Journal of Experimental Psychology: Animal Behavior Processes · The first three experiments...
20
Journal of Experimental Psychology: Animal Behavior Processes Copyright © 1976 by the American Psychological Association, Inc. VOL. 2, No. 2 APRIL 1976 Remembrance of Places Passed: Spatial Memory in Rats David S. Olton and Robert J. Samuelson The Johns Hopkins University Rats were tested on an eight-arm maze in a paradigm of sampling with replacement from a known set of items until the entire set was sampled. The first three experiments demonstrated that the animals performed effi- ciently, choosing an average of more than seven different arms within the first eight choices, and did not utilize intramaze cues or consistent chains of responses in solving the task. The second three experiments examined some characteristics of the rats' memory storage. There was a small but reliable recency effect with the likelihood of a repetition error increasing with the number of choices since the initial instance. This performance decrement was due to interference from choices rather than just to the pas- sage of time. No evidence was found for a primary effect. The data also suggest that there was no tendency to generalize among spatially adjacent arms. The results are discussed in terms of the memory processes involved in this task and human serial learning. When distinctive exteroceptive discrim- inative stimuli are consistently associated with a particular spatial location, rats pref- erentially use these stimuli for discrimina- tion learning, a phenomenon that is usually referred to as "place learning." If place learning can be used to solve a discrimina- tion problem, rats learn very rapidly. If place learning cannot be used to solve a discrimination problem, rats learn slowly and almost invariably adopt a "position habit" or "spatial hypothesis" before find- ing the correct solution. (Relevant literature reviews may be found in Gleitman, 1955; Kimble, 1961, p. 223; Olton & Samuelson, This research was supported in part by National Institute of Mental Health Grant 1-R01-MH-23213 to David S. Olton. The authors express their ap- preciation to E. Blass, J. Deese, H. Egeth, S. Hulse, B. Green, Jr., and R. M. Olton, Jr., for their assistance in the preparation of this manuscript. Requests for reprints should be sent to David S. Olton, Department of Psychology, The Johns Hop- kins University, Baltimore, Maryland 21218. 1974; Sutherland & Mackintosh, 1971; Tol- man, Ritchie, & Kalish, 1946; Woodworth, 1938, p. 124). Also, spontaneous alternation in a T maze is strongly influenced by spatial cues (Douglas, 1966), and during avoidance behavior rats preferentially associate shock with spatial location rather than with visual or auditory cues (Olton, 1973). In spite of the ubiquitous nature of place learning, most experiments have treated place learning as a factor to be controlled and have chosen to assess rats' cognitive abilities by making place learning impos- sible. In contrast, the present experiment was designed to permit place learning and to utilize this learning to assess the capacity of rats to discriminate, remember, and pro- cess information derived from place learn- ing in searching for food. To accomplish this goal, we developed a new experimental testing paradigm, one which requires sam- pling with replacement from a known set of items until the entire set is sampled. In order to establish this procedure, an appa- 97
Transcript of Journal of Experimental Psychology: Animal Behavior Processes · The first three experiments...
Journal of Experimental Psychology:Animal Behavior Processes
Copyright © 1976 by the American Psychological Association, Inc.
VOL. 2, No. 2 APRIL 1976
Remembrance of Places Passed: Spatial Memory in Rats
David S. Olton and Robert J. SamuelsonThe Johns Hopkins University
Rats were tested on an eight-arm maze in a paradigm of sampling withreplacement from a known set of items until the entire set was sampled.The first three experiments demonstrated that the animals performed effi-ciently, choosing an average of more than seven different arms within thefirst eight choices, and did not utilize intramaze cues or consistent chainsof responses in solving the task. The second three experiments examinedsome characteristics of the rats' memory storage. There was a small butreliable recency effect with the likelihood of a repetition error increasingwith the number of choices since the initial instance. This performancedecrement was due to interference from choices rather than just to the pas-sage of time. No evidence was found for a primary effect. The data alsosuggest that there was no tendency to generalize among spatially adjacentarms. The results are discussed in terms of the memory processes involvedin this task and human serial learning.
When distinctive exteroceptive discrim-inative stimuli are consistently associatedwith a particular spatial location, rats pref-erentially use these stimuli for discrimina-tion learning, a phenomenon that is usuallyreferred to as "place learning." If placelearning can be used to solve a discrimina-tion problem, rats learn very rapidly. Ifplace learning cannot be used to solve adiscrimination problem, rats learn slowlyand almost invariably adopt a "positionhabit" or "spatial hypothesis" before find-ing the correct solution. (Relevant literaturereviews may be found in Gleitman, 1955;Kimble, 1961, p. 223; Olton & Samuelson,
This research was supported in part by NationalInstitute of Mental Health Grant 1-R01-MH-23213to David S. Olton. The authors express their ap-preciation to E. Blass, J. Deese, H. Egeth, S.Hulse, B. Green, Jr., and R. M. Olton, Jr., for theirassistance in the preparation of this manuscript.
Requests for reprints should be sent to David S.Olton, Department of Psychology, The Johns Hop-kins University, Baltimore, Maryland 21218.
1974; Sutherland & Mackintosh, 1971; Tol-man, Ritchie, & Kalish, 1946; Woodworth,1938, p. 124). Also, spontaneous alternationin a T maze is strongly influenced by spatialcues (Douglas, 1966), and during avoidancebehavior rats preferentially associate shockwith spatial location rather than with visualor auditory cues (Olton, 1973).
In spite of the ubiquitous nature of placelearning, most experiments have treatedplace learning as a factor to be controlledand have chosen to assess rats' cognitiveabilities by making place learning impos-sible. In contrast, the present experimentwas designed to permit place learning andto utilize this learning to assess the capacityof rats to discriminate, remember, and pro-cess information derived from place learn-ing in searching for food. To accomplishthis goal, we developed a new experimentaltesting paradigm, one which requires sam-pling with replacement from a known set ofitems until the entire set is sampled. Inorder to establish this procedure, an appa-
97
98 D. S. OLTON AND R. J. SAMUELSON
ratus was constructed which required therat to return to the same central choice pointafter each choice so that the entire set ofchoice alternatives (including the ones al-ready chosen) was presented on each trial.The reinforcement contingencies were ar-ranged so that the first choice of eachalternative was reinforced, while subsequentchoices were not, making the optimal be-havior choosing each alternative once untilall alternatives were chosen. This procedureis in distinct contrast to the typical mazelearning task in which choice alternativesare presented sequentially and the animal isfaced with only one subgroup (usually apair) of the entire set of alternatives at anygiven time.
The first three experiments were designedto investigate the general characteristics ofperformance in this paradigm and to assessthe relative importance of extramaze cues,intramaze cues, and response chains in orderto determine the type of strategy used bythe rats. The subsequent experiments werethen conducted to describe the characteris-tics of the memory processes that underlieperformance.
EXPERIMENT 1Method
Subjects. The subjects were six experimentallynaive male albino rats that weighed 300-325 g atthe start of testing.
Apparatus. The apparatus was a radial eight-arm maze as diagrammed in Figure 1. The centerplatform was 34cm wide. All arms were the samelength (86 cm) but various widths (1.3, 2.5, 3.8,5.0, 6.3, 7.5, 9.3, and 10.0 cm). The entire appa-ratus was made of wood and was approximately50 cm above the floor. The testing room was wellilluminated by ceiling lights unless otherwisenoted. There were numerous visual cues scatteredaround the room, including a sink, table, columns,and so on.
Procedure
Exploration. During the first 9 days all rats hadad-lib food and water in the home cage. On eachday they were given approximately 15 min. toexplore the apparatus; no food was present in theapparatus during this time.
Free choice. For the rest of the experiment,all rats were placed on a food deprivation scheduleto bring body weights down to 85% of ad lib;daily feeding took place within i hr. after testing.During the next 10 days (Days 10-19), reinforce-
FIGUEE 1. Diagram of a top view of the apparatus.
ment was provided by pieces of Purina Lab Chowthat weighed approximately .1 g each, At the startof each test session, one piece of food was placedat the end of each arm; reinforcement was notreplaced during the test so that a maximum ofeight pieces of food could be obtained. Each ratwas placed on the center platform of the apparatusand allowed to make 16 choices or given a total of10 min. if 16 choices were not made. A choice wasdefined as the rat proceeding to the end of an arm.
Rebait. During the next 5 days (Days 20-24),reinforcement was replaced on one of the eightarms after the rat had chosen that arm and re-turned to the center platform. The actual armrebaited varied, but in all cases it was an armchosen within the first six choices. If the presenceof food at the end of an arm influenced the rat'schoice of that arm, the rat ought to return to therebaited arm after fewer choices than in the freechoice procedure.
Added odor. During the next 3 days (Days25-27), the apparatus was liberally sprinkled untilwet with either Old Spice or Mennen aftershavelotion. The lotion was applied to both the centerplatform and to all of the arms just prior to testingeach animal. If the odor of food at the end of anarm or some kind of "odor trail" within the appa-ratus influenced the rat's choice behavior, thenchoice accuracy here should be lower than that inthe free choice procedure.
Results
Exploration. During the first day of test-ing, all rats moved freely about the centerplatform of the apparatus but rarely wentonto the arms. Over the next few days theybegan to venture out onto the arms untilthey would readily run from the center plat-
SPATIAL MEMORY IN RATS 99
form to the end of an arm and back again.Typically, the rats would pause at the edgeof the platform before entering an arm andorient toward the end of the arm whilerapidly moving their vibrissae. This orient-ing behavior was followed either by with-drawal to the center of the apparatus andrepetition of the orienting behavior at an-other arm or by a choice of the arm. Onother occasions, the rats would run rapidlyacross the center platform and without pauserun down an arm; this type of behavior wasusually observed only on the first few choices.In almost all cases, once a rat had placedall four legs onto an arm, it proceeded tothe end of the arm before turning around.There was a marked preference for the widerarms, and rats rarely chose the three mostnarrow arms.
Free choice. The general activity level ofthe rats increased markedly from the pre-ceding exploration conditions. At first, somefood pellets were left on an arm even thoughthe rat had placed its nose very close to thepellet and presumably noticed its presence.Within a few days, all food pellets weretaken from the arms and consumed, usuallyin the center platform. As the rats becamemore familiar with the procedure, they wouldoften eat either at the end of an arm or whilerunning along the arms and making sub-sequent choices. As before, rats would oftenpause at the edge of the center platformand orient toward the end of an arm beforechoosing it, and once the rat had all fourlegs on the arm, it almost always proceededto the end before turning around.
All rats rapidly came to choose an averageof more than seven different arms within thefirst eight choices. The mean numbers ofdifferent (i.e., correct) arms chosen withinthe first eight choices for the first 5 andsecond 5 days of this procedure were 5.7(range: 4,2-6.6) and 7.6 (range: 7.2-8.0),respectively.
Rebait. Replacing food on one of the armsafter the original piece of food was obtainedhad no obvious effect on the accuracy ofchoice behavior. Occasionally, the animalsseemed to spend more time than usual sniff-ing and orienting after the food pellet was
replaced, but this behavior was not reliable.No rat ever returned to the rebaited armbefore making at least eight choices, and on5 of the 30 tests (six rats, five tests each)the rebaited arm was never chosen withinthe 16-choice limit. The mean number ofchoices before returning to the rebaited armwas 11,6 (range: 9.6-13.0). Since the ratstended to choose all eight arms again duringChoices 9-16, this performance is very sim-ilar to that expected by chance, indicatingthat the presence of food at the end of an armhad no observable effect on the accuracy ofchoice behavior,
Added odor. The addition of the after-shave lotion had a profound effect on theanimals' general behavior. On the first day,all animals were greatly disturbed by theodor; some were very hesitant to run on thearms and spent most of the time on thecenter platform, while others would oftengo out to •- the end of an arm but refuse topick up the pellet. By the second day, thisbehavior pattern was greatly diminished, andall animals ran readily down the arms andconsumed all the pellets. In contrast to thedisruption of general behavior, there waslittle change in the accuracy of choice be-havior even on the first day. The mean num-ber of correct choices within the first eightchoices was 7.4 (range: 7.3-8.0) over allthree days, indicating that the odor fromfood or some kind of odor trail had noobservable effect on choice accuracy.
All procedures. To simplify the analysisof choice behavior during Choices 9-16, onlythose days on which the animals both (a)obtained all eight pieces of food within thefirst eight choices and (b) continued to makea total of 16 choices were analyzed. For thesix rats, there was a mean of 9.2 (range: 3-11) days on which both Conditions a and bwere met. On these days, the mean numberof different arms chosen during the lasteight choices was 5.9 (range: 4.3-7.2). Thus,even after all the food had been obtainedfrom the apparatus, the rats still tended tochoose each of the arms, although not to theextent as during Choices 1-8. A similarpattern of choice behavior was observed ondays when Conditions a and b were not met.
100 D. S. OLTON AND R. J. SAMUELSON
The order of choices was analyzed todetermine if the rats used specific sequencesto simplify the problem. During all of theprocedures, the rats almost never chose thesame arm twice in succession and this se-quence was found in less than .01% of thechoices. In general, the rats tended not tochoose adjacent arms, although occasionallythere were sequences in which two, three, orfour adjacent arms were chosen in order.The rats often responded in a clockwise orcounterclockwise fashion and seemed to pre-fer the arm 90° from the arm just chosen.None of these patterns or any other patternobvious to the experimenters was regularlyexhibited by the rats, and, most importantly,choice accuracy was routinely high irrespec-tive of the particular order of choices. TableAl presents the raw data for Rat 3 duringDays 10-27. Although various patterns ofchoices appear, there are no patterns that area prerequisite for accurate choice behavior.
Discussion
In the typical sequential maze experiment,decision making at any given choice pointcan be independent of that at any otherchoice point. The rat must decide betweenonly one pair of alternatives (i.e., go left, goright) at a time, and this choice can be madewithout reference to any preceding or sub-sequent choices. In the present experiment,however, the rat always returned to the cen-ter platform after each arm and was thuspresented simultaneously with all the alterna-tives on each choice. Thus the rat cannottreat each successive choice independently.The paradigm may be summarized as onewhich requires sampling with replacementfrom a known set of items until the entireset is sampled.
Rats performed very accurately and with-in 5 days of free choice testing, all animalswere consistently choosing an average ofmore than seven different arms within thefirst eight choices. The high level of accuracyeven on the first 5 days of free choice test-ing suggests that the rats were predisposednot to repeat their choices in this situation,presumably reflecting the tendency towardspontaneous alternation commonly observedin two-choice paradigms in a simple T maze
(Douglas, 1966). Accurate determinationof "chance" performance levels given thisalternation tendency is difficult. If each choiceis completely independent of all others, theexperiment becomes an example of the classicoccupancy problem (Feller, 1950); the prob-abilities of choosing n different arms withinthe first eight choices are as follows: n = 8,p = .0024; w = 7, p = .0673; n = 6, p =.3194; n - 5, p = .4200; n = 4, p = .1680;n = 3, p = .0168; n = 2, p = .0004; n = 1,p = 6 X 10~8. The assumption of choiceindependence for rats is probably not validbecause even on the first few days of explora-tion the rats almost never chose the samearm twice in succession and usually chose atleast five different arms within eight choices.In another experiment (Olton, Note 1) in-fant rats were rewarded for a response toany arm, as in the typical reinforced spon-taneous alternation paradigm. Under theseconditions, the animals chose an average ofsix different arms within the first eightchoices, indicating a strong disposition notto repeat choices to arms even when foodwas obtained on repeated arms as well ason new arms. The presence of a spontaneousalternation tendency does not compromisethe basic argument about the discriminationrequirements of the paradigm made abovebecause in order to exhibit spontaneousalternation, the rat must obviously locateeach arm and determine which arms havebeen chosen and which have not. The onlydifficulty imposed by the presence of spon-taneous alternation tendency is accurateassessment of the relative contribution oflearning and of innate behavioral tendenciesto successful performance, and the presentexperiment was not designed to resolve thisissue.
Rebaiting the arms with food and addingodor to the arms did not markedly influencethe choice accuracy of the animals, demon-strating that cues such as the odor or sightof food at the end of an arm and odor trailsfrom the animals themselves were unimpor-tant in performing the task. A number ofstudies have demonstrated that odor canfunction as a cue for rats in spontaneousalternation (Douglas, 1966), maze learning(Means, Hardy, Gabriel, & Uphold, 1971;
SPATIAL MEMORY IN RATS 101
Wasserman & Jensen, 1969) and other typesof tasks (Slotnick, 1974). In all of the aboveexperiments, however, the salience of theodor was markedly increased either by com-bining odors from many rats all tested underthe same condition or by having all odorsexcept the stimulus odor removed or sup-pressed. The data from the present experi-ment suggest that under more natural con-ditions where the odor trail from a singlerat is deposited among the odor trails frommany other rats exhibiting different be-haviors, odor trails have a minimal effect ondiscrimination behavior.
The number of different arms chosen dur-ing Choices 9-16 was slightly less than thatduring Choices 1-8 but was still routinelyhigh, suggesting that animals tended tochoose each arm even after all the food hadbeen obtained. The results of Experiment 6to be reported below suggest that the slightdecrease in the number of arms chosen dur-ing the second eight choices probably reflectsthe performance decrement that would beexpected in the absence of differential rein-forcement. The results offer further evidencefor the unimportance of intramaze odor cuesbecause after the first eight choices the ratwould have left an odor trail on all arms,while there would be no food on the end ofany arm; nonetheless, the number of differ-ent arms chosen was still high.
Rats have been shown to "chain" specificresponses to form very long specific be-havioral sequences (Hulse, Deese, & Egeth,1975, p. 56). In the present experiment,however, no evidence of chaining was ob-served. Although the rats did occasionallyrepeat specific sequences of choices, thesewere not regularly observed and choiceaccuracy was routinely high irrespective ofthe particular choice pattern emitted. Thedata indicate that performance here is verydifferent from that observed in behaviorchains and that chaining was not utilized asa strategy to solve the task.
In summary, rats perform well in thepresent situation, choosing almost perfectlyduring their first eight choices. Intramazeodor cues and consistent sequences of choicesare both unnecessary for successful perform-ance.
EXPERIMENT 2
The present experiment was designed toprovide additional evidence concerning therelative importance of intramaze cues, extra-maze cues, and choice sequences. To this end,all arms were made equally wide so thatarm width could not serve as a cue; animalswere confined to the center platform bymeans of guillotine doors in order to inter-fere with any sequence of choices that mighthave been in process at this time; arms wereinterchanged so that all extramaze placecues were dissociated from all intramazecues.
MethodSubjects and apparatus. The subjects were the
same as in Experiment 1. The apparatus of Ex-periment 1 was modified in the following manner.All arms were 7 cm wide. A hole 2 cm wide and1 cm deep was drilled at the end of each arm toserve as a food cup. The arms were made so thatthey could be detached from the center platformand interchanged with each other. A wooden com-partment 10 cm high was constructed around theedge of the center platform. Wooden guillotinedoors were placed in the walls of this compart-ment at the entrance to each arm so that theanimal could be confined to the center platform.The top of the compartment was of Plexiglas.
ProcedureFree choice. For 10 days (Days 1-10) the rats
were tested as in the free choice procedure of Ex-periment 1. They were allowed either 16 choicesor 10 min. in the apparatus, whichever came first.
Confinement. For five days (Days 11-15) therats were confined to the center compartment bylowering the guillotine doors when they enteredthe center compartment after their third choice.The doors remained lowered for approximately 1min., after which they were raised and the rat wasallowed to choose freely until a total of 16 choiceshad been made or 10 min. elapsed.
Interchange arms. For IS days (Days 16-30)the rats continued to be confined to the center com-partment after their third choice. During this con-finement, two adjacent arms were interchanged.The actual arms moved varied each day, but inall cases one arm had been chosen during the firstthree choices and the other arm had not, placingspatial location cues and arm cues in oppositionto each other. In the following discussion, "chosenarm" refers to the arm of the interchanged pairwhich had been chosen prior to the arms beinginterchanged, whereas "unchosen arm" designatesthe arm of the interchanged pair which had notbeen chosen prior to the arms being interchanged.Likewise, "chosen spatial location" refers to the
102 D. S. OLTON AND R. J. SAMUELSON
position in the testing room of the arm of theinterchanged pair which had been chosen prior tothe arms being interchanged, and "unchosen spatiallocation" indicates the position in the testing roomof the arm of the interchanged pair which had notbeen chosen prior to the arms being interchanged.To the extent that arm cues (such as odor trail)direct choice behavior, the rats should go to theunchosen arm while repeating their choice ofspatial location. To the extent that spatial locationcues direct choice behavior, the rats should go tothe unchosen spatial location while repeating theirchoice of the arm. Three rats were reinforced forresponding to the unchosen arm following confine-ment, while the other three rats were reinforcedfor responding to the unchosen spatial locationfollowing confinement. Reinforcement on the re-maining six arms was not altered.
Results
Free choice. Making all arms the samewidth had no obvious effect on the accuracyof choice behavior. The mean number of dif-ferent arms chosen during the first eightchoices was 7.7 (range: 7.4-7.9).
Confinement. While confined to the centercompartment the animals were usually active,exploring and moving about the compart-ment for the duration of the confinementperiod. When the guillotine doors wereopened, the rat usually approached andoriented toward the end of the arm it wasfacing. The accuracy of choice behavior wasnot at all affected by the confinement; themean number of different arms chosen in thefirst eight choices was 7,7 (range: 7.0-8.0).
Interchange arms. In all cases, the animalsshowed little, if any, evidence of reaction tothe interchanged arms and continued tochoose almost entirely on the basis of spatiallocation. The rats reinforced for respondingto the unchosen arm performed no differ-ently from the rats reinforced for respondingto the unchosen spatial location and the datafrom both groups were combined for analy-sis. In all but 2 of the 90 tests (6 rats, IStests each), rats responded to the unchosenspatial location prior to responding to theunchosen arm. The mean choice number of .the unchosen spatial location was 6.1 (range:4.1-7.3) and was thus always among thefirst eight choices when food was usuallyobtained, while the mean choice number ofthe unchosen arm was 9.9 (range: 8.5-11.7)and thus always among the second eight
choices when food was usually not obtained.The accuracy of the first eight choices forspatial location did not decrease as a resultof interchanging arms, and the mean numberof different spatial locations chosen withinthe first eight choices was 7.7 (range: 7.6-7.8).
All procedures. The order of choices wasagain analyzed. As in the previous experi-ment, various patterns often appeared on aparticular day, but none of these patternswas necessary for accurate choice behavior.The rats also tended to choose each of theeight different arms on Choices 9-16.
Discussion
The results of Experiment 2 extend andcorroborate those of Experiment 1 describ-ing the types of cues directing the rats'choice behavior. First, the unimportance ofintramaze cues was supported by continuedhigh performance when all arms .werechanged to equal width. Second, the unim-portance of specific continuous responses se-quences was supported by continued highperformance when animals were confined tothe center compartment between choices.Third, the unimportance of intramaze odorcues was strongly corroborated by inter-changing the arms. Those animals reinforcedfor responding to the unchosen spatial loca-tion were unaffected by the presence of anodor trail and those animals reinforced forresponding to an odor trail (i.e., the un-chosen arm) showed no evidence of learningto follow the odor trail.
EXPERIMENT 3
Experiment 3 was designed to replicateand extend the results of the previous ex-periments in three ways. First, the apparatuswas altered so that instead of interchangingjust two arms and leaving any possible intra-maze cues on the center platform unaltered,the entire arm assembly could be rotatedwhile the animal was in the center compart-ment, and then the entire center compart-ment rotated while the animal was on anarm, thus rotating all intramaze cues insteadof just those on two arms. Second, no ex-ploration period took place prior to trainingto obtain a better estimate of the rate of
SPATIAL MEMORY IN RATS 103
learning. Third, new animals were used todetermine the reliability of the previousfindings.
Method
Subjects. The subjects were six experimentallynaive male albino rats that weighed 300-325 g atthe start of testing. All rats were food deprivedto 85% of ad-lib body weight and were maintainedat this level throughout the experiment by feedingthe required amount of Purina Lab Chow withini hr. after testing.
Apparatus. The apparatus was the same as inExperiment 2 except that the arms and centercompartment were suspended independently of eachother so that each could be rotated without dis-turbing the other. Reinforcement consisted of one190-mg Noyes food pellet at the end of each arm.
Procedure
Free choice. For the first 15 days, each rat wasgiven one daily test session until 16 choices weremade or a total of 10 min. had elapsed. On the firstfew days, rats were shaped as necessary to goonto the arms by placing food reinforcement closeto the center platform. On Days 6-15, all foodwas placed in the food cups at the end of thearms.
Confinement. For the next five days (Days16-20), each rat was confined in the center com-partment for approximately 1 min. after the thirdchoice by lowering the guillotine doors after therat entered the center compartment.
Mage rotation. For the next six days (Days21-26), each rat was confined to the center com-partment after the third choice as described above.On Days 1, 3, and 5, while the rat was confinedto the center compartment which did not move atthis time, the arm assembly was rotated 45° andfood was replaced on the three arms alreadychosen on that day so that all arms containedfood. The guillotine doors were then raised and,while the rat was on the arm making its fourthchoice, the center compartment was rotated 45°in the same direction as the arm assembly. Thus,all intramaze cues were rotated 45° with respectto extramaze cues. The rat was allowed to choosefreely until either the five remaining arms or thefive remaining spatial locations were chosen. OnDays 2, 4, and 6, each rat was confined after thethird choice and the maze arms were rotated, butthe arms were returned to their original locationbefore the guillotine doors were raised in order tocontrol for any general influence of the movementof the apparatus on choice behavior.
Results
Free choice. For the first few days, allrats were reluctant to enter the arms and
spent considerable time exploring the centercompartment and orienting toward the endof the arms. Within 8 days, all animalsreadily ran. on the arms, retrieving and con-suming all the food pellets. Their generalbehavior was similar to that reported pre-viously ; the rats often hesitated at a guillo-tine door and oriented toward the end of thearm. Once all four legs were placed on thearm, however, the rat almost always ran tothe end of the arm and placed its nose in'the food cut. No data are reported for thefirst 5 days because the rats often did notleave the center compartment to run downthe arms. As soon as the rats began to runand choose all the arms, they very rapidlydemonstrated the discrimination; the meannumbers of different arms chosen within thefirst eight choices for Days 6-10 and Days11-15 were 7.2 (range: 6.8-7.4) and 7.3(range; 6.8-7.8), respectively.
Confinement. Confinement had no obviouseffect on the animals' behavior. As before,the animals were active during the period ofconfinement, exploring the center compart-ment ; when the guillotine doors were open,they usually oriented down the arm theywere facing. The mean number of differentarms chosen within the first eight choicesduring this period was 7.6 (range: 7.5-7.8).
Maze rotation. To the extent that intra-maze cues control choice behavior, the ratsought to go to the five remaining unchosenarms after maze rotation and repeat choicesto already chosen spatial locations. To theextent that extramaze cues control choice be-havior, the rats ought to go to the five re-maining unchosen spatial locations after themaze rotation and repeat choices to alreadychosen arms. Consequently, the relevant dataare the number of times an already chosenarm was repeated after the maze rotationand the number of times an already chosenspatial location was repeated after the mazerotation. Since three choices were made priorto the maze rotation, the maximum numberof repetitions was three. The mean numberof repetitions of already chosen arms was2.3 (range 1.3-2.7), while the mean numberof repetitions of already chosen spatial loca-tions was .3 (range: .0-1.0). Additionally,the mean number of different spatial loca-
104 D. S. OLTON AND R. J. SAMUELSON
tions chosen within the first eight choiceswas 7.5 (range: 7.2-8.0) on days when themaze was rotated, the same values as on con-trol days.
Discussion
Results of Experiment 3 are consistent withthose of the previous two experiments. Afterthe animals learned to run out onto the arms,their choice accuracy was very high. Manip-ulation of intramaze cues appeared to haveno effect on choice behavior; all animals con-tinued to choose accurately on the basis ofspatial location and readily repeated alreadychosen arms in order to respond to unchosenspatial locations.
EXPERIMENT 4
The results of Experiments 1, 2, and 3consistently demonstrate that the rats did notchoose adjacent arms, did not rely on anytype of intramaze marking, did not chain re-sponses in particular sequences, and did notutilize any other obvious strategy to simplifythe task. Since the testing procedure involvedsampling with replacement, the rats musthave been able to identify each of the eightarms and remember over a period of severalminutes which arms had been chosen andwhich had not. In short, the rats seemed totreat each of the eight spatial locationsseparately.
The purpose of the present experiment wasto investigate the characteristics of the mem-ory for this list of items. In particular, thefollowing questions were addressed: (a) Isthere a performance decrement as the num-ber of choices the animals makes increases?(b) Are correct choices at the beginning ofthe test remembered better than subsequentchoices? (c) Are correct choices at the endof the test remembered better than previouschoices? (d) Is there spatial generalizationfrom one arm to the other so that errors tendto be directed towards correct choices? (e)Is information from correct choices and in-correct choices stored in the same or dif-ferent memory stores ?
MethodSubjects and apparatus. The subjects were 12
experimentally naive male albino rats that weighed
300-325 g at the start of testing. The apparatuswas the same as in Experiment 3.
Procedure
Rats were placed on a food deprivation scheduleto bring body weights down to 85% of ad lib; dailyfeeding took place approximately 1 hr. after testing.Reinforcement was a 45-mg Noyes pellet. At thestart of each test session, one pellet was placed atthe end of each arm; reinforcement was not re-placed during the test so that a maximum of eightpieces of food could be obtained. Each rat wasplaced on the center platform of the apparatus andallowed to make choices until all eight reinforce-ments were obtained or until 10 min. had passed,whichever came first. All rats were tested for 30days, with one test on each day.
Data Analysis
The data analysis for this experiment will be pre-sented in considerable detail because there are anumber of different factors that must be consideredand there are no readily available precedents to useas a model for analysis. For each question, twotypes of analyses will be presented. The first isdesigned to describe the actual performance of theanimal and will provide information about the ob-served behaviors. The second is designed to adjustfor various influences which must be considered inorder to adequately answer the five questions posedin the introduction. In all cases, the procedurechosen for the second analysis is the one that seemsmost appropriate to the question being asked. Butfor any set of raw data and any particular experi-mental question, a number of different analyses arepossible, each based on a different set of assump-tions and with its own set of advantages and dis-advantages (e.g., see Murdock, 1957). Thus, thefirst analysis not only provides information aboutthe way in which the animals actually performedbut also allows the reader to carry out other trans-formations of the data if desired.
All but one of the analyses are based on errors(i.e., returning to an already chosen arm). As inthe previous experiments, the animals performedwell and made relatively few errors. In order toobtain enough observations for an accurate estimateof behavior, the data for each animal were averagedover several days. Since choice accuracy graduallyimproved during the first 10 days and then re-mained stable during the second 10 days, the datawere analyzed in 10-day blocks. Thus, the resultsfrom the first 10 days probably reflect some learn-ing of the task, while the results from the second10 days probably reflect asymptotic performanceof the already learned task. During Days 11-20,one animal performed perfectly so that there wereno errors to analyze. Four other animals had eitherone or two errors, whereas the remaining seven hadfour or more errors. For Days 11-20, the data wereanalyzed separately for »=11 and for » = 7.
SPATIAL MEMORY IN RATS 105
Analyses of variance were for a treatments bysubjects design.
Results
A mean of 4.9 days (range: 3-8) was re-quired before a total of eight choices weremade within the 10-min. test limit. On subse-quent days, animals chose readily, alwayscompleting at least eight choices and usuallyfinishing the day's test within 2 min. Forpurposes of data analysis, Day 1 was definedas the first day on which at least eight choiceswere made.
Choice accuracy: General. The task waslearned rapidly and even during Days 1-5the mean number of different arms chosenwithin the first eight choices was 6.9 (range:6.0-7.3). Performance improved to a meanof 7.5 (range 6.9-7.8) correct responsesduring Days 6-10 and then appeared toreach an asymptote at about 7.6 (range: 7.3-8.0) correct responses within the first eightchoices on Days 11-15 and Days 16-20. Allbut one animal performed better during Days16-20 than during Days 1-5, indicating thatalthough initial performance was accurate, asignificant amount of learning did take place.
Performance decrement: Observed prob-ability of a correct response. Performance ac-curacy was greatest at the beginning of eachday's test and declined slightly but consist-ently during the test. Table 1 presents themeans for the observed probability of a cor-rect response, p(cor)obs, on Choices 2-8 dur-ing Days 1-10 and Days 11-20. Data for thefirst choice are not presented because the firstresponse had to be correct. Data for choicesfollowing Choice 8 are not presented becausethe animals often performed perfectly andtesting was terminated at the end of 8choices. The observed probability of a correctresponse was determined for each animal foreach choice by the formula
p(cor)obs
TABLE 1PROBABILITY OF A CORRECT' RESPONSE
number of correct responsestotal number of responses
X 100.
The p(cor)obs score can range from 100,which indicates a correct response on allchoices, to 0, which indicates an incorrectresponse on all choices. As would be ex-
Choice
Probability
ObservedDays 1-10Days 11-20
TransformedDays 1-10Days 11-20
100 95 95 85 87 81 78100 99 98 96 97 90 80
100 84 86 71 77 70 73100 96 95 91 95 88 77
pected from the data reported in the previousparagraph, p(cor)ob8 was high on all choices.Scores were greatest on the second choiceand steadily declined until the eighth choice(Days 1-10, F(6, 66) = 5.1, p< .01; Days11-20, F(6, 66) = 9.6, p< .01). Scores forall choices were greater during Days 11-20than during Days 1-10, reflecting the im-provement in accuracy reported in the pre-vious paragraph as measured by the numberof correct responses within the first 8 choices.
Performance decrement: Transformeddata. Although the observed probabilities inTable 1 are clear, there is a bias presentwhich tends to exaggerate the magnitude ofthe decline in performance during the test.This bias arises because the chance probabil-ity of a correct response decreases as thenumber of arms already chosen increases.For example, consider a sequence of choicesin which all choices are correct; on the sec-ond choice, seven of the eight arms still con-tain food, on the third choice, six of the armscontain food, and so on. Thus any givenp(cor)obs score may reflect relatively betterperformance after many choices than it doesafter few choices, and the observed prob-ability data tend to exaggerate the decline inperformance that occurs during each test. Tocorrect for this bias, the changing expectedchance probability of a correct response mustbe considered. To this end, p(cor)exp wasdefined as the expected chance probability ofa correct response on any given choice ac-cording to the formula
p(cor)exp
number of arms not chosenX 100,
106 D. S. OLTON AND R. J. SAMUELSON
The observed probability data from Table 2were then transformed for each animal foreach choice, according to the formula below,to provide a measure of the extent to whichthe animals exceeded chance performanceand attained maximum performance on eachchoice:
p(cor)Qb3 — p(cor)exp
100 - p(cor)expX 100.
This formula corrects for the expectedchance performance on each choice and ex-presses performance as a proportion of themaximum performance obtainable, allowingperformance on any given choice to be di-rectly compared with performance on anyother choice. The transformed scores canrange from 100, indicating the maximumnumber of correct responses, through 0, in-dicating random performance with p(cor)ota
= p(cor)exp, to —100, indicating the max-imum number of incorrect responses. [Whenp(cor)obs was less than p(cor)exp, the de-nominator was changed to p(cor)exp — 0.]For example, consider an animal withp(cor)obs = 80. If this performance was onthe third choice and six arms had not beenchosen, the transformed score would be(80-75)/(100-7S) X100 = 20. If thisperformance was on the seventh choice andtwo arms had not been chosen, the trans-formed score would be (80 - 25)/(100—25) X 100 = 73. As the number of correctarms available to the animal decreases, thetransformed score for any given p(cor)obs
increases. Also, as the accuracy of perform-ance increases, the transformed score ap-proaches the asymptote of 100.
Table 1 also presents the transformedscores. An important result is that all scoreswere substantially greater than 0, indicatinggreater than chance performance on eventhe last choice. The same trends that wereapparent in the observed probabilities arealso apparent here. First, performance grad-ually decreased as the number of choices in-creased (Days 1-10, F(6, 66) = 6.5, p <.01; Days 11-20, F(6, 66) = 7.1, p < .01).Second, performance during Days 11-20 wasbetter than performance during Days 1-10.
Sequential position of errors: Generalanalysis. The data in Table 1 indicate that
the likelihood of making an error increasedwith the number of choices made by the ani-mal, but they do not provide informationabout the sequential characteristics of theseerrors. When errors did occur, they werealmost always to arms that had been chosenat the beginning of the day's test and rarelyto arms chosen near the end of the day's test.Four different analyses of these data wereperformed to provide a quantitative measureof this behavior: (a) the observed relativeprobability of repeating each of the first sevencorrect choices, p(rep)obs; (b) correction ofp(rep)obs for the differing number of oppor-tunities to repeat each choice; (c) correctionof p(rep)obs for the possibility of an armpreference; (d) correction of p(rep)0bs forboth differing number of opportunities torepeat each choice and the possibility of anarm preference.
In all of these analyses, the results werethe same. The probability of an error wasgreater for arms chosen early during eachday's test than for arms chosen later duringeach day's test (for Days 1-10 and Days 11-20, ns = 7 and 11, respectively). Analyses ofvariance yielded Fs ranging from 2.7 to 15.9(/> < .05 for 2 tests, p < .01 for 10 tests).To conserve space, each analysis is described,but only the data from Days 1-10 are pre-sented in Table 2. Table A2 presents theraw choice data upon which the analyses arebased.
Sequential position of errors: Observedrelative probability of repetition. The ob-served relative probability of repeating eachof the first seven correct choices was deter-mined. The eighth choice was not consideredbecause testing was terminated after thischoice and there was no opportunity to repeatthe eighth choice. For each animal for eachday, all correct choices were ordered in se-quence from one to seven. Then for each ani-mal, the observed relative probability of re-peating each correct choice was calculatedaccording to the formula
p(rep)ob3
number of repetitions of choicetotal number of repetitions
X 100.
The p(rep)0bs score indicates the relativeprobability of an error to each of the previous
SPATIAL MEMORY IN RATS 107
TABLE 2RELATIVE PROBABILITY OF REPEATING A PREVI-
OUSLY CORRECT CHOICE : DAYS 1-10
Correct choice
Analysis
Observed probabilityCorrection for oppor-
tunitiesCorrection for arm
preferenceCorrection for oppor-
tunities and armpreference
1
36
43
31
40
2
23
30
23
31
3
IS
23
17
27
4
12
23
12
26
S
8
19
9
22
6
6
21
6
25
7
1
S
.8
8
correct choices. Scores can range from 100,indicating that all errors were made by re-peating one particular choice, to 0, indicatingthat no errors were made by repeating thatchoice. The sum of scores for all choicesequals 100. Consider an animal that made atotal of 8 errors in 10 days. If 5 of these er-rors were to the first correct choice, 2 to thesecond correct choice, and 1 to the thirdcorrect choice, then the p(rep)0bs score forcorrect Choices 1, 2, and 3 would be 63, 25,and 13, respectively. The first line of Table 2presents the mean of p(rep)ob8 scores forChoices 1-7 during Days 1-10, As can beseen from Table 2, when errors occurred,they were most likely to be to the first cor-rect choice, and there was a steady declinein the probability of repetitions for each sub-sequent choice.
Sequential position of errors: Adjustmentfor opportunities. There is a bias in the data,as presented in the first line of Table 2, be-cause the number of opportunities availablefor repeating each choice differs. The animalsmade an average of 9.3 choices to obtain alleight pieces of food during Days 1-10 and,thus, had an average of 8.3 opportunities torepeat the first correct choice and a max-imum of 7.3 opportunities to repeat the sec-ond choice, 6.3 for the third, and so on. Anadjusted score, correcting for the number ofopportunities to repeat each choice, was cal-culated for each animal for each choice ac-cording to the formula
X 100.number of choices remaining
Unlike p(rep)0bs, these adjusted scores donot sum to 100 and range from 0 to greater
than 100. The adjusted scores can be 0, in-dicating that no errors were made to thechoice in question, or they can be some posi-tive number, indicating that some errorswere made to that choice: The greater thenumber, the greater the number of errorsmade to that choice. These adjusted scoresalso correct for the number of opportunitiesto repeat each choice and indicate the likeli-hood of repeating a choice any time an erroris made. Consider an animal with p(rep)0i,s
= 40 for the first 10 days. If this score wasobtained for the first choice of each day andthere was a total of 80 choices following thefirst choice (indicating that the animal ob-tained all eight pieces of food in an averageof 9.0 trials), then the adjusted score wouldbe 40/80 X 100 = 50. If this score was ob-tained for the third choice of each day andthere were 50 choices remaining (indicatingperfect performance on Choices 2 and 3),then the adjusted score would be 40/50 X100 = 80. For any given p(rep)ob9 score, theadjusted score increases as the number ofchoices remaining decreases. The second lineof Table 2 presents these adjusted scores.
Sequential position of errors: Adjustmentfor arm preference. If a strong preference forone or more arms developed, the animalmight tend to choose this arm early duringeach day's test. If the arm preference wereconstant throughout the day's test, then theanimal might continue to have a strong tend-ency to respond to this arm. Such a combina-tion of events would lead to the animalrepeating early choices, not because of theposition of the choice within the sequencebut rather because of a preference for thearm. An attempt to evaluate and correct forany possible arm preference was undertakenby determining for each arm the averagesequential position of the first correctresponse to that arm on each day. Each firstchoice of an arm was given a score of 1, eachsecond choice, 2, and so on. All these scoreswere summed and divided by 10 for each 10-day block. The resulting sequential positionscore provides for each arm the average se-quential position of the first correct responseto that arm during the 10 days. For each ani-mal, the scores must total 36 (i.e., 1 + 2 +. . . + 8), The expected chance performance
108 D. S. OLTON AND R. J. SAMUELSON
score for any arm is 4.5 (i.e., [1 + 2 + . . .+ 8]/8). A score of less than 4.5 indicates atendency to choose an arm preferentiallyearly in the day's test; a score of more than4.5 indicates a tendency to choose an armpreferentially late in the day's test. All ani-mals exhibited a preference for some armsover others. The mean (and standard errorof the mean) sequential position score of themost preferred arm was 2.8 (± .06) and ofthe least preferred arm was 6.0 (± .04). Inorder to determine how stable any arm pref-erence might be, a correlation coefficient wascalculated for each animal between the pref-erence scores on each of the eight arms forDays 1-10 and the preference scores on eachof the eight arms on Days 11-20. The cor-relations ranged from —.63 to .89, with anaverage of .46.
The p(rep)ob8 scores of the first line ofTable 2 were adjusted for arm preference byusing sequential position as a weighing fac-tor, multiplying each arm by its sequentialposition score, according to the formula
sequential position score for choice ,sum of sequential position
scores for all choices
Like p(rep)0bs, the adjusted scores canrange from 100, indicating that all errorswere made by repeating that particularchoice, to 0, indicating that no errors weremade by repeating that particular choice, andthe sum of scores for all choices is 100. Un-like p(rep)ollg, the adjusted scores correctfor the relative preference for each arm asmeasured by the sequential position score.Consider an animal that repeats choices toArms 1, 4, and 5 which have sequential posi-tion scores of 1.5, 4.0, and 4.5, respectively,and makes two repetitions of the first choice,both of Arm 1, and two repetitions of thesecond choice, one each to Arms 4 and 5. Theadjusted score for the first choice is [(1.5 +1.5)/11.5] X 100 = 27, the adjusted scorefor the second choice is [(4.0 + 4.5)/11.5]X 100 = 86, and the adjusted scores for allother choices is 0. [The unadjusted p(rep)obs
scores reported as in Table 1 would be:Choice 1, 2/4 X 100 = 50; Choice 2, 2/4 X100 = 50; all other choices = 0.] Thus for
any given p(rep)0b8 score, the greater thepreference for an arm as measured by thesequential position score, the lower will bethe adjusted score. Line 3 of Table 2 presentsthese adjusted scores.
Sequential position of errors: Adjustmentfor arm preference and opportunities. Thedata in the third line of Table 2 were ad-justed to correct for the number of choicesremaining following each correct choice ac-cording to the formula
sequential positionscore for choice
sum of sequential positionscores for all choices
-x 100
number of choices remainingX 100.
The adjusted scores can be 0, indicating thatno errors were made to the choice in ques-tion, or they can be some positive number,indicating that some errors were made tothat choice; the greater the number, thegreater the relative number of errors madeto that choice. These adjusted scores, whichare presented in the fourth line of Table 2,correct for both arm preference and the num-ber of opportunities to repeat each choice andindicate the likelihood of repeating a choiceany time an error is made.
Generalisation among arms. Table 2 pre-sented information about the sequential char-acteristics of errors but did not provideinformation about their spatial character-istics. When errors occurred, they appearedto be randomly distributed among the 8 armsand not preferentially directed near correctarms. An analysis of the spatial distributionof each error was conducted by numberingthe position of the arm on which the erroroccurred as 0, the position of the arms at 45°(immediately adjacent), as 1, the position ofthe arms at 90° as 2, the position of the armsat 135° as 3, and the position of the remain-ing arm at 180° as 4. An error location score,indicating the average distance between therepeated arm and each of the remainingcorrect arms was calculated from the formula
sum of positions of remaining correct armsnumber of remaining correct arms
Error location scores can range from 1.0,indicating that an error was always made to
SPATIAL MEMORY IN RATS 109
an arm adjacent to a correct arm, to 4.0,indicating that an error was always made tothe arm farthest away from a correct arm.Since there were two arms each at 45°, 90°,and 135°, which were assigned values of 1, 2,and 3, respectively, and one arm at 180°,which was assigned a value of 4, chance per-formance would be (2 + 4 + 6 + 4)/7 =2.29. The error location scores for Days 1-10and Days 11-20 (n = 7 and N = 11)ranged from 2.3 (±.03) to 2.5 (±.1).These scores closely approximate the valuesexpected by chance, indicating that whenerrors were made, there was no tendency forthem to be to arms close to correct ones.
Effect of errors on choice accuracy. Table1 demonstrated that there was a generalperformance decrement as the number ofchoices increased within a day but did notconsider whether the previous choices werecorrect or incorrect. A correct choice was,by definition, a new choice for the animal,one which had not been made previously onthat day. An incorrect choice was a repeti-tion of an earlier choice, and thus the animalmade two responses to the same arm. If therat treats the repetition of an arm as if itwere any other choice, then there ought to bean additional performance decrement as aresult of the additional choice, reflecting theresults reported in Table 1. If the rat treatsthe repetition of the arm as if it were anotherinstance of the original choice, then thereought to be no additional performance decre-ment as a result of the error. The questionis whether the rat processes the error as anew piece of information, which would resultin a performance decrement on subsequentchoices or as confirmation of an alreadyacquired piece of information, which mightnot result in a performance decrement. Toanswer this question, an analysis was con-ducted to determine the probability of a cor-rect response on Choice 7 and Choice 8 asa function of the number of prior errors. Foreach rat, test days were grouped accordingto the number of errors made prior to Choice7 for the analysis of choice accuracy onChoice 7 and the number of errors prior toChoice 8 for the analysis of choice accuracyon Choice 8. Only Choices 7 and 8 wereanalyzed because the frequency of errors
prior to other choices was too low to providea sufficient number of observations for analy-sis. The probability of a correct response wasthen determined by the formula
number of correct responsesnumber of observations
Scores can range from 0, indicating that noneof the observed responses were correct, to100, indicating that all of the observed re-sponses were correct. As in Table 1, thesescores are biased because the chance prob-ability of a correct response varies accordingto the choice number and the number of pre-vious errors. Consequently, the same correc-tion as applied in Line 2 of Table 1, wascarried out here according to the formula
number of correct responsesnumber of observations - p(cor)e
1 - p(cor)exp
These scores have the same characteristicsas those presented in the second line ofTable 1.
The results of both analyses indicated thatthere was no systematic change in the prob-ability of a correct response as a function ofthe number of preceding errors. The ob-served trend was in all cases for the prob-ability of a correct response to be lower afteran error than after all correct choices, butthis trend was not reliable.
Discussion
The results of Experiment 4 provide evi-dence about the characteristics of the mem-ory processes the rats use while solving theproblem. First, the data indicate that therewas a progressive decrease in choice accuracyas the number of choices increased (Table 1,Line 1), even when a correction was madefor the expected chance level (Table 1, Line2). Second, when errors occurred, there wasno tendency for them not to be made to thearms chosen on the first few choices of theday (Table 2, Line 1), even when correc-tions were made for the number of oppor-tunities to repeat choices (Table 2, Line 2),arm preference (Table 2, Line 3), and boththe number of opportunities to repeat choicesand arm preferences (Table 2, Line 4).
110 D. S. OLTON AND R. J. SAMUELSON
Third, when errors occurred, there was areliable tendency for them not to be made tothe arms that had just been chosen (Table 2,Line 1), even when the corrections indicatedabove had been taken into consideration(Table 2, Lines 2-4), although this tendencywas relatively weak. Fourth, there was noindication that the performance decrement asa function of increasing numbers of choiceswas lessened by an error; indeed, the trendthat appeared (which was statistically insig-nificant) was for an increase in the perform-ance decrement (i.e., a lowered probabilityof a correct response) after an error. Fifth,there was no indication that the animals con-fused the different spatial locations anddirected their errors toward arms that wereadjacent to correct arms.
Two types of corrections were made onthe observed data. The first considered thechanging chance probability of a correct re-sponse and was straightforward. The secondconsidered the possibility of a preference forparticular arms, and the assumption wasmade in Table 2, Lines 3 and 4, that an armpreference did exist. The low correlationbetween arm preference for Days 1-10 andDays 11-20 raises a question as to the valid-ity of this assumption. One possibility is thatan arm preference did exist but was unstableand changed markedly from one block ofdays to the next. An alternative is that anarm preference did not exist but resultedfrom the limited sample size which wasavailable for analysis. In any case, the cor-rection for arm preference changes the datain only a quantitative and not a qualitativemanner so that the conclusions from thisexperiment remain unaltered. The absolutemagnitude of the tendency not to make errorsto recently made choices must remain inquestion however.
The finding that rats did not tend to makeerrors to arms close to correct arms sug-gests that there was no generalization grad-ient among the arms and that each arm wastreated independently of the others. Theseresults appeared to be contradictory to thoseof Tolman et al. (1946), who reported thatin a "sunburst" maze, rats tended to choosearms close to the one that led to the goal. In
that experiment, however, passage down thecorrect arm was blocked. In the present ex-periment, of course, access to all arms wasavailable at all times. Thus, the paradigmsare substantially different, and the resultscannot be directly compared.
EXPERIMENT 5
Experiment 4 demonstrated that there wasa performance decrement as the number ofchoices increased (see Table 1) but left un-resolved the issue of whether this decrementwas due to the number of choices or just thepassage of time. Experiment 5 was designedto answer this question by forcing theanimals to spend time in the center compart-ment between the fourth and fifth choices.If the performance decrement in Experiment4 was due simply to the passage of time, thenperformance on Choices 5-8 should be im-paired after confinement. If the performancedecrement was due to the choices themselves,then no impairment should be observed.
MethodSubjects and apparatus. The subjects were six
of the animals from Experiment 4 maintainedunder the same conditions. The apparatus was thesame as in Experiment 4.
ProcedureThe procedure was the same as in Experiment 4
except that animals were confined to the centerplatform after the fourth choice. One test wasgiven each day for S days. When the animal re-turned to the center platform following the fourthchoice, the guillotine doors were lowered for atleast 2 min., more than the amount of time thatthe animals usually required to complete the lastfour choices. Following this confinement, the guil-lotine doors were raised and the animals allowedto choose freely until the remaining four reinforce-ments were obtained.
Results
All animals readily adapted to the confine-ment procedure. During the confinementperiod, the animals were active and movedabout the center compartment in the typicallyexploratory fashion of rats. When the guillo-tine doors were raised, the rats usually pro-ceeded to the nearest arm and either chosethat arm or moved to some other arm. Therewas no obvious change in the accuracy of
SPATIAL MEMORY IN RATS 111
performance as a result of the confinementprocedure, and the mean of the number ofdifferent arms chosen within the first 8choices was 7.5 (range: 7.2-7.9) as com-pared with 7.6 (range: 7.3-8.0) during Days11-20 of the previous experiment.
Discussion
Interference theories of forgetting proposethat information is lost because new informa-tion interferes with the storage or retrievalof previous information and not simply be-cause of the passage of time (Hulse et al.,1975). The current evidence supports thenotion that the decrement in performanceobserved here occurs because of the increasednumber of choices and not because of thetime interval between previous choices andthe current choice. The period over whichrats can successfully remember their pre-vious choices was not addressed by the pres-ent experiment, but is obviously much longerthan the few minutes tested here and mayapproach the limits found in spontaneousalternation experiments (Douglas, 1966;Walker, 1956).
EXPERIMENT 6
The results of Experiment 4 demonstratedthat there was a performance decrement asthe number of choices increased on each daybut that this decrement was alleviated be-tween days. These results indicate that theanimals were able to successfully separateeach day's performance from the previousday's performance so that choice accuracyrecovered to its original level at the start ofeach day. The present experiment was de-signed to determine how rapidly recovery ofperformance could take place. To this end,the rats were allowed to choose all eight armsand were then confined to the center platformwhile all eight arms were rebaited. Theywere then allowed to choose all eight armsagain. The results suggest that the animalcould successfully "parse" their learningunder these conditions so that there wasalmost complete recovery of choice accuracybetween the last choice of the precedingseries and the first choice of the followingseries.
Method •
Subjects and apparatus. The subjects were fourrats of Experiment 4 maintained under the sameconditions as described previously. The apparatuswas the same as in Experiment 4.
Procedure
The procedure was the same as in Experiment 4with the following exception: After completingchoices of each of the eight arms, the animals wereconfined to the center platform by the guillotinedoors. All arms were rebaited, a process whichrequired about 1 min., the guillotine doors raised,and the animals allowed to choose freely until alleight arms were chosen again. This procedure wasthen repeated until the animal was given a total offour tests on each of the first 5 days and a total ofeight tests on each of the next S days.
Results and Discussion
On the first day of repeated tests, the ani-mals chose normally on the first test buttended to choose slowly on the subsequenttests. On the following days, the animalsreadily adapted to the procedure and choseat their normal rate on all tests. There wasa slight but consistent decrease in choiceaccuracy from the level demonstrated on thefirst test of the day, but even on the first daythat the animals were given four tests, choiceaccuracy on the last choice was still greaterthan that observed on the first 5 days of Ex-periment 4. The same results were foundwhen the animals were given eight tests oneach day, demonstrating that the rats wereable to successfully separate each test inmemory so that the performance decrementexpected as a function of the increasing num-ber of choices did not appear. Table 3presents these data.
GENERAL DISCUSSION
The general testing paradigm describedhere may be summarized as sampling withreplacement from a known set of items untilthe entire set is sampled. The cognitive re-quirements of the present paradigm arestraightforward; the animal must be ableto locate each of the eight different arms anddetermine whether each arm has been chosen.Numerous strategies are possible to solve thetask. The easiest approach would be to
112 D. S. OLTON AND R. J. SAMUELSON
TABLE 3PERFORMANCE ON CONSECUTIVE TESTS
Four tests Eight tests
First day All days First day All days
Test M SEM M SEM M SEM M SEM
\2345678
7.37.36.86.8
.3
.2
.4
.3
7.87.46.96.9
.1
.3
.1
.1
7.88.07.57.36.87.07.57.5
.1
.0
.2
.4
.4
.3
.2
.2
7.97.67.57.57.37.57.47.3
.04
.1
.03
.0
.2
.2
.2
.1
Note. Table entries are the mean and standard error of the mean of the number of correct choices with the first eight choices ofeach test.
choose adjacent arms; the choice strategycould be stored in long-term memory, andthe arms could be responded to as they ap-peared and would not have to be individuallyidentified or remembered. A slightly morecomplicated approach would be to mark eacharm as it is chosen; the marking systemcould be stored in long-term memory andchoices guided by the number of marks pres-ent at each arm. Many other types of strat-egies are available as well. But the data indi-cate that the rats did not choose adjacentarms, did not rely on any type of intramazemarking, did not chain responses in partic-ular sequences, and did not utilize any otherobvious strategy to simplify the task. Sincethe testing procedure involves sampling withreplacement, the rats must have been ablesimultaneously to identify each of the eightarms and remember over a period of severalminutes which arms had been chosen andwhich had not.
Further analysis of the cues used by ratswas not undertaken in the present experi-ment because it was not relevant to deter-mining the general type of strategy used insolving the problem. Previous research hasindicated that rats in maze tasks utilize al-most any cue available to them (see dis-cussion by Kimble, 1961; Woodworth,1938). Such is probably also the case here,but accurate evaluation of the relative impor-tance of different extramaze, directional, orcompass cues requires a major experimentaleffort and is beyond the scope of the presentpaper.
Other experiments, although slightly dif-
ferent from the present one, have used aprocedure closely resembling sampling withreplacement from a known set of items untilthe entire set is sampled. The most similaranimal experiment is that of Menzel (1973).Monkeys first observed 18 food items beingdistributed around a large enclosed com-pound and were then released to search forthe food. They obtained an average of 12.5food items and did so in a manner whichminimized the distance which had to betraveled. The most similar human experi-ment is the "missing scan" technique ofBuschke (1963). Subjects were successivelypresented with all but one item from a knownlist and then asked to identify the item thatwas missing. They had no difficulty in per-forming accurately with lists of up to 12 items.Thus, rats, monkeys, and humans all demon-strate a substantial memory capacity whentested in a sampling with replacement para-digm. An interspecies comparison of thelimits of the memory capacity is not yetpossible because of differences in procedureamong the experiments and because none ofthe experiments was difficult enough to sub-stantially challenge the subjects' capabilities.Other experiments (Dennis, 1939; Hunter,as analyzed by Miller & Frick, 1949; Tinkel-paugh, 1932; Yerkes & Yerkes, 1928) havealso addressed the issue of memory formultiple events, but the paradigms weredifferent enough from the present one toprevent direct comparison of results.
The rats rapidly attained almost perfectperformance. Even performance on the firstfew days was far above the level expected on
SPATIAL MEMORY IN RATS 113
the basis of mathematical chance. But math-ematical chance assumes that all choices areindependent of each other—an assumptionwhich is undoubtedly invalid. Other experi-ments (Olton, Note 1) have demonstrated asubstantial spontaneous alternation tendency(Douglas, 1966) in the present situation;when all choices of all arms were reinforced,rats still chose an average of approximatelysix different arms within eight choices, aperformance level which is very similar tothat found here during the first 5 days of theexperiment. Thus, even under conditionswhen reinforcement contingencies do notencourage choosing all arms, substantialalternation still occurs. There are severalconclusions to be drawn from these data:First, the rats bring to the experimentalsituation a strong tendency to perform thetask correctly. This is presumably a reflec-tion of spontaneous alternation. Second, thereinforcement contingencies produce an im-provement in the number of correct choices,and animals learn to perform better aftertraining. Third, the presence of a spon-taneous alternation tendency influences onlythe baseline level of performance exhibitedby the rats and not the strategy used tosolve the task; during spontaneous alterna-tion, rats do not choose adjacent arms, chainresponses, or utilize intramaze cues (Olton,Notel) .
The sequential position of errors analysisindicated that the tendency to make an errorby repeating a response to a particular armincreased with the number of responses madeafter that .arm had been chosen, even whencorrections were made for arm preferenceand for the number of opportunities to repeateach choice. Some of this tendency may havebeen due to the physical constraints of theapparatus because animals returning to thecenter platform after choosing an arm weredirected away from the arm just chosen. Anysuch effect would be very minor however.First, the rats often did not go directly toanother arm after returning to the centerplatform; rather, they explored the entranceto several arms, often completing one ormore revolutions around the center platformbefore making a choice. Second, even if therats did tend to respond to arms physically
opposite the one just chosen, following thesubsequent choice they would be facing backtoward the arm they had just previouslychosen. Third, an analysis of responsesequences indicated that although there wasa slight preference for arms at 90° from thechosen arm, such preferences were weak andnot a prerequisite for accurate performance.Thus the tendency for errors to be made byrepeating choices early in the day's test wasrobust and reliable and does not appear tobe due to the physical characteristics of theapparatus.
Taken together, these data suggest severalconclusions about behavior in the presentparadigm. First, the animals had no difficultyidentifying the eight spatial locations andremembering whether they had chosen eachone; although performance was not perfect,even on the eighth choice it was considerablyabove chance. Second, there was a small butconsistent decrement in performance as thenumber of choices increased; this decrementwas due to interference from previouschoices and not just the passage of time.Third, analysis of the effects of prior choiceson the probability of an error indicated thatthe tendency to make an error by repeatinga choice was a function of the number ofchoices since the original correct choice.Fourth, analysis of the effects of priorchoices on the location of errors indicatedthat the eight arms were treated as a separateplace and there was no apparent tendency forspatial generalization among arms. Fifth,information was stored choice by choice andrepetitions of a choice (i.e., errors) werestored in the same manner as a new choice.Sixth, rats were able to store informationfrom sequential tests separately so that therewas no performance decrement of one teston following tests.
Although the present paradigm is opera-tionally different from the usual human para-digms in which primacy, recency, proactiveinterference, and retroactive interferencehave been studied, it is nonetheless concept-ually very similar. Thus the primacy-recencyanalysis of human data is similar to thesequential location of errors analysis pre-sented in Table 2. The equivalent of theprimacy effect in human serial learning
114 D. S. OLTON AND R. J. SAMUELSON
would appear in the present analysis as alowered probability of an error to choicesmade early in the test, while the equivalentof a recency effect would appear as a loweredprobability of an error to choices made nearthe end of the test. No indication of a pri-macy effect was found in any of the analyses.Data from human subjects suggests that theprimacy effect may result from more re-hearsal of items at the beginning of the listthan in the middle; accuracy of recall is afunction of the amount of rehearsal, whichis greatest for items in the beginning of thelist (Rundus, 1971), and when subjects areinstructed to rehearse only the most recentitem, the primacy effect does not appear(Waugh & Norman, 1965). If the absence ofa primacy effect for rodents is a consistentfinding, it may suggest that rats do not re-hearse their choices or that rehearsal is notpreferentially directed toward early choices,in the tasks involved, but equally distributedamong all choices. Alternatively, the primacyeffect in the human data may be due to therequirement of serial ordering of items andthe formation of chains of responses (Hand-ler & Anderson, 1971). In the present para-digm, of course, rats were not required tochain responses and no particular sequencesof choices were preferred over others. Sincehumans almost inevitably order responses,even when asked to randomize them, a testof this hypothesis can probably be conductedonly by testing rats (or some other animalthat does not automatically sequence re-sponses), forcing them to choose in a partic-ular serial order and then determining if aprimacy effect appears.
The equivalent of proactive and/or retro-active interference in human serial learningwould appear in the present analysis as adecrement in performance as the number ofchoices increased. There was a reliable decre-ment in performance as the number ofchoices increased during the day's test, andthis decrement appeared even when a cor-rection had been made for the varying chanceprobability of a correct response. The pres-ent experiment is unable to identify defin-itively whether the performance decrementobserved here is due to proactive or retro-active interference because the way in which
the animals remember which arms have beenchosen and/or which arms have not beenchosen has not yet been determined. Datafrom discrimination learning experimentsstrongly suggest that rats perform discrim-inations largely on the basis of nonreinforce-ment (see discussion in Olton, 1972; Olton,Walker, Gage, & Johnson, Note 2). If suchis the case in the present experiment, thenrats probably remember not to return toarms which have already been chosen be-cause these would result in nonreinforcementif chosen again. The performance decrementwould then be due to retroactive interference.Other evidence also supports the suggestionof retroactive rather than proactive inter-ference. The finding of a recency effect in theabsence of a primacy effect suggests that theanimals are remembering where they havebeen and which arms ought not to be chosen.The fact that animals could successfully over-come the expected performance decrementwhen given four or eight tests on each dayis also compatible with the suggestion ofretroactive interference.
In summary, the results of the presentexperiments are important because theydemonstrate that rats have a remarkablememory capacity for information aboutspatial location and because they provideinformation about some of the characteristicsof the storage process used in this memory.The data also suggest that just as the intro-duction of the "interoceptive" learning para-digm (Garcia, Hankins, & Rusiniak, 1974;Rozin & Kalat, 1971) substantially changedour estimate of rats' learning abilities, so theintroduction of a spatial location paradigmmay change our estimate of rats' cognitivecapabilities.
REFERENCE NOTES
1. Olton, D. S. Response processes and spatialmemory in rats. Manuscript in preparation.
2. Olton, D. S., Walker, J. A., Gage, F. H. Ill, &Johnston, C. T. Choice behavior in rats searchingfor food. Manuscript submitted for publication,1975.
REFERENCES
Buschke, H. Retention in immediate memory esti-mated without retrieval. Science, 1963,140, 56-57.
Deese, J., Hulse, S. H. The psychology of learn-ing. New York: McGraw-Hill, 1967.
SPATIAL MEMORY IN RATS 115
Dennis, W. Spontaneous alternation in rats as anindicator of the persistence of stimulus effects.Journal of Comparative Psychology, 1939, 28,30S-312.
Douglas, R. J. Cues for spontaneous alternation.Journal of Comparative and Physiological Psy-chology, 1966, 62, 171-183.
Feller, W. Probability theory and its applications(Vol. 1). London: Wiley, 1950.
Garcia, J., Hankins, W. G., & Rusiniak, K! W.Behavioral regulation of the milieu interne inman and rat. Science, 1974, 185, 824-831.
Gleitman, H. Place learning without prior per-formance. Journal of Comparative and Physi-ological Psychology, 1955, 48, 77-79.
Hulse, S. H., Deese, J., & Egeth, H. E. The psy-chology of learning. New York: McGraw-Hill,1975.
Kimble, G. A. Hilgard and Marquis' conditioningand learning. New York: Appleton-Century-Crofts, 1961.
Mandler, G., & Anderson, R. E. Temporal andspatial cues in seriation. Journal of ExperimentalPsychology, 1971, 90, 128-135.
Means, L. W., Hardy, W. T., Gabriel, M., & Up-hold, J. D. Utilization of odor trails by rats inmaze learning. Journal of Comparative andPhysiological Psychology, 1971, 76, 160-164.
Menzel, E. W. Chimpanzee spatial memory orga-nization. Science, 1973, 182, 943-945.
Miller, G. A., & Frick, F. C. Statistical behavior-istics and sequences of responses. PsychologicalReview, 1949, 56, 311-324.
Murdock, B. B., Jr. Transfer designs and formulas.Psychological Bulletin, 1957, 54, 313-326.
Olton, D. S. Shock motivated avoidance behaviorand the analysis of behavior. Psychological Bul-letin, 1973, 79, 243-255.
Olton, D. S., & Samuelson, R. J. Decision makingin the rat: response-choice and response-timemeasures of discrimination reversal' learning.Journal of Comparative and Physiological Psy-chology, 1974, 87, 1134-1147.
Rozin, P., & Kalat, J. W. Specific hungers andpoison avoidance as adaptive specializations oflearning. Psychological Review, 1971, 78, 459-486.
Rundus, D. Analysis of rehearsal processes in freerecall. Journal of Experimental Psychology, 1971,59, 63-77.
Slotnick, B. M., & Katz, H. M. Olfactory learning-set formation in rats. Science, 1974, 185, 796-798.
Sutherland, N. S., & Mackintosh, N. J. Mecha-nisms of animal discrimination learning. NewYork: Academic Press, 1971.
Tinkelpaugh, O. L. Multiple delayed reaction withchimpanzees and monkeys. Journal of Compara-tive Psychology, 1932,12, 207-243.
Tolman, E. C., Ritchie, B. F., & Kalish, D. Studiesin spatial learning: I. Orientation and the shortcut. Journal of Experimental Psychology, 1946,36, 13-24.
Walker, E. L. The duration and course of the re-action decrement and the influence of reward.Journal of Comparative and Physiological Psy-chology, 1956, 49, 167-176.
Wasserman, E. A., & Jensen, D. D. Olfactorystimuli and the "pseudo-extinction" effect. Sci-ence, 1969, 166, 1307-1309.
Waugh, N. C., & Norman, D. A. Primary memory.Psychological Review, 1965, 72, 89-104.
Woodworth, R. S. Experimental psychology. NewYork: Holt, 1938.
Yerkes, R. M., & Yerkes, D. N. Concerning mem-ory in the chimpanzee. Journal of ComparativePsychology, 1928, 8, 237-271.
APPENDIX
TABLE AlDAILY SEQUENCE OF FIRST EIGHT CHOICES OF RAT 3 IN EXPERIMENT 1
Day
101112131415
Order
857263548573261558317246687457237365412875436812
Day
161718192021
Order
768124538352174786457138456823756381742586457371
Day
222324252627
Order
765381248657412364572831463518724627351884252371
Note. The numbers in the order columns refer to Arms 1-8 of the radial maze.
(Table A2 is on the next page.)
116 D. S. OLTON AND R. J. SAMUELSON
TABLE A2RAW DATA FROM DAYS 1-10 IN EXPERIMENT 4
Rat
1
2
3
4
Day
12345678910
1234S678910
12345678910
1234S678910
Order
765126783658247123746384575567823451675432183781245656781345623218765435781246163731356754282187573456
671348625783624715857324685174568123687413245463257147867824513683457126781452372345186
7878351246418514687356142712856134145368274253714621734812437518425417836246824673512478251347663467137824531753885264
18431748751826321648718643254523178686431752486321753287654116572843532864174876251365482173
Rat Day
5 12345678910
6 12345678910
7 12345678910
8 12345678910
Order
18127531741647186325253864171642853715387264147482513617534826142836577352846152831746
75861517356842187635184621467825313457826678245137813562478246135682457137135286474623815
761453175287163816542463715421813251847613582547123567143562841725836436825715816724357123468
85326487287417368586174318652851748623862315478164275368725314865714238637254186125473
Rat Day
9 12345678910
10 12345678910
11 12345678910
12 12345678910
Order
24884213715618126747825321256471853773512864713682514713264858751362241537268428637514615247328
18542315143675724543186853658146318572452641873357846127431856217542186571324684815432187658261437
2182617354743185625841627346174258318537462768513726476845132768471523867516443268471565723
1825175346187428651413181782358641541265783613852147816432567762581372581456864137125346124821731785426
Note, The numbers in the order columns refer to Arms 1-8 of the radial maze.
(Received March 17, 1975; revision received September 16, 1975)
