Impact of Physical Activity on Psychological Distress: A Prospective Analysis of an Australian National Sample

Abstract

Objectives. We analyzed the individual-level associations between participation in moderate to vigorous physical activity (MVPA) and psychological distress levels using a large, nationally representative, longitudinal sample and multivariable panel regression models.

Methods. We used 3 waves of panel data from the Household, Income and Labour Dynamics in Australia Survey, consisting of 34 000 observations from 17 000 individuals and covering 2007, 2009, and 2011. We used fixed-effects panel regression models accounting for observable and unobservable confounders to examine the relationships between the weekly frequency of MVPA and summary measures of psychological distress based on the Kessler Psychological Distress Scale.

Results. We found substantial and highly statistically significant associations between the frequency of MVPA and different indicators of psychological distress. Frequent participation in MVPA reduces psychological distress and decreases the likelihood of falling into a high-risk category.

Conclusions. Our findings underscore the importance of placing physical activity at the core of health promotion initiatives aimed at preventing and remedying psychological discomfort.

Moderate to vigorous physical activity (MVPA) is important to people’s lives, with the World Health Organization as well as national and international bodies recommending frequent participation in it.^1,2 Recent analyses of Australian population-level data have endorsed this by showing that MVPA is independently associated not only with general and physical health but also with overall levels of mental health and self-reported life satisfaction.³ However, the relationships between MVPA and other facets of mental health, including levels of psychological distress, have not yet been well established. Psychological distress, understood as the experience of unpleasant feelings or emotions that affect day-to-day functioning, affects a sizable share of the population in developed countries such the United States, the United Kingdom, and Australia^4–7 and is known to lead to more severe mental disorders and physical health issues.^8,9 Consequently, the financial and human costs of psychological distress are non-negligible, and gaining a deeper understanding of the factors that influence individuals’ distress levels is important for the development of efficient public health policies and the devising of effective palliative interventions.

Emerging evidence of an association between MVPA and overall levels of mental health has suggested that associations between MVPA and psychological distress are also likely. Potential effects may run through known physiological, psychological, and social processes. From a physiological point of view, we know that MVPA enhances fitness levels, which in turn regulate physiological stress responses, such as reduced secretion of hormones and lowered blood pressure.¹⁰ From a psychological perspective, MVPA has been linked to reduced arousal and mood enhancement through cognitive distraction and biochemical changes, and to positive health behaviors during periods of high stress (e.g., a lower likelihood to smoke and eat unhealthily).¹¹ Additionally, participation in MVPA tends to increase time spent outdoors, as well as the frequency and quality of social interactions and interpersonal relationships.^12–14 As a result, we would expect MVPA to have the potential to enhance well-being by reducing psychological distress.

Consistent with these theories, results from the limited body of existing empirical research have suggested that there are indeed negative associations between the frequency of MVPA and psychological distress levels. However, these findings have emerged almost exclusively from analyses of small nonprobability samples,^15–21 and the few available studies based on nationally representative samples are cross-sectional (e.g., Scotland²² and Singapore²³). The small, nonprobability nature of the samples used in these studies means that findings are tentative and cannot be generalized to the population as a whole. Their cross-sectional nature means that longitudinal regression techniques that enable more precise estimation of the associations of interest by examining within-individual change over time and minimizing omitted variable bias attributable to unobservable factors cannot be implemented. In fact, undertaking large-scale prospective analyses is often regarded as a necessary step forward in enhancing current knowledge of the associations between MVPA and psychological distress.^6,23

In this article, we fill this gap and add to the literature by establishing the population-level associations between the weekly frequency of MVPA and self-reported levels of psychological distress with a nationally representative Australian panel data set and fixed-effect (FE) panel regression models.

METHODS

Our data set of choice, the Household, Income and Labour Dynamics in Australia (HILDA) Survey, is a large-scale, nationally representative panel survey that collects annual information from the same respondents. Twelve waves of data covering 2001 to 2012 are currently available, with low attrition rates.²⁴ This data set is very useful in examining the relationships between MVPA and psychological distress for 2 reasons: (1) it features a remarkably large sample that is representative of the Australian population, and (2) it contains repeated measures over time of the variables of interest (MVPA, psychological distress, and contextual factors), which allows for more elaborated statistical analysis via assessment of within-individual changes. Information on psychological distress has been collected in waves 7 (2007), 9 (2009), and 11 (2011) of the HILDA Survey. Hence, our analyses are restricted to these 3 time points.

Measures

The key independent variable was the weekly frequency of MVPA, as reported by survey participants. This variable contains responses to a question in a self-completion questionnaire: “In general, how often do you participate in moderate or intensive physical activity for at least 30 minutes? Moderate level physical activity will cause a slight increase in breathing and heart rate, such as brisk walking.” Respondents can choose 1 of the following 6 answers: “not at all,” “less than once a week,” “1 or 2 times a week,” “3 times a week,” “more than 3 times a week (but not every day),” and “every day.” We used this information to derive a set of dummy variables capturing the frequency of MVPA undertaken by survey participants.

Our outcome of interest was self-reported levels of psychological distress, operationalized using the well-established Kessler Psychological Distress Scale.^25,26 This scale consists of a battery of 10 questions designed to capture nonspecific psychological distress and measure depressive symptoms and anxiety disorders. Respondents are asked how often in the past 4 weeks they had experienced different feelings and emotions, including feeling

1. “tired for no good reason,”

2. “nervous,”

3. “so nervous that nothing could calm you down,”

4. “hopeless,”

5. “restless or fidgety,”

6. “so restless that you could not sit still,”

7. “depressed,”

8. “that everything was an effort,”

9. “so sad that nothing could cheer you up,” and

10. “worthless.”

Possible responses are rated on a 5-point Likert scale (all the time, most of the time, some of the time, a little of the time, and none of the time) and can be combined into more informative summary measures. First, reversing and adding scores for the 10 survey items gives an additive index ranging from 10 to 50 known as the K10. Second, the K10 scores can be used to separate the population into 4 risk groups: scores of 10 to 15 take the value 1 (“low risk”), scores of 16 to 21 take the value 2 (“moderate risk”), scores of 22 to 29 take the value 3 (“high risk”), and scores of 30 or higher take the value 4 (“high risk”).²⁷ The resulting outcome variable is an ordered measure with scores ranging from 1 to 4.

Statistical Analyses

To explore the association between the frequency of MVPA and psychological distress, we exploited the panel structure of the HILDA Survey data and estimated within-group FE panel regression models. These models use the repeated observations from the same individuals over time to account for unobserved person-specific factors that might confound the associations and minimize omitted variable bias. Therefore, FE models provide a better picture of the associations between MVPA and psychological distress than is possible in cross-sectional regression.^28,29 Note that it is not possible to retrieve the effect of time-invariant explanatory factors such as gender and socioeconomic background on the outcome variable using FE models, although they are implicitly accounted for by the model.

To model the first outcome variable, the K10 scale, we fit linear FE models for continuous dependent variables. These models estimate how deviations from individuals’ usual behavior and characteristics associate with deviations from their usual outcomes (captured by the individual mean scores in these over time). These take the form

where subscripts i and t denote individual and time; K10 is the measure of psychological distress of interest; MVPA is a set of dummy variables capturing the weekly frequency of physical activity; X is a vector of control variables; β and γ are vectors of coefficients; and e is the usual stochastic error term in regression. The X vector of control variables includes variables known or suspected to influence psychological distress. These variables include the respondent’s age at last birthday measured in years; highest educational qualification ever attained (“university qualification,” “professional qualification,” “school year 12,” “below school year 12”); gross yearly personal income adjusted for inflation using Consumer Price Indices; respondent’s body mass index, calculated as mass in kilograms divided by the square of height in meters; whether the respondent lives alone (“yes,” “no”); whether the respondent currently smokes cigarettes (“yes,” “no”); the respondent’s frequency of sex-based excess alcohol drinking during the past 12 months (“sometimes (less than once a month),” “once a month,” “several times a month,” “never”); whether the respondent reports having a long-term health condition, impairment, or disability that restricts everyday activities (“yes,” “no”); the respondent’s employment status (“employed,” “unemployed,” “not in the labor force”); and the respondent’s number of total weekly work hours, measured as the sum of the usual weekly hours worked in all jobs and the usual weekly hours of domestic labor.

To model the risk categories extracted from the K10, an FE model for ordered outcomes is required, but the literature has only recently provided solutions to achieve this. Two competing specifications have been proposed: the person-specific threshold, ordered FE logit model by Ferrer-i-Carbonell and Frijters³⁰ and the blow-up and cluster, ordered FE logit model by Baetschmann et al.³¹ None of these is readily available in standard statistical packages, and we thus programmed them ourselves. Because there is not yet consensus as to which model is preferable, we fit and discuss the results of both estimation strategies.

In broad terms, the person-specific threshold-ordered FE logit strategy consists of creating a binary variable out of the original ordered variable using a person-specific threshold: Risk values that are equal or higher than the person-specific over-time mean take the value 1, and risk values that are lower than the person-specific over-time mean take the value 0. A binary logistic FE model (also known as a conditional logit model) is then fitted to the resulting dichotomous variable. In the blow-up and cluster, ordered FE logit strategy, one first creates an expanded data set by multiplying every data row as many times as there are potential dichotomizations of the original ordered variable and applying a different dichotomization to each set of duplicated observations. In our case, there were 3 possibilities: (1) risk level 1 takes the value 0, and risk levels 2 to 4 take the value 1; (2) risk levels 1 to 2 take the value 0, and risk levels 3 to 4 take the value 1; and (3) risk levels 1 to 3 take the value 0, and risk level 4 takes the value 1. Again, a binary logistic FE model is fitted to the resulting dichotomous variable, adjusting the standard errors to account for the artificial duplication of rows. For detailed, formal derivations of these estimators, see Ferrer-i-Carbonell and Frijters³⁰ and Baetschmann et al.,³¹ respectively. Despite the complex properties of these methods, the estimated model parameters can be transformed into odds ratios and interpreted quite simply.

RESULTS

The analytical sample of the HILDA Survey in 2007, 2009, and 2011 consisted of 33 918 observations from 17 080 individuals, and thus respondents were observed 1.98 times on average. Descriptive statistics for all variables can be found in Table 1. In 10% of all observations, individuals reported doing no MVPA at all; in 15%, less than once a week; in 24%, once or twice a week; in 16%, 3 times a week; in 22%, more than 3 times a week (but not every day); and in 13%, every day.

TABLE 1—

Descriptive Statistics: Household, Income and Labour Dynamics in Australia Survey, 2007, 2009, and 2011

Variable	Mean (SD) or %	Min.	Max.
Frequency of moderate or intensive physical activity
Not at all	10	0	1
< 1 time/wk	15	0	1
1 or 2 times/wk	24	0	1
3 times/wk	16	0	1
> 3 times/wk (but not every day)	22	0	1
Every day	13	0	1
K10 summary scale	15.5 (6.0)	10	50
K10 risk categories
Low risk	65	0	1
Moderate risk	21	0	1
High risk	10	0	1
Very high risk	4	0	1
Age, y	44.2 (18.1)	15	97
Highest educational qualification
< year 12	24	0	1
Year 12	31	0	1
Professional qualification	16	0	1
University qualification	29	0	1
Gross yearly personal income, $	88 000 (64 000)	0	1 550 000
Body mass index	26.6 (5.7)	12	171
Long-term condition or disability	26	0	1
Current smoker	20	0	1
Sex-based excessive alcohol drinking, past y
Sometimes (< 1 time/mo)	46	0	1
Once a month	24	0	1
Several times a month	10	0	1
Never	20	0	1
Lives alone	15	0	1
Employment status
Employed	66	0	1
Not in the labor force	30	0	1
Unemployed	3	0	1
Weekly work hours, paid and unpaid	35 (21)	0	160

Note. K10 =  Kessler Psychological Distress Scale; Min. = minimum; Max. = maximum. The sample size was n = 17 080, with 33 918 person-year observations.

Raw relationships between this variable and indicators of psychological distress are presented in Figures 1 and 2. Mean responses to each of the 10 items constituting the K10 scale increased with the weekly frequency of MVPA (Figure 1; high scores mean absence of distress). Consequently, the overall scores in the K10 index and the prevalence of higher risk categories decreased with the frequency of MVPA (Figure 2; high scores represent worse outcomes).

FIGURE 1—

Means for items of the Kessler Psychological Distress scale by frequency of moderate to vigorous physical activity: Household, Income and Labour Dynamics in Australia Survey data, 2007, 2009, and 2011.

Note. K10 = Kessler Psychological Distress Scale; MVPA = moderate to vigorous physical activity. Scores relate to the experience of feelings; responses are rated on a 5-point Likert scale (1 = all the time, 2 = most of the time, 3 = some of the time, 4 = a little of the time, and 5 = none of the time).

FIGURE 2—

Frequency of moderate to vigorous physical activity by (a) prevalence of Kessler Psychological Distress scale risk categories and (b) mean scale score: Household, Income and Labour Dynamics in Australia Survey, 2007, 2009, and 2011.

Note. K10 = Kessler Psychological Distress Scale; MVPA = moderate to vigorous physical activity.

This is a preliminary indication that MVPA may influence psychological distress levels. More thorough examination of whether true relationships exist requires multivariable models that control for observable and unobservable confounding factors. FE panel regression models that can achieve this are presented in Table 2. Note that because these models are based on within-individual change over time, individuals who are observed only once do not contribute to model estimation. Similarly, those for whom no change in a certain explanatory variable is ever recorded do not contribute to the estimation of the model parameter associated with that variable.

TABLE 2—

Fixed-Effect Models on Summary Measures of the Kessler Psychological Distress Scale: Household, Income and Labour Dynamics in Australia Survey, 2007, 2009, and 2011

		Risk Categories
Model	K10, b (95% CI)	PST, OR (95% CI)	BUC, OR (95% CI)
Frequency of MVPA
Not at all (Ref)	0.00	1.00	1.00
< 1 time/wk	−0.41** (−0.68, −0.10)	0.84* (0.71, 1.00)	0.88* (0.78, 0.99)
1 or 2 times/wk	−0.83*** (−1.12, −0.53)	0.69*** (0.58, 0.82)	0.76*** (0.67, 0.85)
3 times/wk	−1.14*** (−1.44, −0.81)	0.60*** (0.50, 0.73)	0.68*** (0.59, 0.77)
> 3 times/wk (but not every day)	−1.42*** (−1.72, −1.09)	0.52*** (0.43, 0.63)	0.59*** (0.51, 0.67)
Every day	−1.79*** (−2.14, −1.42)	0.46*** (0.37, 0.57)	0.53*** (0.45, 0.62)
Age, y	−0.01 (−0.05, 0.01)	1.00 (0.98, 1.02)	1.00 (0.99, 1.02)
Education
< year 12 (Ref)	0.00	1.00	1.00
Year 12	−0.23 (−0.97, 0.74)	1.02 (0.62, 1.66)	1.05 (0.72, 1.53)
Professional qualification	−0.12 (−0.62, 0.66)	1.11 (0.80, 1.53)	1.04 (0.80, 1.34)
University qualification	−0.22 (−0.59, 0.31)	1.07 (0.82, 1.40)	0.97 (0.80, 1.18)
Gross personal income, y	−0.00 (−0.02, 0.01)	0.99 (0.98, 1.00)	0.99 (0.98, 1.00)
Body mass index	−0.01 (−0.03, 0.02)	1.00 (0.98, 1.01)	1.00 (0.99, 1.01)
Long-term condition	0.90*** (0.68, 1.07)	1.52*** (1.33, 1.73)	1.42*** (1.30, 1.55)
Current smoker	0.50** (0.17, 0.84)	1.21* (1.00, 1.46)	1.20* (1.04, 1.39)
Sex-based excessive alcohol drinking (past year)
Never (Ref)	0.00	1.00	1.00
Sometimes (< 1 time/month)	0.16 (−0.05, 0.36)	0.96 (0.84, 1.11)	1.00 (0.90, 1.11)
Once a month	0.27 (−0.03, 0.55)	1.12 (0.93, 1.36)	1.12 (0.98, 1.28)
Several times a month	0.47** (0.14, 0.76)	1.28* (1.05, 1.55)	1.22** (1.05, 1.40)
Lives alone	0.68*** (0.27, 1.03)	1.20 (0.98, 1.46)	1.23** (1.05, 1.44)
Employment status
Employed (Ref)	0.00	1.00	1.00
Not in the labor force	0.16 (−0.11, 0.47)	1.04 (0.87,1.24)	1.05 (0.92, 1.20)
Unemployed	−0.07 (−0.60, 0.45)	0.99 (0.77, 1.26)	0.97 (0.81, 1.17)
Weekly work hours, paid and unpaid	−0.05 (−0.11, 0.01)	0.96 (0.93, 1.00)	0.98 (0.95, 1.00)

Note. BUC = blow-up and cluster estimation method; CI = confidence interval; K10 = Kessler Psychological Distress Scale; MVPA = moderate to vigorous physical activity; OR = odds ratio; PST = person-specific threshold estimation method. The sample size was n = 17 080, with 33 918 person-year observations.

*P < .05; **P < .01; ***P < .001.

The first column of results in Table 2 shows the findings from a linear FE model of the K10 summary scale. The model coefficients in this specification give the expected change in the K10 index associated with a within-individual change in the explanatory variables. For MVPA, model coefficients give the difference in outcomes for the same individual at times when he or she falls into a given activity category and at times when he or she falls into the reference activity category (i.e., no MVPA at all). The results indicate that MVPA has a negative, strong, and statistically significant effect on overall psychological distress levels. When an individual participates in MVPA less than once a week, he or she, on average, scores 0.41 units lower (P < .01) on the K10 measure than when he or she engages in no MVPA at all. The analogous estimated effects are −0.83 units (P < .001) for being active once or twice a week, −1.14 units (P < .001) for being active 3 times a week, −1.42 units (P < .001) for being active more than 3 times a week (but not every day), and −1.79 units (P < .001) for being active every day.

The second and third columns of results in Table 2 present the findings from ordered FE models of the K10 risk categories. The model coefficients in these specifications, exponentiated to odds ratios, give the predicted odds of being in a higher risk category associated with a within-individual change in the explanatory variables. For MVPA, the associated odds ratios give the difference in the propensity to be in a higher category of the ordered outcome variable for the same individual at times when he or she falls into a given activity category relative to when he or she falls into the reference activity category (i.e., no MVPA at all). Results in column 2 are for models using the person-specific threshold, ordered FE logit specification. As for the linear model, MVPA has a negative, strong, and statistically significant impact on the risk of experiencing depression or anxiety disorders. When individuals participate in MVPA less than once a week, they have odds of falling into a higher risk category that are 16% lower (P < .05) than those they had when they engaged in no MVPA at all. The analogous estimated effects are −31% (P < .001) for being active once or twice a week, −40% (P < .001) for being active 3 times a week, −48% (P < .001) for being active more than 3 times a week (but not every day), and −54% (P < .001) for being active every day. Results in column 3 use an alternative estimation strategy—the blow-up and cluster, ordered FE logit specification—but are remarkably similar to those of the person-specific threshold, ordered FE logit specification.

Additional regression results that, for brevity, are not shown in the tables (but are available on request) indicated that MVPA is negatively related to all the separate psychological distress markers used to build the K10 summary measures. However, the magnitude of the negative association varied from item to item and was highest for “feeling tired out for no good reasons,” “feeling depressed,” and “feeling that everything was an effort” and lowest for those denoting “feeling so nervous that nothing could calm you,” “feeling restless or fidgety,” and “feeling so restless that you could not sit still.”

Results from all models clearly indicated that MVPA is strongly associated with psychological distress, but the same does not apply to other explanatory variables. We found no independent effects on psychological distress of age, education, income, body mass index, or weekly work hours. Living alone; having a long-term condition, disability, or impairment; excessive drinking; and smoking are, however, significantly related to poorer outcomes.

DISCUSSION

In this article, we have, for the first time in the literature to our knowledge, estimated the prospective, population-level associations between the frequency of MVPA undertaken by individuals and their psychological distress levels. We did so by exploiting the properties of a large-scale, nationally representative Australian panel data set. Our modeling specification, FE panel regression models, maximized the panel structure of these data to control for person-specific unobserved heterogeneity, thus enabling more accurate estimation of the relationships of interest than is possible in a cross-sectional framework.

Our results indicate that substantial and statistically significant associations exist between the frequency of MVPA undertaken by survey respondents and their self-reported psychological distress levels: Higher doses of MVPA unequivocally related to lower distress levels. This association emerged when approximating psychological distress by using either the K10 summary measure or its associated risk categories and also for each of the separate distress markers that composed these summary measures.

In addition, the predicted impacts of MVPA on psychological distress were among the most significant in the models. These results are consistent with findings in previous nonrepresentative, cross-sectional, or smaller studies.^13–21 They are also highly consistent with findings from another survey-based study conducted in Australia.⁶ The latter, however, focused only on older men and was of a cross-sectional nature. An additional interesting finding is that the dose–response relationship between MVPA and psychological distress was more pronounced at lower MVPA frequencies (none–3 weekly sessions) than at higher MVPA frequencies (4 weekly sessions–every day), which is in line with results from other longitudinal research^18,20 and from studies of the relationships between MVPA and overall mental health.³

There are, nonetheless, caveats to our research. First, as have others before us,^3,32 we acknowledge that there are issues with the questionnaire item used to capture MVPA in the HILDA Survey: it does not differentiate activity types, it does not specify exact minutes spent on activities, and it has not been validated via accelerometer or other techniques.³³ Second, we could only use the 3 waves of HILDA Survey data for which psychological distress measures were available. As a result, our FE regression models did not use information from many individuals who did not experience change over time in the variables of interest to estimate the relationships. Finally, reverse causation remains a plausible source of endogeneity: psychological distress may discourage people from engaging in MVPA.

Our findings nevertheless have important implications for health policy. In particular, results reported suggest that physical activity at a moderate to vigorous level might be used as a preventive or remedial coadjuvant treatment for individuals experiencing moderate or high levels of psychological distress or at risk for experiencing such symptoms.^34,35 Findings reported here also suggest that research gaps in our current understandings of the relationships between MVPA and psychological distress must be addressed so that more refined and effective public health policy levers can be devised. In particular, it is important to (1) establish which of the physiological, psychological, and social mechanisms outlined produce the observed associations between MVPA and psychological distress, and (2) further examine the prospective dose–response associations between the frequency of MVPA in different life domains (e.g., leisure time, work, transportation, and the domestic realm) and psychological distress levels.

In conclusion, our research yielded strong, generalizable evidence that MVPA is associated with psychological well-being and that frequent doses of MVPA reduce psychological distress levels and decrease the likelihood of falling into a high-risk distress category. Our findings thus underscore the importance of placing physical activity at the core of public health policies and health promotion initiatives aimed at reducing the prevalence and severity of psychological distress in the Australian population. The patterns we found for Australia are also likely to emerge in other developed countries such as the United States and the United Kingdom. Population-level prospective studies that apply our framework to confirm this are urgently required.