Treatment of Data and statistical analysis
Unless otherwise noted, all effects referred to as statistically significant throughout the article are associated with p-values of < .05, two-tailed. Analyses employed for the purpose of statistical inference are described alongside their respective results.
Only participants who indicated that they had taken the task seriously and who had made at least 50% valid responses during the matching task, of which at least 50% were correct, were concluded in the analysis8. No valid responses were recorded for 8,47% of all trials (3858 trials total). Response times were recorded from the offset of the shape-label pair. Responses made faster than 100ms after stimulus offset were excluded from the analyses. As an upper bound, we used the Tukey criterion for outliers, resulting in a cut-off value of 1126ms9, and the exclusion of one trial. Mean reaction times and accuracies by label, remoteness of past time period and instruction condition are shown in Table 2. For the self-report measures, subscales were recreated based on the respective subscale of the MEQ the original items belonged to. For those subscales achieving satisfactory internal consistencies (defined as Cronbach’s α < .6), mean agreement was computed. Self-report means by remoteness of past time period and instruction condition are shown in Table 2. An overview of all items, subscales and internal consistencies is provided in appendix A.
Table 1
Mean reaction times (in ms) and accuracies (in %) as a function of remoteness group (near past/ “yesterday” & remote past/ “five years ago”, referent label (current self, past self & stranger) and trial type (matching & non-matching). SD in parentheses.
|
|
N
|
Reaction Time
|
Accuracy
|
Remoteness Group
|
Referent
|
|
Matching
|
Non-matching
|
Matching
|
Non-Matching
|
Near Past
|
Present Self
|
58
60
|
538 (100)
553 (107)
|
655 (111)
669 (125)
|
.85 (.17)
.89 (.10)
|
.75 (.16)
.73 (.17)
|
Remote Past
|
Near Past
|
Past Self
|
58
60
|
619 (131)
632 (118)
|
649 (115)
672 (118)
|
.73 (.16)
.75 (.14)
|
.75 (.15)
.73 (.15)
|
Remote Past
|
Near Past
|
Stranger
|
58
60
|
588 (115)
614 (121)
|
633 (109)
650 (112)
|
.80 (.12)
.76 (.12)
|
.85 (.12)
.83 (.14)
|
Remote Past
|
Matching Performance
Matching trials. Mean reaction times and accuracies are shown in Table 1. In a first step, we tested for perceptual prioritization of either present or past selves in general. Since our main aim was to determine whether any matching advantage would occur for the past self over a stranger, we did not include either of the group factors in this initial analysis (for a breakdown of results by remoteness and induction, see the next section “Role of Remoteness and Induction on Mnemonic Experience and Matching Performance”).
To test for perceptual prioritization of either present or past selves, we conducted separate repeated-measures analysis of variance (ANOVA) on reaction times and accuracies for the matching trials only, with referent label (present self, past self, stranger) as a within subjects-factor. To test our hypotheses regarding the graded performance pattern, planned Helmert contrasts were computed between the present self and the combined past self/ stranger conditions, as well as between the past self and stranger condition. There were main effects of label type on both mean reaction times (F 2, 234=97.25, p<.001, partial 𝜂2=.454) and mean accuracies (F1.84, 215.43=49.41, p<.001, partial 𝜂2 =.297). As predicted, performance was highest in the present self, as compared to the past self and stranger conditions (F1, 117 = 157.68, p <.001, partial 𝜂2=.574, for reaction times, F1, 117=9.94, p <.001, partial 𝜂2=.445, for accuracies). Contrary to our prediction, performance was higher in the stranger condition than for the past self conditions, in both reaction times (F 1, 117=19.89, p < .001, partial 𝜂2=.145) and accuracies (F1, 117=8.99, p=.003, partial 𝜂2=.07). The distribution of participant’s mean performance for the present self, past self, and stranger conditions during matching trials is depicted in Figure 2.
Non-matching trials. Mean reaction times and accuracies are shown in Table 1.
Since the SPE appears to be most robust and consistent in matching trials (Sui et al., 2012; Schäfer et al., 2015)10, our main predictions concern the matching trials. However, we also report the non-matching trials, based on two separate repeated-measures ANOVAs for reaction times and accuracies with referent label (present self, past self, stranger) as a within subjects-factor. The main effect of referent label was significant for both reaction times (F2, 234=18.55, p<.001, partial 𝜂2 = .137) and accuracies (F2, 234 = 90.82, p < .001, partial 𝜂2=.437). Bonferroni-Holm-adjusted pairwise comparisons showed that performance was highest in the stranger condition, which differed significantly from the present self (MD= 21ms, p<.001 for reaction times, MD=0.105, p<.001 for accuracies), and the past self condition (MD=19ms, p<.001 for reaction times, MD=.099, p<.001 for accuracies), with no statistically significant difference between the past and present self conditions (MD=1ms, p=.689 for reaction times, MD=.006, for accuracies, p=.482). The distribution of participant’s mean performance for the present self, past self, and stranger conditions during non-matching trials is depicted in Figure 3.
Role of Remoteness and Induction on Mnemonic Experience and Matching Performance
We examined the effects of the group factors remoteness (yesterday vs. five years ago) and reminiscence induction (instructions vs. no instruction) on matching performance and accuracies, and on subjective mnemonic experience as measured by the MEQ. Self-report means by remoteness of past time period and instruction condition are shown in Table 2.
Mnemonic experience. For subjective memory experience, we tested internal consistencies among groups of items that corresponded to the subscales of the MEQ.11 Following the structure of the MEQ, our items were a priori assigned to the subscales Vividness, Accessibility, Valence, Distancing, Emotional Intensity, Visual Perspective and Sensory Detail. Items with their English translations and respective assignment, as well as Cronbach’s α value for each of those subscales shown in Appendix A. Since scale reliability is critical to the possibility of detecting meaningful correlations (see, e.g., Danner, 2016), only subscales reaching a Cronbach’s α of .6 or higher were included in further analyses, leading to the inclusion of four of the seven subscales (Vividness, Accessibility, Valence, and Distances.)
Table 2
Overview of mnemonic experience self-reports by remoteness and induction condition. Means for the subscales with a Cronbach’s α > .60. SDs in parentheses.
|
n (items)
|
Induction
|
No induction
|
Subscale
|
|
Yesterday
(n = 29)
|
5 years ago (n = 29)
|
Yesterday
(n = 29)
|
5 years ago (n = 31)
|
Vividness
|
4
|
3.67 (.87)
|
3.16 (.80)
|
2.59 (.87)
|
2.54 (.92)
|
Accessibility
|
4
|
3.72 (1.01)
|
3.31 (1.20)
|
3.40 (.89)
|
2.77 (.96)
|
Valence
|
4
|
3.61 (.93)
|
3.82 (.71)
|
3.54 (.75)
|
3.70 (.71)
|
Distancing
|
2
|
2.12 (.99)
|
3.92 (.92)
|
2.50 (1.02)
|
3.48 (.84)
|
For each of those four subscales, we conducted 2x2 two-way ANOVAS with remoteness (yesterday vs. five years ago) and reminiscence (instruction vs. no instruction) as between-subject factors, and the respective mean scale value as dependent variables. The reminiscence manipulation had significant effects on self-reports for Vividness (F1, 114= 27.77, p < .001, partial 𝜂2 = .196) and Accessibility (F1, 114= 5.28, p = .023, partial 𝜂2 = .044), with higher vividness and accessibility of the past self for participants who had received reminiscence instructions prior to the experiment. Remoteness of the past self had significant effects on self-reports for Accessibility (F1, 114 = 7.60, p = .007, partial 𝜂2 = .063) and Distancing (F1, 114 = 38.36, p < .001, partial 𝜂2 = .252), with participants reporting the self five years ago as less accessible and more distant than the self yesterday. No main effects of reminiscence instructions were observed on either Valence (F1, 114 = .42, p = .517, partial 𝜂2 = .004) or Distancing (F1, 114 = 2.68, p = .104, partial 𝜂2 = .023). No main effects of remoteness were observed on either Vividness (F1, 114 = 2.92, p = .09, partial 𝜂2 = .025) or Valence (F1, 114 = 1.63, p = .204, partial 𝜂2 = .014). No interactions between reminiscence and remoteness were observed on any of the subscales (F1, 114 = 2.02, p = .158, partial 𝜂2 = .017 for Vividness; F1, 114 = 0.31, p = .580, partial 𝜂2 = .003 for Accessibility; F1, 114 = 0.03, p = .866, partial 𝜂2 < .001 for Valence; F1, 114 = 2.93, p = .589, partial 𝜂2 = .003 for Distancing).
Matching Performance - Matching Trials. For matching performance, we computed two separate mixed-models 2x2x3 ANOVAs on reaction times and accuracies for the matching trials. We included referent label (present self/ past self/, stranger) as a within subjects-factor, and remoteness (yesterday vs. five years ago) and reminiscence (instructions vs. no instructions) as between-subjects factors (for mean reaction times and accuracies broken down by referent condition and remoteness group, see Table 1. Table B1 further breaks down results by reminiscence group). There was a main effect of referent condition on both dependent variables (F2, 228=98.60, p < .001, partial 𝜂2 = .464, for reaction times, F1.86, 212.25=50.13, p < .001, partial 𝜂2 = .305, for accuracies).
For accuracies, there was also an interaction between referent and remoteness (F1.86, 212.25= 4.12, p = .020, partial 𝜂2 = .035). When considering contrasts between referent conditions for each remoteness group individually, the comparison between the past self and stranger conditions was no longer evident in the “five years ago” group (F 1, 59=.60, p = .442, partial 𝜂2 = .010), while all other contrast differences were preserved in both groups(F1, 57 = 23.8, p < .001, partial 𝜂2 = .296 for present self vs. past self/other, “yesterday” group, F1, 57 = 10.81, p = .002, partial 𝜂2 = .159 for present self vs. past self/other “five years ago” group, .F1, 57 = 93.00, p < .001, partial 𝜂2 = .612 for past self vs stranger, “yesterday” group). To understand the source of the interaction effect, we then compared simple differences between individual referent conditions by remoteness group, controlling for α-error accumulation using Bonferroni-Holm-adjustment. Descriptively, there were smaller differences between the present self and the stranger condition (ΔM = .053, SD = .189 versus, ΔM = .125, SD = .121, t = 2.48, p = .015 unadjusted, p = .045 adjusted), and - although negligibly so - between the past and present self condition (ΔM = .123, SD = .163 vs ΔM = .138, SD = .130, t = .64, p = .525, unadjusted = adjusted) in the “yesterday” as compared to the “five years ago” group, while there was a larger difference between the past self and the stranger condition (ΔM = .071, SD = .163 versus ΔM = .013, SD = .129, t = 2.12, p = .036, unadjusted, p = .074, adjusted) for the “yesterday” as compared to the “five years ago” group. When correcting for multiple comparisons, only the group effect on the difference between the present self and the stranger condition was statistically significant. None of the group differences on accuracies for each referent individually was statistically significant (ΔM=-.036; t = 1.38, p = .085 for present self, ΔM=-.021, t = .779, p = .437 for past self, ΔM = .036, t = 1.63, p = .106 for stranger, see Table 1 for means and SDs), even without α-error adjustment.
The two-way interaction between referent and remoteness did not approach significance for the reaction times (F2, 228=0.702, p = .497, partial 𝜂2 = .006). Neither the main effects of remoteness (F1, 114=.74, p = .392, partial 𝜂2 = .006, for reaction times; F1, 114=.12, p = .729 partial 𝜂2 = .001, for accuracies), nor of reminiscence instructions (F1, 114 < .001, p = .995, partial 𝜂2 < .001, for reaction times; F1, 114 = .02, p = .878, partial 𝜂2 < .001, for accuracies) reached significance for either dependent variable. Neither the two-way interactions between remoteness and reminiscence instructions (F1, 114 = 0.334, p = .565, partial 𝜂2 = .003, for reaction times; F1, 114 < .01, p = .991, partial 𝜂2 < .001, for accuracies), nor between referent and reminiscence instructions (F2,228 = 2.72, p = .068, partial 𝜂2 = .023, for reaction times; F1.86, 212.25 = .75, p = .466, partial 𝜂2 = .007, for accuracies) nor the three-way interaction between referent, remoteness and reminiscence instructions (F2, 228 = 1.29, p = .278, partial 𝜂2 = .011, for reaction times; F1.86, 212.25 = .02, p = .974, partial 𝜂2 < .001, for accuracies) reached significance for either dependent variable.
Matching Performance - Non-matching Trials We also computed two separate mixed-models 2x2x3 ANOVASs on reaction times and accuracies for the non-matching trials, including label (present self/ past self/, stranger) as a within subjects-factor, and remoteness (yesterday vs. five years ago) and reminiscence (instructions vs. no instructions) as between-subjects factors (for mean reaction times and accuracies broken down by referent condition and remoteness group, see Table 1. Table B1 further breaks down results by reminiscence group). For both reaction times (F2, 228=18.34, p < .001, partial 𝜂2 = .139) and accuracies (F2, 228 = 89.59, p < .001, partial 𝜂2 = .440), the main effect of referent reached significance. Neither the main effects of remoteness (F1, 114=.69, p = .407, partial 𝜂2 = .006, for reaction times; F1, 114 = .48, p = .491 partial 𝜂2 = .004, for accuracies), nor of reminiscence instructions (F1, 114 = .18, p = .675, partial 𝜂2 = .002, for reaction times; F1, 114 = .01, p = .919, partial 𝜂2 < .001, for accuracies) reached significance for either dependent variable. Neither the two-way interactions between remoteness and reminiscence instructions (F1, 114 = .06, p = .809, partial 𝜂2 = .001, for reaction times; F1, 114 = .13, p = .724, partial 𝜂2 = .001, for accuracies), referent and remoteness (F2, 228 = .77, p = .463, partial 𝜂2 = .007, for reaction times; F2, 228 = .03, p = .973, partial 𝜂2 < .001, for accuracies), nor referent and reminiscence instruction (F2,228 = .40, p = .669, partial 𝜂2 = .004, for reaction times; F2,228 = 1.76, p = .174, partial 𝜂2 = .015, for accuracies), nor the three-way interaction between referent, remoteness and reminiscence instructions (F2,228 = .13, p = .880, partial 𝜂2 = .001, for reaction times; F2,228 = .18, p = .833, partial 𝜂2 = .002, for accuracies) reached significance.
[8] But see also Moseley et al. (2021), and Golubickis et al. (2017) who included trial type (matching vs. non-matching) as a within-subjects factor. We consider potential interpretation of non-matching trials in the discussion.
[9] As described in the method section, our questionnaire differed substantially from the original version of the MEQ and only contained 20 items. This means that for each subscale, there was a lower number of items in our analogue than there are in the MEQ, and that the number of items varied between subscales.
[10] But see also Moseley et al. (2021), and Golubickis et al. (2017) who included trial type (matching vs. non-matching) as a within-subjects factor. We consider potential interpretation of non-matching trials in the discussion.
[11] As described in the method section, our questionnaire differed substantially from the original version of the MEQ and only contained 20 items. This means that for each subscale, there was a lower number of items in our analogue than there are in the MEQ, and that the number of items varied between subscales.