Experiencing Extinction Facilitates Subsequent Acquisition of Positive, but not Negative Patterning in Human Predictive Learning [Annotation]

Dear Dr. Vadillo, Many thanks for considering the manuscript “Experiencing Extinction Facilitates Subsequent Acquisition of Positive, but not Negative Patterning in Human Predictive Learning” for publication in Psicologica. The co-authors (myself included) appreciate the time that you and the reviewers have taken to read the manuscript and think about it carefully. We found all the comments very constructive, and we believe that the manuscript is now a much-improved version of what it was when first submitted. We have addressed all the points highlighted by yourself and the reviewers. Below, you will find a breakdown of reviewers’ main points, followed by our reply to each point explaining how we have addressed the concern/suggestion in the manuscript. Thus, you will find the comments followed by our replies, which start with # in bold. In line with reviewers’ suggestions, we agree that in the previous version our conclusions sometimes went beyond the data. Hence, in the revised version of the manuscript, we have qualified more our conclusions. The manuscript has been revised in full considering all the points raised by you and the reviewers. We hope the readability of the manuscript has also been improved. We again want to thank you and them for their constructive critiques and hope that this version is better suited for the readership of Psicologica. Editor First of all, let me apologize for the unusually long delay in processing your manuscript. As you probably know, the editorial and reviewing system at Psicológica has changed dramatically over the last few months and it has taken me a while to learn how to deal with the new system and find reviewers. I take full responsibility for this delay. I have been fortunate enough to receive comments from two expert researchers in the field. Reviewer 1 has chosen to remain anonymous and, as such, his review is not available on the public repository. You will find a copy of his review pasted at the end of this letter. Reviewer 2 is Tom Beesley and his review is publicly available at https://digital.csic.es/handle/10261/278489. As you can see, both reviewers see merit in your manuscript and think it is a valuable contribution to the field. At the same time, both of them highlight potential shortcomings in the methods, analyses and introduction/discussion. In general, I think that it should be relatively easy to address these concerns in a revised version of the manuscript. I will not reiterate here every single comment made by the reviewers (although I encourage you to take each of them to the heart), but let me highlight some concerns that I find particularly relevant. As noted by Reviewer 1, at some points there is a mismatch between the robustness of the results and the strength of the conclusions drawn from them. Some of the crucial effects and interactions failed to reach statistical significance casting doubts on some of the analyses that follow up on the general ANOVAs. The text should highlight that the conclusions drawn from many of those analyses should be interpreted with extreme caution and considered suggestive more than conclusive. #Thank you for the suggestion. We have modified the conclusions in line with the robustness of the results. In the same vein, I was somewhat surprised by the low number of participants included in each study. Thirty-two participants is a very small sample for a study including a between-groups factor and testing for multiple complex interactions. The fact that this sample size was used in previous studies and sufficed to find significant results is not a terribly convincing argument. It is too late to conduct an a priori power analysis, but this does not prevent the authors from reporting information that will help the reader understand the uncertainty associated with such small samples and any potential implications for the interpretation of the results. It would be useful, for instance, to report what is the smallest effect size that can be detected with 80% power and N=32 in a two-group design. Similarly, limited power casts doubts on the interpretation of non-significant findings. (Do they provide genuine support for the null hypothesis? Or is the evidence simply inconclusive?) It would be extremely useful to complement frequentist stats with Bayesian statistics to address this concern. #Thank you for the suggestion. We have included the BFexc to assess the reliability of the null effect during test phase. In general, the evidence for the null-result is quite convincing in light of Bayesian analyses, despite of the allegedly low sample of participants. Both reviewers have detected typos and grammatical errors. Below I list a few errors found by myself (the authors should take all of them with caution, as my English leaves much to be desired). Beyond correcting these small typos, it would be great if the authors could have the manuscript revised and edited by a native English speaker. Needless to say, this is a minor concern. Typos and minor comments: p. 4 "is a dynamic process that change across learning" -> "changes" p. 4 "For instance, Hall & Pearce (1982)" -> replace & by "and" p. 5 "...but moved to the opposite quadrant during the second phase, and then producing interference." -> "... but moved to the opposite quadrant during the second phase, therefore producing interference. p. 5 "For the group extinction..." -> "For the extinction group..." p. 6 "...the rate of learning about the new cue (P+) is the same in both group and then experiencing... " -> "... in both groups..." p. 6 "Hence, interferences treatments" -> "interference treatments" p. 7 "We decided to conduct a patterning discrimination since it involved a complex discrimination..." -> careful with the verbal tense in the rest of this sentence. p. 21 "visual inspection of the figure 5" -> "visual inspection of Figure 5" p. 24 "This asymmetric might be not entirely surprising" -> "Asymmetry" p. 25 "Although these accounts may serve as speculative at the moment" -> "may be speculative" p. 26 "In summary, there has been a large set of research" -> "a large body of research" p. 27 "Whatever the value of this later speculation is..." -> remove the final "is". #Thank you very much for all these corrections. We have corrected all of them and we have corrected other typos detected in the manuscript. Reviewer 1 Alcalá et al.’s study assessed the facilitatory effect of extinction learning on subsequent learning of new cue-outcome pairings. The authors argue that extinction learning increases uncertainty and prediction error, which should lead to non-specific heightened attention, and, hence, to enhanced learning. In Experiment 1, the effect of extinction learning was evaluated through participants’ learning performance in a positive patterning (Experiment 1a), and a negative patterning (Experiment 1b) discrimination trained in a subsequent learning phase. Experiment 2 was identical except for the fact that participants were trained concurrently in both patterning discriminations. No effect of extinction training on subsequent learning was found in Experiment 1. In experiment 2, an advantage of extinction training over the control condition was found in the positive patterning but not in the negative patterning discrimination. However, no significant effect involving the type of patterning discrimination was found in the overall ANOVA. The authors claim that the results are consistent with the idea that the facilitatory effect of extinction training is conditional upon the level of difficulty of the new learning task. Such effect may be precluded if the learning task is too easy or too difficult. In general, the objectives of the study are relevant to improve our understanding of the learning processes at stake in cases in which interference makes a difference in new learning. The justification of the hypothesis and the objectives is reasonably well elaborated and linked to previous studies and theories of learning, specifically those focused on the role of attention and the effects of interference on subsequent and concurrent learning. The methodology is correct and rigorous enough to provide useful information to make progress in the field. However, I have found some difficulties and concerns that should be addressed. 1. Writing and clarity. The manuscript needs editing for clarity in a good number of places. Although I am not a native English speaker, I detected a lot of mistakes, some of which have rendered the text hardly understandable in a few places. #We have revised the manuscript in terms of readability, corrected phrases and typos. The introduction. 2. The main justification for the study is built on the idea that interference or extinction learning should enhance attention through the increase in uncertainty or in prediction error. According to this, any sound experimental strategy to test the facilitatory effect of extinction on subsequent learning should avoid a) reaching a complete reduction of the prediction error in the experimental group before the new learning task, and b) using a new learning task involving high levels of uncertainty. But the design used included a number of extinction trials in the extinction group that may suffice to (almost) totally reduce the prediction error before facing the patterning discrimination phase. At the same time, all cues and outcomes in the patterning discrimination phase were completely new, which may have increased uncertainty to its maximum level even for the control group. With these considerations in mind, the reader may find some gap between the theoretical framework and the design of the experiments. I think that the manuscript would improve if the authors could find a way to better justify the experimental strategy used. #We agree that both issues are confounded in our design, and both of them can have a relevant role in the final outcome. However, our main goal in this experimental series was to test the faciliatory effect in the context of a complex discrimination, that is, positive and negative patterning. Hence, these two factors are necessary for the goals of this study. Having said that, we now discuss the potential implication of novel cues in the GD, and we have included a paragraph discussing the implication of the number of trials in the GD. We hope this discussion covers to some extent the implication stemming from these two issues and encourage future research to specifically address them. Experiment 1a. Method. 3. In the Participants section (p. 8), the sentence “They reported sensitivity using a n= 32 to effects related with the increase on attentional processing using extinction protocol similar to our protocol.” was included to justify the sample size. I would appreciate if the authors could also include further details such as statistical power and effect size detectable. #Thank you for the suggestion. We have included a sensitive analysis in the paper. 4. The Apparatus section should be titled “Apparatus and stimuli”. #Corrected. Discussion section of Experiment 1. 5. In p. 17, first para., the absence of any facilitatory effect of extinction on subsequent learning of the patterning discrimination is tentatively explained by invoking the use of the “opposites rule”: Two cues presented in isolation precede the opposite outcome to that following the presentation of both cues in compound. But I think that when participants have to solve only one of the patterning discriminations (either the positive or the negative one), the rule may even be much simpler. For example, in the positive patterning discrimination, single cues are never followed by the outcome, whereas compound cues are always followed by the outcome. An advantage of this rule is that the discrimination can be solved without paying attention to the specific cues involved in each training trial. This possibility should be acknowledged in the manuscript. #Thank you for bring up this possibility. We now discussed this possibility in the discussion section of the first experiment (pages 17-18). Experiment 2. 6. One of the motivations for Experiment 2 (see p. 18, last line) was that “the use of both patterning at the same time should preclude the use of the ‘opposite rule’ to solve the discrimination and then the discrimination should be solved by particular set of association between cue(s)-outcome”. This argument should be removed. In fact, the use of the opposites rule is enforced in Experiment 2 to a greater extent compared with Experiment 1. For example, the application of the opposites rule would lead to correctly predict the absence of the outcome from the presentation of Cues M & N in compound provided that one remembers that the same cues presented alone were followed by the outcome (M+, N+). Likewise, the absence of the outcome can be correctly predicted from the presentation of single Cues L and H if one remembers that the presentation of the same cues in compound were followed by the outcome in previous trials. Rather than the opposites rule, the design of Experiment 2 precludes the use of the simpler rule mentioned in my previous comment, which allows the learner to solve the task without storing specific cueoutcome pairing associations. I think that this last idea could work as a strong motivation for Experiment 2. #Thank you for this comment. We have modified the rationale of Experiment 2 in light of this possibility. 7. The Results section, p. 21, last two paras., includes the results of separate group x type of cue ANOVAs within each patterning discrimination condition despite the absence of any significant effect involving patterning. These are considered planned analyses, but the justification offered by the authors is based on a visual inspection of Figure 5. Unless a strong theoretical motivation were provided to expect a significant facilitatory effect of the extinction treatment within the positive but not within the negative patterning condition, the analyses should be considered posthoc, and a correction method should be used to protect the tests performed from an increase in Type-I error. #As mentioned in the introduction, there are previous studies where differences between positive and negative patterning have been reported. Therefore, we think that this difference is enough to justify the analyses in each type of patterning. We have raised this point in the introduction (page 7) and in the results section (page 23). Justifying these analyses by theoretical reasons rather than because the visual inspection of the figure. 8. The authors should avoid making overinterpretations of the results, and should tone down some conclusions in the Discussion section (ps. 22 and 23). One example is the statement “Results reported in Figure 5 are in agreement with the idea that negative patterning may be harder to acquire both in nonhuman (Bellingham et al., 1985; Rescorla, 1972) and human animals ...”. If any, an inspection of Figure 5 suggests just the opposite. Specifically, the control group seems to have acquired the discrimination faster within the negative compared with the positive patterning condition. Another example is the sentence “Whatever the value of this later speculation is, results of Experiment 2 suggest that experiencing extinction facilitates new learning”. In my view, the evidence supporting this conclusion is rather weak, as it is based on a post-hoc analysis within the positive patterning condition after an omnibus ANOVA yielding no significant effects involving the group factor. #In line with this and others comments raised, we have qualified our conclusions. General Discussion. 9. In p. 24, the idea of the negative patterning discrimination being more difficult than the positive patterning discrimination is again invoked to explain why the facilitatory effect of extinction was found within the latter but not within the former condition. I have two concerns in line with my previous comment. First, the lack of a significant interaction effect involving the group factor in the omnibus ANOVA means that the facilitatory effect of group within one patterning condition was not statistically different from the same effect within the other patterning condition. Consequently, it seems that the authors are trying to explain a difference that fell short of statistical significance. Second, any difference in the difficulty of the patterning discriminations should have yielded a significant patterning x type of CS interaction effect, especially in the control group. Specifically, differential expectancy ratings (outcome trials minus no-outcome trials) should be higher in the positive than in the negative patterning. But an inspection of Figure 5 shows a trend towards the opposite direction. #Thank for bring this issue. Indeed, it is true that given the lack of interaction is difficult to sustain differences according to the type of patterning. We have acknowledged this in the GD. 10. In previous studies conducted with humans in the same lab, the facilitatory effect of extinction was detected by using a test phase that included a pre-exposed cue that had been previously subjected to a (latent-inhibition like) treatment that is supposed to reduce attention to the cue. In p. 25 (second para.) the authors discuss the possibility that the use of a patterning discrimination test based on new cues, instead of pre-exposed cues, may have greatly heightened attention, which, in turn, may have rendered the facilitatory effect of extinction learning undetectable. I wonder whether this interesting reflection could be used as an argument to justify the objectives of the study in the introduction. Instead of justifying the study as an attempt to provide evidence of a facilitatory effect of extinction on new learning through heightened attention, such study could be seen as a way of ensuring that the previous results found in their laboratory are not replicable when the test phase includes new cues and a difficult discrimination learning that may be enough to heighten facilitatory attention. I realise that this would entail expecting null statistical results. But this could be overcome by reporting Bayesian analysis testing for null effects. Such null results would lend some additional support to the attentional mechanism postulated to explain their previous results. #Thank you for this suggestion. We have included a bit of this rationale in the introduction. However, the main goal of this study was framed from a different angle and for that reason we did not want to emphasize this rationale during the introduction. However, we agree that this may help to understand the discussion, and consequently we introduced one paragraph in page 6. Reviewer 2 (Tom Beesley) The manuscript presents an examination of how the acquisition and subsequent extinction for one cue affects the learning of subsequent discriminations with entirely different cues. Experiment 1 a and 1b looked at the effect of this training (acquisitionextinction) on positive and negative patterning discriminations, respectively. It was found that the extinction of the cue did not promote differences in learning compared to a control group that received no extinction training. However, when participants were faced with simultaneous positive and negative discriminations in Experiment 2, a difference was observed: those participants who had experienced extinction previously learnt about the positive patterning discrimination more readily (no effect was observed on the negative patterning element of the design). I am generally positive about this paper. It addresses an interesting topic in the field (the consequence of cue ambiguity/uncertainty on subsequent learning and attentional strategies) and the experiments seem well designed. The data presentation is clear and the results straightforward. I think this could make a nice addition to the literature and had just a few suggestions that may be helpful in revising the article. Pg 7. When setting up the rationale for the study, you state “...we believe that patterning may create a better scenario to evaluate the potential facilitation effect of previous extinction, since engaged learn differential associations.” There is an error here in the wording at the end of the sentence. This makes it very difficult to understand the logic of the design at this crucial stage. #Thank you for this comment. We have updated the sentence. Exp 1: It would be good to note the effect sizes you were hoping to obtain, and/or the size of effect observed in previous studies. #We have included a sensitive analysis. Exp 1a and 1b: I wondered whether a Bayesian analysis might be informative here to provide some information on the support you have for the null results. #We have included BF in the analyses. In general terms the BF is quite convincing for the lack of effect of Group factor, either as main factor or in interaction with other factors. All experiments: The summarised data has been uploaded and made available as a .xlsx file. I would recommend that the raw data is also uploaded, ideally in a nonproprietary format such as csv. If there are analysis scripts that have been used (i.e., that convert raw data to summarised data) it would be good to make these available. Pg 11. I wasn’t quite sure what “analyse the files” meant here. Is this a translation error? #We referred to “file” as an expedient in which appeared the name of a person who ate a particular food. We have changed that word by “record” to avoid missunderstandings. Pg 22. You state that the data are consistent with the previous finding that negative patterning discrimination is slower to develop than that of positive patterning. However, your main analysis (pg 21) does not provide the inferential analysis of this (patterning by type of cue interaction?). Please provide this analysis to verify this conclusion. #Thank you for raised this issue. Indeed, data did not provide support for this statement, since the aforementioned interaction was not significant. If anything, as pointed out by R1, negative patterning seems easier to acquire than positive patterning and then this interpretation is difficult to sustain. Hence, we have changed our conclusion in this regard, appealing to previous differences between positive and negative patterning, but making clear that this interpretation should be taken with caution given the lack of statistical support. General discussion: I felt like I came away without a very clear understanding of the mechanisms you were proposing for the effect here. Why do you think extinction leads to the improvements in subsequent discrimination? Is this something to do with a general enhancement of attention? Or is this something about the more abstract knowledge they obtain about how cues operate in the task (e.g., cues can be associated with both the presence and absence of the outcomes). In general, I feel that a more focussed discussion is required that will allow for a great impact on theoretical and experimental development. # Given the lack of strong evidence for the effect of extinction, we think that it is difficult to speak about a specific mechanism, and then, we framed the discussion to stimulate future research rather than to focus on a specific explanatory mechanism.