No evidence of consolidation of evaluative conditioning during waking rest and sleep

* The data and analysis scripts of the reported studies are available on osf.io at https://osf.io/ua398.

ABSTRACT

Research on evaluative conditioning (EC) shows that attitudes can emerge from co-occurrences of stimuli, and accumulating evidence suggests that EC usually depends on memory for these stimulus contingencies. Therefore, processes known to aid memory retention may be relevant for the development of stable attitudes. One such process may be memory consolidation, assumed to be promoted by waking rest and sleep. In two pre-registered experiments, we investigated whether waking rest (vs. cognitive activity, Experiment 1) and sleep (vs. wakefulness, Experiment 2) in between conditioning and measurement of EC, consolidate contingency memory and EC. Contrary to our predictions, waking rest (vs. cognitive activity) promoted neither contingency memory nor EC effects. Sleep (vs. wakefulness) decreased forgetting of contingency memory but crucially, it did not attenuate the impact of counterconditioning on contingency memory. Sleep also did not influence EC effects, nor the reduction of EC by counterconditioning. EC effects in both experiments were predicted by contingency memory. Yet, unexpectedly, EC effects occurred in the absence of contingency memory after waking rest, but neither after sleep nor in the active control conditions. Our findings emphasise a role of contingency memory in EC, but it remains unclear whether this role changes during waking rest.

KEYWORDS:

• Evaluative conditioning
• contingency memory
• consolidation
• waking rest
• sleep

Acknowledgements

We would like to thank Wolfgang Walther for his helpful work in programming the procedure of Experiment 2. Extensive linear mixed model analyses were conducted on the CHEOPS cluster of the RRZK, University of Cologne.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

* The data and analysis scripts of the reported studies are available on osf.io at https://osf.io/ua398.

1 Data were analysed in R (R Core Team, 2019) with the packages ez (Version 3.6.1; Lawrence, 2016) for ANOVAs and BayesFactor (Version 0.0.12–4.2; Morey & Rouder, 2018) for Bayesian analyses. Reported Bayes factors are based on a Cauchy prior centred on zero, r scale = 0.707. Linear mixed model analyses were conducted with lme4 (Bates et al., 2015) and lmerTest (Version 3.1-0; Kuznetsova et al., 2017) and included all fixed and random effects “justified by the [experimental] design” (p. 255; Barr et al., 2013). Generally, predictors were effect-coded (−1, 1), and dummy-coded (0, 1) for simple effects analyses. Effect sizes Cohen’s d were calculated according to Lakens (2013) with the accompanying spreadsheet.

2 We planned to exclude participants from the rest group who fell asleep. This was not noted for any participant but we missed to assess this for seven participants. Since excluding these participants led to very similar results, we decided to retain them.

3 We slightly oversampled to account for drop out and incomplete data.

4 After data collection, we discovered deviations in the duration of stimulus displays for some participants. We suspect that these occurred when additional browser tabs were open during the experiment. Excluding participants for whom this resulted in substantial deviations (>1 s) from the intended display duration (n = 15) did not have any effect on the results. We therefore retained their data in the analyses.

5 Due to a variable publication delay on Prolific Academic, we published Session 2 15–30 min prior to the beginning of the earliest possible time window (8:30 a.m./p.m.).

6 During data collection, we discovered that a programming error led to false feedback and thus increased repetition of pairs in the following cases: Inputs in British English (for the words humour and tumour), inputs including spaces or upper-case letters. We fixed these programming errors and resampled 119 participants to replace participants who were affected by this issue (see second pre-registration https://osf.io/wrm8j).

7 Recalled USs were only considered correct if the exact US word was inputted. We therefore additionally analysed recall performance allowing for spelling mistakes. Results were very similar: Interval condition: F(1, 196) = 24.575, p < .001, ${\eta }_p^2$  = .111, BF₁₀ = 9263.801 (b = 0.487, SE = 0.099, z = 4.922, p < .001; M_sleep = 61.5%, SD_sleep = 26.9%; M_wake = 43.5%, SD_wake = 29.0%); Counterconditioning: F(1, 196) = 22.042, p < .001, ${\eta }_p^2$ = .101, BF10 = 2555.597 (b = 0.199, SE = 0.041, z = 4.856, p < .001; M_{counterconditioned} = 48.6%, SD_{counterconditioned} = 29.8%; M_{non-counterconditioned}= 56.0%, SD_{non-counterconditioned}= 28.5%); Interval condition × Counterconditioning: F(1, 196) = 1.994, p = .160, ${\eta }_p^2$ = .010, BF₁₀ = 0.392 (b = −0.061, SE = 0.041, z = −1.490, p = .136).

Additional information

Funding

This work was supported by the Deutsche Forschungsgemeinschaft under grant GA 1520/2-1 awarded to Anne Gast.