Microlongitudinal design and undirected associations between affect and cognitive processing speed as measured by digital questionnaire response time (DQRT).

(a) EMA protocols and example items and response scales across the three datasets used. Dataset 1 involved the assessment of 16 affective states twice daily (morning and evening) for 8 weeks in a 7- point Likert scale. Dataset 2 included the evaluation of 7 affect-related experiences three times daily (morning, afternoon, and evening) for 6 weeks on an 8-point Likert scale. Dataset 3 included the assessment of 14 depressive symptoms twice daily (morning and evening) for 8 weeks on an 11 -point visual analogue scale. For each time point, DQRT was calculated as the mean time to respond to EMA items. (b) Between- and within-person contemporaneous associations. Plots display results from multilevel vector autoregressive models examining the relationship between affect and DQRT. Between-person partial correlation coefficients (in purple) are the average of two regression parameters indicating how well the mean of each EMA item predicts the mean of DQRT and vice-versa, controlling for temporal (autoregressive and cross-lagged) effects as well as age and gender. Within-person contemporaneous partial correlation coefficients (in blue) were calculated in a second step using the residuals from the initial models and represent the average of two regression parameters, indicating how well each EMA item predicts DQRT and how well DQRT predicts each EMA item in the same time window. Circles represent negative affect items. Triangles represent positive affect items. Filled shapes indicate results that survive FDR-correction for multiple comparisons using an alpha level of P < 0.05. Annotations: Can’t conc.: can’t concentrate, EMA: ecological momentary assessment, DQRT: digital questionnaire response time, FDR: false discovery rate, Med.: median, No interest.: not interested, Non-sig.: non-significant, Obs.: observations, Rep. thought: repetitive thought.

Directed associations between affect and cognitive processing speed as measured by digital questionnaire response time (DQRT).

(a) Affect Granger-causing subsequent DQRT. Plots show standardized beta coefficients with 95% CI derived from multilevel vector autoregressive models assessing the effect of lagged EMA variables on DQRT controlling for mean EMA value across the study period, age, and gender. (b) DQRT Granger-causing subsequent affect. Here, the reverse associations are shown, indicating the beta coefficients with 95% CI derived from multilevel vector autoregressive models assessing the effect of lagged DQRT on EMA variables controlling for mean DQRT value across the study period, age, and gender. Each dataset featured different assessment intervals (12, ≈8, and 12 hours, respectively). Dataset 1 included 753 participants with a total of 65,541 observations (median = 90, range = 56 – 112). Dataset 2 included 173 participants with a total of 15,598 observations (median = 88, range = 63 – 126). Dataset 3 included 195 participants with a total of 16,729 observations (median = 86, range = 56 – 110). To ensure that results were not driven by a minority of participants overlapped across datasets (Supplementary Table 3), these analyses were repeated with unique participants by assigning those in multiple datasets to the smaller dataset and re-estimating the models. Results were consistent (Supplementary Figure 5). Circles represent negative affect items. Triangles represent positive affect items. Filled shapes indicate results that survive FDR correction for multiple comparisons using an alpha level of P < 0.05. Annotations: Can’t conc.: can’t concentrate, CI: confidence interval, EMA: ecological momentary assessment, DQRT: digital questionnaire response time, FDR: false discovery rate, No interest.: not interested, Non-sig.: non-significant, Rep. thought: repetitive thought.

Associations with a gold-standard task measuring processing speed (Trail Making Test: Trails-B) from dataset 3.

(a) Bivariate correlations. Between-person (left panel) and within- person (right panel) Spearman’s rank correlation coefficients between non-detrended mean Trails-B performance and DQRT. (b) Multilevel mixed effects models predicting DQRT by Trails-B. The left panel shows the fixed effect of Trails-B on DQRT using detrended time series and controlling for age and gender. The right panel displays the random effect estimates. (c) Associations between affect and Trails-B (PM assessment). Standardized beta coefficients and SE derived from multilevel mixed effects models predicting Trails-B evening assessments by mean EMA ratings (between-person effect, in purple), EMA ratings at the same assessment time point (within-person contemporaneous effect, in blue), and EMA ratings in the morning (temporal effects, in yellow), controlling for age and gender. (d) Sensitivity analysis testing the association between affect and DQRT restricted to assessment time points concurrent to Trails-B (PM assessments). Standardized beta coefficients and SE derived from multilevel mixed effects models predicting DQRT evening assessments (downsampled to the days when Trails-B was performed) by mean EMA ratings (between-person effect, in purple), EMA ratings at the same assessment time point (within-person contemporaneous effect, in blue), and EMA ratings in the morning (temporal effects, in green), controlling for age and gender. (e) Internal consistency of Trails-B and DQRT. Intraclass-correlation coefficients (single raters, absolute agreement) indicate that within-individual consistency is higher in Trails-B (yellow) compared to DQRT (green). Figures show the range of Trails-B performance (left figure) and DQRT (right figure) for each individual participant ordered by magnitude of mean Trails-B performance and mean DQRT, respectively. All analyses were performed in 163 participants with a total of 3,561 observations (median = 22, range = 14 – 28). Given the exploratory nature of these analyses and the limited number of observations, P-values were not corrected for multiple comparisons (see Supplementary Tables 9 and 10 for details). Annotations: Can’t conc.: can’t concentrate, EMA: ecological momentary assessment, DQRT: digital questionnaire response time, ICC: interclass correlation coefficient, Med.: median, No interest.: not interested, Obs.: observations, Rep. thought: repetitive thought, SE = standard error, ***P uncorrected < 0.001, ***P uncorrected < 0.01, ***P uncorrected < 0.05.

Previous EMA studies using EMA to investigate within-person relationships between affect and cognition

Participants’ demographics

Unique and overlapped participants across datasets

Mean responses in EMA questionnaires

Between-person effects

Within-person contemporaneous effects

Effects on DQRT at time t

Effects on mood at time t

EMA associations with Trails-B PM performance

EMA associations with DQRT PM (Trails-B days)

EMA data distribution in dataset 1

The histograms show the distribution of EMA responses across all participants for each item included in the analysis. The x-axis represents the possible response values in the rating scale, and the y-axis represents the frequency of observations. Analyses were performed on a sample of 753 participants with a total of 65.541 observations (median = 90, range = 56 - 112). EMA: ecological momentary assessment.

EMA data distribution in dataset 2

The histograms show the distribution of EMA responses across all participants for each item included in the analysis. The x-axis represents the possible response values in the rating scale, and the y-axis represents the frequency of observations. Analyses were performed on a sample of 173 participants with a total of 15.598 observations (median = 88, range = 63 - 126). EMA: ecological momentary assessment, Rep. thought: repetitive thought.

EMA data distribution in dataset 3

The histograms show the distribution of EMA responses across all participants for each item included in the analysis. The x-axis represents the possible response values in the rating scale, and the y-axis represents the frequency of observations. Analyses were performed on a sample of 195 participants with a total of 16.729 observations (median = 86, range = 56 - 110). Can’t conc.: can’t concentrate, EMA: ecological momentary assessment, No interest: no interested.

DQRT and Trails-B data

The histograms show the distribution of (a) DQRT and (b) Trails-B across all participants. The x-axis represents seconds, and the y-axis represents the frequency of observations. The bottom panel shows (c) the mean DQRT for each dataset and (d) the mean Trails-B performance for dataset 3 across participants, plotted against the days or number of assessments. Horizontal dashed lines represent the overall sample mean across the assessment period, for both the original (grey) and detrended (colored) time series. The original DQRT and Trails-B time series (grey solid lines) showed a tendency to decrease over time, indicating faster responses (i.e., practice effects). After applying the detrending procedure (see Methods in main text for details), all the time series across datasets became stationary (colored solid lines), as prerequisite for multilevel vector autoregressive models. Dataset 1 included 753 participants with a total of 65.541 observations (median = 90, range = 56 - 112). Dataset 2 included 173 participants with a total of 15.598 observations (median = 88, range = 63 - 126). Dataset 3 included 195 participants with a total of 16.729 observations (median = 86, range = 56 - 110). For Trails-B analyses, we used a subsample of 163 participants from dataset 3, with a total of 3.561 observations (median = 22, range = 14 - 28). DQRT: digital questionnaire response time.

Main results using non-overlapping samples

This figure presents (a) between- and within-person contemporaneous and (b, c) temporal associations between EMA and DQRT including only unique (i.e., non-overlapping) participants across datasets. To rule out the possibility that our results were driven by a minority of participants, those who contributed to more than one dataset (Supplementary Table 2) were assigned to the smaller dataset and models were re-estimated. This allocation resulted in three independent samples. Dataset 1 retained 609 out of 753 total participants (426 female), with 53.328 total observations (median observations per participant = 91, range = 56-112). Dataset 3 retained 132 out of 195 total participants (103 female), with 11.106 total observations (median observations per participant = 84, range = 56-110). Dataset 2 comprised the same participants as reported in the main text (N = 173, 119 female) and is therefore not reproduced here. This sampling approach avoided overlap and ensured the independence of results across datasets. Can’t conc.: can’t concentrate, CI: confidence interval, EMA: ecological momentary assessment, DQRT: digital questionnaire response time, FDR: false discovery rate, No interest.: no interested, Non-sig.: non-significant, Rep. thought: repetitive thought.