Strong mnemonic prediction errors increase cognitive control, attention, and arousal
- Department of Psychology, Stanford University, United States
- Wu Tsai Neurosciences Institute, Stanford University, United States
Figures
Figure 1
Summary diagram of the observed effects of strong mnemonic prediction errors (MPEs) in contrast to weak MPEs (solid lines in orange) and their effects at the trial level (dashed lines in orange).
Figure 2 with 5 supplements
Associative memory task design and memory behavior.
(a) Study phase task design. (b) Match/mismatch test phase task design. (c) Behavior in the match/mismatch retrieval task: d’ (left) and RTs (right) in dark teal for strong pairings and light teal for weak pairings. Frontal theta power (d), posterior alpha power (e), and pupil diameter (f) for strong (top) and weak (bottom) trials, color coded by match/mismatch condition (blue/orange). Subplots on the left depict mean and standard error of z-scored power/size across participants. Mean and standard error of participant-level betas for the effects of Mismatch, Strength, and Mismatch×Strength are depicted on the right. Horizontal lines below beta plots indicate significant clusters (permutation-corrected; all p<0.001).
Figure 2—figure supplement 1
Frontal theta and posterior alpha electrodes.
Depiction of electrodes analyzed to compute frontal theta (green) and posterior alpha power (orange).
Figure 2—figure supplement 2
Behavior in the associative retrieval task in Experiment 1 and Experiment 2.
Associative memory d’ for strong pairings was 3.24±0.83 in Experiment 1 and 2.99±1.00 in Experiment 2. Associative memory d’ for weak pairings was 1.80±0.84 in Experiment 1 and 1.55±0.65 in Experiment 2.
Figure 2—figure supplement 3
Trial-level regression plots, including miss and false alarm trials.
(a) Clusters for β weights for frontal theta ~Mismatch×Accuracy×Strength across 25 subjects. (b) Clusters for β weights for posterior alpha ~Mismatch×Accuracy×Strength across 25 subjects. Near probe onset, there was additionally a main effect of Mismatch (1.976–2.128 s; β=0.06, CI=[0.02, 0.10]; p<0.001) and a Mismatch×Strength interaction (1.952–2.112 s; β=−0.09, CI=[-0.15, –0.02]; p<0.001). These effects may reflect the offset of the delay period Strength effect or anticipation of the oncoming probe. (c) Clusters for β weights for pupil diameter ~Mismatch×Accuracy×Strength across 31 subjects. For statistics on clusters, see Supplementary file 1C.
Figure 2—figure supplement 4
Response-locked trial-level regression plots.
(a) Response-locked frontal theta analyses showed that before responses, there was a main effect of Mismatch (−0.608–0.104 s; β=0.33, CI=[0.24, 0.42]; p<0.001), a main effect of Strength (−0.840 to –0.384 s; β=0.12, CI=[0.05, 0.19]; p<0.001), and a Mismatch×Strength interaction (−0.592–0.056 s; β=−0.36, CI=[−0.47–0.25]; p<0.001). The interaction indicated that the difference in frontal theta between mismatch and match trials was larger for strong compared to weak trials. After responses, there was a main effect of Mismatch (0.736–0.992 s; β=−0.07, CI=[−0.12,–0.02]; p<0.001) that emerged later than a Mismatch×Strength interaction (0.304–0.720 s; β=0.11, CI=[0.04, 0.19]; p<0.001). Together, these results indicate that after responses, the differences in the magnitude of the dip in frontal theta between mismatch and match trials was larger for weak compared to strong trials; differences between mismatch-evoked decreases in frontal theta persisted. (b) Response-locked posterior alpha analyses showed that responses coincided with a drop in posterior alpha power. There were main effects of Mismatch (−0.208 to –0.040 s; β=−0.06, CI=[-0.109, –0.011]; p<0.001; and from –0.016–0.528 s; β=−0.10, CI=[-0.14, –0.06]; p<0.001), a main effect of Strength (−0.344–0.712 s; β=−0.12, CI=[–00.15, –0.09]; p<0.001), and a Strength×Mismatch interaction (−0.040–0.576 s; β=0.14, CI=[0.09, 0.18]; p<0.001). The difference in posterior alpha power between match and mismatch trials was, therefore, larger for strong compared to weak trials. The onset of the interaction was later than that of the Mismatch and Strength effects, indicating strength-modulated changes in attention following mnemonic prediction errors (MPEs) emerged primarily after responses. There was additionally a Mismatch effect from –0.912 to –0.800 s (β=−0.07, CI=[–0.12, –0.01]). (c) Response-locked analyses of changes in pupil diameter revealed that there were main effects of Mismatch (−0.99 to –0.19; β=0.12, CI=[0.04, 0.20], p<0.001; 0.19–0.99 s; β=0.16, CI=[0.08, 0.24]; p<0.001), main effects of Strength (−0.99 to –0.93 s; β=−0.07, CI=[–0.14, 0]; p<0.001; –0.32–0.27, β=−0.17, CI=[-0.25, –0.10], p<0.001; 0.56–0.99 s, β=0.14, CI=[0.06, 0.21], p<0.001), and a Mismatch×Strength interaction (0.10–0.99 s; β=−0.33, CI=[-0.41, –0.25]; p<0.001). The interaction indicates that the difference in pupil size between match and mismatch trials was larger for strong compared to weak trials.
Figure 2—figure supplement 5
Pupil trial-level regression plots separated by experiment.
Effects of Mismatch and Strength on pupil diameter in Experiment 1 (a) and Experiment 2 (b). (a) In Experiment 1, there was a main effect of Mismatch from 2.48 to 4.00 s (β=0.20, CI=[0.11, 0.29], p<0.001), main effects of Strength from 0.90 to 2.84 s (β=−0.13, CI=[–0.21, –0.05]) and 3.27 to 4.00 s (β=0.27, CI=[0.17, 0.37]), and a Mismatch×Strength interaction from 2.62 to 4.00 s (β=−0.31, CI=[–0.40, –0.22]). (b) In Experiment 2, there were main effects of Mismatch from 2.02 to 2.86 s (β=0.16, CI=[0.04, 0.28], p<0.001) and 3.25 to 3.95 s (β=0.19, CI=[0.07, 0.32]), main effects of Strength from 2.69 to 2.78 s (β=−0.11, CI=[–0.22, 0]) and 3.44 to 4.00 s (β=0.18, CI=[0.08, 0.28]), and a Mismatch×Strength interaction from 2.56 to 4.00 s (β=−0.34, CI=[–0.47, –0.21]).
Figure 3 with 4 supplements
Pupil components sensitive to mnemonic prediction errors (MPEs) and their relationships with frontal theta and posterior alpha.
(a) Pupil components extracted by temporal principal components analysis (PCA) with bootstrapped 95% confidence intervals. PC scores for PC3 (b) and PC4 (c) as a function of match/mismatch conditions and memory strength (middle) and as related to reaction times (right). (d) Schematic of observed relationships among frontal theta, posterior alpha, PC3 scores, and PC4 scores. Time windows over which frontal theta and posterior alpha were averaged in trial-level analyses are underlined in yellow and correspond to those displayed in Figure 2. (e) Relationship between frontal theta cognitive control signals and attention signals in posterior alpha. (f and g) Relationships between frontal theta cognitive control signals and PC3 (f) and PC4 (g). (h and i) Relationships between posterior alpha MPE-related attention signals and PC3 (h) and PC4 (i). Quartiles and quintiles are shown for visualization purposes only; data were analyzed using continuous measures as described in the Methods.
Figure 3—figure supplement 1
Retrieval success and cognitive effort (PC1 and PC5) scores.
(a) PC1 scores were higher for stronger memories both at the condition level and the trial level, leading us to interpret PC1 as indicative of the strength of memory retrieval. In particular, PC1 scores were higher for strong compared to weak trials (Supplementary file 1D; effect of Strength) and higher PC1 scores predicted faster match RTs (Supplementary file 1E; effect of PC1 score). The relationship between PC1 scores and RTs was attenuated for mismatch trials (Mismatch×PC1 score interaction), which may be a consequence of the response delay induced by mnemonic prediction errors (MPEs) on mismatch trials (Figure 2c). PC1 accounted for the most variance (28.81%), onset during the cue period, peaked predominantly at the time of probe onset (2.00 s, CI=[1.94, 2.05]; loading=0.90, CI=[0.90, 0.91]), and had a FWHM of 1.75 s (CI=[1.70, 1.78]). Altogether, these findings suggest that PC1 tracks graded retrieval success or strength of memory retrieval. (b) For PC5, the relationship with the memory strength conditions and RTs indicated that this component was a condition-level marker of cue strength and a trial-level marker of cognitive effort during associative retrieval. PC5 scores were higher for strong than weak trials (Supplementary file 1F; main effect of Strength), but in contrast to PC1, higher PC5 scores were associated with longer response times (RTs) (Supplementary file 1G; main effect of PC5 scores). The relationship between PC5 scores and RTs was stronger for weak compared to strong pairings (PC5 score×Strength). Thus, PC5 showed the opposite pattern as PC1, with PC5 scores scaling inversely with trial-level memory strength, suggesting that PC5 reflects a mixture of cue strength (i.e., higher scores for strong cues) and the amount of cognitive effort expended during associative memory retrieval (i.e., inverse relationship with RT). The temporal profile of PC5 is consistent with this interpretation, as PC5 onset during the cue period, peaked during the delay period (1.15 s after cue onset, CI=[1.12, 1.18]; loading=0.56, CI=[0.52, 0.61]) and had a full width at half maximum (FWHM) of 0.98 s (CI=[0.92, 1.04]). This component accounted for 5.66% of variance.
Figure 3—figure supplement 2
Comparison of preparatory attention pupil (PC2) scores computed with and without miss and false alarm trials.
(a) PC2 peaked 0.30 s (CI=[0.29, 0.31]) after cue onset (loading=0.96, CI=[0.95, 0.96]) and declined slowly over the course of the trial, accounting for 25.89% of variance. There was a marginal difference in PC2 scores as a function of Mismatch, but not as a function of Strength (Supplementary file 1H). PC2 scores were marginally associated with longer response times (RTs) (Supplementary file 1I; marginal main effect of PC2 scores). (b) An augmented model using scores computed in a tPCA, including error trials revealed a main effect of Accuracy and an Accuracy×Strength interaction (Supplementary file 1J). The augmented model did not fit the data significantly better than a compact model lacking Accuracy terms (χ2(4)=7.66, p=0.105). Together, these outcomes indicate that the influence of pupil-linked arousal immediately following cue presentation on memory retrieval may influence the likelihood of successful memory retrieval and may impact the precision with which a memory is retrieved.
Figure 3—figure supplement 3
Stimulus detection and mismatch detection (PC6) scores.
PC6 accounted for the least amount of variance (1.83%) and had its most prominent peak during probe presentation (0.58 s after probe onset, CI=[0.55, 0.60]; loading = 0.28, CI=[0.26, 0.30]; FWHM=0.60, CI=[0.56, 0.64]); the second most prominent peak occurred earlier, 0.87 s after cue onset (CI=[0.85, 0.90]; loading = 0.13, CI=[0.11, 0.16]; FWHM=0.45, CI=[0.43, 0.47]); the last peak was at the end of the trial (3.98 s after cue onset, CI=[3.94, 3.99]; loading=0.10, CI=[0.08, 0.11]). The temporal profile of PC6’s loadings suggests that this component reflects responses to changes in luminance driven by stimulus onset; these responses peak after ~500–900 ms. Given that PC6’s loadings were higher for the probe-period response compared to the cue-period response, PC6 may be more sensitive to the identity of the probe than that of the cue. PC6 scores were higher on mismatch than match trials (Supplementary file 1K; main effect of Mismatch) and higher PC6 scores predicted faster RTs (Supplementary file 1L; main effect of PC6 score), indicating that PC6 scores may reflect the amount of attention allocated to the probe and its identity as a match or mismatch. There was additionally a PC6 score×Mismatch interaction, showing that the relationship with response times (RTs) was attenuated for mismatches. Unlike PC3 and PC4, PC6 scores (from the full dataset with error trials) were not fit better by an augmented model including effects of and interactions with Accuracy (χ2(4)=4.31, p=0.366). Misses and false alarms can be accounted for by an erroneous mnemonic prediction (which could give rise to a mnemonic prediction error, MPE) or the absence of a prediction (where the response may reflect guessing). Given ambiguity regarding the experience of MPEs on error trials, and PC6’s lack of sensitivity to accuracy, PC6 may not be fully characterized as an MPE-sensitive component, but rather an initial mismatch response.
Figure 3—figure supplement 4
Association of pupillary mnemonic prediction error (MPE)-related responses to frontal theta and posterior alpha.
Mean pupil responses during strong MPEs are related to frontal theta and posterior alpha responses during strong MPEs, which did not differ with Mismatch. (a) Relationship between frontal theta and mean pupil. (b) Relationship between posterior alpha and mean pupil. Statistics are reported in the main text.
Figure 4 with 1 supplement
Dynamics between cognitive control, attention, and arousal in response to strong mnemonic prediction errors (MPEs).
(a) Schematic of hypothesized relationships: stronger MPEs increase attention/arousal and this effect was hypothesized to be partially accounted for by an increase in cognitive control. (b) A multivariate mediation analysis testing the model diagrammed in (a) with pupil PC3 as the measure of attention/arousal. (c) A multivariate mediation analysis testing the model diagrammed in (a) with posterior alpha as the measure of attention. Paths in mediation models are labeled with the mean of the posterior and 95% credible intervals. Gray histograms accompanying mediation model diagrams depict the posterior distribution for the indirect effect; black points indicate posterior medians and horizontal lines indicate the 66% and 95% highest density intervals. (d) Cross-correlation analysis showing the correlation between lagged frontal theta time series and posterior alpha time series on strong MPE trials. Individual subjects are shown in orange and the mean across subjects is shown in black. The inset in gray is included for illustration of trial-averaged stimulus-locked responses at the example lag of –0.744 s; note that the cross-correlation was conducted on trial-level time series.
Figure 4—figure supplement 1
Cross-correlation of frontal theta and posterior alpha reveals temporal relationship of retrieval-related processes.
The cross-correlation figure is a reproduction of that in Figure 4d; the inset in gray shows the mean frontal theta and posterior alpha time series from Figure 2d and Figure 2e with frontal theta lagging at the max timepoint in the indicated cross-correlation cluster (0.184 s). This cross-correlation analysis of the frontal theta and posterior alpha time series revealed a significant cluster (a) with a positive correlation and (b) with frontal theta lagging (Figure 4d; 0.080–0.184 s, r=0.006, CI=[0.065, 0.068]). This positive association – greater control and less attention – may reflect cognition during memory retrieval, given the positive mean β during the delay-period Strength clusters in Figure 2d and e. The positive lag suggests that at retrieval, changes in attention may precede changes in cognitive control.
Figure 5 with 2 supplements
Mnemonic prediction error (MPE)-driven increases in frontal theta by strong MPEs enhance learning.
(a) Behavioral performance in the surprise recognition task in Experiment 1 and Experiment 2. (b) Higher frontal theta during strong MPEs is associated with a higher likelihood of subsequently remembering the probe ~5 min later (Experiment 1). Points indicate binned values for visualization; smoothed lines reflect model estimates. (c) Trial-level regression models for frontal theta modeling the effects of Subsequent Memory, Strength, and Subsequent Memory×Strength interactions in Experiment 1.
Figure 5—figure supplement 1
Subsequent memory response times (RTs) for mismatch probes in Experiments 1 and 2.
A linear mixed-effects model was used to examine how recognition RTs in the subsequent memory test varied with Strength, Accuracy, Experiment, and their interactions. Pairwise comparisons revealed that correct RTs for strong probes were faster than for familiar probes in the same-day test (ΔStrong-Familiar=−0.050, CI=[–0.079, –0.021], p<0.001) but not 48 hr later (ΔStrong-Familiar=−0.019, CI=[–0.056, 0.019], p=0.583). Correct RTs for strong probes were faster than for new probes in the same-day test (ΔStrong-New=−0.036, CI=[–0.056, –0.015], p<0.001) and 48 hr later (ΔStrong-New=0.061, CI=[0.033, 0.089], p<0.001). Correct RTs between strong and weak probes did not differ in the same-day test (ΔStrong-Weak=0.012, CI=[–0.013, 0.038], p=0.592) or 48 hrs later (ΔStrong-Weak=–0.014, CI=[–0.050, 0.023], p=0.769). These results reveal that an mnemonic prediction error (MPE)-related benefit on subsequent memory RTs diminished when recognition memory was tested 48 hr later. Altogether, subsequent recognition performance revealed that memory was superior for probes that violated strong or weak mnemonic predictions than for familiar items that were not encountered in the context of an MPE. When memory is tested after a 48 hr delay, these effects diminish.
Figure 5—figure supplement 2
Neural substrates underlying learning from mnemonic prediction errors (MPEs).
(a) Trial-level regression models revealed that for posterior alpha, there was a memory outcome×Strength interaction from 1.824 to 2.000 s (β=−0.32, CI=[–0.56, –0.08]) after probe onset such that the difference in posterior alpha power between hits and misses was larger for weak compared to strong trials. There was a main effect of Strength from 1.592 to 1.744 s after cue onset (β=0.134, CI=[0.039, 0.228]). (b-c) For pupil size, there were no windows of time in Experiment 1 or Experiment 2 which predicted subsequent memory. (b) In Experiment 1, there was a main effect of Strength from 0.47 to 1.56 s after probe onset (β=−0.21, CI=[–0.32, –0.09]). (c) In Experiment 2, there were main effects of Strength from 1.26 to 1.67 s (β=−0.23, CI=[–0.42, –0.04]) and 1.91–4.00 s (β=−0.36, CI=[–0.54, –0.18]). (d–e) Subsequent memory effects for mean posterior alpha power and mean pupil diameter during the primary Mismatch×Strength clusters, illustrated by the yellow clusters, were not found. (d) There was no effect of subsequent memory outcome on mean MPE-related posterior alpha main effect of subsequent memory outcome: β=−0.030, CI=[–0.089, 0.030], p=0.326; memory outcome×Strength: β=0.038, CI=[–0.047, 0.123], p=0.377. (e) There was no effect of subsequent memory outcome on mean MPE-related pupil size (main effect of subsequent memory outcome: β=−0.085, CI=[–0.216, 0.047], p=0.205; memory outcome×Strength: β=0.036, CI=[–0.152, 0.225], p=0.705; memory outcome × experiment: β=−0.034, CI=[–0.228, 0.159], p=0.728; no three-way interaction: β=0.113, CI=[–0.164, 0.390], p=0.425). (f) PC1 scores did not predict subsequent memory (no main effect of PC1: O.R.=1.000, CI=[0.852, 1.174], p=0.995; no PC1×Strength interaction: O.R=0.950, CI=[0.753, 1.198], p=0.753; a marginal PC1×Experiment interaction: O.R.=1.246, CI=[0.977, 1.589], p=0.077; no PC1×Strength×Experiment interaction: O.R.=0.862, CI=[0.609, 1.219], p=0.400). (g) PC2 scores did not predict subsequent memory (no main effect of PC2: O.R.=1.018, CI=[0.868, 1.193], p=0.826; no PC2×Strength interaction: O.R.=1.061, CI=[0.841, 1.338], p=0.619; a marginal PC2×Experiment interaction: O.R.=0.814, CI=[0.641, 1.035], p=0.093; no PC2×Strength×Experiment interaction: O.R.=1.103, CI=[0.781, 1.559], p=0.578). (h) PC3 scores did not predict subsequent memory (no main effect of PC3: O.R.=1.079, CI=[0.921, 1.264], p=0.348; no PC3×Strength interaction: O.R.=0.968, CI=[0.766, 1.225], p=0.789; no PC3×Experiment interaction: O.R.=0.918, CI=[0.723, 1.166], p=0.485; no PC3×Strength×Experiment interaction: O.R.=1.067, CI=[0.752, 1.513], p=0.717). (i) PC4 scores did not predict subsequent memory (no main effect of PC4: O.R.=1.030, CI=[0.879, 1.207], p=0.716; no PC4×Strength interaction: O.R.=1.018, CI=[0.806, 1.284], p=0.883; no PC4×Experiment interaction: O.R.=1.086, CI=[0.858, 1.375], p=0.493; no PC4×Strength×Experiment interaction: O.R.=0.834, CI=[0.589, 1.181], p=0.307). (j) PC5 scores did not predict subsequent memory (no main effect of PC5: O.R.=0.912, CI=[0.782, 1.064], p=0.240; no PC5×Strength interaction: O.R.=1.195, CI=[0.947, 1.507], p=0.133; no PC5×Experiment interaction: O.R.=1.178, CI=[0.935, 1.485], p=0.165; marginal PC5×Strength×Experiment interaction: O.R.=0.723, CI=[0.512, 1.022], p=0.067). (k) PC6 scores did not predict subsequent memory (marginal main effect of PC6: O.R.=1.138, CI=[0.979, 1.323], p=0.093; no PC6×Strength interaction: O.R.=0.944, CI=[0.746, 1.194], p=0.631; no PC6×Experiment interaction: O.R.=0.924, CI=[0.737, 1.158], p=0.490; no PC6×Strength×Experiment interaction: O.R.=0.989, CI=[0.697, 1.403], p=0.951).
Tables
Table 1
Hit and false alarm rates.
| % Hits (SD) | % False alarms (SD) | |
|---|---|---|
| Associative retrieval performance | ||
| Strong (4×) | 94.1 (9.9) | 11.1 (13.5) |
| Weak (1×) | 71.0 (18.8) | 17.0 (16.2) |
| Recognition task performance | ||
| Experiment 1 | ||
| Strong (4×) | 75.6 (16.4) | |
| Weak (1×) | 76.7 (18.3) | |
| Familiar item | 43.9 (16.7) | |
| Novel | 15.7 (9.3) | |
| Experiment 2 | ||
| Strong (4×) | 46.4 (18.7) | |
| Weak (1×) | 48.6 (20.6) | |
| Familiar item | 27.6 (15.1) | |
| Novel | 14.2 (8.7) | |
Table 2
PC3 scores as a function of Mismatch and Strength.
| Fixed effects | Estimate | CI | p-value | SE | DF |
|---|---|---|---|---|---|
| (Intercept) | 0.002 | [–0.119, 0.123] | 0.979 | 0.061 | 90.970 |
| Mismatch | 0.079 | [0.024, 0.134] | 0.005 | 0.028 | 148.072 |
| Strength | –0.113 | [-0.177,–0.050] | 0.001 | 0.032 | 114.654 |
| Mismatch×Strength | –0.172 | [-0.240,–0.104] | <0.001 | 0.035 | 10236.929 |
Table 3
PC3 scores as a function of Mismatch, Strength, and Memory Accuracy.
| Fixed effects | Estimate | CI | p-value | SE | DF |
|---|---|---|---|---|---|
| (Intercept) | 0.034 | [–0.087, 0.155] | 0.577 | 0.061 | 92.514 |
| Mismatch | 0.079 | [0.027, 0.130] | 0.003 | 0.026 | 215.623 |
| Accuracy | 0.021 | [–0.104, 0.146] | 0.743 | 0.064 | 655.824 |
| Strength | –0.107 | [-0.168, –0.046] | 0.001 | 0.031 | 123.938 |
| Mismatch×Accuracy | –0.418 | [-0.601, –0.235] | <0.001 | 0.093 | 3369.869 |
| Mismatch×Strength | –0.176 | [-0.244, –0.107] | <0.001 | 0.035 | 11969.179 |
| Accuracy×Strength | –0.169 | [-0.304, –0.035] | 0.013 | 0.068 | 3106.734 |
| Mismatch×Accuracy×Strength | 0.490 | [0.275, 0.706] | <0.001 | 0.110 | 7003.623 |
Table 4
PC3 scores and response times (RTs).
| Fixed effects | Estimate | CI | p-value | SE | DF |
|---|---|---|---|---|---|
| (Intercept) | –0.660 | [-0.805,–0.516] | 0.006 | 0.022 | 1.427 |
| PC3 score | 0.009 | [–0.003, 0.021] | 0.145 | 0.006 | 152.586 |
| Strength | 0.202 | [0.187, 0.217] | <0.001 | 0.008 | 105.059 |
| Mismatch | 0.161 | [0.144, 0.178] | <0.001 | 0.009 | 136.388 |
| PC3 score×Strength | –0.031 | [-0.043, –0.020] | <0.001 | 0.006 | 3083.608 |
| PC3 score×Mismatch | –0.055 | [-0.068, –0.042] | <0.001 | 0.007 | 4019.950 |
| Strength×Mismatch | –0.128 | [-0.146, –0.109] | <0.001 | 0.009 | 10186.180 |
| PC3 score×Strength×Mismatch | 0.064 | [0.046, 0.082] | <0.001 | 0.009 | 10234.438 |
Table 5
PC4 scores as a function of Mismatch and Strength.
| Fixed effects | Estimate | CI | p-value | SE | DF |
|---|---|---|---|---|---|
| (Intercept) | –0.115 | [-0.203, –0.027] | 0.011 | 0.044 | 103.597 |
| Mismatch | 0.275 | [0.224, 0.327] | <0.001 | 0.026 | 10260.365 |
| Strength | 0.422 | [0.379, 0.465] | <0.001 | 0.022 | 10262.071 |
| Mismatch×Strength | –0.477 | [-0.553, –0.400] | <0.001 | 0.039 | 10257.275 |
Table 6
PC4 scores as a function of Mismatch, Strength, and Memory Accuracy.
| Fixed effects | Estimate | CI | p-value | SE | DF |
|---|---|---|---|---|---|
| (Intercept) | –0.137 | [-0.235, –0.040] | 0.006 | 0.049 | 91.302 |
| Mismatch | 0.252 | [0.184, 0.319] | <0.001 | 0.034 | 152.507 |
| Accuracy | 0.747 | [0.589, 0.906] | <0.001 | 0.081 | 457.972 |
| Strength | 0.397 | [0.335, 0.460] | <0.001 | 0.032 | 132.103 |
| Mismatch×Accuracy | –0.488 | [-0.692, –0.284] | <0.001 | 0.104 | 4905.789 |
| Mismatch×Strength | –0.465 | [-0.540, –0.390] | <0.001 | 0.038 | 11923.486 |
| Accuracy×Strength | –0.679 | [-0.827, –0.531] | <0.001 | 0.075 | 3826.939 |
| Mismatch×Accuracy×Strength | 0.538 | [0.300, 0.777] | <0.001 | 0.122 | 7308.225 |
Table 7
PC4 scores and response times (RTs).
| Fixed effects | Estimate | CI | p-value | SE | DF |
|---|---|---|---|---|---|
| (Intercept) | –0.649 | [-0.847, –0.450] | 0.012 | 0.024 | 1.227 |
| PC4 score | 0.081 | [0.071, 0.091] | <0.001 | 0.005 | 154.422 |
| Strength | 0.170 | [0.155, 0.184] | <0.001 | 0.007 | 115.840 |
| Mismatch | 0.140 | [0.124, 0.156] | <0.001 | 0.008 | 142.039 |
| PC4 score×Strength | 0.012 | [0.001, 0.022] | 0.025 | 0.005 | 7330.103 |
| PC4 score×Mismatch | –0.028 | [-0.041, –0.016] | <0.001 | 0.006 | 8214.680 |
| Strength×Mismatch | –0.086 | [-0.104, –0.069] | <0.001 | 0.009 | 10176.523 |
| PC4 score×Strength×Mismatch | –0.013 | [–0.031, 0.005] | 0.169 | 0.009 | 10157.134 |
Additional files
-
Supplementary file 1
Supplementary tables for trial-level regression model outputs for frontal theta, posterior alpha, and pupil and linear mixed effects model outputs for pupil PCs.
(A) Frontal theta trial-level regression model, including miss and false alarm trials. (B) Posterior alpha trial-level regression model, including miss and false alarm trials. (C) Pupil trial-level regression model, including miss and false alarm trials. (D) PC1 scores as a function of Mismatch and Strength. (E) PC1 scores and RTs. (F) PC5 scores as a function of Mismatch and Strength. (G) PC5 scores and RTs. (H) PC2 scores as a function of Mismatch and Strength. (I) PC2 scores and RTs. (J) PC2 scores as a function of Mismatch, Strength, and Memory Accuracy. (K) PC6 scores as a function of Mismatch and Strength. (L) PC6 scores and RTs.
- https://cdn.elifesciences.org/articles/109678/elife-109678-supp1-v1.docx
-
MDAR checklist
- https://cdn.elifesciences.org/articles/109678/elife-109678-mdarchecklist1-v1.docx
Download links
A two-part list of links to download the article, or parts of the article, in various formats.
Downloads (link to download the article as PDF)
Open citations (links to open the citations from this article in various online reference manager services)
Cite this article (links to download the citations from this article in formats compatible with various reference manager tools)
Strong mnemonic prediction errors increase cognitive control, attention, and arousal
eLife 15:RP109678.
https://doi.org/10.7554/eLife.109678.4
