Soc Psychiatry Psychiatr Epidemiol (2001) 36: 348353

Steinkopff Verlag 2001

ORIGINAL PAPER

H. Sexton A. J. Sgaard R. Olstad

How are mood and exercise related? Results from the Finnmark study

Accepted: 26 April 2001

s Abstract Background: Recreational exercise and mood have frequently been correlated in population studies. Although it is often assumed that recreational exercise improves mood, this has not been consistently demonstrated in population studies. Method: The relationship between mood and exercise was studied prospectively in a community sample. A series of synchronous panel models was constructed in two samples (2798 paired observations; sample I =1219, sample II =1498) to examine this relationship in the entire population, for women and men separately, for those with sedentary occupations, for those performing physical labour, and for those who initially showed a more dysphoric mood. Results: Although mood and exercise were correlated, the only directional relationship that could be demonstrated was that recreational exercise had an inconsistently positive effect upon mood in those with sedentary occupations. There was no such relationship between doing physical work and mood. Analyses of those who initially showed higher levels of dysphoria did not uncover any directional relationship between mood and exercise. None of the other subgroups showed any directional effects between mood and recreational exercise, nor did the population as a whole. Conclusion: The relationship between exercise and mood in this population sample appears to be largely correlational in
H. Sexton () Psychiatric Research Centre for Troms and Finnmark RIT-sgard hospital 9005 Troms, Norway Tel.: +47-77 62 75 00 E-Mail: sexton@online.de A. J. Sgaard Center of Preventative Medicine Ullevl University Hospital, Oslo, Norway R. Olstad Department of Clinical Psychiatry Institute of Community Medicine University of Troms Troms, Norway
SPPE 468

nature. This result suggests the need to take a cautious view of the role played by exercise in promoting mood in the general population.

Introduction
Recreational exercise has often been associated with greater well-being, less depression and anxiety and a positive mood in cross-sectional studies of the general population. These effects, however, have generally been small in magnitude [1, 2].Although the findings are consistent with exercise having salutary psychological effects, the actual nature of the relationship between mood and exercise remains an issue [1, 3, 4]. It has been uncertain as to whether the relationship might be a purely correlational one, expressing, at least to some degree, other underlying factors such as the social environment [5], personality characteristics [6], a sense of mastery [7], general physical health [2], or general health habits [8]. The fact that physical fitness in itself has not always been associated with mental health in cross-sectional studies underlines this uncertainty [9]. Longitudinal studies of exercise and mood in the general population have been infrequent and, in these reports, more exercise has not been invariably associated with indicators of better mental health [10]. Although it is often assumed that exercise has a salutary effect upon mood, it is also plausible that a better mood might well result in more recreational activity or exercise. The direction of any possible causal relationship effects between mood and exercise in population samples is still unresolved [4]. Therapeutic exercise, however, has often demonstrated positive effects upon mood in individuals with medical disorders [1113]. Nonetheless, poor physical health may serve to confound the relationship between mood and exercise in the general population [4]. Factors such as age [1], gender [14], the persons initial mood [4] and the nature of the persons occupation may also in-

349

fluence the relationship between exercise and mental health [15]. Thus, for the reasons mentioned above, we suspected that the magnitude of any directional relationship between exercise and mood might be small. We, therefore, used a large population-based prospective data set to explore for directional relationships between exercise and mood. We examined the relationships between mood and exercise for: 1. The entire population, 2. Men and women separately, 3. Those individuals who reported carrying out physical work and those who had a sedentary job, and 4. Those reporting higher levels of dysphoria at baseline.

items for summer and winter), and the intensity of leisure activity (sedentary to rigorous physical activity) was ascertained. Although the items were used as indicators in the latent variable analysis used here, these three variables, when summed, did form an internally consistent scale (Cronbachs range = 0.700.76 at the three time points in this study). Type of work The degree to which the persons work involved physical activity (four categories, from sedentary to heavy physical labour) was ascertained. Health Global subjective health was assessed by four categories (excellent to very poor). As poor physical health might possibly confound any relationship between mood and exercise, those reporting very poor health were removed from the sample (4% and 4.9% of samples I and II, respectively) at baseline, prior to any analysis. s Data analysis The relationship among the variables was first explored with principal components analysis, and a subsequent varimax rotation of the components, as a preliminary step to the construction of structural models. For each sample, only those individuals who had answered the depression question at both time points were included in the modelling analyses. Missing values in the other relevant questions were, with one exception, minimal (less than 2 %), and were replaced by the mean value. There were a considerable number of individuals who had not responded to the question about sleep problems (yes/no) in the first data collection (1987), as a result of a weakness in the data collection routine. The resultant missing values were imputed using the EQS [19] routine, which allows for their estimation based on the other mood-related variables measured at the same time point. Construction of the latent variable panel models The modelling of latent constructs or factors requires the use of at least two measured variables (or indicators) of the latent construct. When, as here, the latent constructs rather than the individual items are of primary interest, it is a common practice to sum the individual items representing a particular concept into two or more item parcels. Item parcels often result in variables more in harmony with the assumptions made in these models (that the observed variables are normally distributed and continuous, while responses to the individual questions are, as here, often ordinal). The item parcels are then used as observed indicators of the latent construct (or factor) in the model. The two indicator variables for the latent variable, Mood, were constructed by summing the items related to mood, i. e. depression, sleep disturbance, and loneliness into one item parcel and summing current coping and current life satisfaction into another. For the indicators of the latent variable, Exercise, exercise frequency and exercise intensity were used as the two indicators of recreational exercise. Simultaneous (synchronous) panel models rather than crosslagged panel models were used, as the effect of exercise upon mood appears within a few months during clinical exercise programmes, while the time span between our data collections was 3 years. Therefore, simultaneous path models were deemed more likely to detect any directional effects between mood and exercise that might be present. In addition, preliminary analyses had not demonstrated any significant cross-lagged path coefficients. Models were constructed using structural equations modelling (Fig. 1) with maximum likelihood estimations and robust standard errors. Exercise and mood were allowed to freely correlate at time 1. The unrelated (error) variances of the indicators of mood and physical activity were also allowed to correlate from time 1 to time 2. The model fit criteria of the confirmatory fit index, CFI a frequently used measure of the adequacy with which the structural model represents the observed data and the overall model chisquare test were used to determine the adequacy, or fit, of the model.

Subjects and methods

s Sample The data base consisted of survey data collected at three time points (1987, 1990, and 1993) from the adults residing in five small rural communities in northern Norway. This data was a part of the Finnmark Study [16, 17]. The samples are described in detail in an earlier report [18]. In brief, all residents aged between 40 and 62 years and a random sample of persons aged between 20 and 39 were invited to participate in the surveys. Attendance rates were 77 %, 74 % and 70 % for each of the three surveys, respectively. The structural equations modelling programme, EQS 5.5 [19], that we used required complete data. We wanted to retain the largest possible number of observations in the data. Therefore, we chose to view the data base as consisting of two separate data collections of two waves each; sample I (1987 and 1990) and sample II (1990 and 1993). This allowed for the retention of considerably more paired observations than if we had only included those persons with complete data at all three time points (the requirement of a three-wave model). However, as would also be the case in a three-wave model, the two data collections do not represent independent samples. Rather, the data gathered in the two time periods was from many of the same individuals who had again responded to the re-survey (909 individuals responded at all three time points). About 4% reported poor health (range 4.04.9 %) and were removed from the sample at baseline (prior to any analysis). The first two data waves (sample I) then consisted of 1219 paired observations (individuals), while sample II comprised 1498 paired observations. s Variables Mood The variables measuring mood consisted of verbally anchored items, which assessed coping (three categories), loneliness (three categories) and current life satisfaction (four categories) at all three time points. Also assessed were depression and sleep problems (two categories each, yes or no) [16]. Three of the items had been taken and modified from the General Health Questionnaire (GHQ).A few of the GHQ items have been shown to adequately capture a large proportion of the total variance of the original instrument [20]. Although the latent variable modelling used in this study makes use only of the common variance of the items (indicators) describing the participants mood, the items, when summed, did form an internally consistent scale (Cronbachs range 0.680.70 at the three time points in this study). Exercise The frequency with which respondents exercised for at least 20 min was recorded (four categories, from seldom to daily, with separate

350 Models in which the CFI values were greater than 0.95 [21] and the chi-square of the overall model fit was not significant at the P < 0.01 level were judged to provide an accurate description or model of the observed data. Critical ratios, CR, (a z-test; the parameter divided by its standard error) were used to determine whether or not a path contributed significantly to the model. Men and women were first analysed separately, as it has been suggested that there could be sex-specific differences in the relationships between exercise and psychological distress [2224]. Finally, the data from both waves were combined to determine whether increasing the number of observations would demonstrate very small effects that were not otherwise significant. The population samples were also divided into those with sedentary work (no more than light physical activity at work) and those doing physical labour, and separate models were constructed for both groups. Our reasoning was that, if physical activity in itself influenced mood, then the effect of recreational physical activity (exercise) upon mood might be apparent only for those having sedentary work. We thought that physical work alone could possibly provide a sufficient amount of exercise to provide any mood-enhancing effects that might be inherent in exercise. If this proved to be the case, then recreational exercise might not have any additional influence upon mood among those with heavy physical work. It seems relevant to mention here that the social status of doing heavy physical work is generally high in this subarctic rural area, even though it has previously been found that the correlation between recreational exercise and mood persists even after adjusting for any socio-economic difference among the individuals studied, including work status [1, 7]. Lastly, we modelled separately the 20 % of those who initially reported the highest levels of depressive symptoms, in order to determine whether those with more initial distress responded differentially to recreational exercise than did the sample as a whole.

Results
s Sample characteristics
The two samples were quite similar in their demographic characteristics. In sample I, 48.5 % were women, mean age 48.4 (SD 8.9, range 2062) years, 73 % married or cohabiting, 14.8 % not married, and 11.3 % previously married. The frequencies of the responses to the key variables used in the modelling analyses varied slightly over the 6-year period, and are reported elsewhere [17]. Recreational exercise and the degree to which the participants work involved physical activity were fairly stable at all time points. Recreational exercise intensities and frequencies varied little between the data collection waves. Recreational exercise intensity reported for those who attended the survey in 1990 were: sedentary 19.1 %, light exercise 68.2 %, moderate exercise 11.6 %, and rigorous exercise 1.1 %. Recreational exercise frequencies were (second data collection, 1990): seldom/never 29.8 %, once a week 20.7 %, several times a week 38 %, and daily 11.5 % in the summer. Winter exercise frequencies were nearly identical with those reported in the summer.

s Modelling exercise and mood

The confirmatory fit indexes (CFI) and overall model scaled chi-square statistics indicated very satisfactory model fits for all the models that we tested. The variables
Fig. 1 Synchronous panel model for all paired observations (N = 2879). Path coefficients are standardized. Superscript f indicates a fixed path coefficient in the model. Standardized path coefficients > 0.09, P < 0.05 (z-test). All other models tested in this study have the same form. Mood is negatively scaled, i. e. a better mood has lower values

351 Table 1 Selected simultaneous effects panel models of mood and recreational exercise (see Fig. 1 for a diagrammatic representation of the models) Samplea nb Model 2 P (df) 21.61 (12) 15.63 (12) 17.34 (12) 16.39 (12) 20.61 (12) 13.21 (12) 0.5 0.2 0.1 0.2 0.05 0.4 CFIc Significant simultaneous path coefficients between mood and exercise none Exercise Better mood none Exercise Better mood none none Critical P ratio (z-test) 2.21 2.15 < 0.05 < 0.05

All observationsd Sedentary work, Ie Sedentary work, IIe Sedentary work, all Hard physical work, all More dysphoric, allf More dysphoric and sedentary workf
a

2879 1055 978 2033 846 564

1.0 1.0 0.997 0.998 0.992 0.998

404 14.28 (12) 0.3

0.995 none

There were no differences when men and women were analysed separately, nor was physical activity at work related to mood. All those reporting poor health were removed from the data set b Paired observations c Comparative fit index d All paired observations when Finnmark I and II were combined e Finnmark I and II, respectively; all those with work that involved at least considerable walking and lifting were removed from this subsample f The approximately 20 % reporting the highest initial levels of depressive symptoms were examined separately

that were used in the analyses were normally distributed (mean univariate kurtosis 0.18 and 0.15 for men and women, respectively). The robust standard errors did not differ substantially from the normally estimated standard errors. Model fits and the paths of primary interest are shown in Table 1. Levels of exercise and mood were relatively stable over a 3-year period, as attested to by the fairly high standardised path coefficients between the variables from time 1 to time 2 (Fig. 1). Mood and exercise were negatively correlated [r = 0.13, CR (z-test; parameter/SE) = 4.46, P < 0.0001] at time 1, i. e. more exercise was associated with a better mood. However, the models did not demonstrate any directional relationship between mood and recreational exercise when the entire sample was examined (all CR P > 0.05). This was true in both data waves, for men and women when they were analysed separately, as well as when all observations were combined (2879 observation pairs) and analysed together. There was no relationship between physical activity levels at work and mood for either men or women, when analysed either together or separately (model scaled 2 =3.07, df=4, P =0.5, CR between work activity levels and mood, P > 0.05 when all paired observations were analysed together). However, when only those with sedentary occupations were examined, recreational exercise exerted a small positive effect upon mood in sample I (CR = 0.177/0.08 =2.21, P < 0.05), but not in sample II.When both samples I and II were analysed together, the effect was significant (CR = 0.137/0.064 =2.15; P < 0.05). The effect was of small magnitude (standardised path coefficient 0.074). For those with heavy physical work, there was no significant relationship between recreational exercise and mood. The 20 % (n = 564) who registered a higher initial level of dysphoric mood were modelled separately in order to determine whether they responded more positively to higher levels of recreational exercise. Neither

this subgroup nor those with initial dysphoric mood and sedentary work (n = 404), showed any directional relationship between mood and recreational exercise.

Discussion
Exercise and mood were correlated in this moderately large prospective population-based survey. However, the evidence that regular recreational exercise positively influenced mood was limited. There was a suggestion that the mood of those with sedentary occupations might be positively related to recreational exercise, although the size of the effect was small and inconsistent. This was the only evidence of a causal or directional relationship between exercise and mood that we found. We had initially thought that recreational exercise might only show a positive effect on mood in those with otherwise sedentary life styles. That is, we had suspected those individuals who were engaged in physically active occupations might not derive additional benefit from more physical activity during their leisure time. There was no evidence that a better mood led to more recreational activity in this sample. The sample consisted of persons attending repeated general population health surveys. Selection biases in the samples are discussed elsewhere [18]. In brief, the sample who attended the repeated health surveys and conscientiously completed the questionnaires tended to have a higher proportion of married individuals than those who were initially asked to participate in the survey. The participation rate was also lower for the younger individuals in the communities (under 40 years old). Thus, the samples used here somewhat over-represent the more stable and older elements of the population. A limitation in this study is the quantity and quality of the exercise and psychological variables that were available for analysis. The availability of anxiety and other psychological measures as well as the inclusion of

352

more extensive physical activity measures would have served to improve the study. Nevertheless, the large number of observations available serve to compensate for these limitations by increasing the studys overall power. A second limitation in this data set is the 3-year time span between the data collections. A shorter interval would have been desirable, although the use of simultaneous path models serves to compensate for this weakness. The question arises as to why the results of population studies relating exercise and mood have been inconsistent with clinical studies. Those who have found a relationship in the general population have noted a low correlation between them [1, 2]. So far a causal or directional relationship has not been demonstrated. Yet, the positive effect of many clinical exercise programmes upon mood has been well-documented. One possible explanation could be that health survey data are generally collected for other purposes than to study the relationship of mood and exercise. This could tend to disaggregate the psychological and the physical activity components that are present in clinical exercise programmes in a way that is not easily accomplished in formal exercise programmes. In fact, it would seem that these results lend support to the notion that elements other than the exercise itself are important in explaining the positive effects of exercise programmes upon mood [24]. Further, a variety of non-exercise, activity-based interventions have generally been as effective as exercise in improving mood [22, 2527]. It is possible that exercise does have other, more positive, effects upon the mood of highly symptomatic individuals [28]. Nevertheless, the fact that we found no directional relationship between mood and recreational activity in those who reported an initially more dysphoric mood would seem to weigh against this possibility. The immediate anxiety-reducing and mood-enhancing effects that are often experienced by exercisers [23, 2932] might also contribute to the effects of an exercise programme. Any immediate effects of exercise upon mood would not be expected to have much impact on survey data, such as the data we have studied here.

References
1. Stephens T. Physical activity and mental health in the United States and Canada: evidence from four population surveys. Prev Med 1988; 17: 3547 2. Camacho T, Roberts R, Lazarus N, Kaplan G, Cohen R. Physical activity and depression: evidence from the Alameda County study. Am J Epidemiol 1991; 134: 220231 3. Stewart A, Hays R, Wells K, Rogers W, Spritzer K, Greenfield S. Long-term functioning and well-being outcomes associated with physical activity and exercise in patients with chronic conditions in the medical outcomes study. J Clin Epidemiol 1994; 47: 719730 4. Gauvin L, Spence J. Physical activity and psychological well-being: knowledge base, current issues, and caveats. Nutr Rev 1996; 54: 5365 5. Gauvin L, Rejeski W. The exercise induced feeling inventory: development and initial validation. J Sport Exerc Psychol 1993; 15: 403423 6. Hamid N. Positive and negative affectivity and maintenance of exercise programmes. Percept Mot Skills 1990; 70: 478 7. Farmer M, Locke B, Moscicki E, Dannenberg A, Larson D, Radloff L. Physical activity and depressive symptoms: the NHANES I epidemiologic follow-up study. Am J Epidemiol 1988; 128: 13401351 8. Frederick T, Frerichs R, Clark V. Personal health habits and symptoms of depression at the community level. Prev Med 1988; 17: 173182 9. Buchman B, Sallis J, Criqui M, Dimsdale J, Kaplan R. Physical activity, physical fitness, and psychological characteristics of medical students. J Psychosom Res 1991; 35: 197208 10. Cooper P, Ford D, Mead L, Klag M. Exercise and depression in mid-life: a prospective study. Am J Public Health 1997; 87: 670673 11. Beniamini Y, Rubenstein JJ, Zaichkowsky LD, Crim MC. Effects of high-intensity strength training on quality-of-life parameters in cardiac rehabilitation patients. Am J Cardiol 1997; 80: 841846 12. Courneya, KS, Friedenreich CM. Physical exercise and quality of life following cancer diagnosis: a literature review. Ann Behav Med 1999; 21: 171179 13. Reuter I, Engelhardt M, Stecker K, Baas H. Therapeutic value of exercise training in Parkinsons disease. Med Sci Sports Exerc 1999; 31:15441549 14. Muraki S, Maehara T, Ishii K, Ajimoto M, Kikuchi K. Gender differences in the relationship between physical fitness and mental health. Ann Physiol Anthropol 1993; 12: 379384 15. Cohen S, Schwartz J, Bromet E, Parkinson D. Mental health, stress, and poor health behaviors in two community samples. Prev Med 1991; 20: 306315 16. Fylkesnes K, Sgaard AJ, Henriksen N. Finnmark Heart Study: list of variables, English translation, crosstabulations and methodological issues. Finnmark III (1987/88) and Finnmark IV (1990). Memo. Troms: Institute of Community Medicine, University of Troms, 1992 17. Henriksen N, Sgaard AJ, Fylkesnes K. The Finnmark Intervention Study. Design, methods and effect of a 2 year communitybased intervention. Eur J Public Health 1995: 5: 269276 18. Olstad R, Sexton H, Sgaard AJ. The Finnmark study. Social support, social network and mental distress in a prospective population study. Soc Psychiatry Psychiatr Epidemiol 1999; 34: 519525 19. Bentler PM, Wu JC. EQS for Windows User Guide. Encino, California: Multivariate Software, Inc., 1995 20. Jacobsen BK, Hasvold T, Hoyer G, Hansen V. The General Health Questionnaire: how many items are really necessary in population surveys? Psychol Med 1995; 25: 957961 21. Hu L, Bentler PM. Cutoff criteria for fit indexes in covariance structure analysis: Conventional criteria versus new alternatives. Struct Equation Modeling 1999; 6:155 22. Berger BG, Owen DR. Mood alteration with yoga and swimming: aerobic exercise may not be necessary. Percept Mot Skills 1992; 75: 13311343

Conclusion
The results suggest that the relationship between mood and recreational exercise is largely correlational rather than directional, at least in this population-based sample.We believe that these findings suggest a need for taking a rather cautious view of the role that exercise, in itself, plays in influencing mood in the general population.
s Acknowledgement The authors are grateful for the co-operation of the National Health Screening Service, Oslo, in collecting the data. We also wish to thank the County Health Officer and the general practitioners in Finnmark for all their help and practical support.

353 23. Morris M, Salmon P. Qualitative and quantitative effects of running on mood. J Sports Med Phys Fitness 1994; 34: 284291 24. Desharnais R, Jobin J, Cote C, Levesque L, Godin G. Aerobic exercise and the placebo effect: a controlled study. Psychosom Med 1993; 55: 149154 25. Long BC. Aerobic conditioning (jogging) and stress inoculation intervention: an exploratory study of coping. Special Issue: exercise and psychological well-being. Int J Sport Psychol 1993; 24: 94109 26. Martinsen EW. Physical activity and depression: clinical experience. Yrjo Jahnsson Foundation VIII Medical Symposium: Depression: preventive and risk factors (1992, Porvoo, Finland). Acta Psychiatr Scand 1994; 89 [377 Suppl]: 2327 27. Stein PN, Motta RW. Effects of aerobic and nonaerobic exercise on depression and self-concept. Percept Mot Skills 1992; 74: 7989 28. Williams P, Lord SR. Effects of group exercise on cognitive functioning and mood in older women. Aust N Z J Public Health 1997; 21: 4552 29. Yeung RR.The acute effects of exercise on mood state. J Psychosom Res 1996; 40: 123141 30. Altchiler L, Motta R. Effects of aerobic and nonaerobic exercise on anxiety, absenteeism, and job satisfaction. J Clin Psychol 1994; 50: 829840 31. McGowan RW, Talton BJ, Thompson M. Changes in scores on the profile of mood states following a single bout of physical activity: heart rate and changes in affect. Percept Mot Skills 1996; 83: 859866 32. Raglin JS. Exercise and mental health. Beneficial and detrimental effects. Sports Med 1990; 9: 323329

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.