![Sample our Behavioral Sciences journals, sign in here to start your access, latest two full volumes FREE to you for 14 days]({"type":"image" , "src" : "/sda/110801/TOC-Subject-Sample-2021-Behavioral-Sciences.jpg"})
ABSTRACT
Investigating the recall process of autobiographical memories (AMs) and, particularly, the order in which AMs are recalled has the potential to shed light on the organisation of autobiographical memory. However, research on order effects in the recall of AMs is relatively rare. Moreover, to date, no study addressed the question of where emotion fits into the organisation. The present study aimed to close this gap by examining whether emotional valence serves as one organising principle. Data come from 117 older adults (M = 74.11; SD = 7.06) who reported up to 39 AMs. The use of a multivariate multilevel model with autoregressive effects allows us to analyse the order effect within one person, as well as how the order effect differs between persons. The results replicated a temporal first-order effect that has been shown in previous studies and moreover, demonstrated a temporal second-order effect. Furthermore, our results indicated an emotional first-order effect that was even stronger than the temporal first-order effect and an emotional second-order effect. In addition, both first-order effects differed reliably between persons. Thus, the present study emphasises the need for considering emotion in current theoretical formulations of autobiographical memory and also of considering individual differences in the order of AMs recalled.
Disclosure statement
No potential conflict of interest was reported by the author(s).
Notes
1 Unfortunately, no information whether lifetime period associated memories were always thematically related was given in the study.
2 Note, however, that there is no clear criterion of how fast the recall of an AM should be in order to guarantee that it is located at the lowest event-specific knowledge level.
3 Effect size calculated based on the t-statistics reported in the article.
4 Originally, formal education was measured using the highest graduation. The German “Hauptschulabschluss”, “Realschulabschluss”, “Fachhochschulreife”, and “Abitur” graduations correspond to 9, 10, 12, and 13 years of formal education, which we used here to calculate years of education.
5 Because the estimation of an autoregressive model becomes less precise if the number of measurements per person is small (Jongerling et al., Citation2015), we decided to use a cue word paradigm. This grounded on the assumption that presenting cue words would result in a greater number of retrieved AMs per participant.
6 A methodologically more sound approach would not use the individual mean, but the random intercept. It is well-known that the random intercept is a better estimate (in the sense of a smaller mean square error) than the individual mean (e.g., Bryk & Raudenbush, Citation1992). However, subtracting the random intercept either requires a two-step procedure or more complex models. Moreover, if the number of outcome measurements is large and / or the random intercept variance is comparatively small (or, equivalently, if the ICC is low), the difference between the simple and the more sound approach is typically small. To illustrate, the Lag-1 emotional valence effect was 0.1025 (SE 0.0204) in Model 4, while it would be 0.1143 (SE 0.0197) using random intercept within person centering – implying that in our data the difference is minimal.
7 The double subscript of n makes explicit that participants can have varying numbers of measurements of the characteristics of AMs.
8 AMs are categorized into positive, negative and neutral based on their emotional valence ratings on the 5-point scale (1 and 2 = negative, 3 = neutral, 4 and 5 = positive).
9 More formally, the difference in −2LL of two nested models is approximately χ²-distributed with degrees of freedom given as the difference in the number of parameters estimated. For Models 0 and 1, the difference in −2LL is 26064 − 25588 = 476. Because Model 1 involves the estimation of three additional parameters (two random intercept variances plus their correlation) compared to Model 0, the difference in degrees of freedom is 3. Of course, χ²(df = 3) = 476 is (highly) significant, showing that Model 1 described the data much better than Model 0.
10 The complete table () of correlations among random effects can be found in the web-based additional material.
11 Strictly speaking, it may, of course, be that the first-order effects we found are due to other variables not considered in this study.
12 To clarify whether a higher order autoregressive model would describe the data better, we also estimated third order and fourth order models (not shown in ). In these models, none of the Lag-3 or Lag-4 effect estimates (be it for emotional valence or MSA) did come close to statistical significance. Note that increasing the number of lags inevitably reduces the number of outcome variable measurements available for data analysis.