Preregistration: Freeing capacity in WM through the use of LTM representations

• Main Author: Bartsch, Lea
• Format: Text
• Language: unknown
• Published: Open Science Framework 2021
• Subjects: Cognitive Psychology; Social and Behavioral Sciences; Psychology; FOS Psychology; Akureyri; Levy; Rubin; The Gate; University of Akureyri
• Online Access: https://dx.doi.org/10.17605/osf.io/cya64; https://osf.io/cya64/

Preregistration: Freeing capacity in WM through the use of LTM representations Experiment 3 Lea Bartsch1 & Peter Shepherdson2 1: University of Zurich, Cognitive Psychology 2: University of Akureyri Background and Research Question In the first two experiments of this preregistered study (see OSF) we aimed to investigate whether the presence and use of LTM representations frees capacity for maintaining additional information in WM. In the first phase of our first experiment, we presented participants with word pairs for them to encode into LTM (LTM learning phase). Subsequently, they completed trials of a WM task, also involving word pairs. Crucially, the pairs presented in each WM trial consisted of varying numbers of new pairs (LTM unavailable) and the previously learned LTM pairs. The results of this first experiment provided evidence that cued recall performance in the WM task was unaffected when memory set size increased through the addition of LTM-available pairs but deteriorated when set size increased through adding LTM-unavailable (new) pairs. In the second Experiment we not only varied WM loads across two levels (2, and 4 pairs), but also varied LTM loads across three levels of LTM load (0, 2, and 4 pairs). Recall performance deteriorated with LTM load for low WM load (2 pairs) but remained superior to performance with the same set size of only LTM-unavailable (new) pairs. In contrast, LTM load did not affect recall performance at higher WM load (4 pairs). This added to the evidence of the first experiment suggesting that individuals can outsource workload to LTM to optimize performance, but also speaks towards a WM system with a flexible gate to LTM that can be opened or closed depending on the current cognitive need. In order to investigate the flexibility of this gate, and to provide evidence concerning whether subjects always rely on LTM representations when they are available, in Experiment 3 we intend to introduce proactive interference to the WM test phase. Proactive interference (PI) occurs when previously acquired knowledge disrupts the acquisition of new information. In essence, “old” information prevents the recall of “new” information. Episodic LTM is prone to proactive interference, so if subjects solely and inflexibly draw on the available episodic LTM representations in our paradigm, this should lead to detrimental effects in conditions with PI, but performance enhancements in conditions utilising LTM but without PI should persist. By contrast, if subjects rely on the LTM representations only when the information in LTM is likely to be helpful, no PI effects should be evident. However, this may also reduce or abolish the LTM facilitation effects we previously found, depending on participants’ ability to flexibly recognise and make use of potentially-helpful LTM representations. A similar approach has been taken for visual memory (Oberauer, Awh, & Sutterer, 2017): In an initial learning phase, participants acquired LTM for 120 object-color associations. In a subsequent WM test, they had to remember three object-color conjunctions in each trial. Of the three objects, two had been associated with a color in the learning phase. When an old object's learned color matched the one presented in the WM test, WM performance was facilitated compared to WM for the color of a new object (proactive facilitation). However, when the learned color mismatched the one in the WM test, there was no hint of proactive interference. This pattern of effects is predicted from the assumption that the contribution of LTM to WM is controlled by a flexible mechanism that draws on LTM if and only if the information in LTM is likely to be helpful rather than harmful---for instance, if no WM representation is accessible for the probed item, such that the alternative to relying on an LTM representation is relying on no memory at all. Though our previous experiments have gone beyond this in showing that the availability of LTM representations enhances memory for novel stimuli (i.e., the words presented for the first time in the WM phase), it is unclear what effect a situation in which LTM representations are not reliably helpful will have on participants’ tendency to outsource memory load to LTM. Thus, here we want to extend the findings of Experiment 2, which hinted at a flexible gate in WM, which opens to LTM in case representations of memoranda were available. Specifically, we aim to investigate how flexible this gate is, whether WM can draw on LTM only if it is helpful, or whether the LTM available pairs will be subject to PI. Methods Participants We will first collect data of 30 subjects who did not participate in Experiment 1 or 2, online. We will then run a first analysis and check the evidence (indexed by the Bayes Factor, BF) for differences between our conditions of interest. Our goal is to report BF ≥ 3 for or against interaction effects of LTM load with the PI condition, and all main effects (WMload, LTMload, and PI). If we do not reach the targeted BF after the initial data collection, we will add bouts of 10 participants and re-run the analyses. We will stop data collection once the BF ≥ 3, or once we have collected acceptable data from 100 participants. Participants will be replaced (i.e., their data considered unacceptable) if: (a) their overall response accuracy across all conditions is greater than 2 standard deviations below the overall mean; (b) they do not complete all experimental conditions; or (c) if they do not comply with the instructions. In order to assess the latter, data from participants who spend less than 20 seconds reading the instructions will be excluded from the analysis. We will also exclude data from any trials with response times < 500 ms, as participants are unlikely to be able to read the cue word and identify the appropriate response in such a short duration. Only participants whose mother tongue is German, aged between 18-35 years, will take part in the experiment. Participants will have to indicate consent prior to the study and will be debriefed at the end. The experimental protocol is in accordance with the Institutional Review Board of the Psychology Institute from the University of Zurich. Materials and Procedure Again, the study consists of 2 phases: a LTM learning phase in which subjects are presented with 24 pairs of concrete words, and a subsequent WM task phase in which either 2, 4, or 6 word-pairs will be presented sequentially, for 4 seconds. The WM task in the present study will be an immediate memory test in which participants remember arbitrary word pairs (e.g., dog–tooth, tree–bottle) and are tested with a four-alternative forced choice procedure. One of the words will be presented as a memory cue, and participants can choose one out of four response options: the target (previously paired with the cue), another item which was paired with a different word in the same trial (within trial intrusion probe), a LTM lure (in case of PI condition: the word which was originally paired with the cue, in case of no PI: a word from another LTM pair) or a new item. The WM trials either consist of new pairs only (LTM load 0, WM load 2, or 4), or comprise new pairs of their respective WM load plus the two LTM pairs (LTM load 2), adding up to the respective set sizes 2, 4, 6. Further, in 2/3 of the trials with LTM of 2, interference is introduced in two different ways (see Figure 1): either the LTM pairs will be rearranged, (word 1 and 2 are from two different LTM pairs), or a word from an LTM pair is bound to a new item (new associate). Investigations of PI have traditionally used a paired-associate learning procedure, in which interference is created by holding cues constant, with the responses being changed between two lists (A–B, A–D). Performance in this interference condition is compared with that in a control condition for which both cues and responses are changed between lists (A–B, C–D). We follow this implementation of PI and add the implementation of a new associate to investigate the influence if only the binding and the cue is stored in LTM (compared to when both words and the binding are in LTM). The detailed design can be seen in Figure 1. In each session, stimuli will consist of 320 words randomly drawn from a set of 878 concrete German nouns. Planned analyses WM task performance Bayesian generalized linear mixed models. The data of the WM task will be analysed using Bayesian generalized linear mixed models (BGLMM) implemented in the R package brms (Bürkner, 2017, 2018). The dependent variable will be the correctness of each response in each condition per participant. Correct responses are defined as recalling the target item compared to choosing the list lure, LTM lure, or new item. Therefore, we assume a Bernoulli data distribution predicted by a linear model through a logit link function (i.e., a repeated-measures logistic regression). The fixed effects are WM load (2 vs. 4), and condition (LTM load 0 vs. LTM load 2/ intact vs. LTM load 2/PI) as well as their interaction. Following the recommendation of Barr and colleagues (Barr, Levy, Scheepers, & Tily, 2013; see also Schielzeth & Forstmeier, 2009) we will implement the maximal random-effects structure justified by the design; by-participant random-intercept and by-participant random-slope for WM load and condition. In addition, we will estimate the correlation among the random-effects parameters. We will decompose the main effect of condition into two pairwise comparisons of the LTM load 2 conditions to the baseline (LTM load 0), in order to estimate evidence for a facilitatory effect of intact pairs and evidence for an interference effect of changed pairs: Evidence for a facilitatory effect would become evident in an advantage of the LTM load 2/ intact condition to the baseline. In contrast, performance in the LTM load 2/PI conditions should be worse than the baseline, should proactive interference affect performance.. The regression coefficients will be given moderately informative Cauchy priors with scales between 0.3 and 2. These scales will be used because they define a default prior analogous to that proposed by Rouder et al. (2012) for the General Linear Model. Specifically, this prior assigns its probability mass approximately equally over effect sizes on the predictor scale that translate into effects between -0.5 and 0.5 on the p(correct) scale when starting from p(correct)=0.5 as baseline. We will use completely non-informative priors for the correlation matrices, so-called LKJ priors with shape parameter 1. We will calculate Bayes Factors to estimate the strength of evidence for the null or the alternative hypothesis. For instance, with the BF we can calculate the evidence for the effect of WM load (BF10) by comparing the evidence for a model including this factor against an intercept-only model that serves as the null model. Additionally, we can calculate evidence against a difference between the WM loads (BF01), where BF01 = (1/BF10). A BF10 larger than 1 gives evidence for an effect, a BF10 lower than 1 yields evidence against an effect and hence evidence for the null hypothesis. A BF10 of 10 indicates that the data are 10 times more likely under the alternative hypothesis than under the null hypothesis. Usually, BFs > 3 are regarded as providing substantial evidence for one hypothesis over the other. We will use an MCMC algorithm (implemented in Stan; Carpenter et al., 2017) that estimates the posteriors by sampling parameter values proportional to the product of prior and likelihood. These samples are generated through 4 independent Markov chains, with 1000 warmup samples each, followed by 50000 samples drawn from the posterior distribution which are retained for analysis. Following Gelman and colleagues (2013), we will confirm that the 4 chains converge to the same posterior distribution by verifying that the rhat statistic – reflecting the ratio of between-chain variance to within-chain variance – is < 1.05 for all parameters, and we will visually inspect the chains for convergence. Finally, we will use the bayes_factor function in the brms package, which implements the bridge sampler (Gronau, Singmann, & Wagenmakers, 2017), for computing the BFs. As a next step of the analyses, we will focus on performance for new pairs only and investigate whether performance in remembering those new pairs varies between trials which included intact or rearranged/new associated LTM pairs and trials that did not. The fixed effects are again WM load (i.e., 0, 2 or 4) by condition condition (LTM load 0 vs. LTM load 2/ intact vs. LTM load 2/PI). This analysis will help us distinguish between possible explanations for a main effect of condition in the overall analysis. If a hypothetical performance advantage in the LTM load 2/ intact trials in that analysis results from WM workload being outsourced to LTM, we would expect to find a similar effect here. On the other hand, if an advantage results from generally better memory for LTM pairs (vs. new pairs), no effect should be present in this analysis. Furthermore, in order to investigate whether proactive interference effects differ between the two PI manipulations (rearranged vs. new associated pairs), we will compute planned pairwise comparisons between the two conditions for both all pairs as well as new pairs only. This analysis will help us determine whether the way we manipulate PI (if only the binding and the cue is stored in LTM compared to when both words and the binding are in LTM) determines the size of the effect. Predictions Hypotheses We will evaluate the effect of WM load (i.e., 2, or 4) by condition (LTM load 0 vs. LTM load 2/ intact vs. LTM load 2/PI) on recall performance of word pairs in a WM task. We aim to assess the likelihood ratio (i.e., BF) of two hypotheses: H1 The gate between WM and LTM is always open: recall performance in the WM task is facilitated by the presence of intact LTM pairs. Further, performance is prone to proactive interference from LTM in case pairs are rearranged or associated with new words. H2 The gate between WM and LTM flexibly opens and closes, depending on the helpfulness of LTM representations: recall performance in the WM task is facilitated by the presence of intact LTM pairs, and unaffected by proactive interference. References Barr, D. J., Levy, R., Scheepers, C., & Tily, H. J. (2013). Random effects structure for confirmatory hypothesis testing: Keep it maximal. Journal of Memory and Language, 68(3), 255–278. https://doi.org/10.1016/j.jml.2012.11.001 Bürkner, P.-C. (2017). brms: An R package for Bayesian multilevel models using Stan. Journal of Statistical Software, 80(1), 1–28. Bürkner, P.-C. (2018). Advanced Bayesian multilevel modeling with the R package brms. R Journal, 10(1), 395–411. https://doi.org/10.32614/rj-2018-017 Carpenter, B., Gelman, A., Hoffman, M. D., Lee, D., Goodrich, B., Betancourt, M., … Riddell, A. (2017). Stan : A Probabilistic Programming Language. Journal of Statistical Software, 76(1). https://doi.org/10.18637/jss.v076.i01 Gelman, A., Carlin, J. B., Stern, H. S., Dunson, D. B., Vehtari, A., & Rubin, D. B. (2013). Bayesian data analysis, 3rd edition. Chapman & Hall/CRC. Gronau, Q. F., Singmann, H., & Wagenmakers, E.-J. (2017). Bridgesampling: An R package for estimating normalizing constants. ArXiv Preprint ArXiv:1710.08162. Oberauer, K., Awh, E., & Sutterer, D. W. (2017). The role of long-term memory in a test of visual working memory: Proactive facilitation but no proactive interference. Journal of Experimental Psychology: Learning Memory and Cognition, 43(1), 1–22. https://doi.org/10.1037/xlm0000302 Schielzeth, H., & Forstmeier, W. (2009). Conclusions beyond support: Overconfident estimates in mixed models. Behavioral Ecology, 20(2), 416–420. https://doi.org/10.1093/beheco/arn145