JOURNAL OF THE EXPERIMENTAL ANALYSIS OF BEHAVIOR

REVIEWERS' COMMENTS ON TIMBERLAKE'S BEHAVIORREGULATION AND LEARNED PERFORMANCE

The preceding paper by William Timberlakeon behavioral regulation and set-points has beenthrough several cycles of review and resubmis-sion. After the paper was accepted, the reviewerswere asked to reconsider and revise their originalreviews in ways that would clarify the issues thatTimberlake had addressed in the final version.Six such reviews follow, in an order correspond-ing to their original chronology. Our intention inpublishing them is to present efficiently and con-structively some alternative views of the in-teresting and important problems that Timber-lake addresses.

J. A. Nevin,Former Editor and ActionEditor on the manuscript

P. N. Hineline,Editor

[The following is an extract from one of theoriginal reviews of this article. The revised articleaddresses many of the original criticisms and thiscommentary represents the unanswered residual.It is offered here for the reader's consideration asa constructive counterpoint. I hope that this com-mentary may make a contribution to futurethought in this area in the same way that theoriginal review may have contributed in somesmall measure to the improvements contained inthe present version. I retain great respect for theauthor and his thoughtful analysis contained inthis important acticle.]

In this article William Timberlake attempts toexpand on regulatory theory of performance byincluding consideration of multiple set-points.He illustrates these concepts by reference to twosets of data. The author has given birth to severalnew concepts; some are healthy but others couldbenefit from further intellectual nurture. I am ingeneral agreement with the regulatory approach.My most fundamental concern centers on theidea of set-points.A set-point conveys the idea of an ideal condi-

tion, most valued amount, or "normal" level.This notion makes most sense when there are

only a few cardinal set-points that dictate adjust-ments in other aspects of performance when con-fronted with environmental constraints. For ex-ample, food and water intake might be two set-points that "drive" response rate, meal frequency,and meal size. Timberlake notes that certain ofthese other measures are more modifiable thanothers, so again he postulates secondary set-points for these. I find this form of analysis tohave several shortcomings. First, I find it dif-ficult to adhere to a set-point theory when eventhe cardinal measures (i.e., daily intakes) are notdefended at all costs. Economically speaking,none of these measures is perfectly inelastic; theydo vary in elasticity. It is unclear from thecriteria for a set-point whether elastic demandthat accompanies decreases in instrumental per-formance in the face of increasing constraint(e.g., increasing fixed-ratio requirement) wouldimply a set-point or not. For example, brainstimulation will reinforce instrumental perfor-mance to a high degree under conditions ofminimal contraint but shows a highly elastic(steeply sloping) demand curve and decreasingresponse-rate function under increasing con-straint. By contrast, the opportunity to eat foodpellets will not maintain nearly as high a rate ofperformance under minimal constraint but willshow a highly inelastic (minimally sloping) de-mand curve and increasing respone-rate functionunder increasing constraint (Hursh & Natelson,1981). It seems more reasonable to think in termsof a continuum of elasticity or resistance tochange rather than in terms of discrete categoriesof molar and response-pattern set-points, par-ticularly because the baseline level of perfor-mance is a poor predictor of the elasticity of de-mand under increasing constraint (see Hursh,1980). The usual way of plotting demand curvesin log-log coordinates allows comparison ofslopes of change without concern for differencesin initial level.A second related problem has to do with the

mathematical imprecision of the terms resistanceand flexibility as used by the author. How doesone precisely compare the resistance of two mea-

376

1984, 41, 376-382 NUMBER 3 (MAY)

REVIEWERS' COMMENTS

sures that differ in flexibility? For example, Rat 1

in Figure 3 shows a more flexible (i.e., highermaximum) bar-press rate than meal size, but it isdifficult to judge which one is more resistant(i.e., reaches its final level more rapidly). Visualinspection can be deceptive in such a case. Incontrast, Nevin has more precisely definedresistance to change in discussions of response

strength and behavioral momentum as the slopeof the function (Nevin, Mandell, & Atak, 1983).The economic-demand-curve approach offers a

similar but more inclusive analysis because it ex-

plicitly allows for changes in elasticity (nonlineardemand curves in log-log coordinates) with in-creases in constraint. Elasticity is usually theslope of the demand curve within a restrictedrange or the derivative of the function at a point(see "mixed elasticity," Hursh, 1980). A separateconcept refers to the level of demand, which issimply the placement of the curve in relation tothe origin.

Third, the measurement of set-points clearlysuggests that set-points reflect an underlying "in-stigation." Accurate measurement of the set-point requires great care to maintain a constantlevel of this "instigation" throughout the experi-ment. This kind of object language when refer-ring to relational processes can lead to an

unintended search for the entity named and,worse, explanation of changes in performance byreference to changes in the "thing." This notioncould imply for some that variations in elasticityor resistance to change for the same activity indifferent experiments point to uncontrolled levelsof "instigation." It would seem more fruitful tolook for contextual (i.e., environmental) dif-ferences such as differences in substitutable alter-natives or differences in the overall economiccontext (i.e., open vs. closed). Furthermore, it isnot clear to me whether differences in "instiga-tion" should lead to differences in the baselinelevel (set-point), differences in resistance tochange, or differences in maximum change "flex-ibility." That may be an empirical question butbecause "instigation" is not directly measured, itseems less susceptible to empirical analysis.

Finally, I find a set-point analysis arbitrary inthat it takes a measure of intake under some idealcondition (e.g., no contingency) as intrinsicallymore valuable or perfect than intake under more

constrained (less ideal) conditions. This ap-

proach seems doomed from the start becausemost species evolved under conditions of con-

straint; I find it implausible that they would,

nevertheless, be structured to approach someimaginary set-point attainable only under zeroconstraint. What makes one set of conditions orcombination of performances more "ideal" thanany other? I am reminded of data showing thatrats live longer when somewhat food deprivedthan when allowed to eat ad lib. Most of us ac-cept the notion that laboratory animals overeatwhen given free access to food in a confined cage.The point here is that one must be careful whentaking some initial value as the ideal and explain-ing everything else in relation to that value. Onecan make any situation seem ideal when examin-ing deviations as percentage of baseline, but thatpicture is really artificial. Much of that problemis avoided by concepts such as elasticity of de-mand or Nevin's response strength because thereferent is the slope of a function rather thansome ideal point on the function. The author iscognizant of this limitation of the "free-baseline"measure of set-points; I see it as a fundamentalproblem with the entire set-point approach.

Steven R. HurshWalter Reed Army Institute of Research

To borrow from Timberlake, this commentbegan as a modest attempt, one of my first man-uscripts for JEAB as a guest reviewer. The sec-tions of that version of his paper which I judgedto merit publication have been retained, and I amtherefore satisfied that I probably made a soundfavorable recommendation. However, other sec-tions have been expanded (rather than shortenedor deleted) and it now seems that there is as muchto criticize about the paper as to applaud.On the positive side, Timberlake makes sev-

eral cogent arguments in rebuttal to Allison's(1981a, 1981b) critiques of optimal behavior-regulation theory in the section, "Behavioral set-points in optimal regulation." His point thatdeprivation (instigation) was probably not heldconstant as the ratio requirement was varied inthe Collier experiments seems pertinent enough,but one might add that manipulation of the ratiowas also apparently confounded with time in theexperiment, further precluding meaningful con-clusions on simple methodological grounds.Also, the points Timberlake raised concerningthe concurrent-ratio data are convincingarguments that these data do not (and perhapsnever even possibly could) clearly contradict theminimum-deviation hypothesis. Finally, Tim-

377

REVIEWERS' COMMENTS

berlake's treatment of the issue of molar versusmolecular set-points seems to be cutting newground in proposing that both types are impor-tant in determining behavior. Although this is aninteresting idea that deserves further investiga-tion, I fear his conjecture and/or the assumptionsthat underly it may not be testable. Timberlakedid not make it clear how one substantiates as anexplanatory device the existence of all the rele-vant set-points (and their relative importance) ina way that avoids circularity. Still, I think thesection, "Response-pattern set-points and molarregulation." is intriguing and will make a signifi-cant contribution to the theoretical discussion ofthe molar/molecular set-point problem.On the negative side, the paper may leave

some misapprehensions and disagreements unre-solved. Among these are the relation of behavior-regulation theory to reinforcement theory, theevaluation of the advantages and limitations ofbehavior-regulation theory, and conceptual dif-ferences within the basic regulatory approach.

Timberlake seems to hold three different viewsof the relationship between behavior-regulationand reinforcement theories. Behavior-regulationtheory is alternately viewed as an opposing alter-native, a descendent or refined reformulation, oran independent but compatible (even comple-mentary) hypothesis to reinforcement theory.When viewed as an opposing alternative, Tim-berlakes's interpretation of reinforcement theoryis not generally convincing. For example, he sug-gests that extra-experimental manipulation of ac-cess to food is the analog in reinforcement theoryto the idea of constrained instigation as a predic-tor of the reinforcing or punishing effects ofeating. However, I would suggest that reinforce-ment theory's assumption of cross-situationalgenerality of the reinforcing effectiveness of astimulus (or response, from Premack's perspec-tive), modulated by reinforcement contingencies,provides similar predictive generality. In thesame way as in behavior-regulation theory. thisassumption allows prediction of reinforcementand punishment effects independent of extra-experimental operations so long as these condi-tions remain unchanged between estimation/prediction sessions and test sessions.The discussion of the advantages and limita-

tions of behavior-regulation theory is also confus-ing. The point about the regulatory approach be-ing an a priori theory does not seem to be an ad-vantage over other views, but rather merely aprerequisite to serious consideration. Indeed,if

his claim is correct, one wonders why behavior-regulation theory is unable to specify all relevantset-points a priori. Perhaps the most confusingaspect of the discussion of advantages involvesthe claims of uniquely predicting bitonic func-tions between ratio requirement and responserate. There seems to be an incompatibility be-tween the ability to predict bitonic functions forresponding and the assumption that the defini-tion of a set-point requires a direct function be-tween responding and the challenge to expressionof instigation. Can one have it both ways and stillargue that the theory is well formulated?

Timberlake's discussion of conceptual dif-ferences within the basic regulatory approachcould be clearer. I fail to see the problem withPremack's formulation, although Timberlakesuggests that it misleadingly emphasizes responsedeprivation over response strengthening. Also,the dispute over the appropriate baseline, singleresponse or paired response, seems only to pdintup weaknesses in the case for behavior regula-tion. It was not made clear whether differences inthe type of baseline were critical to the accuracyof potentially discriminating predictions.My final point, alluded to earlier, is that

behavior-regulation theory may not yet be wellformulated. The theory assumes that the notionof instigation for a response is critical forunderstanding learning and performance butcannot now predict what effects should occurfrom varying it. Even if behavior-regulationtheory is well formulated within the boundariescurrently set for it in predicting behavior, Ipredict it will not gain in acceptance until it moreclosely competes with the much more substantialset of applications addressed by more conven-tional views. Such a change will depend uponbetter specification of the regulation hypothesis.

David CaseUniversity of California, San Diego

In my review of an earlier version of thismanuscript, I was critical of Timberlake's discus-sion of both response-pattern set-points and mo-lar set-points.This draft, however, satisfies manyof my objections to his postulation of response-pattern set-points. On the other hand, I continueto be skeptical about the usefulness of molarbehavioral set-points as explanatory concepts. Iwill discuss these two types of set-points in turn.

I originally thought that the postulation of a

378

REVIEWERS' COMMENTS

set-point for duration of drinking bouts in theMarwine and Collier (1979) experiment wassuperfluous. Now, however, it seems thatTimberlake's use of the term set-point carries onlythree implications: (1) For each subject there issome bout duration of intermediate size that ispreferred over either longer or shorter durations.A subject's preference presumably declines mon-otonically with increasing deviations from thisbout duration. (2) As a consequence, a subject ismotivated to minimize deviations from thispreferred bout size. As Timberlake puts it, "thedisparity between expression and instigationbecomes a driving force for learned perfor-mance." (3) Because drinking-bout duration isnot the only factor motivating a subject'sbehavior, there is no reason to expect that thisbout duration will remain unchanged in all con-ditions where this is physically possible. For in-stance, it is not surprising that bout duration in-creased with increasing ratio size, because theadvantages of maintaining the preferred boutduration had to be weighed against the disadvan-tages of more bar presses being required for eachperiod of access to water. I find each of theseassumptions reasonable, and if the concept of aresponse-pattern set-point involves no otherassumptions than these, I have no objection to it.

In contrast, I have some serious doubts aboutthe usefulness of molar behavioral set-points andof molar regulatory theories in general. All of themolar regulatory theories Timberlake cites sharethe assumption that there is something specialabout the total durations or amounts of behaviorin a (multiple) baseline situation, and that sub-jects (1) are capable of measuring these totaldurations with a fair degree of accuracy, (2) do infact go to the trouble of measuring these totals inboth baseline and contingency sessions, and (3)are motivated to preserve these durations asclosely as possible in a constrained situation. Ifind all three of these assumptions implausible.Data on animal timing (e.g., Gibbon, 1977) andcounting (e.g., Hobson & Newman, 1981) sug-gest that standard deviations increase at least asrapidly as the means of the events to be countedor timed. These data challenge the first assump-tion, for they suggest that subjects' estimates ofmolar events will be quite inaccurate. The abun-dant data on the effects of delayed reinforcement(e.g., Ainslie, 1975) show that animals aremotivationally very short-sighted, and these datarun counter to the second and third assumptions.To make my objections to molar regulatory

theories more clear, imagine an hour-longbaseline session in which a rat drinks water for600 s, presses a lever for 10 s, and spends the re-maining 2990 s performing other types ofbehavior. In a subsequent contingency session inwhich 20 s of drinking is allowed for every 50lever presses, the rat spends 400 s pressing thelever and 500 s drinking. According to molarregulatory theories, the baseline results representthe most preferred distribution for this subject,and its behavior in the contingency sessionrepresents the best possible compromise underthe circumstances. Molar regulatory theoriesassume, in effect, that the animal has as its goal,"accumulate 400 s of lever pressing, 500 s ofdrinking, and 2700 s of other behavior," becausethis is the best possible compromise in the con-tingency condition. In short, molar regulatorytheories assume that this long-term goal is adirect determinant of the subject's behavior in thecontingency session.

I suggest that the typical animal subject has nosuch long-term goal. Indeed, if we must speak ofgoals at all (and the concept of a set-point seemsintimately related to the concept of a goal), then Imaintain that the subject has no constant set ofgoals in an experiment of this kind. To useTimberlake's terminology, the "conditions of in-stigation" vary tremendously from the beginningof the session to the end, both in baseline andcontingency conditions. At the beginning of thebaseline session, the rat's motivation to drink ishigh, and much of its time is spent drinking.Toward the end of the session, the now-satiatedrat spends little time drinking. At the beginningof the contingency session, the rat's motivation topress the lever is higher because this behaviormakes drinking possible, but it is not highenough to allow as much drinking as occurredearly in the baseline session. Because there is lessconsumption of water, the subject's motivation todrink (and consequently its motivation to pressthe lever) does not decline as quickly as in thebaseline session.

In a recent chapter, I presented evidence con-sistent with this scenario (Mazur, 1982, Figure3-1 1), and I argued that it is not long-term goalsbut the moment-to-moment "values" of the dif-ferent types of behavior that govern an animal'stime allocation in such situations. I have madethese points for a number of years (Mazur, 1979,1982), however, and I do not expect the ad-vocates of molar regulatory theories to acceptthem now. All I ask is that they begin to collect

379

REVIEWERS' COMMENTS

and report information on within-session trendsand moment-to-moment patterns of behavior intheir experiments. All too often, the data for eachactivity such as drinking or lever pressing are

reduced to a single number for an entire session,or for several sessions (as in Timberlake's Figures2 and 3). When experiments are controlled bycomputer, it is not difficult to collect much more

detailed information. The reporting of suchsmall-scale behavior patterns would be extremelyvaluable to those of us who believe that the totaldurations or amounts of behavior across an en-

tire session tell relatively little about variablesthat control an animal's behavior.

James E. MazurHarvard University

This paper serves a valuable function in pro-

viding a clear statement of behavior-regulationtheory. Timberlake points out how the use ofbehavioral set-points provides the possibility ofmore precise specifications of "learned perfor-mance in behavior." The strongest section of themanuscript is its assessment of the Allison (1981)paper. Readers of the earlier paper will find Tim-berlake's response cogent, persuasive, and de-serving of publication.

I am not sanguine about the value of this paperfor understanding behavior, nor of its ultimateimpact. In an era in which it seems almost every

laboratory is producing its own theory aimed atdescribing some large segment of operant behav-ior, it has become easy to ignore everyone else'stheories. Devotees of molar regulatory theoriesappear to be a particularly insular and -rightlyor wrongly- ignored band of theoreticians.Thus, several reviewers of earlier versions of thispaper professed little hope for the ultimate im-pact of such theories. I too have reservationsabout the predictive power and usefulness of thisapproach. A related problem, and one whichTimberlake appears to acknowledge, concerns

the apparent lack of criteria, independent ofbehavior, for determining what is and what is nota set-point. We are told an organism has a set-point because a systematic change occurs in a

behavioral measure under increased schedulechallenge. Why does that systematic change oc-

cur? Because the animal is minimizing the devia-tion from a set-point. As I see it, there seems tobe a degree of circularity here (and one less trac-table than that facing reinforcement theory

several decades ago, because regulatory theory isfar more complex). Given that there is no way toidentify a set-point independently of behavior,and given that there appears no way to specify inadvance which of several putative set-points ismore influential, the theory appears overlylimited in terms of predictive power.

Edmund FantinoUniversity of California, San Diego

This paper sets out to clarify what the authorsees as misunderstandings of behavior-regulationtheories of performance. Specifically, he pro-poses to explain the nature and importance of theapproach, to define and describe the notion ofbehavioral set-point, to discuss the place of op-timality in such theories, and to discuss molarversus molecular set-points. He assumes that allthese have been misunderstood or misapplied.

Although the paper is logically arranged andgenerally clear, I am not sure that it really addsvery much that is new with respect to behavior-regulation theories. As one who has not workeddirectly on the issue, I found little that made theapproach any more (or less) defensible or thatenhanced or altered my understanding of it.Consequently, I am not enthusiastic aboutrecommending acceptance, even though I foundthe paper a nice exposition of the behavior-regulation approach. Perhaps I am incorrect, butit is my impression that most workers in the ex-perimental analysis of behavior believe that thebehavior-regulation approach to identifying, apriori, situations in which one will observe rein-forcement (or punishment) is viable and impor-tant. Thus, I am not sure it is in need of muchdefense. And obviously I was not convinced bythe author that the approach has been attacked tothe point that it needs to be defended. A revision,then, should make the case for what seems to meto be largely a reiteration of previous work.

It is not entirely clear whether the regulationapproach is offered as an explanation of rein-forcement effects or as a completely independentnotion. If the latter is intended, it needs to bemore fully defended.The largest portion of my original review was

devoted to issues of clarity. I found several issuesand concepts not clearly presented. The authorresponded effectively to most of my comments.Included in my confusion were the concepts of"instigation," "linkage," "set-point," "severity of

380

REVIEWERS' COMMENTS

challenge," and "minimal constraint." The revi-sion makes what is meant by "instigation" and"set-point" more easily understood, and refer-ences to "linkage" and "severity of challenge" havebeen minimized. Also, the notion of "minimalconstraint" has been dealt with. In the revision,however, the concept of "instigation" stilltroubles me. For example, it is stated that in-stigation refers to the combined effects ofvariables both within and outside the test sessionthat lead to engaging in certain activities. Yet it islater claimed that "whether an organism isdeprived of a commodity outside the experimen-tal session is not . . . interesting." This certainlyseems contradictory to me. I also am still a bitconfused by what constitutes a "set-point." Itseems that any activity, including latency to eat ina novel environment, that changes as a functionof constraint can be viewed at least potentially as

a "set-point."The situation in which the author is trying to

determine relative contributions of molecularand molar set-points is one in which the two are

perfectly confounded. Even in the Marwine andCollier (1979) study, total contingent respondingeither is not constrained or is only minimallyconstrained. Once the subject completes one

ratio, it can drink forever (theoretically). Adiscussion is needed of the kinds of tests (if theyexist) that would allow one to separate molar andmolecular set-point contributions.

Marc BranchUniversity of Florida

Theoretical approaches may differ in terms ofhow they organize or conceptualize interestingvariables and effects. The usual move within theresponse-strengthening approach has been togroup into separate conceptual categories vari-ables that alter the potential of an event to func-tion as a reinforcer or punisher (establishingoperations; Michael, 1982) and variables thatdetermine the effect of a reinforcer of givenpotency on the emission rate of a response (con-tingency operations). The establishing operationthat we are most familiar with is, of course, fooddeprivation, which is an operation made outsidethe experimental session and one that has rela-tively prolonged and stable effects. It is recog-nized, however, that many significant establish-ing operations have relatively fleeting effects andare generated by events within the experimental

session (Herrnstein, 1977; Michael, 1982). Theresult of any experimental procedure is viewed asthe result of the various establishing operations(intended and unintended), the various contin-gencies, and the various discriminative-stimuluseffects (e.g., Logan, 1964). It seems an openquestion whether the regulation-based theories,which conceptualize and organize these variablesdifferently, will be more successful than response-strengthening-based theories in accounting forinteresting aspects of behavior in different situa-tions. At the present time, however, there seemsto be no compelling reason to assume that theo-ries developed within the response-strengtheningtradition are incapable of dealing effectively withsuch data. Clearly, theorists in this tradition havebeen sensitive to the kinds of effects discussedhere and have been interested in learning moreabout the various kinds of establishing opera-tions.

In discussing molar and molecular set-points,it would seem important to emphasize that termslike set-point and regulation refer to concepts thatmight or might not turn out to be useful inorganizing data. There seems to be no reason, inprinciple, why the same data could not bedescribed using the concepts of the traditionalresponse-strengthening approach. The reinforc-ing potency of a bite of food in a restaurant prob-ably is a function of both long-term variables(e.g., the time since I've last eaten, the totalcalories that I've consumed during the last week,etc.) and short-term variables (e.g., the size ofeach bite and the time between each bite). If thewaiter snatches my plate whenever I pauselonger than five seconds between bites, I prob-ably would come to pause less than usual. Wecould say that my eating more rapidly introducesa punishing effect (I don't like to eat so fast) butthat this punishment is less severe than wouldresult from my plate being taken away and frommy having to order and pay for a new dinner.That these facts could be described in terms ofmolar and molecular set-points is not at issue.The point, simply, is that the data do not have tobe described in those terms. What could bedescribed as evidence that molar set-points aremore strongly defended than molecular set-points (e.g., I'll change my eating tempo beforeI'll sacrifice my regular-sized meals) could bedescribed in terms of different response strengthsdue to different punishment/reinforcement com-binations.Of course, if we could really predict in advance

381

382 REVIEWERS' COMMENTS

what set-points are going to be defended and howthe set-points are weighted when circumstancesplace them in conflict, then the advantage of theregulation approach would be apparent. How-ever, it is clear from the discussion that we arenot yet in a position to do this. We even see thatcomplex aspects of behavior such as the tendencyto sample might be regulated. In view of thesekinds of complexities and uncertainties, it seemsfair to wonder how real the advantages of theregulation approach are relative to those of thetraditional response-strengthening approach.

Richard L. ShullUniversity of North Carolina, Greensboro

REFERENCES

Ainslie, G. (1975). Specious reward: A behavioraltheory of impulsiveness and impulse control.Psychological Bulktin, 82, 463-496.

Allison, J. (1981a). Paired baseline performance asa behavioral ideal. Journal of the Experimental Anal-ysis of Behavior, 35, 355-366.

Allison, J. (1981b). Economics and operant condi-tioning. In P. Harzem & M. D. Zeiler (Eds.),Advances in analysis of behaviour: Vol. 2. Predictability,correlation and contiguity (pp. 321-353). Chichester:Wiley.

Gibbon, J. (1977). Scalar expectancy theory andWeber's law in animal timing. Psychological Review,84, 279-325.

Herrnstein, R. J. (1977). The evolution of behavior-ism. American Psychologist, 32, 593-603.

Hobson, S. L., & Newman, F. (1981). Fixed-ratio-counting schedules: Response and time measuresconsidered. In M. L. Commons & J. A. Nevin(Eds.), Quantitative analyses of behavior: Vol. 1.Discriminative properties of reinforcement schedules(pp. 193-224). Cambridge, MA: Ballinger.

Hursh, S. R. (1980). Economic concepts for theanalysis of behavior. Journal of the Experimental Anal-ysis of Behavior, 34, 219-238.

Hursh, S. R., & Natelson, B. H. (1981). Electricalbrain stimulation and food reinforcement dissoci-ated by demand elasticity. Physiology & Behavior,26, 509-515.

Logan, F. A. (1964). The free behavior situation. InD. Levine (Ed.), Nebraska Symposium on Motivation(Vol. 12, pp. 99-128). Lincoln: University ofNebraska Press.

Marwine, A., & Collier, G. (1979). The rat at thewaterhole. Journal of Comparative and PhysiologicalPsychology, 93, 391-402.

Mazur, J. E. (1979). The value-averaging model:An alternative to "conservation of behavior."Jour-nal of Experimental Psychology: General, 108, 43-47.

Mazur, J. E. (1982). A molecular approach to ratioschedule performance. In M. L. Commons, R. J.Herrnstein, & H. Rachlin (Eds.), Quantitative anal-yses of behavior: Vol. 2. Matching andnmaximizingaccounts (pp. 79-110). Cambridge, MA: Ballinger.

Michael, J. (1982). Distinguishing between discrim-inative and motivational functions of stimuli. Jour-nal of the Experimental Analysis of Behavior, 37,149-155.

Nevin, J. A., Mandell, C., & Atak, J. R. (1983).The analysis of behavioral momentum. Journal ofthe Experimental Analysis of Behavior, 39, 49-59.