Will things feel better in the morning? A time-of-day analysis of mental health and wellbeing from nearly 1 million observations

Table of Contents

Overall Summary

Study Background and Main Findings

This study investigated the diurnal variation in self-reported mental health and well-being (MHW) using data from the UCL COVID-19 Social Study, a large longitudinal panel study. The analysis included 49,218 adults in England, with an average of 18.5 observations per person. Linear mixed-effects models were used to analyze the data, controlling for various demographic, socioeconomic, and health-related covariates. The key finding was a nonlinear relationship between time of day and MHW, with individuals generally reporting better MHW in the morning (lowest depressive and anxiety symptoms, highest happiness, life satisfaction, and sense of life being worthwhile) and worse MHW around midnight. The difference between early morning and midnight could be as high as 10% of the standard deviation for depressive and anxiety symptoms, and up to 15% for well-being measures. There was also evidence of associations with day of the week (with more variation during weekends) and season (with better MHW in the summer). The association between time of day and MHW was moderated by day of the week, with different patterns observed for depressive and anxiety symptoms on Tuesdays and Wednesdays compared to other days.

Research Impact and Future Directions

The study provides strong evidence for diurnal variation in self-reported mental health and well-being (MHW), showing a consistent pattern of better MHW in the morning and worse MHW around midnight. It also demonstrates associations with day of the week and season, and a moderation effect of day of the week. However, it is crucial to note that these findings demonstrate correlation, not causation. The study design does not allow for conclusions about whether time of day *causes* changes in MHW, or vice-versa.

The study's large sample size and longitudinal design are major strengths, allowing for the examination of within-person variations. The use of standardized measures for depression and anxiety, and the inclusion of multiple aspects of well-being, enhance the study's validity. The weighting procedure increases the generalizability of the findings to the English population. The practical implications are substantial, suggesting that time of day, day of week, and season should be considered in research design, intervention delivery, and public health service provision. For example, mental health services might need to adjust staffing levels or resource allocation based on expected diurnal and seasonal variations in demand.

Despite these strengths, the study acknowledges key uncertainties. The non-random nature of the sample and potential biases in weighting limit the generalizability beyond the English population. The study also highlights the possibility of reverse causality, meaning that individuals' MHW might influence *when* they choose to complete the survey, rather than time of day directly influencing MHW. Future research should use methods like ecological momentary assessment (EMA) to address this limitation and explore directionality. The study provides clear guidance for researchers and practitioners, emphasizing the importance of accounting for temporal factors, but also acknowledges the need for further investigation.

Critical unanswered questions remain. What are the specific mechanisms underlying the observed diurnal variations? Are there other moderators, such as latitude, climate, or cultural factors, that influence these patterns? How do individual differences in chronotype (e.g., morningness-eveningness) interact with time of day to affect MHW? While the methodological limitations, particularly the potential for reverse causality, do not fundamentally invalidate the conclusions about *associations*, they do significantly affect the interpretation of *causal* relationships. Further research is needed to disentangle the complex interplay of factors influencing MHW across different time scales.

Critical Analysis and Recommendations

Diurnal Variation in MHW (written-content)
The study found a nonlinear relationship between time of day and MHW, with outcomes generally best in the morning and worst at night. This finding is significant because it suggests a consistent pattern across multiple measures of mental health and well-being, highlighting the potential importance of circadian rhythms and daily routines.
Section: Results
Appropriate Statistical Analysis (written-content)
The study utilized linear mixed-effects models, appropriate for analyzing longitudinal data with repeated measures, and tested for nonlinear trajectories and interactions. This is a strength because it allows for a more nuanced understanding of the relationships between time and MHW, accounting for individual differences and potential complexities.
Section: Methods
Seasonal Effects on MHW (written-content)
The study found consistent evidence for a seasonal effect, with MHW being significantly better in the summer compared to winter across all outcome measures. This has practical implications for public health service provision, suggesting a potential need for increased resources during winter months.
Section: Results
Reverse Causality Limitation (written-content)
The study acknowledges the possibility of reverse causality, meaning that individuals' MHW might influence when they choose to complete the survey. This is a crucial limitation because it prevents strong causal conclusions about the impact of time of day on MHW.
Section: Discussion
Unspecified Random Effects Structure (written-content)
The study did not specify the random effects structure of the linear mixed-effects models (e.g., random intercepts, random slopes). This omission is important because it limits the transparency and reproducibility of the statistical analysis, making it difficult for other researchers to fully understand and replicate the study.
Section: Methods
Inconsistent Reporting of Statistical Significance and Effect Sizes (written-content)
The Results section did not consistently report p-values for all reported associations, and effect sizes were not provided. Including these would allow readers to directly assess the statistical significance and magnitude of *all* findings, improving interpretability and comparability with other studies.
Section: Results
Figure 2 Lacks Clear Error Bars (graphical-figure)
Figure 2, while presenting differences by day, season, and year, lacks visually distinct error bars representing the 95% confidence intervals. Adding clear error bars would allow for a more immediate visual assessment of statistical significance.
Section: Results
Lack of In-depth Discussion of MHW Component Differences (written-content)
The Discussion section could benefit from a more in-depth discussion of *why* differences exist between MHW components (hedonic/eudemonic vs. social well-being) and what they signify. This would help readers understand the nuances of MHW variations and stimulate further research.
Section: Discussion

Section Analysis

Abstract

Key Aspects

Strengths

Suggestions for Improvement

Introduction

Key Aspects

Strengths

Suggestions for Improvement

Methods

Key Aspects

Strengths

Suggestions for Improvement

Results

Key Aspects

Strengths

Suggestions for Improvement

Non-Text Elements

Table 1 Sample characteristics (n=49218)
Figure/Table Image (Page 3)
Table 1 Sample characteristics (n=49218)
First Reference in Text
After weighting, the sample reflected population proportions, with 50.8% women, 34.2% of participants with high education and 14.6% of ethnic minority (table 1).
Description
  • Key aspect of what is shown: Table 1 presents the characteristics of the study participants. It shows the distribution of several demographic and socioeconomic variables within the sample, both before and after a statistical process called 'weighting'. Weighting is a technique used to adjust the sample data to better reflect the proportions of these characteristics in the broader population of England, thereby reducing potential biases. The characteristics include age groups (e.g., 18-29 years), gender (woman, man), ethnicity (non-White, White), education level (low, medium, high), employment status (employed, other), area of living (rural, urban), living status (alone, not alone), and the presence of diagnosed physical and mental health conditions (yes, no). For instance, before weighting, 76.4% of the sample were women, but after weighting, this was adjusted to 50.8% to match the population proportion. Similarly, the percentage of participants with a high level of education (degree or above) decreased from 68.1% to 34.2% after weighting, and the percentage of ethnic minority participants increased from 5.9% to 14.6% after weighting.
Scientific Validity
  • Appropriateness of Weighting: The table presents essential information regarding the sample demographics and the impact of weighting. The use of weighting is appropriate given the non-random nature of the sample and is crucial for improving the generalizability of the findings to the English population. Including both unweighted and weighted percentages allows the reader to assess the extent of the adjustments made.
  • Variable Selection and Methodology: The choice of variables included in the table is relevant to the study's objectives, as these factors are known to influence mental health and well-being. The use of entropy balancing weights is a statistically sound method for adjusting the sample to match population proportions.
  • Transparency: It would be beneficial to include information on the source of the population proportions used for weighting (e.g., specific census data or national survey). This would further enhance the transparency and replicability of the study.
Communication
  • Clarity and Referencing: The table is clearly labeled and provides sufficient information for the reader to understand the characteristics of the sample. The inclusion of both 'Before weighting (%)' and 'After weighting (%)' columns is crucial for understanding the impact of the weighting procedure. The reference to the table in the results section clearly directs the reader to the relevant information.
  • Enhancements: The table could benefit from a clearer explanation of what 'Employment: other' entails. Providing a few examples within the table caption or a footnote would enhance understanding.
Figure 1 Predicted changes over time from linear mixed-effects model for each...
Full Caption

Figure 1 Predicted changes over time from linear mixed-effects model for each outcome (covariates fixed at reference values): (a) depressive symptoms, (b) anxiety symptoms, (c) happiness, (d) life satisfaction, (e) worthwhile and (f) loneliness. std, standardised.

Figure/Table Image (Page 4)
Figure 1 Predicted changes over time from linear mixed-effects model for each outcome (covariates fixed at reference values): (a) depressive symptoms, (b) anxiety symptoms, (c) happiness, (d) life satisfaction, (e) worthwhile and (f) loneliness. std, standardised.
First Reference in Text
Figure 1 shows the predicted trajectories over time for each outcome and their 95% confidence intervals (full results are in online supplemental table S3).
Description
  • Key aspect of what is shown: Figure 1 presents a series of six graphs (subplots) showing how different mental health and well-being measures are predicted to change over time. Each graph represents a different outcome variable: (a) depressive symptoms, (b) anxiety symptoms, (c) happiness, (d) life satisfaction, (e) feeling worthwhile, and (f) loneliness. The x-axis represents time, presumably hours of the day (although this should be explicitly stated in the figure or caption). The y-axis represents the level of each outcome variable, expressed as standardized values ('std'). Standardization is a statistical process that transforms data to have a mean of 0 and a standard deviation of 1, allowing for comparison across different scales. The lines in each graph represent the predicted trajectory of the outcome variable over time, based on a 'linear mixed-effects model'. This is a statistical model that accounts for both fixed effects (covariates, which are held constant at 'reference values' – meaning a specific, typical level) and random effects (individual variations between people). The shaded areas around the lines represent the 95% confidence intervals, indicating the range within which the true value of the outcome variable is likely to fall 95% of the time.
Scientific Validity
  • Model Appropriateness: The use of a linear mixed-effects model is appropriate for analyzing longitudinal data with repeated measures, as it accounts for both within-individual and between-individual variability. Fixing covariates at reference values allows for a clear visualization of the predicted trajectories, but it's important to acknowledge that this simplifies the complex interplay of factors influencing mental health outcomes.
  • Statistical Rigor: The inclusion of 95% confidence intervals is crucial for assessing the statistical significance and uncertainty of the predicted trajectories. The reference to online supplemental table S3, which contains the full results, is also important for transparency and allows readers to examine the underlying statistical output.
  • Model Specification: The figure would benefit from a more detailed description of the model specification in the methods section or online supplemental material. This should include information on the specific fixed and random effects included in the model, as well as any assumptions made about the data.
Communication
  • Clarity and Visual Presentation: The figure effectively communicates the predicted trajectories of mental health outcomes over time. The use of separate subplots for each outcome (depressive symptoms, anxiety symptoms, happiness, life satisfaction, worthwhile, and loneliness) allows for a clear visualization of each trajectory. The inclusion of 95% confidence intervals provides a measure of uncertainty around the predictions.
  • Contextual Information: The caption clearly states that the covariates are fixed at reference values, which is important for interpreting the figure. However, it would be helpful to explicitly state what these reference values are, either in the caption or in the figure legend. This would allow the reader to understand the specific context for the predicted trajectories.
  • Labeling: While the figure shows the overall trends, the y-axis labels (e.g., 'Depressive symptom (std)') are not very informative for a general audience. It would be beneficial to provide more descriptive labels or include a brief explanation of what 'std' refers to in the caption (i.e., standardised values).
Figure 2 Coefficients and 95% confidence intervals from linear mixed-effects...
Full Caption

Figure 2 Coefficients and 95% confidence intervals from linear mixed-effects model showing average differences by day, season and year for each outcome: (a) depressive symptoms, (b) anxiety symptoms, (c) happiness, (d) life satisfaction, (e) worthwhile and (f) loneliness. Reference categories were Sunday (day), winter (season) and 2020 (year).

Figure/Table Image (Page 5)
Figure 2 Coefficients and 95% confidence intervals from linear mixed-effects model showing average differences by day, season and year for each outcome: (a) depressive symptoms, (b) anxiety symptoms, (c) happiness, (d) life satisfaction, (e) worthwhile and (f) loneliness. Reference categories were Sunday (day), winter (season) and 2020 (year).
First Reference in Text
There was some inconsistent evidence that day of the week was associated with MHW (figure 2).
Description
  • Key aspect of what is shown: Figure 2 consists of six plots (a-f), each corresponding to a different mental health outcome measure: (a) depressive symptoms, (b) anxiety symptoms, (c) happiness, (d) life satisfaction, (e) feeling worthwhile, and (f) loneliness. Each plot displays 'coefficients' derived from a 'linear mixed-effects model'. In this context, a coefficient represents the average difference in the outcome variable for a specific day of the week, season, or year, compared to a 'reference category'. Reference categories are a baseline against which other categories are compared. Here, the reference categories are Sunday for day of the week, winter for season, and 2020 for year. The figure also shows 95% confidence intervals, which provide a range of values within which the true effect is likely to lie with 95% certainty. If the confidence interval crosses zero, it suggests that the difference is not statistically significant at the 0.05 level (meaning there's more than a 5% chance that the true difference is zero).
Scientific Validity
  • Model Appropriateness: The use of a linear mixed-effects model is appropriate for analyzing the data, as it accounts for the repeated measures nature of the data and allows for the estimation of average differences by day, season, and year. However, it's important to consider potential confounding factors that may not be fully accounted for in the model.
  • Statistical Rigor: The inclusion of 95% confidence intervals is crucial for assessing the statistical significance of the findings. The figure allows the reader to visually assess whether the differences are statistically significant by examining whether the confidence intervals cross zero.
  • Interpretation of Results: The interpretation of the coefficients should be cautious, as they represent average differences and may not reflect the experiences of all individuals. It would be helpful to report the standard errors or p-values associated with the coefficients in the online supplemental material.
Communication
  • Clarity and Visual Presentation: The figure effectively presents the average differences in mental health outcomes by day of the week, season, and year, relative to the reference categories (Sunday, winter, and 2020). The use of separate subplots for each outcome variable allows for a clear comparison of the effects across different domains of mental health and well-being.
  • Interpretability: The figure caption is clear in stating the reference categories, which is essential for interpreting the coefficients. However, it would be beneficial to include a brief explanation of what a 'coefficient' represents in this context. For example, 'A positive coefficient indicates that the outcome is higher on that day/season/year compared to the reference category.'
  • Visual Clarity and Error Bars: The figure could be improved by adding error bars representing the 95% confidence intervals. While the caption mentions their inclusion, they are not visually distinct in the current format, making it difficult to assess the statistical significance of the differences. Also, the axis labels are not very informative and lack units.
Figure 3 Predicted changes over time from linear mixed-effects model with...
Full Caption

Figure 3 Predicted changes over time from linear mixed-effects model with moderation by day of the week for each outcome (covariates fixed at reference values): (a) depressive symptoms, (b) anxiety symptoms, (c) happiness, (d) life satisfaction, (e) worthwhile and (f) loneliness. Solid lines representing significantly different patterns to Sunday (p<0.05). std, standardised.

Figure/Table Image (Page 6)
Figure 3 Predicted changes over time from linear mixed-effects model with moderation by day of the week for each outcome (covariates fixed at reference values): (a) depressive symptoms, (b) anxiety symptoms, (c) happiness, (d) life satisfaction, (e) worthwhile and (f) loneliness. Solid lines representing significantly different patterns to Sunday (p<0.05). std, standardised.
First Reference in Text
Only the interactions between time and day of the week were statistically significant and so were included in the final model (online supplemental table S4). Figure 3 shows the predicted patterns over time separately by day of the week for each outcome, setting other covariates at their reference values.
Description
  • Key aspect of what is shown: Figure 3 presents a series of six graphs showing how different mental health and well-being measures change over time, taking into account how these changes differ depending on the day of the week. This is known as 'moderation'. Each graph (a-f) represents a different outcome variable: (a) depressive symptoms, (b) anxiety symptoms, (c) happiness, (d) life satisfaction, (e) feeling worthwhile, and (f) loneliness. The x-axis represents time, presumably hours of the day. The y-axis represents the level of each outcome variable, expressed as standardized values ('std'). Each graph contains multiple lines, each representing a different day of the week. The caption indicates that 'covariates' (other factors that could influence mental health) are held constant at 'reference values'. The caption also notes that 'solid lines represent significantly different patterns to Sunday (p<0.05)'. This means that for those days with solid lines, the way the outcome variable changes over time is statistically different from how it changes on Sunday. The 'p<0.05' indicates the level of statistical significance; it means that there is less than a 5% chance that the observed difference between that day and Sunday is due to random chance.
Scientific Validity
  • Model Appropriateness: The use of a linear mixed-effects model with moderation is appropriate for examining how the relationship between time of day and mental health outcomes varies across different days of the week. By only including statistically significant interaction effects (time and day of week), the model is parsimonious and focuses on the most relevant relationships.
  • Statistical Details: The figure effectively presents the predicted patterns based on the model. However, it would be helpful to provide more details about the statistical tests used to determine significant differences from Sunday (e.g., post-hoc tests with adjustments for multiple comparisons).
  • Interpretation of Results: The interpretation of the interaction effects should be cautious, as they represent average differences and may not reflect the experiences of all individuals. It's important to acknowledge the limitations of the model and the potential for unobserved confounding factors.
Communication
  • Visualization of Interaction Effects: The figure effectively visualizes the interaction between time of day and day of the week on mental health outcomes. Presenting each day's trajectory separately allows for a clear comparison of how the diurnal patterns differ across the week.
  • Highlighting Statistical Significance: The use of distinct line styles to highlight days with significantly different patterns from Sunday (p<0.05) is helpful for identifying statistically meaningful variations. However, the figure would benefit from a clearer explanation of how statistical significance was determined (e.g., which post-hoc tests were used).
  • Axis Labels and Units: The y-axis labels (e.g., 'Depressive symptom (std)') remain somewhat uninformative. Providing more descriptive labels or including a brief explanation of 'std' in the caption (i.e., standardized values) would improve accessibility for a broader audience. It would also be helpful to indicate the time units on the x-axis.

Discussion

Key Aspects

Strengths

Suggestions for Improvement

↑ Back to Top