Can perceived changes in autobiographical memories’ emotionality be explained by memory characteristics and individual differences?

ABSTRACT

Recalling autobiographical memories (AMs) is closely tied to emotional experience. However, the emotionality associated with an event can change from occurrence to recall. Autobiographical memories show fixed affect (i.e., no change in emotionality), fading affect (i.e., decrease in emotional intensity), flourishing affect (i.e., increase in emotional intensity), and flexible affect (i.e., change of valence). The present study used mixed-effects multinomial models to predict perceived changes in positive and negative valence as well as intensity. Initial intensity, vividness, and social rehearsal were entered into the models as event level predictor variables, whereas rumination and reflection were entered into the models as participant level predictor variables. Analyses were based on 3950 AMs reported by 352 participants (18–92 years old) in response to 12 emotional cue-words. Participants rated the emotionality of each memory from the perspective of event occurrence and event recall. Only the predictors on the event level meaningfully distinguished between memories that stayed fixed in affect and memories that showed fading, flourishing, or flexible affect (R² values ranging from .24 to .65). The present results highlight the importance of considering different aspects of AMs and the ways they change emotionally to fully understand emotional experiencing in autobiographical memory.

KEYWORDS:

• Autobiographical memory
• fading affect bias
• fixed affect
• flourishing affect
• flexible affect

Acknowledgements

I would like to thank Daniel Zimprich and Justina Pociūnaitė for their valuable feedback and support and for allowing me to use their data. Additionally, I would like to thank the Office for Gender Equality Ulm University, who included me in their financial support programmes for female researchers.

Disclosure statement

No potential conflict of interest was reported by the authors.

Data availability statement

The data that support the findings of this study are available from the corresponding author, Sophie Hoehne, upon reasonable request.

Ethical statement

The present study did not meet the suggested criteria for a full application for ethical approval provided in a checklist by the ethics committee of Ulm University at the time of data collection. Participants were provided with detailed information about the study and gave their written consent.

Notes

1 To determine the necessary sample size, typically an a priori power analysis is conducted. However, both the analysis model and the data structure defy any straightforward power analysis. First, the model is multivariate, such that to determine the necessary sample size, one would have to specify the effect of each predictor variable on the outcome variable(s) and the associations among predictor variables. Second, the model is a generalized linear mixed model, which complicates power analyses even more due to the non-linear link between the data scale (probabilities) and the model scale (logits). Third, the presence of two levels of analyses would require additional assumptions concerning the amount of total variance located on the event level and the participant level. Taken together, these intricacies effectively render even a power analysis based on simulations so specific and dependent on pre-assumptions that it could very easily lead an author astray. Therefore, in estimating the necessary sample size, my aim was to have enough observations (ca. 100 AMs) even in the smallest affect change category (flexible positive affect). Based on previous studies (e.g., Gibbons & Rollins, 2016; Ritchie et al., 2009; Walker & Skowronski, 2009), I calculated with 2-3% AMs showing flexible positive affect, such that an event level sample size of approximately 4000 AMs resulted.

2 Indeed, the four affect change categories can be ordered according to direction and strength of change in AMs’ emotional intensity. An alternative approach would, thus, be to use an ordered logit model (e.g., McCullagh, 1980), which would treat the categories either in a descending or ascending order (Wolf & Zimprich, 2018). However, treating the four affect change categories as ordered would also imply that the different AM characteristics are ordered across the categories. Using an ordered logit model would imply, for example, that initial intensity of AMs should be smallest in the flourishing affect change category, still small for fixed affect, medium within fading affect, and highest within the flexible affect change category (or vice versa). In contrast, Gibbons and Rollins (2016) found the lowest initial intensity ratings for flourishing affect memories, followed by flexible, fading, and fixed affect memories. To avoid these restrictions through ordering, I chose a multinomial approach, which is able to treat the different categories as nominal (e.g., Long, 1997).

3 Transformed back to the probability one gets $p_{\mathrm{FadeP}}={\displaystyle \frac{\exp \left({\text{β}}_{0\mathrm{FadeP}}\right)}{1+\exp \left({\text{β}}_{0\mathrm{FadeP}}\right)+\exp \left({\text{β}}_{0\mathrm{FlourishP}}\right)+\exp \left({\text{β}}_{0\mathrm{FlexibleP}}\right)}}$ $={\displaystyle \frac{\exp \left(-2.025\right)}{1+\exp \left(-2.025\right)+\exp \left(-1.080\right)+\exp \left(-4.405\right)}=.089}$.

4 In generalized linear mixed models marginal and conditional models are distinguished. The observed (marginal) distribution of the outcome variable is different from the distribution of the outcome variable conditional on the random effects (Stroup, 2012). With increasing random variance, the difference increases between these two distributions. In the present study, the fixed effects estimates in Model 0 for both positive and negative valence ratings represent effects that are conditional on random effects (Zimprich & Wolf, 2018).

5 A useful calculation to interpret and compare the estimates is transforming them into an odds ratio (OR) for a standardized factor change (Long, 1997). For the effect of initial intensity on fading positive affect one gets $\mathrm{OR}=\exp (\text{β }\mathrm{Initial}\mathrm{Intensity}\mathrm{FadeP}\times \mathrm{SD}\mathrm{Initial}\mathrm{Intensity}P)=\exp (0.683\times 2.50)=5.52$ indicating that for a standard deviation change in initial intensity, the odds for this AM to show fading compared to fixed positive affect increase by a factor of 5.52.

Additional information

Funding

The author reports that there is no funding associated with the present work.