The relationship between change in routine and student mental wellbeing during a nationwide lockdown

Abstract

Objective

During March 2020, the UK entered a national lockdown, causing a sudden change in undergraduate students’ routines. This study uses this event to investigate the impact routine change had on students’ mental wellbeing; in particular looking at depression, anxiety, sleep disturbance, and eating behaviors.

Method

Participants reported their daily routine timings (waking, breakfast, lunch, evening meal and bedtime) and activities (e.g. exercise amount, time with friends, time studying, etc) on a typical Monday, Wednesday and Saturday during term time and lockdown. Additionally they completed the PROMIS measures of anxiety, depression, and sleep disturbance, and the Eating Pathology Symptom Inventory.

Results

Lockdown saw small but significant shifts in routine timing (on average 1.5 h) However, there was no clear overall pattern of relationships between mental wellbeing and routine structure or magnitude of routine change. There was some evidence of changes in amount of exercise relating to reported anxiety.

Discussion

These findings are consistent with the current literature reporting lockdown effects on behavior. Routine timings shifted, but this change was small and largely did not affect the mental wellbeing reported by undergraduate students. The change in amount of exercise posed by lockdown did appear to be an important factor in wellbeing, and more research should focus on the wellbeing implications of closing places for exercise.

Keywords:

• Lockdown
• eating behaviors
• stability
• instability
• anxiety
• mood
• routine and structure
• COVID-19 pandemic

Introduction

Everyday life has many specific timing demands, such as a particular time to wake up, a particular time to be at work or study, or a particular time to eat. These schedules can be highly consistent over time, and so individuals develop routines which allow them to minimize unpredictable variability in everyday life. Unpredictability is a factor in the development of mental health difficulties, particularly in young adults (Jivanjee, Kruzich, & Gordon, 2009). Consequently, routines could have a role in creating a stable life experience. In contrast to the public discussion on the benefits of a routine for mental wellbeing (such as Blurt Team, 2018), there has been relatively little empirical research studying the effects of routine structure. There has been research showing the benefits to family routines for adolescent wellbeing (Koome, Hocking, & Sutton, 2012) and, in general, routines promote increased engagement with known correlates of mental wellbeing, such as frequent exercise (Teychenne et al., 2020), a healthy diet (Wattick, Hagedorn, & Olfert, 2018), and healthy sleep patterns (Di Benedetto, Towt, & Jackson, 2020).

During the 2020 COVID-19 pandemic, universities in the United Kingdom closed their physical campuses between March and July and, in doing so, disrupted students’ regular routines when external schedules (e.g. university schedules) became less relevant to individuals’ direction of their time. As undergraduate students are already at high risk of mental health difficulties (see Eisenberg, Gollust, Golberstein, & Hefner, 2007), a sudden change to social, work, and study routines may have particular consequences for these individuals. The current study investigates the relationship between the change in University students’ self-reported routine between typical “term time” and the national “lockdown” and their mental wellbeing (eating, depression, anxiety, and sleep) and serves as a quasi-experiment into the effect of change of routine on the mental health of students.

Several studies have investigated the general impact of pandemic-related policy decisions on populations’ mental health. In particular, the World Health Organization (2020) has drawn attention to the broad range of potential risks to mental wellbeing affecting the general population. Recent research on Italian school children has shown that lockdown did change routines, and this was accompanied with worsening mental wellbeing (Cellini, Di Giorgio, Mioni, & Di Riso, 2021). Research on Chinese undergraduate populations indicated that reports of anxiety during the COVID-19 pandemic were more serious than during the 2013 SARS pandemic (Cao et al., 2020; Gambin et al., 2021) and, in the UK, there have been concerns raised about the possible mental health demands of social isolation on vulnerable populations (Lopes & Jaspal, 2020). These assertions have been supported with findings that students who live alone and have less direct contact with social support networks are at a higher risk of developing mental health conditions during lockdowns (Elmer, Mepham, & Stadtfeld, 2020; Xiang et al., 2020; Zhang, Wu, Zhao, & Zhang, 2020). One explanation as to why students might be particularly affected by the pandemic is increased uncertainty about their routine (Dwivedi, Kaur, Shukla, Gandhi, & Tripathi, 2020). Additionally, evidence from the US showed that students became more sedentary when universities suspended classes, which co-occured with increased anxiety and depression (Huckins et al., 2020). However, there is also evidence that these changes in perceived stress and anxiety may not be of clinical significance. In a UK longitudinal study (Kwong et al., 2021) and a Dutch ecological momentary assessment of students (Fried, Papanikolaou, & Epskamp, 2022), it was found that changes in anxiety and mental wellbeing were present but not of clinical significance. In general, current research into the impact of lockdown-induced routine change on mental health is a mixed picture, with remaining questions about the practical significance of mental health changes and the magnitude of activity change.

In qualitative research, individuals with a history of eating disorders have specifically referenced the lockdown’s disruption to their routines as a source of concern for their wellbeing (Branley-Bell & Talbot, 2020; Clark Bryan et al., 2020). It has been suggested that even those without a history of maladaptive eating could develop problematic eating behaviors due to routine changes during the lockdown (Rodgers et al., 2020). A national Australian survey highlighted increases in self-reports of various problematic eating behaviors, such as restricting, bingeing, and purging food (Phillipou et al., 2020). The typical student population is at elevated risk of eating disorders in general, with most students reporting disrupted eating behavior in response to the stress of other contexts (e.g. exam season; Deliens Clarys, De Bourdeaudhuij, & Deforche, 2014). Given that mental health conditions in student populations are related to lower quality of life (Jenkins, Ducker, Gooding, James, & Rutter-Eley, 2021) and university attainment (Andrews & Wilding, 2004), research in this population is needed to mitigate these long-term effects.

Drawing on the above literature, we hypothesize that greater change in routine will relate to poorer mental wellbeing. We operationalize elements of routine as the typical timing of bedtime and mealtimes, as well as the amount of time dedicated to exercise, socializing, work and study. Wellbeing is assessed using measures of depression, anxiety, sleep disturbance and eating pathology symptoms.

Method

Participants

For our principal analysis, we were interested having sufficient participants to test the relationship between change in routine and mental health variables. We estimated our sample size based on at the smallest effect size of interest of r = .20. We chose this effect size due to typical effects in the psychological literature, where personality research generally finds r = .21 (Fraley & Marks, 2007), social and personality psychology finds r = .19 (Gignac & Szodorai, 2016) and Funder and Ozer (2019) refer to r = .20 as the typical effect size in research. Thus, setting our smallest effect size of interest to r = .20 allows us to detect effects in line with field norms. At a conservative 95% power and a literature-typical target alpha of .05, we aimed for at least N ≥ 350 to detect our effect. However, due to delays in data collection and concerns about the decline in quality of retrospective self-reports of routine as time moved on, we ended data collection after three months (before analyzing any data), with a sample size of N = 235. Whilst smaller than we had intended to collect, this sample size is sensitive to detecting r = .23, assuming the same power (95%) and alpha (5%) targets.

Participants were recruited through online advertisements such as on the social media platforms (Instagram and Facebook) using the social networks of research assistants and other study advertising websites. Recruitment was conducted during the March to June 2020 lockdown in the UK. We targeted undergraduate students from different UK universities to ensure variation in institutions and geography whilst maintaining the cultural norms of UK study culture. The final sample had a majority of participants who identified as female (n = 178), a mean age of 21.45 (SD = 5.02) and varied in their year of study (n_FirstYear = 90, n_SecondYear = 73, n_ThirdYear = 60, n_FourthYear = 12). Students were studying a variety of courses, but most common were courses with a psychology component (n = 92). Most students were meat-eaters (n = 192), with n = 12 pescatarians, n = 19 vegetarians, and n = 6 vegans. Most of the participants lived in private accommodation with partners or friends (n = 112) or university halls (n = 74) during typical term time and most changed their living situation to be living in their family home during the lockdown (n = 192), a significant change in place of residence (χ²(9) = 66.99, p < .001, V = .31).

Procedure and materials

The study was conducted online using Qualtrics. Participants first reported their demographic information and their routine before responding to the wellbeing questionnaires which were presented in a random order. Descriptive statistics for all variables can be found in Table 1.

Table 1. Descriptive Statistics for Wellbeing.

Variable	Mean (SD)	Skewness	Cronbach’s α
PROMIS measures
Depression	2.89 (1.02)	0.07	.94
Anxiety	2.96 (0.97)	−0.06	.93
Sleep	3.10 (1.02)	0.01	.92
EPSI measures
Body dissatisfaction	1.91 (1.11)	0.01	.91
Binge eating	1.38 (0.83)	0.51	.86
Cognitive restraint	1.54 (1.11)	0.58	.80
Purging	0.19 (0.37)	2.47	.88
Restricting	1.15 (0.99)	0.81	.89
Excessive exercise	1.30 (0.97)	0.45	.85
Negative attitudes toward obesity	0.84 (0.83)	0.93	.86
Muscle building	0.46 (0.51)	1.38	.73

Demographic information

After providing informed consent, participants were asked to report their age in years and their gender identity. They were then asked about their diet type (meat-eater, pescatarian, vegetarian or vegan), where they were living during lockdown and typical term time, the title of the degree they were studying, and their current year of study.

Routine behavior

To define our participants’ routines, we collected data on the timing of events that happen within a day and are repeated across days. To get an overall picture of a “routine” we drew on wake-up, to-bed and meal timings paced throughout the day. This allowed us to draw on multiple time points throughout the day for a better estimate of consistency of behavior, which we called “routine timings.” However, there are other activities in a routine which may not happen at specific times. The durations of times they spent doing activities like exercise, study, or work, which may not happen in such fixed structured ways, we recorded in our “routine durations” variable. This allowed the respondents more freedom with how they could report engagement in an activity which might be broken up by other events in the day.

Data on participants’ routine timings was collected by asking them to consider a typical Monday, Wednesday and Saturday during typical term time (specifically, they were asked to consider their routine during February 2020, i.e., pre-Lockdown) in a retrospective report and during the nationwide lockdown (with the example of their routine in April 2020) when university courses were running remotely. These days were chosen to be a diverse range of days on students’ social and study calendars. It was emphasized that we were not looking for an exact computed average of their routine but for “just a good estimate of your usual schedule.” For routine timings, participants reported the time (using a 24-h clock; HH:MM) they typically (i) would wake up, (ii) go to bed, (iii) have breakfast, (iv) lunch, and (v) an evening meal. All the mealtime options had a “do not have” reporting option. There were some issues with reporting timings in a 24-h clock format, which required cleaning. For example, a participant reported waking up at 07:00, going to bed at 23:30, eating breakfast at 07:30, lunch at 11:00 but an evening meal at 06:00. In cases where this happened, and only the cases where the timings were impossible and out by 24-h, three researchers agreed on adjusting the 24-h clock for consistency. For transparency, a complete list of these data edits can be found on the OSF page for this study. For analysis, the reported times were converted to “minutes after midnight.”

For the duration of routine activities (which might not have fixed timings), for each day and timing, participants reported how much time (in minutes) they spent doing (i) light (“activity that involves movement but does not greatly raise your heart rate,” examples included walking leisurely), (ii) moderate (“exercise that raises your heart rate and feels like work”) and (iii) vigorous (“activity that is active exercise and will leave you breathless”) exercise. Finally, they were asked how much time they would spend (iv) interacting with friends (“in person or via eHangouts”), (v) in lecture environments, (vi) conducting independent study (“non-timetabled university related work”) and (vii) doing paid work or volunteering.

All timing and duration data were cleaned for outliers (z > 3.00) before analysis.

Eating Pathology Symptoms Inventory (EPSI, Forbush et al., 2013)

This 45-item scale was used to measure a diverse range of maladaptive eating behaviors. It covers the following areas: Body Dissatisfaction (seven items); own body image, for example, “I did not like how my body looked.” Binge Eating (eight items); consuming large amounts of food e.g. “I ate when I was not hungry.” Cognitive Restraint (three items); resisting consumption, e.g. “I counted the calories of food I ate.” Purging (six items); expulsion after consuming food, i.e., “I used diuretics in order to lose weight.” Restricting (six items); reducing food intake such as “People would be surprised if they knew how little I ate.” Excessive Exercise (five items); compulsive high-intensity activity, “I felt that I needed to exercise nearly every day.” Negative Attitudes Toward Obesity (five items); “I thought that obese people lack self-control.” Finally, Muscle Building (five items) looked at a desire for a muscular body, for example, “I used muscle building supplements.” Participants were asked to respond on a scale from “never” (coded as 0) to “very often” (4). Sample and scale descriptive can be found in Table 1.

PROMIS emotional distress—anxiety short form (PROMIS-Anx, Cella et al., 2007)

This non-diagnostic tool is a seven-item scale assesses levels of anxiety over the previous seven days. The questions included items such as whether they had felt fearful or worried recently. They were asked to report on a scale from “never” (1) to “always” (5). Sample descriptive statistics can be found in Table 1.

PROMIS emotional distress—depression short form (PROMIS-Dep, Cella et al., 2007)

This eight-item scale assesses self-reported depression-like behavior over the last seven days (see Table 1 for descriptives). Responses are collected on a scale five-point scale from “never” (1) to “always” (5).

PROMIS sleep disturbance—short form (PROMIS-Sleep, Buysse et al., 2010)

Participants responded to this eight-item scale on disturbed sleep. The total response to the PROMIS-Sleep is the product of responses to seven-items asking about their experience of sleep issues on a scale of “not at all” (1) to “very much” (5). Items 5, 6, and 7 are intended for a frequency response (“never” to “always”) and it was in error that we did not change these response scales, however, the meaning is preserved using the “not at all” to “very much” scale and we retain this scale for analysis. The PROMIS-Sleep contains a further item based on the quality of sleep on a scale from “very poor” (0) to “very good” (4). See Table 1 for sample descriptive statistics.

Data and analyses

All our data, code and supplemental materials can be found on the Open Science Framework: https://osf.io/wqvcu/

This study collected five routine timing variables (times of the day) and seven routine duration variables (numbers of minutes) for the three days for a typical term time and lockdown. We further collected 11 wellbeing variables.

To describe participants’ routine, we also computed the number of hours participants spend out of bed (their time to bed minus their time of waking). Using this variable and the other routine variables, we report the size differences between the two time points (term time and lockdown time) on routine using indices of mean difference (Cohen’s d from “MOTE,” Buchanan, Gillenwaters, Scofield, & Valentine, 2019) and we additionally correlate the term time and lockdown timings to illustrate sample-level stability in routines beyond absolute change. We test for change in having a meal (1) or not (0) for each of the meal times during typical and lockdown time using McNemar tests (using base R) with the effect size Cohen’s g (using “rcompanion,” Mangiafico & Mangiafico, 2017). Tests for consistencies over days use intraclass correlations (for scale data) and Kuder-Richardson-20 (for binary data), computed using the “psych” package (Revelle, 2019).

For the main inferential tests, we correlate the two classes of routine change variables with the wellbeing variables. To avoid issues with inflated type I error where we conduct multiple tests (t tests and correlations), inference of notable correlations is drawn from meeting a conservative α = .001. We then use a linear model containing the PROMIS and EPSI items to predict change in routine, to investigate what facets of eating psychopathology are most strongly associated with larger change in routine. Where a change in routine is binary (such as eating meals/not) binomial linear models are used. Inference from linear models will be drawn from a literature typical α = .05.

Results

Routine timings

Table 2 describes the routine timings (expressed as minutes after midnight) and the minutes participants spend out of bed for each day during lockdown and term time. The values in this table (with the exception of minutes out of bed) can be expressed as times, for example participants were, on average across the three days, waking up at 09:15 (544.74/60 = 9.25 h past midnight) and going to bed at midnight (1449.32/60 = 24.15 or 00:09). Table 2 also demonstrates the significant consistency (Intraclass Correlation Coefficients; ICCs) in routine times across the week; a participant who wakes up earlier on Monday, is one who wakes up earlier on Wednesday and Saturday. This was the case in both term time and lockdown, albeit with lockdown having more consistency.

Table 2. Descriptive Statistics, Day-day Consistency (ICC) and Comparisons (Difference [d] and Correlation [r]) between Term Time and Lockdown Times of Activity and Minutes Spent Out of Bed.

	Term time			Lockdown			Comparison
Variable	Mean (SD)	N	Skew	Mean (SD)	N	Skew	d	r
Wake up time
Monday	507.66 (80.37)	231	0.21	575.5 (122.83)	229	0.12	−0.55*	.37*
Wednesday	536.35 (93.08)	230	0.19	580.04 (119.76)	230	0.09	−0.36*	.30*
Saturday	590.22 (106.47)	230	−0.20	600.67 (119.47)	223	−0.26	−0.11	.45*
	ICC= .61*			ICC= .95*
Bed time
Monday	1419.30 (91.28)	229	0.45	1461.18 (130.35)	228	0.32	−0.34*	.45*
Wednesday	1455.00 (116.27)	230	0.38	1457.14 (126.66)	224	0.33	−0.02	.35*
Saturday	1473.67 (114.14)	229	0.16	1471.08 (127.59)	222	0.21	0.02	.50*
	ICC= .79*			ICC= .97*
Minutes out of bed
Monday	912.37 (93.55)	228	0.08	880.93 (106.67)	228	0.07	0.30*	.46*
Wednesday	922.40 (113.81)	229	0.39	872.01 (105.01)	224	−0.06	0.39*	.29*
Saturday	882.76 (108.58)	228	0.32	866.05 (111.35)	220	0.17	0.18	.41*
	ICC= .66*			ICC= .93*
Breakfast time
Monday	545.40 (81.60)	161	0.63	625.66 (114.20)	159	0.50	−0.55*	.46*
Wednesday	572.44 (95.22)	160	0.31	621.60 (114.76)	150	0.53	−0.35*	.54*
Saturday	611.15 (100.77)	148	−0.29	644.27 (100.88)	143	0.47	−0.26*	.62*
	ICC= .66*			ICC= .95*
Lunch time
Monday	792.84 (51.37)	208	0.89	815.62 (79.12)	203	0.85	−0.25*	.30*
Wednesday	802.99 (78.41)	206	0.76	812.99 (77.31)	201	0.90	−0.13	.39*
Saturday	824.04 (79.28)	203	0.75	820.73 (74.82)	193	0.85	0.02	.48*
	ICC= .62*			ICC= .94*
Evening meal time
Monday	1135.11 (79.35)	227	0.97	1148.24 (79.40)	222	0.49	−0.16	.37*
Wednesday	1121.30 (81.35)	223	0.25	1145.48 (82.70)	219	0.26	−0.29*	.36*
Saturday	1159.55 (83.48)	224	0.69	1153.98 (81.83)	221	0.42	0.05	.40*
	ICC= .78*			ICC= .94*

Note: ICCs are two-way random effects intraclass correlation coefficients on the three days for that variable.

*Indicates significance at conservative p< .001.

As can be seen in the n values in Table 2, many participants did not report eating breakfast, and some did not report eating lunch or an evening meal. Table 3 focuses on this in more detail, reporting the percentage of participants eating the three meals across the three days during term time and lockdown. Using a binomial linear model, we investigated if the EPSI domains predicted routinely eating breakfast (reported a breakfast time for all six occasions n = 100, 42.55% of sample). This model only weakly, albeit significantly, explained variance in routine breakfast eating (R²_Adj = .05, p = .011). This was explained by the weak negative predictor of Restricting behavior (estimate = −.08, p = .045) with all other predictors being weaker and nonsignificant (estimates ≤ | .08 |, p = .053). This suggests that many individuals do not eat breakfast but this is not related to problematic beliefs about eating.

Table 3. Percentage of Participants Eating Breakfast, Lunch, and Dinner on the Three Days during Term Time and Lockdown with Tests of Consistency (KR20) and Difference (McNemar’s Test).

Has meal	Term time (%)	Lockdown (%)	McNemar’s χ^2 (Cohen’s g)
Breakfast
Monday	68.51%	68.09%	0.00 (−0.01)
Wednesday	68.94%	64.68%	1.13 (−0.07)
Saturday	62.98%	62.55%	0.00 (−0.01)
	KR20= .78	KR20= .92
Lunch
Monday	90.64%	87.23%	1.53 (−0.13)
Wednesday	89.36%	87.23%	0.52 (−0.08)
Saturday	87.66%	82.98%	2.56 (−0.14)
	KR20= .66	KR20= .85
Evening meal
Monday	97.87%	94.47%	6.13 (−0.50)
Wednesday	95.32%	94.04%	0.31 (−0.12)
Saturday	95.32%	95.32%	0.00 (0.00)
	KR20= .83	KR20= .83

Notes: No McNemar tests were significant at p< .001. KR20 is the Kuder-Richardson test for consistency in binary data across the three days.

Change in routine timings

In general, there were small, significant differences in the routine timings between term time and lockdown. Participants’ routines started and finished later (see d, Table 2). However, the magnitude of difference in routine change was notably small, with an average absolute d value of change of M_|d| = 0.24. Furthermore, there was consistency in the order of participants’ routine timings (see r, Table 2). That is, the participants who were waking up later in term time were those waking up later in lockdown. Overall, participants showed consistent behavior over the days and between term time and lockdown. We computed absolute (net) change in routine scores as a measure of routine disruption for each timing per each day (i.e., absolute result of Termtime-Monday Evening Meal minutes after midnight—Lockdown-Monday Evening Meal minutes after midnight). These 18 (change for six variables across three days) variables were evaluated for consistency using intraclass correlation coefficients, and it was found that participants whose routine was disrupted on one timing was highly likely to experience disruption at another timing (ICC = .85, p < .001). With this information, we retained the average of these absolute changes in timing to create a Routine Timing Change variable. This variable was highly skewed (skew = 2.15) due to few high cases of routine change (Max = 396 min, Z = 6.73). Outliers with Z > 3.00 were cleaned from the computed variable (n = 4), leaving the Routine Timing Change variable with a mean of 82.73 (an average of 1 h and 23 min disruption) and a standard deviation of 36.70 min (skew = 0.69).

Table 3 shows no notable change in whether participants ate the selected meals or not between term time and lockdown. Participants were consistent in eating breakfast or not across the three days. It should be noted that, on average, across the times and days, only 65.96% of the sample report eating breakfast compared to an average of 87.52% reporting eating lunch and 95.39% reporting eating an evening meal.

Routine timings and mental wellbeing

All 396 possible pairwise correlations between routine timings across each day during term time and lockdown for all mental wellbeing variables can be found in Supplemental Table 1 (accessible at: https://osf.io/wqvcu/). The overall pattern is a clear absence of evidence for a relationship between the psychological variables and timing of routine elements for term time or lockdown on each day of the week.

To assess the relationship between the amount of routine change and the mental wellbeing of participants, we built a linear model where Routine Timing Change was predicted by the PROMIS and EPSI factors (see Table 4). Overall, the linear model significantly explained a small proportion of variance in Routine Timing Change. This was primarily explained by the EPSI Restricting domain (see Table 4), with participants who report eating less food experiencing more disruption between their term time and lockdown routines. There was weak evidence that Muscle Building-oriented diet behavior was also related to increased routine disruption. None of the PROMIS scales or the other EPSI variables was significantly related to routine change. None of the pairwise correlations met our conservative multiple correction criteria for significance, and all effects were small (all r ≤ | .18 |).

Table 4. A Linear Model Predicting Routine Timing Change Simultaneously Using Mental Wellbeing Variables and Pairwise Correlations between Wellbeing Variables and Routine Change.

	Routine timing change
Variable	Linear model estimate (exact p)	Correlation (exact p)
	R^2Adj= .07, p= .006
PROMIS measures
Depression	0.29 (.943)	.15 (.026)
Anxiety	−0.49 (.894)	.08 (.241)
Sleep	3.42 (.245)	.16 (.015)
EPSI measures
Body dissatisfaction	0.22 (.944)	.08 (.230)
Binge eating	6.00 (.102)	.15 (.020)
Cognitive restraint	−3.80 (.199)	−.09 (.190)
Purging	−4.07 (.574)	.04 (.573)
Restricting	8.58 (.004)**	.18 (.008)
Excessive exercise	−1.51 (.631)	−.01 (.905)
Neg. att. tow. obesity	−0.47 (.890)	.08 (.217)
Muscle building	11.62 (.030)*	.18 (.008)

Notes. No correlation effects were significant at the conservative p< .001

*p< .05, **p< .01 for linear model predictors only

Routine duration

The descriptive statistics and consistencies of the routine duration variables can be found in Table 5. In general, these were highly positively skewed distributions, with most reports being of little to no minutes on each activity on those days. The lockdown variables were further skewed toward no minutes of activity. For further analysis, we combined the durations in groups of similar behaviors. This included a total weekly minutes spent exercising (total minutes of light, moderate, and vigorous exercise across the three days) during term time (M_TermExercise = 333.04, SD = 194.07) and lockdown (M_{LockdownExercise} = 225.63, SD = 182.24). Similarly, the total minutes on nonsocial obligations (total minutes of work/volunteering, class work and study time across the three days) was computed for term-time (M_{TermObligation} = 690.12, SD = 338.14) and lockdown (M_{LockdownObligations} = 340.88, SD = 358.17). Finally, the total minutes of time spent with friends per week were computed for term time (M_TermFriends = 593.67, SD = 384.94) and lockdown (M_TermFriends = 241.20, SD = 271.62).

Table 5. Descriptive Statistics, Day-day Consistency (ICC) and Comparisons (Difference and Relationship) between Term Time and Lockdown Times of Activity and Minutes Spent Out of Bed.

	Term time			Lockdown			Comparison
Variable	Mean (SD)	N	Skew	Mean (SD)	N	Skew	d	r
Light exercise
Monday	61.46 (45.45)	230	1.03	45.21 (48.19)	230	1.74	0.32*	.27*
Wednesday	64.43 (58.27)	228	1.86	41.41 (41.77)	228	1.40	0.34*	.20
Saturday	73.32 (69.94)	225	1.54	49.38 (53.56)	226	1.62	0.32*	.31*
	ICC= .76*			ICC= .89*
Moderate exercise
Monday	30.54 (30.67)	228	1.28	27.37 (42.65)	231	2.85	0.07	.16
Wednesday	35.87 (39.63)	231	1.52	20.59 (27.35)	232	1.59	0.32*	.17
Saturday	33.28 (43.63)	227	1.54	21.64 (30.90)	230	1.61	0.32*	.31*
	ICC= .57*			ICC= .76*
Vigorous exercise
Monday	22.76 (33.85)	230	1.61	13.76 (21.48)	229	1.52	0.22*	.13
Wednesday	24.63 (36.19)	230	1.52	12.87 (21.30)	226	1.68	0.28*	.22*
Saturday	15.95 (29.15)	227	2.07	10.67 (19.12)	225	1.75	0.15	.33*
	ICC= .59*			ICC= .87*
With friends
Monday	194.36 (139.81)	233	1.08	70.80 (93.54)	230	2.12	0.78*	.15
Wednesday	200.97 (155.88)	232	0.71	81.54 (99.81)	232	1.93	0.72*	.26*
Saturday	200.66 (169.69)	229	0.96	89.57 (102.35)	229	1.80	0.64*	.26*
	ICC= .81*			ICC= .91*
Class time
Monday	177.97 (116.80)	232	1.02	15.66 (35.18)	225	2.17	1.33*	.14
Wednesday	12.97 (32.29)	225	2.49	12.97 (32.29)	225	2.49	0.66*	.21
Saturday	1.61 (7.97)	227	5.60	0.86 (5.49)	225	7.00	0.06	.36*
	ICC= .06			ICC= .59*
Study time
Monday	101.82 (73.58)	226	1.42	107.47 (113.23)	231	1.02	−0.05	.27*
Wednesday	91.82 (104.65)	228	1.18	91.82 (104.65)	228	1.18	−0.01	.19
Saturday	92.76 (89.98)	229	1.17	58.50 (90.44)	228	1.69	0.33*	.39*
	ICC= .63*			ICC= .83*
Work/volunteering time
Monday	33.25 (82.62)	225	2.97	50.81 (138.22)	233	3.03	−0.07	.06
Wednesday	18.04 (67.35)	223	4.27	18.04 (67.35)	223	4.27	0.25*	.12
Saturday	104.86 (179.54)	230	1.47	35.10 (119.03)	229	3.79	0.34*	.14
	ICC= .40*			ICC= .85*

Note: ICCs are two-way random effects intraclass correlation coefficients on the three days for that variable.

*Indicates significance at conservative p< .001.

Change in routine durations

Table 5 reports the changes in numbers of minutes between term time and lockdown for the individual reported variables. The pattern was consistent with the routine timings; consistent durations of activity with small, yet significant, differences in amounts of minutes. Participants were doing less of all variables, and those who did more of an activity in term time were those who did more during lockdown. This was reflected in the weekly total measures too. There were small decreases (and moderate consistencies) between term time to lockdown in the minutes of exercise per week between term time (t(234) = 6.40, p < .001, d = 0.42, r = .43, p < .001) and the number of minutes spent on nonsocial obligations (t(234) = 11.28, p < .001, d = 0.74, r = .43, p < .001). There was a larger decrease and less consistency in the weekly amount of time spent with friends (including in online hangouts) in term time and lockdown t(234) = 12.86, p < .001, d = 0.84, r = .25, p < .001). For further analysis, absolute change scores between term time and lockdown were retained.

Routine durations and mental wellbeing

As reported in Table 6, there were very few correlations between the mental wellbeing variables and durations of routine activities. The notable correlations meeting a conservative significance level were the relationship between increased minutes of exercise and self-reported EPSI Excessive Exercise in both term time and lockdown. This suggests that relationships between compulsive exercise engagement and minutes of activity persisted into the lockdown. Additionally, participants working less during lockdown reported higher depression scores. During term time, those who scored higher on the EPSI Purging reported spending less time with their friends. Overall, the message of these correlations is limited relationships between mental wellbeing and reported minutes of activity per week.

Table 6. Pairwise Correlations between Weekly Routine Duration Summary Variables and Wellbeing.

	Correlations for term time / lockdown
Variable	Weekly exercise	Weekly obligations	Weekly time with friends
PROMIS measures
Depression	−.05 / −.13	−.02 / −.25*	−.14 / −.03
Anxiety	−.01 / −.07	.08 / −.08	−.08 / −.05
Sleep	−.11 / −.15	−.01 / −.10	−.09 / −.10
EPSI measures
Body dissatisfaction	.06 / −.07	.09 / −.07	−.13 / −.06
Binge eating	.02 / −.12	.15 / −.12	−.06 / .03
Cognitive restraint	.13 / .11	−.01 / .15	−.17 / −.13
Purging	.03 / −.02	−.04 / −.09	−.27* / −.17
Restricting	.08 / −.10	−.09 / −.11	−.01 / .10
Excessive exercise	.30* / .33*	.10 / .03	−.02 / .09
Neg. att. tow. obes.	.24* / .02	−.04 / .01	.03 / .10
Muscle building	.11 / .06	.07 / .07	.01 / −.01

*Is significant at a conservative p< .001.

Linear models were built to assess the relationship between the magnitude of absolute change in routine durations and the mental wellbeing variables. A model using the PROMIS and EPSI domains to predict magnitude of exercise routine change explained a significant amount of variance (R²_Adj = .10, p < .001), with the PROMIS Anxiety score being the strongest predictor (estimate = 51.00, p = .004). This suggests that the anxiety reported by participants was most reflective of the magnitude of change they had experienced in their routine exercise duration. The EPSI Muscle domain weakly positively explained the magnitude of change in this model as well (estimate = 43.48, p = .049). All other predictors in this model were weaker (estimates ≤ | 29.71 |, p ≥ .057).

Another model using the mental wellbeing variables did not significantly explain variation in magnitude of change in minutes of obligations (R²_Adj = .05, p = .059). As was the case for the relationship between the mental wellbeing variables and magnitude of change in spending time with friends (R²_Adj = .01, p = .603)

Discussion

The current study investigated the relationship between mental wellbeing and the magnitude of change in routine between typical term time and lockdown in a UK student population. The results suggest that students’ routines (in terms of duration and exact timings) did change, but these effects were small and ordinally consistent (those who woke up earlier relative to others continued to do so). Mental wellbeing variables (depression, anxiety, sleep and eating) were largely unrelated to routine timings or routine durations. Furthermore, the magnitude of change in routine was largely unrelated to mental wellbeing, with some exceptions. First, participants who reported more restricted calorific intake and those who reported desiring a more muscular build were those who experienced more disruption in the timing of their routines. Second, participants who experienced the greatest change in amount of exercise were those who were experiencing the most anxiety at the time of their self-reporting.

Our findings align with other recent papers demonstrating a shift in routine timing for those in education. Much like Cellini et al. (2021) and Gruber, Saha, Somerville, Boursier, and Wise (2020) found in their school samples, our sample was, on average, experiencing a 1 h and 23 min later routine. Here, the students were reducing their “out of bed” hours, but concrete elements of their routine, such as eating meals, did not change as a function of lockdown. There were also stable individual differences in routine change, that is those who started their routines earlier during term time were those starting earlier in lockdown. Across the time, we observed that a notable number of participants were not eating some meals, particularly breakfast. We found that this was unrelated to ESPI scores, suggesting that future research into routine structure could look at the reasons why people do not include breakfast as part of their routine beyond mental wellbeing.

We were limited by self- (and for term time, retrospective-) reports on routine timings. Despite encouraging participants to check their calendars, it is possible that they did not accurately recall their routine. This could be improved through the use of momentary assessment research techniques to follow participants in real-time during potential future lockdowns. We also did not consider the phenomenological impact of these small changes in routine. It might be the case that participants are experiencing a change in their mental health beyond formalized scales of mental wellbeing. In addition, our sample largely consisted of participants who identified as female and were psychology students. We did not record the ethnicity, nationality, or location of our sample either. Hence, there are limitations on the generalizability of our sample to the wider student base in the UK.

The broad pattern of a lack of relationships between mental wellbeing and change in routine was surprising. If participants were at least demonstrating bias or constancy in self-reporting, we could expect some correlations between the sleep timing variables and the PROMIS sleep measure, for example. However, this was not the case in our data as, by and large, the participants who scored higher on the wellbeing scales were not reporting any more or less routine change. As such, our work is similar to the findings that suggest only minimal impact on mental wellbeing due to lockdown (i.e., Fried et al., 2020). Future research could better study resilience factors in student populations to understand how disruption to routine was not having a large impact on wellbeing. In our models we did find that anxiety was the strongest predictor of experiencing more routine change in exercise, indicating the importance of physical activity for managing mental wellbeing. This is similar to other recent large scale research which has indicated that declines in physical activity were related to poorer mental health indicators (Faulkner et al., 2021). Further research could focus on these specific effects and understand the relationship between access to places for exercise and mental health during lockdowns. We did not ask for detail on the location and style of exercising that our participants typically engaged in, and this could also be addressed in dedicated future research as these findings have important policy implications.

Conclusion

In this study we found that students’ routines were different yet consistent within individuals between term time and lockdown. Furthermore, we found no overall relationship between routine (and routine change) and mental wellbeing. However, more research is needed to look at the detail and experience of these routiness. Longitudinal ecological momentary assessment data may be useful in future studies of how future lockdowns may affect routines. Moreover, our research only investigates routines during three days in a week rather than looking at variance over longer periods of time. Our findings are similar to other recent research showing that those who experienced more disruption in amount of exercise were those who were more anxious, perhaps highlighting more consideration for the impacts on wellbeing from disrupted access to exercise facilities.

Acknowledgments

This study is made possible thanks to the contributions of data collectors. Additional thanks go to: Mahboob Miah, Dionne Wallace, Sian Valentine and, Simon Wood. This is a First Real Interactions and Engagement in Naturalistic Designs (FRIENDs) lab collaboration.

Data availability statement

Author note and open science practice: All our data and code can be found on the Open Science Framework: https://osf.io/wqvcu/

Additional information

Funding

The author(s) reported there is no funding associated with the work featured in this article.