used.Inmodernphysics,lightisrepresented,accordingtothetheorychosen,eitherasdiscreteparticles called quanta or as continuous waves. Logically, light cannot be both at thesame time. Thus, you might think that at least one of these two theoretical views mustnecessarily be false.The physicist tolerates this ambiguity (although perhaps not cheer-fully) and uses whichever representation-quantum or wave--is more appropriate.Instead of flatly stating that a theoryis true, the scientist is much more likelyto statethat it is supported substantially by data, thereby leaving open the possibility that newdatamaynotsupportthetheory.Althoughscientistsdonotstatethatatheoryistruetheymustoftendecidewhich ofseveral theories is best.Asnoted earlier,explanationsare tentative; nevertheless, the scientist still needs to decide which theory is best fornow.To do so, explicit criteria are needed for evaluating a theory.Four such criteriaare parsimony,precision, testability, and abilitytofit data.One important criterion was hinted at earlier when we stated that the fewer thestatements in a theory, the better the theory.This criterion is called parsimony, orsometimes Occam's razor, after William of Occam. If a theory needs a separate state-ment for every result it must explain, clearly no economy has been gained by thetheory. Theories gain power when they can explain many results with few explanatoryconcepts.Thus, if two theories have the samenumber of concepts, the one that can
14 PART 1 FUNDAMENTALS OF RESEARCH that when he or she fi nally makes the turn it is too late and oncoming traffi c cannot avoid an accident. The advantage of a good theory is that it produces many generalizations. Theories of attention not only deal with aging drivers but make generalizations about many other practical situations such as operating airplanes and nuclear power plants, to say nothing of more abstract predictions to be tested in laboratories. For example, many theories of attention would predict that talking on your cell phone while you are driving would be dangerous, and indeed laboratory research suggests that it is (Steayer & Drew, 2007). However, common-sense generalizations are not productive because, even if they are correct, they do not create new generalizations. So theories are more effi cient in advancing scientifi c inquiry. While hypothesis testing is the dominant methodology used in experimental psychology, there are other points of view. Most theories in psychology are verbal and qualitative so that mathematical predictions are hard to come by. However, if a formal model can be generated either mathematically or by computer simulation, then it becomes possible to estimate parameters of the model. Parameter estimation is superior to hypothesis testing and curve fi tting (Kantowitz & Fujita, 1990), and as psychology evolves as a science, estimation will supplement, and perhaps eventually replace, hypothesis testing. Indeed, there is a new movement in the philosophy of science, called naturalism, that criticizes current methodologies such as hypothesis testing, and its tentacles have reached the shores of psychological science (Proctor & Capaldi, 2001). Naturalism suggests that methodological criteria are not fi xed for eternity based on logical premises, but can change and evolve (just like theories) on pragmatic grounds. Evaluating Theories The sophisticated scientist does not try to determine if a particular theory is true or false in an absolute sense. There is no black-and-white approach to theory evaluation. A theory may be known to be incorrect in some portion and yet continue to be used. In modern physics, light is represented, according to the theory chosen, either as discrete particles called quanta or as continuous waves. Logically, light cannot be both at the same time. Thus, you might think that at least one of these two theoretical views must necessarily be false. The physicist tolerates this ambiguity (although perhaps not cheerfully) and uses whichever representation—quantum or wave—is more appropriate. Instead of fl atly stating that a theory is true, the scientist is much more likely to state that it is supported substantially by data, thereby leaving open the possibility that new data may not support the theory. Although scientists do not state that a theory is true, they must often decide which of several theories is best. As noted earlier, explanations are tentative; nevertheless, the scientist still needs to decide which theory is best for now. To do so, explicit criteria are needed for evaluating a theory. Four such criteria are parsimony, precision, testability, and ability to fi t data. One important criterion was hinted at earlier when we stated that the fewer the statements in a theory, the better the theory. This criterion is called parsimony, or sometimes Occam’s razor, after William of Occam. If a theory needs a separate statement for every result it must explain, clearly no economy has been gained by the theory. Theories gain power when they can explain many results with few explanatory concepts. Thus, if two theories have the same number of concepts, the one that can 59533_02_ch01_p001-023.indd 14 9533_02_ch01_p001-023.indd 14 3/4/08 11:47:15 PM /4/08 11:47:15 PM