Pigeons’ demand and preference for specific and generalized … · Pigeons were studied in a...

19
PIGEONSDEMAND AND PREFERENCE FOR SPECIFIC AND GENERALIZED CONDITIONED REINFORCERS IN A TOKEN ECONOMY LAVINIA TAN AND TIMOTHY D. HACKENBERG REED COLLEGE Pigeonsdemand and preference for specific and generalized tokens was examined in a token economy. Pigeons could produce and exchange different colored tokens for food, for water, or for food or water. Token production was measured across three phases, which examined: (1) across-session price increases (typical demand curve method); (2) within-session price increases (progressive-ratio, PR, schedule); and (3) concurrent pairwise choices between the token types. Exponential demand curves were fitted to the response data and accounted for over 90% total variance. Demand curve parameter values, P max , O max and a showed that demand was ordered in the following way: food tokens, generalized tokens, water tokens, both in Phase 1 and in Phase 3. This suggests that the preferences were predictable on the basis of elasticity and response output from the demand analysis. P max and O max values failed to consistently predict breakpoints and peak response rates in the PR schedules in Phase 2, however, suggesting limits on a unitary conception of reinforcer efficacy. The patterns of generalized token production and exchange in Phase 3 suggest that the generalized tokens served as substitutes for the specific food and water tokens. Taken together, the present findings demonstrate the utility of behavioral economic concepts in the analysis of generalized reinforcement. Key words: generalized conditioned reinforcement, token reinforcement, demand analysis, progressive ratio schedules, concurrent schedules, behavioral economics, pigeons, key peck Skinner (1953) defined generalized reinforcers as reinforcers established and maintained via relations with two or more reinforcers. General- ized reinforcers are ubiquitous in everyday human behavior. Money is the most common example, but Skinner also discussed the impor- tance of generalized reinforcement to social and verbal behavior; signs of attention and approval from others are closely coupled with a wide range of other reinforcers and therefore acquire important generalized reinforcing functions. Add to this the important role of generalized reinforcement in applied settings, in the form of token economies, and it becomes clear that the topic of generalized reinforcement is of enor- mous conceptual and applied significance. The expansion from one to multiple rein- forcers is functionally significant in that it also expands the number of relevant motivational operations. According to Skinner (1953), and to subsequent interpretations, generalized reinforcers are assumed be more effective than specific-type reinforcers, related to only a single deprivation, because they recruit multiple motivational conditions, any of which may be present at the time of the response. Less dependent on specific deprivation states, gen- eralized reinforcers should be more effective than specific reinforcers across a wider range of circumstances. The assertion is both plausible and testable. Yet in the 60þ years since Skinners initial conceptualization, the empirical analysis of generalized reinforcement has scarcely begun. Despite the importance of the topic, little is known about even the most basic functions of generalized reinforcers. A broad aim of the present study was to expand the analysis of generalized reinforcement by comparing gen- eralized reinforcers to specific reinforcers at various costs and in various motivational con- ditions. Grounding the analysis in concepts and methods from behavioral economics, rein- forcer efficacy was defined in terms of demand, response output, and preference. Pigeons were studied in a token economy, in which different token types could be earned and exchanged for different commodities. The research builds on a recent study by DeFulio, Yankelevitz, Bullock, and Hackenberg (2014), in which pigeons could produce and exchange Author Notes: Research supported by NIDA Grant R01 DA026127. We are indebted to Greg Wilkinson for technical support and to Carol Franceschini for comments on an earlier version of the paper. Correspondence to: Timothy D. Hackenberg, Psychology Department, Reed College, 3203 SE Woodstock Blvd, Portland OR 97202-8199, Phone: 503-459-4623, E-mail: [email protected] doi: 10.1002/jeab.181 JOURNAL OF THE EXPERIMENTAL ANALYSIS OF BEHAVIOR 2015, 104, 296314 NUMBER 3 (NOVEMBER) 296

Transcript of Pigeons’ demand and preference for specific and generalized … · Pigeons were studied in a...

Page 1: Pigeons’ demand and preference for specific and generalized … · Pigeons were studied in a token economy, in which different token types could be earned and exchanged for different

PIGEONS’ DEMAND AND PREFERENCE FOR SPECIFIC AND GENERALIZED CONDITIONEDREINFORCERS IN A TOKEN ECONOMY

LAVINIA TAN AND TIMOTHY D. HACKENBERG

REED COLLEGE

Pigeons’ demand and preference for specific and generalized tokens was examined in a token economy.Pigeons could produce and exchange different colored tokens for food, for water, or for food or water.Token production was measured across three phases, which examined: (1) across-session price increases(typical demand curve method); (2) within-session price increases (progressive-ratio, PR, schedule); and(3) concurrent pairwise choices between the token types. Exponential demand curves were fitted to theresponse data and accounted for over 90% total variance. Demand curve parameter values, Pmax,Omax anda showed that demand was ordered in the following way: food tokens, generalized tokens, water tokens,both in Phase 1 and in Phase 3. This suggests that the preferences were predictable on the basis of elasticityand response output from the demand analysis. Pmax and Omax values failed to consistently predictbreakpoints and peak response rates in the PR schedules in Phase 2, however, suggesting limits on aunitary conception of reinforcer efficacy. The patterns of generalized token production and exchange inPhase 3 suggest that the generalized tokens served as substitutes for the specific food and water tokens.Taken together, the present findings demonstrate the utility of behavioral economic concepts in theanalysis of generalized reinforcement.

Key words: generalized conditioned reinforcement, token reinforcement, demand analysis, progressiveratio schedules, concurrent schedules, behavioral economics, pigeons, key peck

Skinner (1953) defined generalized reinforcersas reinforcers established and maintained viarelations with two or more reinforcers. General-ized reinforcers are ubiquitous in everydayhuman behavior. Money is the most commonexample, but Skinner also discussed the impor-tance of generalized reinforcement to social andverbal behavior; signs of attention and approvalfrom others are closely coupled with a widerange of other reinforcers and therefore acquireimportant generalized reinforcing functions.Add to this the important role of generalizedreinforcement in applied settings, in the formoftoken economies, and it becomes clear that thetopic of generalized reinforcement is of enor-mous conceptual and applied significance.

The expansion from one to multiple rein-forcers is functionally significant in that it alsoexpands the number of relevant motivationaloperations. According to Skinner (1953), andto subsequent interpretations, generalized

reinforcers are assumed be more effectivethan specific-type reinforcers, related to onlya single deprivation, because they recruitmultiple motivational conditions, any of whichmay be present at the time of the response. Lessdependent on specific deprivation states, gen-eralized reinforcers should be more effectivethan specific reinforcers across a wider range ofcircumstances.

The assertion is both plausible and testable.Yet in the 60þ years since Skinner’s initialconceptualization, the empirical analysis ofgeneralized reinforcement has scarcely begun.Despite the importance of the topic, little isknown about even the most basic functions ofgeneralized reinforcers. A broad aim of thepresent study was to expand the analysis ofgeneralized reinforcement by comparing gen-eralized reinforcers to specific reinforcers atvarious costs and in various motivational con-ditions. Grounding the analysis in concepts andmethods from behavioral economics, rein-forcer efficacy was defined in terms of demand,response output, and preference.

Pigeons were studied in a token economy, inwhich different token types could be earnedand exchanged for different commodities. Theresearch builds on a recent study by DeFulio,Yankelevitz, Bullock, and Hackenberg (2014),in which pigeons could produce and exchange

Author Notes: Research supported by NIDA Grant R01DA026127. We are indebted to Greg Wilkinson fortechnical support and to Carol Franceschini for commentson an earlier version of the paper.Correspondence to: Timothy D. Hackenberg, Psychology

Department, Reed College, 3203 SE Woodstock Blvd,Portland OR 97202-8199, Phone: 503-459-4623, E-mail:[email protected]: 10.1002/jeab.181

JOURNAL OF THE EXPERIMENTAL ANALYSIS OF BEHAVIOR 2015, 104, 296–314 NUMBER 3 (NOVEMBER)

296

Page 2: Pigeons’ demand and preference for specific and generalized … · Pigeons were studied in a token economy, in which different token types could be earned and exchanged for different

three different token types, singly and incombination: (1) specific-food tokens (greentokens exchangeable for food), (2) specific-watertokens (red tokens exchangeable for water), and(3) generalized tokens (white tokens exchange-able for either food or water). The fixed ratio(FR) token-production price for specific andgeneralized tokens was manipulated underboth food or water deprivation, in separateconditions of the study.Token production and exchange came un-

der control of the deprivation conditions: thepigeons earned mainly food tokens under fooddeprivation and water tokens under waterdeprivation. The generalized tokens that wereearned were exchanged nearly exclusively forthe restricted commodity. As the FR prices oftoken production were systematically increasedacross blocks of sessions, token productiondeclined, characteristic of typical demandfunctions, for both specific and generalizedtokens in both deprivation contexts. The rate ofdecline, however, was somewhat higher forspecific than for generalized tokens, especiallyunder food deprivation conditions.In economic terms, the own-price elasticity

values were less negative for generalized thanfor specific token types. Moreover, the slopes ofthe cross-price functions, indicating the degreeto which the constant-price token came tosubstitute for the increasing-price token, weremore positive for generalized than for specifictokens. Pigeons produced roughly similarnumbers of food tokens and generalized tokenswhen the prices were low (FR 1), but came toproduce substantially more generalized tokensas the price of the food tokens increased andthe price of the generalized tokens remainedlow. Thus, the generalized tokens came tosubstitute for the food tokens, one importantfunction of generalized reinforcers. The samegeneral pattern occurred in the context ofwater deprivation – the generalized tokenssubstituted for water tokens – though to asomewhat lesser degree.Due to the restricted range of prices explored

by DeFulio et al. (2014), it was not possible toevaluate demand functions using more quanti-tative criteria. One goal of the presentstudy, therefore, was to compare demand andpreference for different token types usingmore quantitatively precise methods. Usingprocedures similar to those employed byDeFulio et al., pigeons could produce and

exchange red tokens for water, green tokens forfood, and white tokens for either food or water.Using methods adapted from Madden, Sme-thells, Ewan, and Hursh (2007) for comparingqualitatively different reinforcers, the presentstudy examined demand, response output, andpreference for the token types under variousconditions of food and water deprivation.In Phase 1, demand for each token type was

assessed by increasing price (token productionFR) across sessions, and evaluated using Hurshand Silberberg’s (2008) exponential model:

logQ ¼ logQ 0 þ kðe−aQ0C−1Þ;where demand (Q) is consumption (or numberof reinforcers obtained) as a function of unitprice (P, FR requirement); Q0 is the consump-tion at the lowest price. The slope of thefunction (a) expresses the sensitivity to pricechanges, or elasticity. Other informative param-eters include Pmax, the maximum price beforethe decline in consumption at which thefunction turns from inelastic to elastic (slope¼−1.0), and Omax, the peak response output. Thethree token types were evaluated in relation tothese key parameters from the model.In Phase 2, demand was assessed by changing

token-production price within a session, accord-ing to a progressive ratio (PR) schedule thatincremented with each reinforcer. This yieldedadditional measures—last completed FR, orbreakpoint, and peak response rates—withwhich to compare the token types and theirdemand functions. In Phase 3, the differenttoken types were available concurrently forblocks of sessions in pairwise choices, yieldingan additional measure—relative preferencebetween token types—with which to compareto the demand measures. Together, thesephases provided a comprehensive analysis ofdemand and preference for specific andgeneralized token types.Bringing this type of analytic focus to the

generalized token economy will provide a morecomplete picture of the functions and rela-tive efficacy of specific and generalized tokensthan prior research. If generalized reinforcersare indeed more effective than more specificreinforcers, as is commonly assumed, thenthe generalized tokens, relative to thespecific tokens, should generate higher levelsof demand intensity (token production at thelowest price), lower levels of demand elasticity

GENERALIZED CONDITIONED REINFORCEMENT 297

Page 3: Pigeons’ demand and preference for specific and generalized … · Pigeons were studied in a token economy, in which different token types could be earned and exchanged for different

(sensitivity to price changes), elevated responseoutput (Pmax, Omax, peak response rate), andgreater preference (choice proportion).

The multiple converging measures of rein-forcing efficacy (demand intensity, demandelasticity, breakpoint, peak response rate, pref-erence) are important in light of prior researchshowing less than perfect agreement amongcommon indices of reinforcer efficacy (e.g.Bickel & Madden, 1999; Bickel, Marsch, &Carroll, 2000; Madden et al., 2007). For exam-ple, Madden et al. found, with rats as subjectsand food and water reinforcers, that differentmeasures of reinforcer efficacywerenot stronglycorrelated: higher PR breakpoints and demandpeak response rates were obtained for food thanwater, but the rats did not show correspondingpreferences when both food and water wereavailable concurrently. Conversely, JohnsonandBickel (2006) found, with human subjects andmoney and cigarettes as reinforcers, that relativePmax and Omax values correlated with PR break-points and peak response rates, and predictedpreference between cigarettes and money (seealso Bickel & Madden, 1999).

These different results may be due toreinforcer interactions. In economic terms,food and water are complementary reinforcers(consumption of one reinforcer decreases withprice increases of the other reinforcer) whereascigarettes and money are independent rein-forcers (consumption of one reinforcer isunrelated to price changes of the otherreinforcer). Perhaps the close coupling ofdifferent measures of reinforcer efficacy seenwith human subjects in the studies by Bickel andcolleagues is limited to situations with indepen-dent reinforcers. Or, perhaps the findings aredue to substitution, a different type of reinforcerinteraction, in which consumption of a con-stant-priced commodity increases with the priceof an alternate commodity (e.g., increasedconsumption of tea with increases in the priceof coffee). Generalized reinforcers such asmoney are effective in part because they serveas functional substitutes for a wide range ofother reinforcers; or more specifically, theyenable a system of exchange between substitut-able reinforcers. In the present economicconditions, this translates into increased pro-duction of generalized reinforcers with priceincreases for more specific reinforcers.

The present study was designed to explorein greater depth these types of reinforcer

interactions in a generalized token economy.First, unlike the DeFulio et al. (2014) study,some conditions in the present study wereconducted in a fully closed economy in whichdaily intake of food and water occurred withinthe experimental sessions, designed to enhancethe generalized functions of the tokens. Sec-ond, using known complementary reinforcers(food and water) but including a generalizedtype with the potential to function as asubstitute for these more specific reinforcers,we directly assessed the substitutability functionof the generalized tokens. And finally, themultiple converging measures enabled a multi-faceted assessment of the functions of specificand generalized tokens, and the degree towhich the different indices of reinforcerefficacy converge. Together, the methodsprovide quantitatively precise ways to assesssome basic functions of generalized condi-tioned reinforcers in a motivationally complexbut analytically tractable environment.

Method

SubjectsSix male White Carneau pigeons served as

subjects. The pigeons were obtained fromDouble T Farms (Glenwood, IA), and wereapproximately 2 years of age and experimen-tally na€ıve at the start of the experiment. Theywere housed individually in stainless steel homecages in a colony room with a 12-hr light/darkcycle. Most conditions were conducted in aclosed economy, in which daily intake of foodand/or water occurred within daily sessions,occasionally supplemented with postsessionaccess, as described below.

ApparatusPigeons were trained and tested in custom-

made operant chambers measuring 50.8 cm by57.15 cm by 45.72 cm. An Elo1 touchscreen,measuring 22.86 cm by 30.48 cm, with a pixelresolution of 1024 by 768, was located at thefront of the chamber, situated 11.43 cm abovethe wire mesh floor. Illumination was providedby a houselight located at the top of thechamber. Access to Purina Nutriblend1 foodpellets was provided via a mechanicalhopper and opening located underneath thetouchscreen, 24.13 cm from each side of thechamber. Food delivery was signaled by the

298 LAVINIA TAN and TIMOTHY D. HACKENBERG

Page 4: Pigeons’ demand and preference for specific and generalized … · Pigeons were studied in a token economy, in which different token types could be earned and exchanged for different

onset of the hopper light. Water was deliveredusing a MED-PC1 pump, via a plastic tube,which fed into a 4 cm� 3 cm� 2.3 cm plasticreservoir located underneath the right side ofthe touchscreen, the center of which was15.24 cm from the right hand side of thechamber. Water delivery was signaled bythe onset of a single LED light located abovethe container. Programs were created in VisualStudio 2008©, and run using Dell1 computersand Windows 7 OS©. White noise was played onan external speaker located behind eachtouchscreen during all sessions to mask exter-nal noise. An external exhaust fan providedventilation for all the chambers.

PretrainingThe pigeons were trained to produce and

exchange green tokens (exchangeable forfood), red tokens (exchangeable for water),and white tokens (exchangeable for eitherfood or water). Figure 1 shows the locations of

token production keys and tokens for each ofthe three commodities on the touchscreen.The houselight was illuminated for the dura-tion of the session and was only turned offduring the delivery of the food or waterreinforcers. All training sessions ended after60min or 50 reinforcers had been delivered,whichever came first. Each training conditioncontinued until reliable responding had beenacquired.The order in which training occurred for the

food and water tokens was counterbalancedacross groups of pigeons: three pigeons (126,1851, 764) were trained first with the green(food) tokens, and second with the red (water)tokens, while the other three pigeons (1770,1750, 136) were trained first with red (water)tokens, and second with green (food) tokens.Pigeons in both groups received trainingwith the white (generalized) tokens last; com-plete counterbalancing was not feasible, asthe generalized condition necessarily followedthe earlier conditions in which the specific

Fig. 1. Location of all stimuli on the touchscreen for experimental conditions. (Note: all of these were never visiblesimultaneously).

GENERALIZED CONDITIONED REINFORCEMENT 299

Page 5: Pigeons’ demand and preference for specific and generalized … · Pigeons were studied in a token economy, in which different token types could be earned and exchanged for different

functions of the green and red tokens wereestablished. These conditions will be describedseparately.

Food token training. Food was unavailableoutside of the training sessions except, asneeded, to maintain body weights to at least85% free feeding weights. (Once the experi-mental conditions began, no supplementalfeeding was necessary.) Water was unavailableduring the sessions, but continuously availableoutside the session in home cages. Pigeons werefirst trained to eat from the food hopper and topeck a green square presented in the middle ofthe touchscreen; each peck produced 4.5-saccess to food. When pigeons were reliablypecking the green key, responding was placedunder a FR 1 schedule, in which each peckproduced food. Pigeons were then trained topeck a token, a flashing green square (sized1.47 cm2 that flashed on/off). The location ofthe token varied randomly within an array of10� 4 squares, presented on the left hand sideof the screen) for food reinforcement. Whenthis token-exchange response had been estab-lished, token-production responding was estab-lished by presenting the flashing green token asa consequence of a peck on a solid green squarelocated in the center of the screen. When thistoken-production response had been estab-lished, the FR token production requirementsincreased (as described below, Phase 1: FRDemand).

Water token training. Water was unavailableexcept during the training sessions; food wascontinuously available in home cages outsidethe sessions. Pigeons were trained to peck a redsquare presented on the left hand side of thescreen for 2.8 s of water reinforcement (equiva-lent to 0.51 mls, an amount tested andconfirmed to be sufficient for pigeons toconsume with the water reservoir we used).For some pigeons, the water reinforcementcontingencies evoked beak dragging, ratherthan individual key pecks, and this resulted inerrors in recording responses. Consequently,the size of the red square was systematicallyreduced until responses became more discrete.The size of the red square was then graduallyincreased back to the original size, while thepecking topography remained discrete. Thebeak dragging topography did not reemerge forthe remainder of the study. Training thenfollowed the same procedure described abovefor food, except using water reinforcers and red

tokens instead of food reinforcers and greentokens.

Generalized token training. Food and waterwere unavailable outside of the sessions; all dailyintake therefore occurred within the sessions.Given the prior training with the green foodtokens and the red water tokens, no explicittraining was needed to establish responding forwhite (generalized) tokens. A single peck on awhite flashing token turned it off and produceda green and red token on the immediate leftand right side of the white token, respectively. Apeck on the red token turned off both tokensand produced a water reinforcer; a peck on thegreen key turned off both tokens and produceda food reinforcer. The pigeon could thuschoose which reinforcer (water or food) forwhich to exchange a white token.

Experimental ProcedureThe experiment consisted of three phases,

with the reinforcing efficacy of the three tokentypes assessed within each phase, either sepa-rately or in combination. In the first phase,demand functions were produced for each ofthe three token types separately, by increasingFR response requirements across successivesessions. In the second phase, PR breakpointsand peak response outputs were obtained foreach token type separately by increasing FRrequirements within a session. In the third andfinal phase, preference was assessed usingpairwise combinations of the token types. Thesephases will be described separately.

Phase 1: FR Demand. Demand functionswere obtained for the two specific token types(food and water) with the order counterbal-anced across pigeons. As in the trainingconditions, when the demand functions werebeing generated for one commodity, thepigeons had free access to the other commodityoutside the session. Three-hour experimentalsessions were conducted daily. The FR sched-ules increased through the progression 1, 5, 15,30, 50, 90, 150, 300, 500, 900, 1500, with each FRvalue presented for 5 successive days. Theprogression was discontinued if a pigeon failedto earn more than five reinforcers for threeconsecutive sessions; this was followed by animmediate return to unlimited access to foodand water. On the rare occasion in which theseconditions occurred, it was brought about byweak responding under Water conditions. The

300 LAVINIA TAN and TIMOTHY D. HACKENBERG

Page 6: Pigeons’ demand and preference for specific and generalized … · Pigeons were studied in a token economy, in which different token types could be earned and exchanged for different

three-session rule we adopted was based onprior research with water restriction in pigeons(Jenkins & Moore, 1973; Lumeij, 1987; Martin& Kollias, 1989), and therefore was considereda reasonable limit.The demand functions for specific food and

water tokens were then repeated, but with FRchanges occurring each session rather thanafter five sessions. All other aspects of theprocedure were identical to those describedabove for the 5-day method. Once the progres-sion had been completed once, reinforcerswere switched (from food to water, or viceversa) and the progression repeated with thealternate reinforcer. Because there was nosignificant difference in data between the1-day and 5-day demand tests (see Results),the remaining conditions with the white(generalized) tokens used the 1-day methodonly. These generalized token conditions wereconducted in a closed economy, in which dailyintake of both food and water occurred withinthe 3-hr daily experimental sessionsPhase 2: Progressive Ratio. Pigeons were

tested under a PR schedule, in which theresponse requirement increased followingeach reinforcer according to the followingequation (adapted from Madden et al. 2007):

PR ¼ ð10�ðe0:08xÞÞ−3;where x was the trial number (number ofobtained reinforcersþ 1). Three blocks of five3-hr sessions were run for each token type,interspersed by 2 days of free food and water toregain baseline levels of intake, and one day ofdeprivation. Pigeons experienced the sameorder of all three token types as in the demandphase for better comparability between thesetwo phases. The procedure was otherwiseidentical to the previous condition, except forthe following procedural modifications toincrease water intake in some conditions withgeneralized tokens. Due to low levels of water-token exchange and water intake in the firstseveral sessions with generalized tokens, mea-sures were taken to increase daily water intake,while still carrying through with the PRmanipulation. First, the duration of the waterdispenser was increased from 2.8 s to 4.8 s(increasing water volume to 0.87ml per rein-forcer). Second, 10ml of water was providedimmediately following any sessions in whichwater was not consumed.

Phase 3: Preference. To test for preference,the different token types were arranged con-currently within a session, as pairwise choices, inthe following order: Food (green) versus Water(red), Food (green) versus Generalized (white), andWater (red) versus Generalized (white). All food andwater was earned during the experimentalsessions, similar to the generalized tokenconditions, described above. Within each com-parison, the FR price of each token wasincreased systematically (and together) acrossblocks of three sessions, using the sameprogression as in the FR Demand assessmentin Phase 1. Completing the FR requirements onone key turned off both token production keysand presented a single flashing token within therespective array: green for food, red for water,and white for either food or water (as describedabove). This condition ended when the end ofthe FR progression had been reached, or iffewer than six water reinforcers were obtainedper session over three consecutive sessions (thelatter criterion was met just twice, once each forPigeons 126 and 1770 in the Food vs. Watercondition). The pigeons were then returned tofree access to food and water prior to the nextblock of sessions. This free-access phase lastedfor 11 days between Conditions 1 and 2, and9 days between Conditions 2 and 3.

Results

Phase 1. FR Demand AssessmentFigure 2 shows normalized demand curves

for food (left) and water (right) under both the5-day and 1-day methods, averaged across all sixpigeons. Quantitative fits were performed usingHursh and Silberberg’s (2008) exponentialdemand curve, with Q0, a free to vary, and kset as a constant, calculated for each individual,and across all three commodities, as (maximumlog consumption –minimum log consumption)þ 0.5. Sessions where consumption was 0 wereexcluded from analyses. Dependent samplet-tests with individual data were used tocompare the 5-day and 1-day methods withrespect to the following fitted parameters of themodel: R2, Q0, a, Q Pmax andOmax. For FR valuesthat were replicated over multiple days, meanconsumption was used for model fits. Nosignificant differences were found betweenthe 1-day and 5-day conditions on any of thesemeasures. Given the lack of consistent differ-ence, the final condition with generalized

GENERALIZED CONDITIONED REINFORCEMENT 301

Page 7: Pigeons’ demand and preference for specific and generalized … · Pigeons were studied in a token economy, in which different token types could be earned and exchanged for different

tokens was conducted under the 1-day methodonly. To maintain comparability across thetoken types, all results were analyzed using the1-day data only.

Figure 3 shows normalized demand functionsfor all six pigeons and token types. Parametervalues from demand curve model fits forEquation 1 can be seen in Table 1 (absolutefrequencies of responses and reinforcers areshown in Appendix A in the online supplemen-tal materials). Normalized reinforcer magni-tude units (q) were calculated using theequation q¼ 100/B, where B is consumptionat the lowest price (FR 1). Consumption at eachFR value was multiplied by q to obtain normal-ized consumption; normalized price was calcu-lated as FR/q. The functions are well describedby the exponential model, accounting for anaverage of 94% of total variance overall, andbetween 93-96% total variance across the threecommodities.

Despite the good overall fits, the functions fordifferent token types differ in consistent ways.To begin with, the functions differ in elasticity(a). Figure 4 (top left panel) shows individualand group mean a values. The mean valueswere consistently highest (suggesting greaterelasticity) for water (M¼ 1.83E-05, SD¼ 8.31E-0.6), followed by generalized (M¼ 8.62E-06,SD¼ 4.03E-06), and lastly, food (M¼ 5.22E-06,SD¼ 1.67E-06). (Elasticity values differed sig-nificantly, F(2,10)¼ 16.28, p< .001). A TukeyHSD test found that food tokens, but notgeneralized tokens, were significantly moreinelastic than water tokens (mean differencebetween water and food¼ 1.31E-05, p< .001,and mean difference between water and

generalized¼ 9.67E-06, p< .01). There was nosignificant difference in elasticity for food andgeneralized tokens.

The bottom left panel of Figure 4 shows Q0,consumption at the lowest price. Unlike elastic-ity, Q0 was highest for generalized tokens(M¼ 1008.64, SD¼ 483.96), followed by food(M¼ 721.4, SD¼ 448.4) and then water(M¼ 91.02, SD¼ 12.47). These values differedsignificantly, F(2,10)¼ 14.48, p< .005. A TukeyHSD test found consumption of water tokenswas significantly lower than both food tokens(mean difference¼ 630.4, p< .01) and general-ized tokens (mean difference¼ 917.98,p< .01).

The functions also differed in Pmax values,reflecting different degrees of price tolerance.The top right panel of Figure 4 shows individualand group mean Pmax values. These values wereconsistently greatest for food tokens(M¼ 330.70, SD¼ 97.83), followed by general-ized (M¼ 214.50, SD¼ 81.59), followed bywater tokens (M¼ 99.37, SD¼ 33.21). A repea-ted measures ANOVA found a significantdifference in Pmax values (F(2,10)¼ 29.43,p< .001) and a Tukey HSD test confirmedsignificant differences between all pairwisecombinations of commodities (p< .01).

The functions also differed in peak responseoutput (Omax) values (bottom right panel ofFig. 4). In general, the Omax values showed thesame pattern as Pmax values at both theindividual and group level; a repeatedmeasuresANOVA found a significant effect of commod-ity, F(2,10)¼ 29.39, p< .01.Omax values for foodtokens were consistently highest, followed bygeneralized, followed by water; Tukey HSD tests

Fig. 2. Fully normalized demand curves using group average data from the 5-day and 1-day food and water conditions.Normalized consumption is plotted as a function of normalized price for food and water in the left and right panels,respectively.

302 LAVINIA TAN and TIMOTHY D. HACKENBERG

Page 8: Pigeons’ demand and preference for specific and generalized … · Pigeons were studied in a token economy, in which different token types could be earned and exchanged for different

confirmed that all these differences weresignificantly different (p< .01).

Phase 2. Progressive RatioFigure 5 shows the PR breakpoints, averaged

across all three blocks, for each token type foreach pigeon, across the 15 sessions per

condition (14 sessions in the Generalized PRphase). Absolute response and reinforcer datafor each pigeon are shown in Appendix B in theonline supplemental materials. Averaged acrosspigeons, the highest breakpoints were obtainedfor food tokens (M¼ 279.63, SD¼ 118.75),followed by water tokens (M¼ 214.26, SD¼ 75.67), then generalized tokens (M¼ 206.60,

Fig. 3. Rapid (1-day) demand curves. Normalized consumption plotted as a function of normalized price for food(squares), water (circles) and generalized (triangles), for each individual.

GENERALIZED CONDITIONED REINFORCEMENT 303

Page 9: Pigeons’ demand and preference for specific and generalized … · Pigeons were studied in a token economy, in which different token types could be earned and exchanged for different

SD¼ 37.24). However, a one-way repeated mea-sures ANOVA found no significant effect ofcommodity (F¼ 1.71, n.s.). This is most likelydue to the inconsistency in the group andindividual breakpoint functions. Breakpoints forfood tokens were markedly greater than waterand generalized tokens for four of the sixpigeons (1851, 1770, 1750, 136) but wereapproximately similar to generalized tokensfor the remaining subjects (126, 764). Break-points for water tokens were noticeably greaterthan food and generalized tokens for Pigeons764 and 1770.

Peak response output patterns in Phase 2resembled those seen in the PR breakpoints.Mean peak response outputs were greatest forfood (M¼ 3244.37, SD¼ 1476.41), followed bywater (M¼ 2408.54, SD¼ 904.95), and thengeneralized tokens (M¼ 2321.31, SD¼ 1113.55).There was no significant difference foundbetween these (F¼ 1.86, n.s.).

Phase 3. Preference AssessmentFigure 6 shows the proportion of the differ-

ent token types produced for each condition,averaged across all prices. Absolute frequenciesof responding and reinforcement for eachpigeon are shown in Appendix C in the onlinesupplemental materials. A strong preferencefor food tokens can be observed in the Foodversus Water (F-W) and Food versus Generalized(F-G) comparisons, and a strong preference forgeneralized tokens in the Water versus General-ized (W-G) comparisons. A repeated measuresANOVA found a significant difference betweenproportion of tokens produced between allthree commodities, averaged across pairingsF(2,17)¼ 144.9, p< .001. Bonferroni post-hoctests showed all commodities differed signifi-cantly from each other (p< .05). AdditionalBonferroni tests were conducted to examineeffects within and across pairings. Preferencefor the available token types were signifi-cantly different within each pairing (p< .05).No significant differences were found for thefood and water tokens across conditions,suggesting preference for these commoditiesremained constant regardless of whether it wasavailable concurrently with the other primaryreinforcer or the generalized reinforcer. Con-versely, preference for generalized tokens wasdependent on the concurrently available tokentype; preference for generalized tokens was

Tab

le1

Parameter

values

for1-da

yex

ponen

tial

deman

dcu

rves

fitted

toaveragegrou

pan

dindividu

alda

ta.

k

R2

Normalized

P max

Omax

Q0

a

Globa

lFo

odWater

Gen

Food

Water

Gen

Food

Water

Gen

Food

Water

Gen

Food

Water

Gen

Group

3.97

92.4

91.8

97.3

90.9

572.4

116.2

249.7

1823

9.6

3620

.115

945.8

755.23

84.17

539.76

2.33

E-06

1.15

E-05

5.34

E-06

126

3.2

97.6

97.2

96.1

98.4

457.3

108.6

241.1

1518

7.8

3606

.680

0872

2.78

79.89

540.91

3.54

E-06

1.49

E-05

6.71

E-06

1851

3.43

97.3

96.7

95.4

98.8

207.6

45.6

132.9

6844

.115

02.7

4381

.789

9.73

102.47

1249

.22

7.30

E-06

3.33

E-05

1.14

E-05

764

3.32

91.4

86.7

90.6

96.1

367.1

137.6

237.8

1215

3.8

4557

.178

73.3

992.42

92.07

1195

.36

4.26

E-06

1.14

E-05

6.58

E-06

1770

3.16

93.2

9395

.889

.823

3.4

102.4

226.8

7760

.334

05.2

7540

.621

0.69

100.36

391.63

7.03

E-06

1.60

E-05

7.24

E-06

1750

395

.596

.599

.293

.631

3.5

78.0

112.3

1044

4.3

2599

.537

40.1

186.84

99.37

974.90

5.53

E-06

2.22

E-05

1.54

E-05

136

3.53

90.7

85.9

9893

.140

5.4

123.9

336.0

1330

8.2

4068

.511

030.5

1316

71.94

1699

.81

3.64

E-06

1.19

E-05

4.39

E-06

Mean

3.27

94.3

92.7

95.9

95.0

330.7

99.4

214.5

1094

9.7

3289

.970

95.7

721.41

91.02

1008

.64

5.22

E-06

1.83

E-05

8.63

E-06

SEM

0.08

1.21

2.11

1.21

1.41

39.9

13.6

33.3

1319

.82

447.08

1091

.53

183.06

5.09

197.57

6.82

E-07

3.39

E-06

1.65

E-06

304 LAVINIA TAN and TIMOTHY D. HACKENBERG

Page 10: Pigeons’ demand and preference for specific and generalized … · Pigeons were studied in a token economy, in which different token types could be earned and exchanged for different

Fig. 4. Individual and group mean a, Q0, normalized Pmax and Omax values for 1-day demand curves.

Fig. 5. Mean progressive ratio breakpoints for each pigeon for food (dark gray series), water (black series) andgeneralized (light gray) commodities. Error bars show� 1 SD.

GENERALIZED CONDITIONED REINFORCEMENT 305

Page 11: Pigeons’ demand and preference for specific and generalized … · Pigeons were studied in a token economy, in which different token types could be earned and exchanged for different

Fig. 6. Average total proportion of tokens produced in the Food-Water, Food Generalized andWater-Generalized conditionsin the preference demand assessment.

306 LAVINIA TAN and TIMOTHY D. HACKENBERG

Page 12: Pigeons’ demand and preference for specific and generalized … · Pigeons were studied in a token economy, in which different token types could be earned and exchanged for different

significantly greater when presented with waterthan with food.Figure 7 shows the proportion of genera-

lized tokens exchanged for food and water forthe F-G and W-G conditions, averaged acrossFR replications. (Note that the proportions donot add to 1 for certain FR values, as pigeons

occasionally stopped responding and did notexchange all available tokens before the sessionended.) In the F-G condition, the majority ofgeneralized tokens were exchanged for water,whereas in the W-G condition they were mostlyexchanged for food; the generalized tokensthus served as a substitute for the commodity

Fig. 7. Average total proportion of generalized tokens exchanged for water or food in the Food-Generalized (solid lines)and Water-Generalized (dotted lines) conditions of the preference assessment.

GENERALIZED CONDITIONED REINFORCEMENT 307

Page 13: Pigeons’ demand and preference for specific and generalized … · Pigeons were studied in a token economy, in which different token types could be earned and exchanged for different

not available. Generalized tokens were occa-sionally exchanged for the commodity that wasconcurrently available; however this occurredonly at the low FR values. As price increased, theproportion of generalized tokens exchangedtended to decrease.

Convergence Among MeasuresPhases 1 and 2. To assess the convergence of

different measures of reinforcer efficacy fromPhases 1 and 2, Spearman rank-order correla-tions were calculated for each commodity:normalized Pmax, PR breakpoints, Omax, andpeak response output. For food, the correlationsbetween normalized Pmax and PR breakpointsandOmax values and peak response outputs wereequal and moderate but not significant (Spear-man’s s¼ 0.66). Correlations for water wereidentical for both pairs ofmeasures but were alsonot significant (Spearman’s s¼ 0.6). An evenweaker relationship was found for generalizedtokens; the correlation betweennormalizedPmaxand PR breakpoint was 0.31, and 0.38 for Omaxand peak response output.

Averaged across pigeons, both normalizedPmax and average PR breakpoint were greatestfor food tokens, but there was little consis-tency between these two measures for waterand generalized tokens; the normalized Pmaxvalue for generalized tokens was muchgreater than for water, but the averagebreakpoint was lower, though not signifi-cantly different, from the average breakpointobtained for water.

Normalized Omax, calculated by multiplyingnormalized Pmax by consumption at Pmax, failedto successfully predict the rank order of peakresponse output for any of the token types.Both values were greatest for food tokens(Omax¼ 138196.6 and peak response output¼ 3248.34), but mean peak response outputwas slightly higher for water (M¼ 2381.40,SD¼ 895.82) than generalized tokens (M¼2321.31, SD¼ 1113.55)—opposite to the patternseen in the Omax values (water normalizedOmax¼ 3125.52 and generalized normalizedOmax¼ 43089.60).

The success with which normalized Pmaxpredicted PR breakpoints for individual datavaried depending on the token type. Normal-ized Pmax successfully predicted greater PRbreakpoints for food than for generalizedtokens for all six pigeons, for food than for

water for four of the six pigeons (excluding 764and 1770), and for generalized than for waterfor three of the six pigeons (excluding 764,1770 and 1750). However, a Fisher’s exact testshowed that normalized Pmax did not predict PRbreakpoints for all commodities significantlybetter than chance (p¼ 0.26, n.s.). NormalizedOmax predicted peak response output correctlyfor four out of six pigeons when comparingfood and water (excluding 764 and 1770), forthree of six pigeons when comparing waterand generalized (excluding 764, 1770 and 136)and for all six pigeons when comparingfood and generalized. A Fisher’s exact testshowed normalized Omax failed to predict peakresponse output significantly better thanchance (p¼ 1, n.s.). On the whole, then, themain performance measures in Phase 2 (break-points, peak response rates) were not wellpredicted by the demand curve parametersfrom Phase 1.

Phases 1 and 3. To compare the demandfunction data from Phase 1, conducted in asingle-schedule context, with that from Phase 3,conducted in a concurrent-schedule context,we computed the following Phase-3 parametersfrom Equation 1: Pmax, Q0 and a. These areshown in Figure 8 for each token type in thecontext of each of the other two. Demand curveparameters for Phase 3 data can be seen inTable 2. Hursh and Silberberg’s (2008) modelprovided a good account of the data, account-ing on average for at least 90%of the variance inresponding.

Characteristics of the fits were largely consis-tent with the ordinal data from Phase 1. First,demand was most inelastic (lower a values) forfood tokens, followed by generalized tokensand water tokens (compare the right panels ofFig. 8 and the upper left panel of Fig. 4). Arepeated measures ANOVA found nosignificant effect of condition (F(2,10)¼ 3.62,ns) or commodity (F(1,5)¼ 1.93, ns), but didfind a significant interaction (F(2,10)¼ 19.6,p< .001); demand for food tokens wasmore inelastic than the other commodities,t(5)¼ 10.32, p< .001 with water, and t(5)¼ 8.83, p< .001, with generalized. Demand forgeneralized tokens was significantly moreinelastic than water tokens in theW-G condition(t(5)¼ 3.29, p< .05) but when pitted againstwater tokens, was significantly more elastic thanthat of food tokens (t(5)¼ 8.69, p< .002), andalso significantly more elastic than water tokens

308 LAVINIA TAN and TIMOTHY D. HACKENBERG

Page 14: Pigeons’ demand and preference for specific and generalized … · Pigeons were studied in a token economy, in which different token types could be earned and exchanged for different

when bothwere pitted against food (t(5)¼ 7.09,p< .001).Second, also consistent with the Phase 1 data,

Pmax values were largest and approximatelysimilar for tokens that were exchangedprimarilyfor food in all three conditions (compare leftpanels of Fig. 8 with upper right panel of Fig. 4).Pmax values were greatest for food tokens in the F-W condition (M¼ 286.23, SD¼ 39.90), for foodin the F-G condition, and generalized tokens inthe W-G condition were second highest andsimilar (M¼ 239.77, SD¼ 35.60 andM¼ 237.27,SD¼ 62.22, respectively). A repeated measuresANOVA, with commodity pair and individualcommodity as factors, foundno significant effectof pairing on Pmax values (F(2,10)¼ 3.90,

p¼ 0.05), but did find a significant effect ofcommodity (F(1,5)¼ 34.0, p< .01), and a signif-icant interaction (F(2,10)¼ 58.1, p< .001). Thissuggests Pmax values for food, water and general-ized tokens differed significantly and the effectof commodity was also dependent on thealternative commodity available.Unlike Phase 1, demand intensity, Q0, was

ordered in the following way: food, generalized,water (compare the middle panels of Fig. 8 withthe lower left panel of Fig. 4). Q0 of food tokensat FR 1 was significantly greater than water,t(5)¼ 4.60, p< .01, and generalized tokens,t(5)¼ 5.68, p< .01. Q0 values for generalizedtokens were significantly greater than watert(5)¼ 4.61, p< .01 in the W-G condition. Q0

Fig. 8. Normalized Pmax (left column), Q0 (middle column) and a (right column) values obtained from demand curvefits in the preference assessments for Food-Water (top row), Food-Generalized (middle row) andWater-Generalized (bottom row)conditions.

GENERALIZED CONDITIONED REINFORCEMENT 309

Page 15: Pigeons’ demand and preference for specific and generalized … · Pigeons were studied in a token economy, in which different token types could be earned and exchanged for different

values for generalized tokens in the W-Gcondition were significantly greater than Q0values for food tokens in the F-W condition(t(5)¼ 4.70, p< .01) and F-G conditions(t(5)¼ 3.33, p< .05), but did not differ signifi-cantly from Q0 values for water in the W-Gcondition (t(5)¼ 1.23, n.s.). A repeated mea-sure ANOVA found Q0 differed significantlybetween the paired commodities (F(2,10)¼ 13.0,p< .001); this was due to a significant interactionbetween condition and commodity (F(2,1)¼ 23.0, p< .001), and not commodity alone(F(1,5)¼ 1.05,ns).

Pairwise comparisons were conducted toevaluate differences in Pmax as a function ofcommodity within each condition. Pmax valueswere significantly higher for food tokens whenpitted against both water tokens (t(5)¼ 7.62,p< .001) and generalized tokens (t(5)¼ 9.29,

p< .001). Pmax values for water tokens weresignificantly lower than generalized tokenswhen they occurred together (t(5)¼ 5.95,p< .005). The Pmax values obtained for general-ized tokens and water tokens when they bothoccurred with food, were significantly different(t(5)¼ 5.74, p< .01).

The ordinal rankings of the Pmax and aparameters of the model were thus consistentacross Phases 1 and 3, but Q0 differed acrossthe two phases. Only the former could there-fore be used to predict ordinal preferences inPhase 3: food tokens> generalized tokens>water tokens. Preference for generalized tokensdepended on the concurrently available alter-native, substituting for food when occurring inthe context of the water token alternative andfor water when occurring in the context of thefood alternative.

Table 2

Demand curve parameter values for preference assessment (Phase 3).

k

R2Normalized

Pmax Q0 a

Bird Food Water Food Water Food Water Food Water

126 3.07 0.99 0.96 235.3 88.8 378.65 204.08 7.18E-06 1.90E-051851 3.28 0.96 0.76 173.18 55.91 784.43 94.01 9.14E-06 2.83E-05764 3.24 0.97 0.9 383.64 107.68 918.9 197.37 4.17E-06 1.48E-051770 3.23 0.89 0.95 253.13 78.91 749.53 243.53 6.82E-06 2.03E-051750 3.05 0.89 0.94 309.85 98.57 260.54 102.54 5.49E-06 1.73E-05136 3.28 0.98 0.89 362.29 98.56 623.3 126.58 4.36E-06 1.60E-05

Mean 3.19 0.95 0.9 286.23 88.07 619.23 161.35 6.19E-06 1.93E-05SE 0.04 0.02 0.03 32.81 7.58 103.38 25.22 7.75E-07 1.98E-06

Food Gen Food Gen Food Gen Food Gen

126 3.15 0.99 0.99 208.34 66.42 450.71 216.81 7.92E-06 2.48E-051851 3.27 0.98 0.98 150.7 34.16 562.07 143.75 1.05E-05 4.65E-05764 3.18 0.97 0.96 297.97 51.81 630.99 257.98 5.47E-06 3.15E-051770 3.26 0.92 0.98 282.33 58.35 752.7 185.93 5.63E-06 2.73E-051750 3.01 0.97 0.99 249.08 62.17 268.89 125 6.92E-06 2.77E-05136 3.2 0.99 0.99 250.21 57.87 663.74 155.57 6.47E-06 2.80E-05

Mean 3.18 0.97 0.98 239.77 55.13 554.85 180.84 7.16E-06 3.10E-05SE 0.04 0.01 0 21.84 4.64 70.58 20.33 7.67E-07 3.22E-06

Water Gen Water Gen Water Gen Water Gen

126 3.22 0.96 0.98 64.46 238.22 281.4 657.48 2.50E-05 6.75E-061851 3.46 1 0.97 18.83 151.17 126.79 1073.84 7.98E-05 9.94E-06764 3.41 0.83 0.72 89.08 346.9 129.42 1442.59 1.71E-05 4.39E-061770 3.24 0.83 0.89 30.68 115.03 146.86 1052.26 5.21E-05 1.39E-051750 3.08 0.97 0.96 70.26 224.41 101.76 300.65 2.40E-05 7.50E-06136 3.3 0.97 0.94 71.86 347.91 112.51 997.48 2.18E-05 4.51E-06

Mean 3.29 0.93 0.91 57.53 237.27 149.79 920.72 3.66E-05 7.83E-06SE 0.06 0.03 0.04 11 39.49 27.06 160.5 1.00E-05 1.48E-06

310 LAVINIA TAN and TIMOTHY D. HACKENBERG

Page 16: Pigeons’ demand and preference for specific and generalized … · Pigeons were studied in a token economy, in which different token types could be earned and exchanged for different

Discussion

A broad aim of the present study was todevelop a method for analyzing basic mecha-nisms of generalized reinforcement, a topiclong overdue for an experimental analysis.Using concepts and models from behavioraleconomics, we sought to put a sharper quanti-tative point on Skinner’s (1953) assertion thatgeneralized reinforcers are more effective thanmore specific conditioned reinforcers, in part,by more rigorously defining reinforcer efficacy—in relation to demand, response output, andpreference and as a function of price andeconomic context.The comparisons of the different token types

were quantified in relation to Equation 1,Hursh and Silberberg’s (2008) essential valuemodel. Taken as a whole, the functions wereremarkably consistent with the model, account-ing for more than 90% of the total variance.These results support a growing body of dataconsistent with Equation 1 (e.g., Barrett &Bevins, 2012; Bentzley, Fender, & Aston-Jones,2013; Cassidy & Dallery, 2012; Christensen,Silberberg, Hursh, Huntsberry, & Riley, 2008).A chief benefit of the model is that itsummarizes reinforcer efficacy with a singlemetric (a), expressing price and consumptionin standardized units. This conceptualization isespecially useful in comparisons of differentreinforcer types, including different tokentypes, because it allows direct and quantitativelymeaningful comparisons.The token types differed, based on the

various indices of efficacy. With respect todemand elasticity, a values were consistentlylowest (most inelastic) for food tokens, followedby generalized tokens, then by water tokens.Food tokens also had higher Pmax and Omaxvalues, and higher breakpoints in the PR phase,than the other token types. In contrast, thegeneralized tokens were consistently higherthan either of the specific tokens with respect toQ0 values only, reflecting greater consumptionat the lowest price. This pattern of differentialproduction of generalized tokens was notseen, however, when the tokens were availableconcurrently with food tokens in Phase 3. Infact, the opposite pattern occurred: Higherlevels of food than generalized tokens wereproduced at the lowest prices. Q0 values inPhase 1 did not predict either Q0 values orpreferences in Phase 3.

In a majority of conditions, then, food tokensappeared to be of greater efficacy than eithergeneralized or water tokens. In comparing thetoken types, however, it is important to considersome differences in the methods used togenerate the specific and generalized demandfunctions. First, in the generalized tokeneconomy, a second response was required todeliver either a food or water reinforcer afterthe generalized token had been earned, slightlyincreasing the number of responses (anddelays) to reinforcement. These delays betweenthe presentation of the generalized token andthe exchange for food or water were typicallyquite short, and therefore made a negligiblecontribution to the overall reinforcer delays.The degree to which this extra response (ordelay) was a factor probably varied acrossprocedures. In the individual demand functionruns of Phase 1, the marginal costs of anadditional response were relatively negligible atthe higher FR prices, whereas in the concurrentdemand-function runs of Phase 3, with equalprice changes on the specific and generalizedtoken types, earning and exchanging specifictokens was always a less expensive and moreefficient method of obtaining that reinforcerthan earning and exchanging generalizedtokens. This is perhaps to be expected, giventhat concurrent FR schedules tend to generatestrong preferences for the lower priced alterna-tive (Herrnstein & Loveland, 1975).It is also possible that the extra response

required under generalized token conditionsincreased the likelihood of choice “errors” (i.e.,the water key was pecked by accident when foodwas actually the intended target, or vice versa).Such outcomes may reduce the efficacy of thegeneralized tokens, resulting in higher demandfor food tokens. We saw little evidence of this,however. By the time of the generalized tokenphase the pigeons all had extensive historieswith the specific tokens in specific contexts(green/food and red/water), and food andwater exchanges were under good stimuluscontrol. Moreover, the orderly patterns ofexchange in Phase 3 showed that behaviorwas sensitive to the motivational conditions:Generalized tokens were consistently ex-changed for food when food was restrictedand for water when water was restricted (seeFig. 7). This suggests that even if exchangeerrors occurred occasionally, they did not play amajor role in the results.

GENERALIZED CONDITIONED REINFORCEMENT 311

Page 17: Pigeons’ demand and preference for specific and generalized … · Pigeons were studied in a token economy, in which different token types could be earned and exchanged for different

A second important difference concerns themotivational conditions. In the demand func-tions for the specific tokens in Phases 1 and 2,daily intake of the target commodity (food orwater, depending on the condition) occurredwithin the session, and therefore approximateda closed economy. The alternate commodity, onthe other hand, was freely available outside thesession, and therefore approximated an openeconomy. In the generalized token economy,daily intake of both food and water occurredwithin the session, and was thus a more fullyclosed economy than the specific token econ-omy. Thus, unlike the conditions when demandfor food and water tokens was assessed, in whichdemand was centered on a single commodityand a single deprivation, demand for general-ized tokens occurred in a context of twocommodities and two deprivations.

The double deprivation conditions usedwhen testing generalized tokens were arrangedto maximize the potential generalized func-tions of the tokens, making them less depen-dent on a single deprivation; by necessity,however, it created a more dynamic economywith many potential interactions betweenreinforcers. In the semi-open economies inwhich the alternate commodity is freely avail-able outside the session, most or all behavior isallocated to the target commodity. In the moreclosed generalized token economies, on theother hand, the tokens are produced andexchanged for the both food and water.

The preference conditions (Phase 3) wereconducted under the more fully closed econo-mies surrounding the generalized token con-ditions in Phases 1 and 2. These permit morebalanced comparisons of the different tokentypes across economic conditions, while addingan additional measure (relative preference) tothe analysis. These conditions produced consis-tent preferences for food over both water andgeneralized tokens, and for generalized overwater tokens. This preference hierarchy isunderstandable in terms of the substitu-tability function of the generalized tokens.When generalized tokens were presented con-currently with water tokens, demand for theformer was greater and more inelastic thanwhen they were presented concurrently withfood tokens; Pmax values and Q0 values weremuch higher, and a values much lower forgeneralized token demand functions whenchoosing between generalized tokens and

water, than with food. This would be predictedif the generalized tokens were functioning assubstitutes for food tokens in the Water vs.Generalized conditions, and as substitutes forwater tokens in the Food vs. Generalized con-ditions. In other words, the generalized tokensserved as water tokens when they were pittedagainst food, and as food tokens when they werepitted against water.

This type of functional substitution effectcan also be seen in Figure 8. In the pairwisechoices between water tokens and generalizedtokens, the majority of generalized tokenswere exchanged for food, whereas in choicesbetween food tokens and generalized tokens,most generalized tokens were exchanged forwater. The generalized token demand functionis really a composite of two functions: general-ized tokens exchanged for food and general-ized tokens exchanged for water. Viewed in thisway, the ordering of the demand functions, withthe generalized token function in between thefood and water functions, makes functionalsense.

The consistent preference for the foodtokens over the other token types in thepairwise choices in Phase 3 of the present studywas predictable on the basis of the Pmax and aparameters from Phases 1 and 3. Whenpresented concurrently in Phase 3, food tokenswere preferred over generalized tokens andwater tokens, and generalized tokens werepreferred over water tokens. The consistencybetween preference and demand obtained witheach commodity separately is in line withprevious research showing that preferencecould be predicted by consumption in demandanalysis in isolation (Bickel & Madden, 1999;Jacobs and Bickel, 1999; Johnson & Bickel,2006). To date, however, this type of predict-ability of preferences based on demand indicesis limited to humans and independent rein-forcers, such as money versus drugs. Thepresent findings expand the analysis to adifferent species (pigeons) and a differenteconomic context (fully closed economy).

While this type of consistency across mea-sures is at least broadly consistent with a unitaryconstruct of relative reinforcing efficacy, otheraspects of the present results run against thisview. Rank-order Pmax values from Phase 1 didnot predict rank-order breakpoints in Phase 2;neither did Omax values from Phase 1 predictpeak response rates in Phase 2. These aspects of

312 LAVINIA TAN and TIMOTHY D. HACKENBERG

Page 18: Pigeons’ demand and preference for specific and generalized … · Pigeons were studied in a token economy, in which different token types could be earned and exchanged for different

the results, along with others reported in theliterature (Madden et al., 2007), suggest lessthan perfect agreement among various mea-sures of the sort required by unitary concep-tions of relative reinforcer efficacy. As morerefined methods are developed for analyzingthe multiple functions of reinforcers, we mayfind even less convergence among measures.Ultimately it may be more productive to viewreinforcer efficacy not as a unitary constructreflected in the different indices, but rather, asa description of the measures that comprise it(e.g., demand, preference, peak responserates). The relationship between the measuresis therefore a question to be worked outempirically, rather than assumed. One of thebenefits of a behavioral economic approach isthat it provides a theoretically coherent frame-work for describing the interrelationshipsamong these various measures of reinforcerefficacy.The topic of generalized reinforcement falls

squarely in the middle of conceptual issues inthe study of conditioned reinforcement, such aswhether stimuli correlated with reinforcementare properly considered conditioned rein-forcers (Williams, 1994) or as discriminativestimuli, or signposts (Shahan, 2010). Thepresent study was not designed to distinguishbetween these two accounts, though pastresearch on token reinforcement has shownboth consequent/reinforcing and antecedent/signaling functions of tokens (Hackenberg,2009). So too, it is likely that tokens in thepresent study served both functions. However,because token presentations and exchangeperiods were so closely coupled in the presentprocedures, the signaling functions cannot beclearly separated from the reinforcing func-tions. In prior research designed to disentanglestimulus functions of tokens, however, Bullockand Hackenberg (2015) showed the separateinfluences of reinforcing, discriminative, andeliciting functions of tokens. When contingenton behavior, tokens served reinforcing func-tions; when arranged as signals for an operantcontingency, they served discriminative func-tions; and when arranged as signals temporallycorrelated with exchange periods but withoutan operant contingency, they served elicitingfunctions. The particular function (or func-tions) of the tokens was shown to depend on thecontingencies in which they were embedded.Expanding these methods to the study of

generalized reinforcement will enable an evensharper specification of the mechanisms ofgeneralized reinforcement.In conclusion, the present study advances the

analysis of generalized reinforcement in severalkey ways. We developed a more rigorous andquantitatively precise way to assess specific andgeneralized reinforcers. We enhanced thegeneralized functions of the tokens by embed-ding them in a closed token economy, wheretheir interactions with other reinforcers andmotivational conditions could be systematicallyexplored, and the underlying mechanismsbetter understood. Expanding the numberand type of reinforcers and motivational con-ditions will generalize the token economy evenfurther, as the generalized tokens becomemoreakin to human monetary currencies. Develop-ing these types of generalized token econo-mies will bring within reach a wide range ofeconomic variables never before studied in thelaboratory, and with it, exciting new possibili-ties for an experimental analysis of economicbehavior.

References

Barrett, S. T., & Bevins, R. A. (2012). A quantitative analysisof the reward-enhancing effects of nicotine usingreinforcer demand. Behavioural Pharmacology, 23(8),781–789.

Bentzley, B. S., Fender, K.M., & Aston-Jones, G. (2013). Thebehavioral economics of drug self-administration: Areview and new analytical approach for within-sessionprocedures. Psychopharmacology, 226(1), 113–125.

Bickel, W. K., & Madden, G. J. (1999). A comparison ofmeasures of relative reinforcing efficacy and behavioraleconomics: Cigarettes and money in smokers. Behav-ioural Pharmacology, 10, 627–637.

Bickel, W. K., Marsch, L. A., & Carroll, M. E. (2000).Deconstructing relative reinforcing efficacy and situat-ing the measures of pharmacological reinforcementwith behavioral economics: a theoretical proposal.Psychopharmacology, 153, 44–56.

Bullock, C. E., & Hackenberg, T. D. (2015). The severalroles of stimuli in token reinforcement. Journal of theExperimental Analysis of Behavior, 103, 269–287.

Cassidy, R. N., & Dallery, J. (2012). Effects of economy typeand nicotine on the essential value of food in rats.Journal of the Experimental Analysis of Behavior, 97,183–202.

Christensen, C. J., Silberberg, A., Hursh, S. R., Huntsberry,M. E., & Riley, A. L. (2008). Essential value of cocaineand food in rats: tests of the exponential model ofdemand. Psychopharmacology, 198, 221–229.

DeFulio, A., Yankelevitz, R., Bullock, C., & Hackenberg,T. D. (2014). Generalized conditioned reinforcementwith pigeons in a token economy. Journal of theExperimental Analysis of Behavior, 102, 26–46.

GENERALIZED CONDITIONED REINFORCEMENT 313

Page 19: Pigeons’ demand and preference for specific and generalized … · Pigeons were studied in a token economy, in which different token types could be earned and exchanged for different

Hackenberg, T. D. (2009). Token reinforcement: A reviewand analysis. Journal of the Experimental Analysis ofBehavior, 91, 257.

Herrnstein, R. J., & Loveland, D.H. (1975).Maximizing andmatching on concurrent ratio schedules. Journal of theExperimental Analysis of Behavior, 24, 107–116.

Hursh, S. R., & Silberberg, A. (2008). Economic demandand essential value. Psychological Review, 115, 186–198.

Jacobs, E. A., & Bickel, W. K. (1999). Modeling drugconsumption in the clinic using simulation procedures:demand for heroin and cigarettes in opioid-dependentoutpatients. Experimental and Clinical Psychopharmacol-ogy, 7, 412–426.

Jenkins, H. M., & Moore, B. R. (1973). The form of theauto-shaped response with food or water reinforcers.Journal of the Experimental Analysis of Behavior, 20,163–181.

Johnson, M. W., & Bickel, W. K. (2006). Replacing relativereinforcing efficacy with behavioral economic demandcurves. Journal of the Experimental Analysis of Behavior, 85,73–93.

Lumeij, J. T. (1987). Plasma urea, creatinine and uric acidconcentrations in response to dehydration in racingpigeons (Columba Livia Domestica). Avian Pathology, 16,377–382.

Madden, G. J., Smethells, J. R., Ewan, E. E., & Hursh, S. R.(2007). Tests of behavioral-economic assessments ofrelative reinforcer efficacy II: economic complements.Journal of the ExperimentalAnalysis of Behavior,88, 355–367.

Martin, H. D., & Kollias, G. V. (1989). Evaluation of waterdeprivation and fluid therapy in pigeons. Journal of Zooand Wildlife Medicine, 20, 173–177.

Shahan, T. A. (2010). Conditioned reinforcement andresponse strength. Journal of the Experimental Analysis ofBehavior, 93, 269–289.

Skinner, B. F. (1953). Science and human behavior. New York:Macmillan.

Williams, B. A. (1994). Conditioned reinforcement: Neglec-ted or outmoded explanatory construct? PsychonomicBulletin & Review, 1, 457–475.

Received: July 15, 2015Final Acceptance: November 16, 2015

Supporting Information

Additional supporting information may befound in the online version of this article atthe publisher’s web-site.

314 LAVINIA TAN and TIMOTHY D. HACKENBERG