False Memories for Ending of Events

Memories are not perfect recordings of the past and can be subject to systematic biases. Memory distortions are often caused by our experience of what typically happens in a given situation. However, it is unclear whether memory for events is biased by the knowledge that events usually have a predictable structure (a beginning, middle, and an end). Using video clips of everyday situations, we tested how interrupting events at unexpected time points affects memory of how those events ended. In four free recall experiments (1, 2, 4, and 5), we found that interrupting clips just before a salient piece of action was completed, resulted in the false recall of details about how the clip might have ended. We refer to this as “event extension.” On the other hand, interrupting clips just after one scene had ended and a new scene started, resulted in omissions of details about the true ending of the clip (Experiments 4 and 5). We found that these effects were present, albeit attenuated, when testing memory shortly after watching the video clips compared to a week later (Experiments 5a and 5b). The event extension effect was not present when memory was tested with a recognition paradigm (Experiment 3). Overall, we conclude that when people watch videos that violate their expectations of typical event structure, they show a bias to later recall the videos as if they had ended at a predictable event boundary, exhibiting event extension or the omission of details depending on where the original video was interrupted.

In experiment 1, during the encoding session, we collected measures on predictability of the videos.Specifically, we asked participants to rate how confidently they felt they could predict what would happen next in the video (after it finished).In a post-hoc analysis we examined whether our predictability measure predicted probability of making an extension error after accounting for the condition difference between the Incomplete and Complete videos.Therefore, we fitted a mixed effects logistic regression model as (Extension ~ Condition + SeenBefore + Predictability*Condition + (Condition| Subject) + (1|Video)).We used brms package to estimate the model parameters using Bayesian inference, which also allowed us to use Bayes Factors to compare between models with and without including Predictability.
For both models we used a Bernoulli distribution with a logit link function resulting in standard logistic mixed effect model.Models were fitted with uninformative (flat) priors on the fixed slope effects and default student-t distribution priors for the variance components as implemented in brms package.We ran four chains each with 10000 iterations to estimate the models using 1000 warm-up samples and examined chain convergence.Using the bayes_factor function in brms we compared the model including predictability and condition interaction to a model that did not include predictability as a regressor.We found evidence in support of the null model that BF01 = 9.42, suggesting that our predictability measure does not explain variance in the extension errors after accounting for the condition factor.

Descriptive relationships between variables
For completeness, we report association between vividness, confidence, and predictability in the separate experiments.Note these are post-hoc analyses, unrelated to our main hypotheses.Where we report correlations, they were performed by taking into consideration the multilevel structure of the data.We used the 'correlation' package recently developed by Makowski and colleagues (2020).

Experiment 1
First, we examined the pairwise correlations between vividness, confidence and predictability.This was done for all trials.We found that vividness and confidence were highly correlated.Furthermore, vividness and confidence were also correlated to predictability to a smaller extend.Focusing on the remembered trials that entered the final model, we observed high correlation between vividness and confidence.Predictability was not strongly correlated with either vividness or confidence.Although we note that the correlations were significant.Given the high correlation between vividness and confidence we computed a combined measure.This was done simply by averaging for each trial the vividness and confidence.This was referred to as Subjective_Memory score.We then computed a logistic mixed effect model predicting extension errors from the condition, Subjective_Memory score and their interaction (Extension ~ Condition + SeenBefore + Subjective_Memory + Condition*Subjective_Memory + (Condition | Subject) + (1 | Video)).
First, we examined the correlation between vividness and confidence.Note in this experiment we did not have predictability ratings.We found that vividness and confidence were highly correlated (r = 0.82; 95% CI [0.80 0.83]; t1423 = 53.43,p < 0.001).Therefore, we again combined them into a single measure of Subjective_Memory, by averaging them together.

Experiment 3
For experiment 3 we examined correlations between the continuous memory error measure (referred to as Memory_Error in the main manuscript) and vividness, confidence, and predictability.We note that in this experiment, participants provided vividness and confidence ratings after each recognition trial.In other experiments, subjective measures were collected before the recall trials.
We observed a strong positive correlation between vividness and confidence.Memory_Error was also correlated with confidence and vividness, but not with predictability.See Supplementary Table 5.Given the high correlation between vividness and confidence we again combined them into a single Subjective_Memory measure.We did not further explore predictability since it was not correlated with Memory_Error and showed weak correlations with the other measures.

Experiment 1 :
Predicted extension errors for different predictability and condition levels.Plot shows predicted values of extension errors of fitted model (Extension ~ Condition + Seen_Before + Predictability*Condition + (Condition | Subject) + (1|VideoID)).Line represents average effects and shaded area represents confidence intervals.Supplementary Figure3.Experiment 1: Estimated Posterior distributions for the condition and predictability effects.Dark shaded area shows median of the posterior distribution, the grey shaded area shows the 95% Highest Density Interval.Distributions centered at 0 suggest no effect of the predictor on extension memory errors.Note x-axis is shows logOdds.