Backward Conditioning: A Reevaluation of the Empirical ...

Psychological Bul le t in1981. Vol. 89, No. I, 163-175

Copyright 1981 by the American Psychological Association, Inc.0033-2909/81/8901-OI63S00.75

Backward Conditioning:A Reevaluation of the Empirical Evidence

Marcia L. Spetch, Donald M. Wilkie, and J. P. J. PinelUniversity of British Columbia, Vancouver, British Columbia, Canada

There is an apparent discrepancy between the widespread view that backwardconditioning does not occur and the experimental evidence which suggests thatit does. Backward pairing of conditioned and unconditioned stimuli frequentlyhas resulted in effects similar to those produced by forward pairing, and theresults of several recent experiments have established that such effects cannotbe attributed to factors other than stimulus pairing per se. Surprisingly, evensome of the earlier experiments that provided the basis for the current skepticismconcerning backward conditioning provide evidence of its existence. The failureto recognize backward conditioning as a legitimate phenomenon seems to reflecttheoretical biases rather than a paucity of empirical evidence. Thus backwardconditioning and its properties merit renewed interest and rcexamination.

In a recent experiment by Keith-Lucasand Guttman (1975), rats that received asingle footshock followed by the presentationof a toy hedgehog subsequently avoided thehedgehog. Such a finding probably wouldnot surprise one unfamiliar with the animallearning literature, although the imaginationof the experimenters might. As both Hudson(1950) and Keith-Lucas and Guttmanpointed out, common sense would lead oneto expect animals to have the ability to re-spond defensively to a novel stimulus de-tected after a sudden aversive event. An an-imal that sighted an unfamiliar predatorfollowing an abortive attack surely wouldnot submit to another attack (i.e., a forwardpairing of the predator and pain) before re-acting defensively.

It is therefore interesting that the occur-rence of such backward conditioning wouldnot be predicted from a survey of the animallearning literature. The prevailing view isthat conditioned responses (CRs) develop asa result of stimulus pairing only when theconditioned stimulus (CS) precedes the un-

This research was supported in part by a scholarshipand grants from the Natural Sciences and EngineeringResearch Council of Canada. S. Gray, P. Pasquali, andF. Vallc provided many valuable comments.

Requests for reprints should be sent to Marcia L.Spetch, University of British Columbia, Department ofPsychology, 2075 Wcstbrook Mall, Vancouver, BritishColumbia, Canada V6T 1Y7.

conditioned stimulus (US), In most textbookdiscussions of classical conditioning (e.g.,Hulse, Deese, & Egerth, 1975; Kimble,1961; Millenson & Leslie, 1979; Osgood,1953; Schwartz, 1978), the unqualified con-clusion is that conditioning does not occurif the CS follows the US (but cf. Fantino& Logan, 1979). Opinions on what does oc-cur as a result of such backward pairings,however, are divided. Mackintosh (1974)and Terrace (1973), among others, believethat backward pairings have little effect andthat the occasional instances of apparentCRs can be attributed to factors such aspseudoconditioning or sensitization ratherthan to stimulus pairing. The wide accep-tance of this view is illustrated by the factthat backward pairings have been frequentlyused as a baseline control in evaluating theconditioning resulting from forward pairings(Kalish, 1954; Moeller, 1954; Spence &Runquist, 1958). In contrast, Moscovitchand LoLordo (1968), for example, suggestedthat backward pairings result in conditionedinhibition (i.e., a behavior change oppositeto that produced by forward pairings).

What is the empirical basis for the beliefthat the outcomes of forward and backwardpairing differ? Pavlov (1927/1960, p. 27)reported the first failure to demonstratebackward conditioning, and his conclusionthat backward conditioning does not occurhas been defended in subsequent reviews of

163

164 M. SPETCH, D. WILKIE, AND J. PINEL

the literature. Cautela (1965) reviewed sev-eral American backward conditioning stud-ies and concluded, "In all of the precedingstudies backward conditioning was not ob-tained or was attributed to pseudocondition-ing (by the investigators themselves or byothers)" (p. 140). In perhaps the most au-thoritative of contemporary surveys of theconditioning literature, Mackintosh (1974)concluded, "All in all there is little reasonto accept the reality of backward condition-ing" (p. 60). Only one major review providesan exception to this general conclusion. Af-ter reviewing evidence from several Sovietand American experiments, Razran (1956)asserted that backward conditioning pro-duced genuine CRs and was not a case ofpseudoconditioning. His review was eitherlargely ignored or criticized (e.g., Mackin-tosh, 1974), however, and has had little im-pact on the current status of backward con-ditioning within the American learningliterature.

Why were Keith-Lucas and Guttman(1975) able to demonstrate a phenomenonthat has apparently eluded other investiga-tors for so many years? Keith-Lucas andGuttman attributed their success to the na-ture of their CS. The hedgehog CS of theirexperiment was a complex stimulus that per-haps resembled a natural predator of the rat,whereas the CSs in previous attempts todemonstrate backward conditioning typi-cally consisted of arbitrary stimuli such aslights and tones. They suggested that thebackward conditioning established in theirexperiment reflected "an evolved capabilityof the rat and many similar species to as-sociate certain classes of stimulus objectswith exteroceptive pain" (p. 475).

Although such an interpretation mesheswell with the recent interest in adaptive spe-cializations in learning (e.g. Rozin & Kalat,1971) and may account for some of the fail-ures of previous investigators to demonstratebackward conditioning, other recent exper-iments (Heth & Rescorla, 1973; Mahoney& Ayres, 1976; Wagner & Terry, 1975)have successfully demonstrated backwardconditioning with traditional CSs such aslights and tones. Thus a closer examinationof the discrepancy between these recent

demonstrations of backward conditioningand the widespread belief that such condi-tioning does not occur is warranted.

Empirical Requirements for aDemonstration of Backward Conditioning

Is there an unequivocal demonstration ofbackward conditioning? The answer ob-viously depends on the particular criteriaused as a basis for the evaluation. In eval-uating whether a phenomenon exists, thecriteria should be strictly empirical ones.Empirical criteria require that the phenom-enon can be demonstrated and can be at-tributed to the experimental manipulationin question (in this case to the backwardpairings of a US and a CS). Theories of con-ditioning often incorporate additional crite-ria that may be used to evaluate whether thephenomenon supports the theory. But suchtheoretical criteria should not be used toevaluate evidence for the existence of a phe-nomenon because the same evidence may beevaluated differently: positively using thecriteria of one theory, negatively using cri-teria of a different theory. Four empiricalrequirements are basic to a demonstrationof backward conditioning.

1. The most obvious requirement is, ofcourse, the backward pairing of the US andCS. There is considerable disagreement inthe literature as to what constitutes back-ward pairing. In some classification schemes(e.g., Kimble, 1961), all procedures in whichCS onset follows US onset are defined asbackward pairing. In this article we reservethe term backward pairings for only themost unambiguous of such cases—ones withno temporal overlap between the US and theCS. Evidence from only those experimentsin which CS onset follows US offset (im-mediately or after a delay) are considered.According to traditional views, however,none of the procedures in which CS onsetfollows or even occurs simultaneously withUS onset would be expected to produce con-ditioning.

There have been several recent reports ofbackward conditioning of taste aversions (cf.Logue, 1979). Although these may representgenuine instances of backward conditioning,

BACKWARD CONDITIONING 165

the problem of specifying the time course ofthe US (toxicosis) in the taste aversion par-adigm makes it difficult to determine whethera true backward pairing has been used. Forthis reason, we do not evaluate evidence forbackward conditioning of taste aversions.

2. Evidence must be provided that theobserved behavioral change is a direct con-sequence of the backward pairing of the USand the CS. Thus the experiments must in-clude controls that enable the investigatorto specify the degree to which behavioralchanges resembling CRs may have resultedfrom exposure to the US or CS rather thanto their pairing. Two processes that could beconfused with backward conditioning in in-appropriately controlled studies have beenidentified. Pseudoconditioning is a changein behavior to a CS that results from priorexposure to the US (Grether, 1938). Forexample, an animal that has just received apainful shock may exhibit a fear responseto a subsequent novel stimulus. If this occurs,regardless of whether this stimulus has beenpaired with shock, the fear response can beattributed to pseudoconditioning. Sensiti-zation is an augmentation of an uncondi-tioned response (UR) to a CS as a result ofprior exposure to the US (Grant & Adams,1944) or to the CS (Groves & Thompson,1970). Sensitization controls are particularlyimportant whenever the naturally occurringresponses (i.e., URs) to the CS topograph-ically resemble the CRs that are establishedthrough pairings with the US.

Random occurrence of both the CS andUS (Rescorla's 1967, "truly random controlprocedure") generally is held to be the mostappropriate control procedure for both pseu-doconditioning and sensitization. This pro-cedure is impractical, however, when con-ditioning occurs in a few trials. For example,with only two random presentations of theCS and US, it is possible to end up with twoforward or two backward pairings. For stud-ies of backward conditioning in which con-ditioning is asymptotic in a few trials, thebest control procedure is one in which theCS and US are explicitly unpaired (i.e., theCS and US are both presented, but neverclose in time). Discriminative conditioningin which the US is paired with one CS but

not another is a version of this explicitlyunpaired control (cf. Rescorla, 1967).

The explicitly unpaired procedure is pref-erable to either the CS-alone or US-aloneprocedures because it controls for effects ofprior exposure both to the CS and to the US.Rescorla (1967) argued that the explicitlyunpaired procedure introduces a negativeCS-US contingency that may endow the CSwith inhibitory properties. He also argued,however, that the same negative CS-UScontingency is present in the backward pro-cedure. Since any inhibitory tendencies thatmight result from such a contingency shouldbe reflected equally in both groups, the un-paired procedure seems to be an appropriatecontrol for backward conditioning.

3. The interval between trials must belong enough to rule out the possibility thatapparent backward conditioning effects arethe result of delayed forward pairings of theCS presented on one trial and the US pre-sented on the following one. Alternativelyunpaired control groups that are exposed tothe same intervals between the CS and thenext US presentation also can be used toassess the possible effects of such delayedforward pairing. Of course, the most effec-tive way to completely rule out delayed for-ward pairing as a factor in backward con-ditioning is to limit conditioning to a singleUS-CS pairing.

4. To rule out inhibitory conditioning,backward pairing of a US and a CS mustresult in a behavioral change to the CS thatis qualitatively similar to that produced byforward pairing, although the magnitude ofthis change and its temporal course need notbe the same as that produced by a forwardpairing. Furthermore, there is no require-ment as to the nature of such a change (e.g.,that the CR resemble the UR).

Backward conditioning as defined heredoes not specify the nature of the resultinglearning. Issues such as whether the subjectlearns a backward association between theCS and the US or a forward association be-tween the CS and some memory of the USare theoretical rather than empirical ques-tions. As such, they are not relevant to anevaluation of the empirical evidence forbackward conditioning.

166 M. SPETCH, D. WILKFE, AND J. PINEL

Recent Demonstrations of BackwardConditioning

Although many of the earlier backwardconditioning experiments did not include thenecessary control procedures to providestrong evidence for the existence of back-ward conditioning, four recent experiments,including that of Keith-Lucas and Guttman(1975), appear to meet all of the necessaryempirical criteria.

Keith-Lucas and Guttman

The experiment by Keith-Lucas and Gutt-man (1975) was a systematic replication ofan earlier study by Hudson (1950). Hudsonshocked rats while they were eating, thendropped a bundle of pipe cleaners into thechamber. In a subsequent test, the ratsavoided the pipe cleaners. Keith-Lucas andGuttman repeated Hudson's study, using atoy hedgehog as the backward CS and in-cluding controls for nonpairing factors. Theyexposed four groups of rats to the followingsequence of events. A novel stimulus panelcontaining a sugar pellet was inserted intoone end of the chamber. As soon as the ratremoved the pellet, a 750-msec, 1-mA foot-shock was administered, the chamber lightswere extinguished, and the panel was re-moved. After a blackout of 1, 5, 10, or 40sec (1-, 5-, 10-, and 40-sec backward groups),the lights were turned on and a toy hedgehogwas presented for 1 min. Rats in two controlgroups underwent the same procedure ex-cept that exposure to either the shock(hedgehog group) or the hedgehog (shockgroup) was omitted.

AH the subjects were tested 20 hours later.During the 10-min test, the panel and thehedgehog were presented at opposite endsof the chamber; their locations were reversedhalfway through the test.

Five classes of behavior (location, normalapproach, elongated approach, retraction,and pellet consumption) were recorded toassess avoidance of the panel (forward CS),the hedgehog (backward CS), or the shocklocation. This assessment revealed that the1-, 5-, and 10-sec backward groups avoidedthe backward CS (hedgehog) rather than theforward CS (panel) or the shock location.

The 40-sec backward group and shock con-trol group avoided the forward CS (stripedpanel) rather than the hedgehog; the hedge-hog control group did not differentially avoideither stimulus.

Thus, following a single backward pairingwith a US-CS interval of 10 sec or less,avoidance was conditioned to the CS (hedge-hog) that followed shock. Factors other thanstimulus pairing were clearly ruled out bythe absence of comparable effects in the con-trol groups and in the group that receiveda backward pairing with a 40-sec US-CSinterval.

Wagner and Terry

Wagner and Terry (1975) hypothesizedthat manipulations that ensure the continuedrehearsal of a US should result in more ro-bust backward conditioning. Assuming thatmore rehearsal would occur with a "sur-prising" US than with an "expected" US,they designed an experiment to demonstratethat robust and enduring backward condi-tioning could be produced by maintainingthe surprise value of the US over the courseof the backward pairings. The surprise valueof a US was established during preliminarydiscrimination training sessions in which oneforward stimulus (CS+) was always followedby the US and another forward stimulus(CS~) was never followed by the US. Con-sequently, a US that followed the CS+

should be expected; a US that followed theCS" should be unexpected or surprising.Backward pairings were then conductedwith two new stimuli: one paired with anexpected US, the other with a surprising US.

Wagner and Terry tested their hypothesison rabbits in an eyelid conditioning para-digm. During initial discrimination sessions,a CS+ and CS' consistently preceded theoccurrence or nonoccurrence, respectively,of a 100-msec, 5-mA eye shock. Once theblink CRs consistently occurred to the CS+

but not to the CS~, 48 backward condition-ing and testing sessions were conducted. Forthe experimental group, backward pairingtrials and test trials, one of each per session,were embedded in the regular training trials.On some days, one backward CS (CSE) waspresented 500 msec after a CS+-US se-


quence, that is, after an expected US; on theremaining days, the other backward CS(CSS) was presented 500 msec after aCS~-US sequence, that is, after a surprisingUS. A 500-msec flashing light or vibrotactilestimulus served as the backward CSs; theirdesignation as CSE and CSS was counter-balanced across subjects. A control groupwas included to assess the effect of simpleexposure to a surprising US: Subjects re-ceived the same schedule of expected andsurprising US trials, but the surprising UStrials were not followed by a backward CS.For both groups, each session ended with atest trial in which either the CSS or the CSEwas presented alone and the occurrence ofblink CRs was recorded. The same criterionfor recording blink CRs was used for boththe forward and the backward CSs.

Wagner and Terry found clear evidencefor backward conditioning to the CS thatfollowed a surprising US. The subjects dis-played significantly more blink CRs to theCSS than to the CSE—significantly morethan the control subjects displayed in re-sponse to either stimulus. Moreover, thesedifferences increased over the 24 tests witheach stimulus.

Possible pseudoconditioning or sensitiza-tion effects of exposure to a surprising USwere ruled out by the control subjects thatreceived the same schedule of surprising USsbut responded much less to the CSS than didthe rabbits in the experimental group. Fur-thermore, the counterbalanced pairing ofone backward CS with a surprising US andanother CS with an expected US can beviewed as a variation of the discriminativeconditioning control procedure. The factthat the rabbits made significantly more re-sponses to the CSS than to the CSE estab-lished that neither exposure to the backwardCS prior to testing nor any forward condi-tioning effects with the US of the next trialcould account for the results.

Heth and Rescorla

Heth and Rescorla (1973) investigatedsimultaneous and backward fear condition-ing in rats using a conditioned punishmentparadigm. Four groups of rats were exposedon consecutive days to two conditioning ses-

sions in which a total of 20 2-sec tone-lightcompound CSs and 20 4-sec, ,5-MA shockUSs were presented. Each day the rats inone group received 10 simultaneous pairingsin which CS onset occurred 2 sec after USonset. The subjects in two other groups re-ceived 10 backward pairings each day; CSonset occurred 0 sec or .5 sec after US offset.The intertrial interval varied between 4 and8 min. The control group received 20 CSpresentations the 1st day and 20 US presen-tations the 2nd day.

During a subsequent test session, the pun-ishing effects of the CS on previously estab-lished food-reinforced bar pressing were as-sessed. The degree of suppression producedby the response-contingent CS presentationsserved as the measure of conditioned fear.In an earlier experiment reported in thesame article, 20 forward pairings of the sameUS and CS had resulted in a high degree ofresponse suppression by the CS.

Although response suppression was great-est in the simultaneous condition, the barpressing of the rats in the two backwardgroups was suppressed significantly morethan that of the control rats during the first10 min of the test. The two backward groupsdid not differ significantly from each other.

Mahoney and Ayres

Mahoney and Ayres (1976) investigatedthe fear-eliciting properties of a tone CS thathad been paired a single time in a forward,simultaneous, or backward manner with ashock US. The suppression of licking by thethirsty rats during the presentation of thetone was the index of conditioned fear.

Baseline drinking rates were assessed forall six groups of rats before the conditioningsession. During the conditioning session, thesubjects in each group were presented witha single 4-sec tone CS and a 4-sec, 1-mAfootshock US. For the two forward groups,tone offset preceded shock onset by 0 sec or4 sec; for the simultaneous group, the toneand shock occurred simultaneously; for thetwo backward groups, tone onset followedshock offset by 0 or 4 sec. Rats in the sixthgroup served as control subjects; the tonewas presented 3 min before shock onset. Theeffect of each of these pairings on licking


was assessed during 2 subsequent test days.The CS was presented after the 100th lick,and the time taken to complete 10 more lickswas the dependent measure.

Although forward and simultaneous pair-ings produced more suppression than thebackward pairing, the licking of the rats thathad received a backward pairing was sup-pressed significantly more than was the lick-ing of the control subjects. Subjects in thetwo backward groups displayed no signifi-cant difference in suppression.

Burkhardt (1980) replicated the majorfeatures of these results and demonstratedthat the magnitude of suppression producedby a backward pairing increased as US in-tensity increased.

Summary

Each of the preceding experiments satis-fies all of the empirical criteria and thusqualifies as a demonstration of backwardconditioning. In each experiment a back-ward US-CS pairing produced a statisticallysignificant change in behavior that resem-bled the effect of forward pairing of the samestimuli. Each studies included control pro-cedures that ruled out the possibility that thechange in behavior was due to factors otherthan the backward pairings, such as pseu-doconditioning or sensitization. Finally, inall four studies the possibility that the effectwas due to a delayed forward pairing of theCS with the US of a following trial was ruledout: Heth and Rescorla (1973) used longintertrial intervals (4-8 min); Wagner andTerry (1975) used both long intertrial in-tervals and a discriminative conditioningcontrol procedure; and in two studies (Keith-Lucas & Guttman, 1975; Mahoney & Ayres,1976), backward conditioning occurred aftera single pairing.

Early Studies of Backward Conditioning

The discrepancy between the recent dem-onstrations of backward conditioning andthe prevalent view that backward condition-ing does not occur suggests that a reeval-uation of the early studies of backward con-ditioning, which serve as a basis for thecurrent view, is warranted.

Pavlov (1927/1960, p. 27) initially con-cluded that CRs could not be established ifthe CS was presented after the US. Thisconclusion was based on unpublished exper-iments by Krestovnikov in which dogs weregiven many (374 or 427) US-CS pairings;in a subsequent test, the CS presented alonefailed to elicit a CR. In a later lecture (pp.391-394), however, Pavlov stated that hisearlier position was a "probable error" (p.391). He described the results of experi-ments by Pavlov, Kreps, Podkopaev, Pro-rokov, and Koupalov in which only a fewbackward pairings were administered: "Thehitherto neutral stimulus when now testedalone revealed undoubted conditioned prop-erties" (p. 393). He then suggested that sinceKrestovnikov's experiments assessed onlythe effects of multiple trials, his earlier con-clusions representeda brilliant illustration of the danger of too hasty gen-eralizations. We imagined that if it were a true condi-tioned reflex it would increase in intensity upon repe-tition of the combination, and not diminish and fi-nally vanish as happens in these experiments,(p. 394)

Still later, Pavlov (1928, p. 381) restated hisconclusion that weak CRs develop as a resultof initial backward pairings, adding thatwith repeated pairings the CS becomes in-hibitory. Thus Pavlov's final position was notthat the backward procedure produces noconditioning but rather that it has "a doubleeffect: at first temporarily it assists in theformation of the conditioned reflex, and thendestroys it" (1932, p. 93).

The existence of the "double effect" firstreported by Pavlov has been confirmed insubsequent studies. Several investigators havereported the appearance of CRs followinga few backward pairings (e.g., Switzer,1930; Wolfle, 1930), whereas those assessingthe effects of multiple backward pairingshave found either no conditioning (e.g.,Bernstein, 1934; Porter, 1938) or condi-tioned inhibition (e.g., Moscovitch & Lo-Lordo, 1968). Spooner and Kellogg (1947)administered different numbers of backwardpairings and reported that CRs appearedfollowing initial pairings but diminishedwith repeated pairings. Heth (1976) also re-ported that backward pairings produced abiphasic acquisition function, with the CS


controlling excitatory tendencies after 10pairings and inhibitory tendencies after 160pairings. Unfortunately, many of the earlystudies did not include the necessary controlsto rule out the possibility that factors otherthan stimulus pairing were producing theapparent CRs during the initial backwardpairings. A few studies, however, did includesuch controls.

Mowrer and Aiken (1954), for example,demonstrated backward conditioning usinga conditioned punishment paradigm. Onegroup of rats received a single backwardpairing of a shock and light on each of 5successive days. On the 6th day the light wasmade contingent on a previously establishedbar-press response; suppression of this re-sponse served as a measure of conditioning.Suppression in these rats was significantlygreater than rats that had received unpairedCS and US presentations. Matsumiya (1960)replicated these results. The Heth and Res-corla (1973) study was in fact a systematicreplication of these earlier studies.

Two studies have demonstrated backwardconditioning of the galvanic skin response(GSR). In both, backward pairing producedsignificantly greater changes in the GSRthan did unpaired presentations of the CSand US (Champion & Jones, 1961; Zeiner& Grings, 1968). Thus empirical evidencefor backward conditioning does exist in theearly literature, and this evidence is not onlyconsistent with but anticipates the findingsof recent backward conditioning studies.

The frequent reports in the early literatureof unsuccessful backward conditioning ex-periments also deserve attention. As men-tioned earlier, several experimenters failedto observe CRs following multiple backwardpairings (e.g., Pavlov, 1927/1960; Porter,1938). Others who reported that apparentCRs occurred early in training concludedthat these were not genuine CRs becausetheir occurrence declined with repeated pair-ings (e.g., Spooner & Kellogg, 1947). Andsome experimenters (Grether, 1938; Pro-kasy, Hall, & Fawcett, 1962) reported thatbackward groups do not differ from pseu-doconditioning groups. Because the resultsof these studies are frequently cited to sup-port the conclusion that backward condi-tioning is not a genuine phenomenon (Cau-

tela, 1965; Gormezano & Moore, 1969;Mackintosh, 1974), it is worthwhile to care-fully examine the evidence provided by someof them.

An experiment by Spooner and Kellogg(1947) is probably the most commonly cited.Spooner and Kellogg compared forward, si-multaneous, and backward conditioning ofthe finger withdrawal response in humansubjects. Three features of their data led totheir conclusion that backward and simul-taneous conditioning are entirely differentphenomena from forward conditioning.

1. Differences in response latency.Spooner and Kellogg found that the CR la-tencies of the backward and simultaneousgroups were shorter and more variable thanthose of the forward groups and suggestedthat these differences reflect "a basic dis-tinction between the two processes" (Spooner& Kellogg, 1947, p. 331). When one ex-amines the mean latencies for all groups,however, one finds even greater differencesin latencies between forward groups withdifferent interstimulus intervals. As Pavlov(1927/1960, p. 88) reported, response la-tency varied as a direct function of the in-terval between the CS and the US. Thusshorter latencies in the backward and si-multaneous groups may simply represent anextension of this general principle. Never-theless, if differences in response latencyconstitute evidence for different processes,one is forced to conclude that different pro-cesses also underlie forward conditioning atdifferent interstimulus intervals.

2. Differences in shock threshold. TheUS intensity was adjusted for individual sub-jects throughout the experiment to preventadaptation to the shock, and it was foundthat the US intensity required to elicit un-conditioned finger withdrawal was higher forthe backward and simultaneous groups thanfor the forward groups. Although Spoonerand Kellogg (1947) presented this as evi-dence that forward and backward condition-ing are different phenomena, no statisticalanalysis was performed to test the signifi-cance of these differences, and no evidencewas offered that the US threshold is relatedin any way to the conditioning process.Moreover, because these differences werepresent from the first test sessions, it is likely


that the groups were not evenly matchedwith respect to initial threshold levels.

3. Differences in progress. When thefrequency of CRs was plotted as a functionof trials, differences between the forward,backward, and simultaneous groups emerged.Whereas the forward groups showed an in-crease in CR frequency as a function oftraining, the backward and simultaneousgroups showed a decrease and eventual ces-sation of the CR with repeated pairings.Essentially, these results confirm those re-ported by Pavlov. Unlike Pavlov (1927/1960, p. 393), however, Spoonerand Kellogg(1947) concluded that backward and simul-taneous conditioning were not "true condi-tioning." This conclusion was reached inspite of the fact that their experiment in-cluded no control groups to assess the effectsof factors other than stimulus pairing. With-out such controls, it is impossible to deter-mine whether the CRs produced by thebackward pairings reflect genuine condition-ing. Nevertheless, their tenuous conclusionthat backward conditioning is a fundamen-tally different process from forward condi-tioning frequently has been cited (e.g., Os-good, 1953; Terrace, 1973).

Subsequent authors further complicatedthe issue by exaggerating the claims ofSpooner and Kellogg (1947) regarding thepossibility that backward conditioned CRsmight be due to pseudoconditioning. Spoonerand Kellogg's actual conclusion was that"it is possible that backward and simulta-neous CRs are special cases of pseudocon-ditioning, although our data do not whollysupport such a view" (p. 334). Cautela(1965) claimed, however, that "they con-cluded that backward-conditioned responsesprobably are special cases of pseudocondi-tioning" (p. 139), and Gormezano and Moore(1969) stated that Spooner and Kellogg"clearly articulated the notion that forwardand backward conditioning are fundamen-tally different processes with the latter beinga form of pseudoconditioning" (p. 145). Sur-prisingly, Bugelski (1956) claimed thatSpooner and Kellogg found that "any re-sponses that appear early in training mightbe attributed to sensitization" (p. 129), eventhough there was no mention of sensitizationin Spooner and Kellogg's article.

An earlier experiment by Grether (1938)has also been referenced frequently to sup-port the view that backward conditioningmay be attributed to pseudoconditioning,Grether presented two monkeys with 10backward pairings of a US and a CS. Forone subject, the US was flashlight powderthat exploded 2 feet (.6 m) in front of themonkey; for the other, the US was a snake"blowout" that uncurled to 6 inches (15.24cm) from the monkey's nose and produceda rattling noise. For both, the sound of anelectric bell served as the CS. Two additionalmonkeys served as pseudoconditioning con-trol subjects and received 10 presentationsof the US alone (exploding powder for one;the snake blowout for the other). When sub-sequently tested with the bell, Grether re-ported, all subjects "reacted violently."

Although, as Grether (1938) suggested,this experiment demonstrates the need forpseudoconditioning control groups, it by nomeans provides evidence against the possi-bility that backward conditioning is genuineconditioning. First, "reacting violently" isnot a quantifiable measure with which tomake fine distinctions between the groups.Second, because the control subjects appar-ently responded maximally in terms of thisqualitative measure, the comparison be-tween the backward and pseudoconditioningtreatments may have suffered from a ceilingeffect. The possibility of such a ceiling effectcannot be ruled out in the absence of evi-dence that any experimental treatment, suchas forward pairings, would produce greaterresponding than that shown by the pseudo-conditioning subjects.

In fact, Prokasy et al. (1962) includedsuch a forward conditioning group in a sub-sequent failure to demonstrate differencesbetween backward conditioning and pseu-doconditioning groups. They compared GSRconditioning for forward, simultaneous,backward, randomly paired, US-only, andCS-only groups. None of the conditioninggroups, including the forward one, differedsignificantly from the groups that had re-ceived random pairings or US-alone presen-tations. Nevertheless, even these results havebeen used to argue that backward condi-tioning does not exist (e.g., Mackintosh,1974).


In summary, these experiments by Grether(1938), Spooner and Kellogg (1947), andProkasy et al. (1962), which have been citedfrequently as evidence that backward con-ditioning does not exist, in fact provide littleif any support for the view that backwardconditioning is not possible.

Previous Reviews of BackwardConditioning

Our evaluation of the earlier empiricalevidence for backward conditioning suggestsa conclusion that is similar to that reachedby Razran (1956) in his review of the Rus-sian and American literature but is clearlyat odds with the influential conclusions ofCautela (1965) and Mackintosh (1974).This discrepancy appears to stem in partfrom differences in the criteria used.

Cautela

Cautela (1965) evaluated the evidence forbackward conditioning in terms of two the-oretical views of conditioning: a "substitutestimulus" definition and a "modifying stim-ulus" definition. He used criteria derivedfrom the former definition to evaluate clas-sical conditioning studies and criteria de-rived from the second to evaluate studiesthat used operant conditioning proceduresto assess the effect of stimulus pairing.

According to the substitute stimulus view,conditioning is said to have taken place onlyif a neutral stimulus, paired with a stimulusthat always elicits a particular response,elicits that same response: The response tothe CS must resemble the response to theUS, Cautela required as well that CR prob-ability "increase with repeated reinforce-ments and remain relatively stable whensome maximum strength is reached. . . .Unstable or temporary conditioned re-sponses would be attributed to pseudocon-ditioning" (p. 136).

Because backward conditioning is maxi-mal following the initial trials and may dis-appear with repeated pairings (Spooner &Kellogg, 1947), it is readily apparent whyCautela's view of conditioning would leadto the conclusion that backward conditioningdoes not occur. Indeed, after reviewing the

classical conditioning studies, Cautela con-cluded that

only one study seems unquestionably to have fulfilledthe criteria for the substitute stimulus definition of con-ditioning. In all other studies, the conditioned responseswere unstable, did not become stronger with practiceor were called pseudoconditioning by some investigator,(p. 141)

Cautela then reviewed studies that usedan operant conditioning procedure to assessthe effect of prior backward pairings of a CSand a US, basing his evaluation on criteriadetermined by the modifying stimulus def-inition of conditioning. According to thisview, conditioning is "the process of modi-fying the effects of a stimulus on the be-havior of organisms by associating it witha second stimulus. . . . The stimulus caninhibit as well as facilitate response proba-bility" (p. 136). This definition is less re-strictive than our definition of backward con-ditioning because it encompassesdemonstrations of inhibitory conditioning(i.e., behavioral change produced by back-ward pairings opposite to that produced byforward pairings of the CS and US). Fiveof the six such operant studies reviewed byCautela provided evidence of backward con-ditioning, according to his modifying stim-ulus criteria. Of these, four were demon-strations of inhibitory conditioning.

In the final section of Cautela's review, itbecomes clear that, in addition to the criteriamade explicit from the start, Cautela hadsome implicit criteria based on assumptionsabout the nature of the association resultingfrom backward pairings. Six studies metCautela's explicitly stated criteria for back-ward conditioning (the one substitute-stim-ulus study and five modifying-stimulus stud-ies). Yet, rather than accepting the evidencefrom them, Cautela dismissed all six on thegrounds that all had used noxious stimuli asthe US. He first suggested that "the appar-ent necessary involvement of a noxious stim-ulus in backward conditioning supports thesensitization hypothesis" (p. 141). He thenproposed an alternative "duration-of-pain"hypothesis to account for the appearance ofCRs following backward US-CS pairings:"Cases of reported backward conditioningwith the use of a noxious stimulus are really


cases of simultaneous conditioning of the CSand pain perception" (p. 141).

Interestingly, Cautela proposed these hy-potheses to explain the appearance of CRsfollowing a backward pairing, even thoughfour out of six of the studies that met hiscriteria for backward conditioning actuallydemonstrated conditioned inhibition. Thus,although his duration-of-pain hypothesismight explain at a theoretical level the de-velopment of backward conditioning, it doesnot explain why repeated backward pairingswould produce an effect that is opposite tothat produced by repeated forward pairings.

Cautela concluded, "Before either the sen-sitization hypothesis or the duration-of-painhypothesis can be ruled out, backward con-ditioning will have to be obtained withoutthe use of a noxious stimulus" (p. 143). Thisconclusion is erroneous on two accounts.First, even when noxious stimuli are used,sensitization can be ruled out by the use ofsensitization control groups. Second, the du-ration-of-pain hypothesis is a theoretical ex-planation of the results of backward pair-ings. As we state in the introduction, anempirical demonstration of backward con-ditioning does not require evidence that theassociation learned be backwards. If onewere to make such a requirement, backwardconditioning could not be demonstrated withcertainty, regardless of the stimulus used.Any stimulus, noxious or rewarding, can behypothesized to produce perceptual after-effects, the duration of which for an animalcannot be determined with certainty. Thuseven a demonstration of backward condi-tioning with an appetitive US would not ruleout the possibility that the underlying as-sociation was a forward one between the CSand aftereffects of the US.

Although Cautela's review appeared priorto many of the more well-controlled dem-onstrations of backward conditioning, it isapparent that none of the experimental ev-idence for backward conditioning to datewould satisfy both his explicit and his im-plicit criteria. In fact, not even widely ac-cepted demonstrations of forward condition-ing (e.g., Peterson, Ackil, Frommer, &Hearst, 1972; Wasserman, 1973) meet allof the criteria derived from the stimulus-sub-stitution definition of conditioning.

Mackintosh

Mackintosh (1974) did not make explicitthe criteria he used. He gave two reasons,however, for rejecting the apparently suc-cessful backward conditioning studies: (a)the numerous failures to replicate successfulexperiments and (b) the problems arisingfrom the use of the GSR as a CR.

In view of the number of successful dem-onstrations of backward conditioning, fail-ures to replicate successful studies do notconstitute sufficient evidence for questioningits existence, especially when several suchfailures (e.g., Grether, 1937; Prokasy et al.,1962) were poorly controlled or provided lit-tle direct evidence against the effectivenessof backward conditioning or both. The prob-lems with the GSR, however, do deservesome consideration.

Mackintosh stated that, in a test for back-ward conditioning,

the CS presented alone is a novel event, and the occur-rence of a GSR on the first few test trials is most simplythought of as a component of the orientation reaction.In support of this . . . conditioning is maximal on thefirst few test trials and thereafter tends to de-cline, (p. 59)

Although this argument is potentially viable,a close examination of the studies thatMackintosh cited to support his interpreta-tion reveals that this possibility may havebeen ruled out by the control data reportedin the studies. For example, Champion andJones (1961) reported conditioning of theGSR following a small number of backwardor forward pairings of a shock and a tone.Three groups of people received seven for-ward CS-US pairings, seven backwardUS-CS pairings, or seven unpaired presen-tations of the US and CS. The forward andbackward groups were tested with the CSalone following one, three, five, and sevenpairings. The subjects in both groups re-sponded more on all tests than did the un-paired control subjects, who were tested fol-lowing the same number of US and CSpresentations. Because the contribution oforientation reactions to the GSR should havebeen the same in all groups, the higher in-cidence of responding in the backward groupcompared with the unpaired group cannotbe attributed to orientation reactions alone.


In a footnote Mackintosh stated:An additional complication (deriving from the use ofstudents as subjects) has been noted by Zeiner andGrings (1968): subjects that reported expecting the CSto signify something showed a markedly higher level ofGSR responding than did less suspicious subjects,(p. 59)

Zeiner and Grings first replicated the Cham-pion and Jones (1961) study and found sig-nificantly greater GSR CRs in subjects whohad received backward pairings than in sub-jects in an unpaired control group. They thendivided the backward group into twosubgroups on the basis of verbal reports andresponses to a questionnaire regarding theperceived significance of the CS. A post hoccomparison of the data of these subgroupsrevealed that subjects who attributed signif-icance to the CS had displayed greater GSRCRs than had those who did not think theCS was significant. This is an interestingcorrelation, but it has no bearing on theoriginal finding that subjects in the back-ward group responded more than those inthe unpaired control group. First, becausethe assessment of perceived significance ofthe CS was conducted after the conditioningprocess, it is impossible to determine whethersubjects who attached more significance tothe CS were more suspicious to start withor whether they perceived the CS as signif-icant because of the conditioning process.Even if one wished to infer from this findingthat suspicious subjects are more easily con-ditioned, this would not constitute evidenceagainst the efficacy of backward pairings inproducing such conditioning.

Razran

In his review of Soviet and Americanbackward conditioning experiments, Razran(1956) concluded:

On the whole, the analyzed evidence is unmistakable indemonstrating that B.C. [backward conditioning] is nota case of pseudoconditioning, but is a genuine CR-as-sociative manifestation, and that stable backward CR'scan be obtained and maintained under favorable exper-imental conditions, (p. 67)

Razran did not enumerate the criteria heused to evaluate the evidence that led to thisconclusion. In fact, it appears that his onlycriteria were that a backward procedure was

used and that the backward procedure re-sulted in the appearance of CRs. He ac-cepted as evidence of genuine backward con-ditioning all studies that observed CRs as aresult of backward pairings, regardless ofwhether controls for factors other than thebackward pairings were included. In fact,several of the studies that Razran cited asthe most conclusive demonstrations of back-ward conditioning (Spooner & Kellogg,1947; Wolfle, 1930, 1932) did not includethe necessary control groups to rule out pseu-doconditioning or sensitization effects. Thus,although we agree with Razran's conclusionthat backward conditioning is a genuine phe-nomenon, we do not agree with the evidenceon which his conclusion was based. Razran'sunqualified acceptance of evidence from un-controlled studies is probably why his reviewhas had little impact on views of backwardconditioning in the American literature.

Possible Frameworks for BackwardConditioning

The existence of backward conditioningis incompatible with several traditional viewsof classical conditioning (e.g., Hull, 1943)and poses problems for contemporary viewsthat emphasize the predictive function of theCS in the process of learning—for example,the information hypothesis (Egger & Miller,1962) or the contingency theory (Rescorla,1967). Thus the empirical evidence for back-ward conditioning provides a theoreticalchallenge that should stimulate new or mod-ified views of conditioning processes.

On the other hand, a few contemporaryperspectives do seem able to account forbackward conditioning. For example, thebiphasic effect of backward, pairings, withexcitatory conditioning occurring after ini-tial pairings and inhibition occurring afterrepeated pairings, is consistent with Wagnerand Terry's (1975) postepisodic rehearsaltheory. They proposed that associations be-tween the CS and US depend on the simul-taneous rehearsal of the two events in short-term memory and that more post-US re-hearsal occurs when the US is surprising.During initial backward pairings, the US isrelatively surprising, and post-US rehearsalwould be in progress when the CS is pre-


sented; excitatory conditioning would occurbecause of the joint rehearsal of the twoevents. On later trials the US would becomeexpected on the basis of contextual cues, andpost-US rehearsal would be attenuated. ThenCS rehearsal in the absence of US rehearsalwould cause the CS to undergo extinction,and with repeated pairings the CS mightacquire inhibitory properties.

Both the existence of backward condition-ing and some of its prominent features mightalso be viewed profitably in terms of recentapproaches that emphasize the role for phy-logenetic factors in learning. Fantino andLogan (1979) summarized the major prem-ises that underly this "natural history" per-spective:

(1) Intricate interactions occur between ontogenetic andphylogenetic factors in learning. The nature of theseinteractions is determined by the ecological demandsplaced on the species in question. (2) Learning processesoccur with tremendous diversity, which is governed bythe nature of the situation that renders learning ofadaptive benefit. (3) Only by understanding how learn-ing functions to maximize reproductive success in theorganism's natural environment can we arrive at com-plete explanations of the nature of behavior change. Forit is here that the demands for survival are defined,(p. 386)

Most successful demonstrations of back-ward conditioning to date have several com-mon features: the use of noxious stimuli asUSs, the small number of US-CS pairings,the unpredictability of the US or all three.Keith-Lucas and Guttman (1975) demon-strated particularly robust backward con-ditioning in rats using a CS similar to a nat-ural predator. The contribution of suchfeatures to the effectiveness of backwardpairings would be expected if the processthat mediates backward conditioning evolvedas an adaptation to attacks by conspecificsor predators. An animal that could associatesudden, aversive stimulation with a subse-quent novel stimulus would gain clear re-productive advantage.

Regardless of the particular theoreticalframework in which backward conditioningis considered, the empirical evidence for itsexistence can no longer be ignored. Thus thetime for disputing whether backward con-ditioning is possible is past; it is time, in-stead, for systematic exploration of the con-ditions under which it occurs and the

variables that affect the magnitude and du-ration of the effect.

References

Bernstein, A. L. Temporal factors in the formation ofconditioned eyelid reactions in human subjects. Jour-nal of General Psychology, 1934, 10, 173-197.

Bugelski, B. R. The psychology of learning. New York:Holt, 1956.

Burkhardt, P. E. One-trial backward fear conditioningin rats as a function of US intensity. Bulletin of thePsychonomic Society, 1980, 15, 9-11.

Cautela, J. R. The problem of backward conditioning.Journal of Psychology, 1965, 60, 135-144.

Champion, R. A., & Jones, J. E. Forward, backward,and pseudoconditioning of the GSR. Journal of Ex-perimental Psychology, 1961, 62, 58-61.

Egger, M. D., & Miller, N. E. Secondary reinforcementin rats as a function of information value and reli-ability of the stimulus. Journal of Experimental Psy-chology, 1962, 64, 97-104.

Fantino, E., & Logan, C. A. The experimental analysisof behavior: A biological approach. San Francisco:Freeman, 1979.

Gormezano, L, & Moore, J. W. Classical conditioning.In M. H. Marx (Ed.), Learning: Processes. Toronto,Ontario, Canada: Collier-Macmillan, 1969.

Grant, D. A., & Adams, J. K. "Alpha" conditioning inthe eyelid. Journal of Experimental Psychology,1944, 34, 136-147.

Grether, W. F. Pseudoconditioning without paired stim-ulation encountered in attempted backward condi-tioning. Journal of Comparative Psychology, 1938,25, 91-96.

Groves, P. M., & Thompson, R. F. Habituation: A dual-process theory. Psychological Review, 1970, 77,419-450.

Heth, C. D. Simultaneous and backward fear condi-tioning as a function of a number of CS-UCS pairings.Journal of Experimental Psychology: Animal Be-havior Processes, 1976, 2, 117-129.

Heth, D. C., & Rescorla, R. A. Simultaneous and back-ward fear conditioning in the rat. Journal of Com-parative and Physiological Psychology, 1973, 82,434-443.

Hudson, B. B. One-trial learning in the domestic rat.Genetic Psychology Monographs, 1950, 41, 99-145.

Hull, C. L. Principles of behavior. New York: Appleton-Century-Crofts, 1943.

Hulse, S. H., Deese, J., & Egerth, H. The psychologyof learning. New York: McGraw-Hill, 1975.

Kalish, H. I. Strength of fear as a function of the numberof acquisition and extinction trials. Journal of Ex-perimental Psychology, 1954, 47, 1-9.

Keith-Lucas, T., & Guttman, N. Robust single-trialdelayed backward conditioning. Journal of Compar-ative and Physiological Psychology, 1975, 88,468-476.

Kimble, G. A. Hilgard and Marquis' conditioning andlearning. New York: Appleton-Century-Crofts, 1961.

Logue, A. W. Taste aversion and the generality of thelaws of learning. Psychological Bulletin, 1979, 86,276-296.


Mackintosh, N. J. The psychology of animal learning.New York: Academic Press, 1974.

Mahoney, W. J., & Ayres, J. J. B. One-trial simulta-neous and backward fear conditioning as reflected inconditioned suppression of licking in rats. AnimalLearning and Behavior, 1976, 4, 357-362.

Matsumiya, Y. The effects of US intensity and CS-USpattern on conditioned emotional response. JapanesePsychological Research, 1960, 2, 35-42.

Millenson, J. R., & Leslie, J. C. Principles of behavioralanalysis. New York: Macmillan, 1979.

Moeller, G. The CS-UCS interval in GSR conditioning.Journal of Experimental Psychology, 1954, 48,162-166.

Moscovitch, A., & LoLordo, V. M. Role of safety in thePavlovian backward fear conditioning procedure.Journal of Comparative and Physiological Psychol-ogy, 1968, 66, 673-678.

Mowrer, O. H., & Aiken, F. G. Contiguity vs. drive-reduction in conditioned fear: Temporal variations inconditioned and unconditioned stimulus. AmericanJournal of Psychology, 1954, 67, 26-38.

Osgood, C. E. Method and theory in experimental psy-chology. New York: Oxford University Press, 1953.

Pavlov, I. P. Conditioned reflexes: An investigation ofthe physiological activity of the cerebral cortex(G. V. Anrcp, Ed. and trans.). New York: Dover,1960. (Originally published, 1927.)

Pavlov, I. P. Lectures on conditioned reflexes (W. H.Gantt, trans.). New York: International, 1928.

Pavlov, I. P. The reply of a physiologist to psychologists.Psychological Review, 1932, 39, 91-127.

Peterson, G. B., Ackil, J. E., Frommer, G. P., & Hearst,E. Conditioned approach and contact behavior to-wards signals for food or brain-stirnulation reinforce-ment. Science, 1972, 777, 1009-1011.

Porter, J. M., Jr. Backward conditioning of the eyelidresponse. Journal of Experimental Psychology, 1938,23, 403-410.

Prokasy, W. F., Hall, J. F., & Fawcett, J. T. Adaptation,sensitization, forward and backward conditioning,and pseudoconditioning of the GSR. PsychologicalReports, 1962, 10, 103-106.

Razran, G. Backward conditioning. Psychological Bul-letin, 1956, 53, 55-69.

Rescorla, R. A. Pavlovian conditioning and its propercontrol procedures. Psychological Review, 1967, 74,71-80.

Rozin, P., & Kalat , J. W. Specific hungers and poisonavoidance as adaptive specializations of learning. Psy-chological Review, 1971, 78, 459-486.

Schwartz, B. Psychology of learning and behavior. NewYork: Norton, 1978.

Spence, K. W., & Runquist, W. N. Temporal effectsof conditioned fear on the eyelid reflex. Journal ofExperimental Psychology, 1958, 55, 613-616.

Spooner, A., & Kellogg, W. N. The backward condi-tioning curve. American Journal of Psychology,1947, 60, 321-334.

Switzer, C. A. Backward conditioning of the lid reflex.Journal of Experimental Psychology, 1930, 13,76-97.

Terrace, H. S. Classical conditioning. In J. A. Nevin& G. S. Reynolds (Eds.), The study of behavior:Learning, motivation, emotion, and instinct. Glen-view, 111.: Scott, Foresman, 1973.

Wagner, A. R., & Terry, W. S. Backward conditioningto a CS following an expected vs. a surprising UCS.Animal Learning and Behavior, 1975, 3, 370-374.

Wasserman, E. A. Pavlovian conditioning wi th heat re-inforcement produces stimulus directed pecks inchicks. Science, 1973, 181, 875-877.

Wolfle, H. M. Time factors in conditioning finger with-drawal. Journal of General Psychology, 1930, 4,372-378.

Wolfle, H. M. Conditioning as a function of the intervalbetween the conditioned and the original stimulus.Journal of General Psychology, 1932, 7, 80-103.

Zeiner, A., & Grings, W. W. Backward conditioning:A replication with emphasis on conceptualizations bythe subject. Journal of Experimental Psychology,1968, 76, 232-235.

Received January 30, 1980

Backward Conditioning: A Reevaluation of the Empirical ...

Documents