Changes in Leisure Activities and Dimensions of Depressive Symptoms in Later Life: A 12-Year Follow-Up

Abstract

Purpose of Study:

Although leisure activities benefit the mental health of the elderly population, the effect of changes in leisure activities on dimensions of depressive symptoms remains unclear. This investigation examined the influences of changes in intellectual, social, and physical activities between waves on four dimensions of depressive symptoms at follow-up.

Design and Methods:

Random effects modeling was utilized with data from a nationwide longitudinal study conducted in Taiwan. The study data comprised 6,942 observations from 2,660 older adults over a 12-year period.

Results:

The results suggested that changes in physical activities contributed to depressive symptoms which reflected positive affect in the later wave. Increased social activities between waves predicted higher positive affect and lower interpersonal difficulties scores at follow-up. Increased intellectual activities between waves did not substantially affect any domain of depressive symptoms. In contrast, declines in intellectual activities between waves predicted higher scores in three depressive symptoms domains, including depressed mood, somatic symptoms, and interpersonal difficulties.

Implications:

Engagement in a varied range of activities benefits mental health among elders more than participation in any single type of activity among elders. Reducing physical activities can lower positive affect, while the adverse effect may be balanced by increasing social activities. Also, the impact of decreasing intellectual activities on the interpersonal difficulties domain of depressive symptoms may be offset by increasing social activities.

Leisure activity, Depressive symptom, Random effects modeling

Maintaining an active lifestyle is considered as essential to successful aging ( Havighurst, 1961 ; Rowe & Kahn, 1997 ). Participation in leisure activity has been found associated with physical, mental, social, and overall well-being. For instance, physical well-being includes reduced mortality ( Agahi & Parker, 2008 ), fewer functional impairments ( Janke, Payne, & Van Puymbroeck, 2008 ; Strain, Grabusic, Searle, & Dunn, 2002 ), and increased perceived health ( Werngren-Elgstrom, Brandt, & Iwarsson, 2006 ). Mental well-being includes better cognitive functioning ( Hughes, Flatt, Fu, Chang, & Ganguli, 2013 ; Wang et al., 2013 ), more positive affect ( Klumb, 2004 ), enjoyment ( Chen & Fu, 2008 ), happiness ( Menec, 2003 ), and lack of depressed mood ( Arai et al., 2007 ; Lu, 2011 ; Sheung-Tak, Chow, Yu, & Chan, 2012 ), negative affect ( Everard, 1999 ), depressive symptoms ( Lindwall, Rennemark, Halling, Berglund, & Hassmén, 2007 ; Lu, 2011 ) or anxiety ( Pasco et al., 2011 ). Social well-being refers to satisfactory social relationships ( Litwin, 2000 ) while overall well-being includes higher levels of subjective well-being ( Simone & Haas, 2013 ), quality of life ( Silverstein & Parker, 2002 ), and life satisfaction ( Nimrod, 2008 ). Among these substantial merits, leisure activities are more strongly associated with depressive symptoms than with physical (e.g., self-rated health) and overall (e.g., life satisfaction) well-being ( Ritchey, Ritchey, & Dietz, 2001 ).

Depression is one of the most common chronic mental health issues in the elderly globally. However, numerous older individuals do not meet the criteria of depression but also suffer a certain degree of depressive symptoms ( George, 2011 ). In Taiwan, the prevalence of depressive symptoms among community dwelling elderly people ranges from 25.7% to 34.8% ( Liu et al., 1997 ; Lu, Liu, & Yu, 1998 ), which is higher than figures reported in other Chinese communities ( Lim et al., 2011 ). Symptoms of depression experienced in old age are consistently related to increased risk of physical and cognitive impairment, as well as suicide ( Ku, Fox, Chen, & Chou, 2012 ). Therefore, how to reduce depressive symptoms through early detection and effective interventions is critically important for policy makers and professionals serving older adults in Taiwan and elsewhere.

The manifestation of depressive symptoms in later life is complex. By applying factor analytic work with the Center for Epidemiological Studies-Depression (CES-D) scale, Radloff (1997) noted that depressive symptoms comprise four symptom clusters, including depressed affect, somatic symptoms, positive affect, and interpersonal difficulties. Older adults in different cultural contexts tend to express their psychological distress differently ( Krause & Laing, 1992 ; Lim et al., 2011 ). For example, Chinese elders generally reference their emotional distress using the language of somatic complaints such as headache or tiredness ( Lim et al., 2011 ). Meanwhile, older black and white Americans are more likely to exhibit depression in the form of depressed mood and interpersonal problems ( Krause & Liang, 1992 ). Thus, researchers interested in examining the influence of depressive symptomology that occurs during later life should consider the multidimensional nature of depressive symptoms. Looking solely at summary depressive symptom scores may overlook important information nested beneath varied facets of depressive symptomatology ( Krause & Liang, 1992 ).

Although growing evidence supports the protective effect of leisure activities on depressive symptoms, to our knowledge, no study has yet examined the relationship between different types of leisure activities and dimensions of depressive symptoms. Given previous research involving older adults, different leisure activities exert complex effects on mental health outcomes. For instance, Everard, Lach, Fisher, and Baum (2000) found that continuing to engage in low-demand leisure activities was associated with higher scores in the mental health domain of the 12-Item Short-Form Health Survey (SF-12), but no such association existed for social and high-demand leisure activities. Hilleras, Jorm, Herlitz, and Winblad (1999) concluded that physical activities were significantly associated with positive affect, but not negative affect. Neither intellectual nor social activities significantly influenced positive or negative affect.

A small number of longitudinal studies on the relationship between leisure activity and depression have yielded mixed findings. Glass, Mendes de Leon, Bassuk, and Berkman (2006) found longitudinal protective effects of productive and social activities on depressive symptoms. Engagement in activities such as walking and talking on the phone was associated with lower initial status and smaller longitudinal growth in depressive symptoms ( Hong, Hasche, & Bowland, 2009 ; Lampinen & Heikkinen, 2003 ). In contrast, Morgan and Bath (1998) found no significant longitudinal association between social activities and depression. Additionally, lower involvement in outdoor activities predicted increased depression 4 years later, while engagement in indoor activities was not related to current and future depression.

The lack of consistency in typology and leisure activity classification makes meaningful comparison among studies difficult. Consequently, the longitudinal relationship between various leisure activities and symptoms of depression remains unclear. This study investigates the longitudinal effects of engagement in different types of leisure activities on subsequent dimensions of depressive symptomology, using data from a nationally representative sample of elderly Taiwanese over a 12-year period. The research questions were: (a) Do changes in intellectual, social, and physical activities between waves influence subsequent levels of depressive symptom? (b) If so, do changes in intellectual, social, and physical activities between waves influence different dimensions of depressive symptoms in the subsequent wave?

Method

Sample

The study data come from a nationwide panel study: “Survey of Health and Living Status of the Elderly” (SHLSE) in Taiwan. The study began in 1989 with face-to-face interviews of a representative sample of 4,049 adults aged 60 years old and over, taken through probability sampling. The age distribution of the baseline survey resembles the population at the data collection point. Respondents were reinterviewed at 3 or 4 yearly intervals. Totals of 4,049; 3,155; 2,669; 2,310; 1,743; and 1,268 respondents completed the 1989, 1993, 1996, 1999, 2003, and 2007 interviews, respectively. Since 5 out of the 10 depressive symptom items, the dependent variable of this study, were not included in the 1993 survey, the analytical data came from the waves in 1996, 1999, 2003, and 2007. Additionally, proxies were not asked any subjective questions (i.e., depressive symptoms) and hence were excluded. Finally, 6,942 observations from 2,660 participants were used for analyses.

Measures

Dependent Variable: Dimensions of Depressive Symptoms

Depressive symptoms were measured with 10-item shortened version of the CES-D developed by Radloff (1977) . Respondents were asked how often they had experienced these feelings in the past week. Responses for each item ranged from 0 to 3, where 0 = never , 1 = rarely or only one day , 2 = sometimes or 2–3 days to 3 = always or 4 days or more . Based on Radloff (1977) , the 10 depressive symptom items can be divided as (a) depressive affect (“I felt sad,” “I felt lonely,” and “I felt depressed”); (b) somatic symptoms (“I did not feel like eating,” “My sleep was restless,” “I felt that everything I did was an effort,” and “I could not get “going”“); (c) positive affect (“I enjoyed my life” and “I felt very happy”); and (d) interpersonal difficulties (“People were unfriendly.”). The two items representing “positive affect” were reverse recoded so that higher scores reflected lower levels of positive affect.

To test the original four-factor structure of the CES-D, confirmatory factor analysis (CFA) was employed with baseline data, using principal component estimation with a varimax rotation ( Bandalos & Finney, 2010 ; Stevens, 2009 ). As can be seen from Table 1 , 8 out of 10 items had factor loadings greater than 0.6 on the factors proposed from the theory. Two items were excluded (“I felt that everything I did was an effort” and “I could not get ‘going’”) because they revealed greater factor loadings on the “depressed affect” subscale rather than the subscale they are proposed to be (“somatic symptoms”). This four-factor structure of depressive symptoms explains 69.75% of variability.

Table 1.

Confirmatory Factor Loadings for 10-Item Depressive Symptom Measures ( N = 3,923)

	Mean	SD	Depressed affect	Somatic symptoms	Positive affect	Interpersonal difficulties
I felt sad	0.31	0.75	.775
I felt lonely	0.43	0.87	.769
I felt depressed	0.59	0.93	.644
I felt that everything I did was an effort	0.34	0.76			(Excluded)
I could not get “going”	0.30	0.73			(Excluded)
I did not feel like eating	0.45	0.84		.817
My sleep was restless	0.65	1.02		.724
I enjoyed life	1.13	1.22			.907
I was happy	1.43	1.25			.904
People were unfriendly	0.11	0.42				.954
Percent of variance			25.74%	17.22%	16.90%	9.89%

Mean scores for each depressive symptom domain at each wave were then computed separately. The average scores were calculated if respondents answered more than half of the subscale items; otherwise, they were counted as missing values.

Independent Variables: Leisure Activity

Respondents were asked to assess their frequency of participation in eight leisure activities. Responses for each activity ranged from 1 to 4, where 0 = never , 1 = less than once monthly , 2 = 2–3 times monthly , 3 = 1–2 times weekly to 4 = almost everyday .

To cluster these activities, baseline data were used to perform exploratory factor analysis (EFA) ( Bandalos & Finney, 2010 ; Stevens, 2009 ). Three factors with factor loadings exceeding 0.5 were obtained through principal component extraction utilizing varimax rotation. The current study adopted the categorization from Adams, Leibbrandt, and Moon (2011) and labeled these three factors as (a) intellectual activity (“watching TV,” “reading newspaper,” and “playing Chinese chess”); (b) social activity (“chatting”); (c) physical activity (“walking,” “individual-based exercise,” and “group-based exercise”; shown in Table 2 ). Gardening had loadings lower than 0.4 in all three factors, and therefore was excluded ( Stevens, 2009 ). This three-factor solution of leisure activity explains 52.39% of variability.

Table 2.

Exploratory Factor Loadings for Leisure Activity Measures ( N = 3,923)

	M	SD	Intellectual	Social	Physical
Watching TV	2.75	0.73	.500
Reading newspaper	1.13	1.41	.717
Playing Chinese chess	0.17	0.56	.653
Chatting	1.59	1.31		.863
Walking	1.81	1.28			.572
Individual-based exercise (e.g., jogging, mountain climbing)	0.68	1.14			.781
Group-based exercise (e.g., dancing, playing Tai-chi)	0.27	0.81			.690
Gardening	0.74	1.22	(Excluded)
Percent of variance			17.40%	14.71%	20.28%

Mean scores for each leisure activity type at each wave were computed separately. Responses with information on more than half of the factor items were included; otherwise, they were treated as missing values. The former mean scores of each leisure activity type were then subtracted from the later wave to reflect changes in leisure activity participation between waves. For each leisure activity type, two dummy variables were then created to represent increasing and decreasing activity levels from previous wave, respectively. Older individuals who maintained the same activity levels between waves served as a reference group.

Control Variables

All analyses controlled for variables including hazard of attrition, age, education, ethnicity, gender, marital status, self-rated health, cognitive impairment, and activities of daily living (ADL).

Hazard of attrition: Accumulative attrition becomes inevitable when gathering data from the same sample on multiple occasions, especially when the target sample comprises an older population ( Tuma, 2004 ). As suggested by Heckman (1979) (also see Berk, 1983 ; Hao, 2008 ), cox proportional hazards models were employed to create a time-varying attrition risk indicator that reflects the probability of a study participant dropping out at follow-up. Control variables in this study, including age, education, gender, self-rated health, marital status, cognitive impairment, and ADL acted as predictors for attrition. The attrition risk indicator represents the relative likelihood of each respondent changing from remaining in the sample to attrition between waves, compared to a hypothetical 65-year-old Taiwanese male with no education, bad self-rated health, no cognitive impairment, and ADL defect. Adding the attrition risk indicator into models provides control over disturbances arising from nonrandom attrition that are associated with predictors of the attrition risk indicator.

Age was measured continuously in years. Education was measured by the years of schooling completed. Gender (1 = male ; 2 = female ) and marital status (0 = not married ; 1 = married ) were measured as a binary variable. Self-rated health was assessed by asking how respondents feel about their health status, ranging from 1 = very bad to 5 = very good .

Cognitive impairment was measured using the 10-item Short Portable Mental Status Questionnaire (SPMSQ) ( Pfeiffer, 1975 ). Study participants were asked to identify their address, location, date of interview, day of the week of interview, age, mother’s maiden name, the names of the current and previous presidents, their date of birth, and to backwards from 20 using intervals of three. Each correct response was given a score of 0, while incorrect responses received a score of 1. The summated scores for the 10 items were then calculated, with higher score indicating greater cognitive deficit.

ADL scale was used to evaluate the level of functional dependency. Respondents were asked if they have difficulty performing the tasks including bathing, dressing/undressing, eating, getting off the bed/standing/sitting on a chair, moving around the house, and toileting. Each indicator ranges from 0 = no difficulty to 3 = unable to do at all . The summated scores for the six items were then calculated. The higher scores referred greater degrees of functional impairment.

Analytical Approach

Multiple imputation (MI) was applied via the ICE procedure using Stata 11.0 software. The use of MI can address the potential bias resulting from missing data and maximize the utilization of all possible information ( Acock, 2005 ). Missing values for variables at each wave were estimated based on all variables used in analyses in the same wave. At last, five imputed data sets were created, which is considered sufficient to restore the natural variability in the missing data and produce unbiased parameter estimates ( Acock, 2005 ; Rubin, 1987 ; Schafer & Graham, 2002 ).

Random effects models, using the maximum likelihood method (ML), were employed to investigate the effects of changes in leisure activity on the dimensions of depressive symptoms. The random effects method allows intercepts to vary between individuals and allows residuals to vary among observations, making this method a superior means of dealing with panel data. For individuals are measured multiple times, each has a separate intercept representing between-individual differences in the outcome with a control of all independent variables. The random residuals represent the within-individual variance in predicting the outcome under the function of both time-constant and time-varying independent variables over time ( Petersen, 2004 ; Rabe-Hesketh & Everitt, 2007 ; Rabe-Hesketh & Skrondal, 2005 ).

The following model is one of the preferred models for explicating causal relationship between variables suggested by Finkel (1995) . Theoretically, both the lagged independent ( X_{t −1} ) and dependent variables ( Y_{t −1} ) directly affect the current values of dependent variable ( Y_t ). Therefore, the model should be corrected by controlling the values of X_{t −1} and Y_{t −1} to estimate the “real” influence of the predictor $(\Delta X)$ on the outcome variable ( Y_t ) ( Finkel, 1995 ; Twisk, 2003 , p. 108). The other advantage of adding Y_{t −1} to the equation is that this can partially control the potential influence from omitted variables ( Finkel, 1995 ; Menard, 2002 ). To summarize, this study assumes that changes in leisure activities between waves (∆X) will affect Y_t , after controlling for X_{t −1} and Y_{t −1} .

$$Y_t={\beta }_0+{\beta }_2{Y}_{t-1}+({\beta }_1+{\beta }_3)X_{t-1}+{\beta }_1\Delta X+{\epsilon }_t$$

A series of random effects models were implemented to address the research questions. First, each leisure activity type was added to the models separately to investigate its effect on each dimension, and on the CES-D scale. Second, all three types of leisure activities were added to the models simultaneously to further test their effects on each dimension and on the CES-D scale when adjusting for the influences of the other two activity types.

Results

Table 3 lists the characteristics of study participants from the 1996 to 2007 surveys. The average age of participants was 80.76 years old in 1996. Except for 2007, the subjects were consistently more male and married in marital status over time. The results of a series of simple random effect estimations revealed that the participants had significantly worse self-rated health, worse cognitive functioning, and declining physical status with the passage of time. Regarding depressive symptoms, respondents exhibited significantly decreasing positive affect and increasing somatic symptoms over time. Levels of depressed affect, interpersonal difficulties and total depressive symptoms changed little over time. As for leisure activities, participants reported decreased intellectual and physical activities over time, while their social activity levels remained stable across waves.

Table 3.

Range, Means, Standard Deviations, and Characteristics of the Older Participants

	1996 ( N = 2,408)	1999 ( N = 2,069)	2003 ( N = 1,474)	2007 ( N = 991)
Hazard of attrition (ratio)	27.73 (6.13)	30.42 (6.23)	33.83 (5.81)	37.30 (5.63)
Age	80.76 (5.02)	83.42 (4.82)	86.56 (4.10)	89.24 (3.73)
Education (0–17)	4.49 (4.61)	4.63 (4.64)	4.93 (4.70)	5.29 (4.80)
Gender: Male	56.1%	55.3%	54.6%	52.8%
Self-rated health (1–5)***	3.05 (1.09)	2.94 (1.07)	2.92 (1.05)	2.88 (0.97)
Married	60%	56.1%	50.9%	41.8%
Cognitive impairment (0–10)***	1.31 (1.83)	1.34 (1.93)	1.41 (1.84)	1.90 (2.15)
ADL (0–3)***	0.06 (0.31)	0.10 (0.37)	0.11 (0.41)	0.15 (0.47)
Dimensional depressive symptoms
Depressed affect (0–3)	0.48 (0.76)	0.49 (0.77)	0.44 (0.73)	0.46 (0.72)
Positive affect (0–3)***	1.15 (1.21)	1.05 (1.12)	1.04 (1.12)	2.09 (1.04)
Somatic symptoms (0–3)***	0.66 (0.83)	0.67 (0.85)	0.65 (0.85)	0.72 (0.87)
Interpersonal difficulties (0–3)	0.17 (0.56)	0.21 (0.60)	0.13 (0.51)	0.14 (0.50)
Total depressive symptoms (0–3)	0.66 (0.65)	0.64 (0.66)	0.59 (0.62)	0.60 (0.62)
Leisure activities
Intellectual(0–3)***	1.30 (0.62)	1.32 (0.64)	1.26 (0.62)	1.28 (0.60)
Physical (0–3)***	1.54 (1.32)	1.53 (1.34)	1.48 (1.34)	1.53 (1.31)
Social (0–3)	0.79 (0.67)	0.78 (0.66)	0.76 (0.62)	0.77 (0.62)

Notes: ADL = Activities of Daily Living .

*** p < .001.

Table 4 lists the percentage changes in the three types of leisure activities between waves. Among the three types of leisure activities, respondents were more likely to reduce physical activities between waves (ranged from 27.8 to 31.3), followed by social activity (ranged from 25.9 to 30.3), and lastly, intellectual activity (ranged from 20.2 to 26.9). In contrast, respondents were more likely to increase social activities between waves (ranged from 26.7 to 23.4), followed by physical activity (ranged from 20.6 to 24.2), and finally, intellectual activity (ranged from 13.8 to 18.2).

Table 4.

Changes in Leisure Activity Between Waves (%)

	1996–1999	1999–2003	2003–2007
Intellectual activity
Decrease	20.6	26.9	20.2
Increase	18.2	13.8	14.6
Physical activity
Decrease	27.8	29.6	31.3
Increase	24.2	20.6	21.2
Social activity
Decrease	27.5	30.3	25.9
Increase	26.7	23.4	24.7

Effects of Changes in Intellectual Activities on Dimensions of Depressive Symptoms

Table 5 summarizes the effects of changes in intellectual activities between waves on the four CES-D subscales and the full scale. Controlling for prior depressed affect level, prior intellectual activity level, hazard of attrition, age, education, gender, self-rated health, marital status, cognitive and physical impairments, declines in intellectual activities between waves were associated with higher depressed affect ( b = .119, p < .001), somatic symptoms ( b = .097, p < .001), interpersonal difficulties ( b = .079, p < .001), and total CES-D ( b = .099, p < .001) scores in the subsequent wave. No significant association was observed between decreased intellectual activities and positive affect scores. Increased intellectual activities between waves were only related to lower CES-D ( b = −.047, p < .05), but did not substantially affect any of the depressive symptom subscales in the later wave.

Table 5.

Changes in Intellectual Activities and Dimensions of Depressive Symptoms (Separate Model)

	Depressed affect		Somatic symptoms		Positive affect		Interpersonal difficulties		Total CES-D
	Estimate	SE	Estimate	SE	Estimate	SE	Estimate	SE	Estimate	SE
Prior dependent variable	0.259***	0.016	0.253***	0.023	0.043**	0.016	0.201***	0.016	0.468***	0.013
Prior intellectual activities	−0.076**	0.024	−0.057*	0.028	0.108*	0.043	−0.063***	0.020	−0.066***	0.018
Hazard of attrition	0.043	0.023	0.037	0.027	−0.119**	0.041	0.053**	0.018	0.042*	0.017
Age	−0.046*	0.022	−0.037	0.026	0.085*	0.039	−0.058**	0.018	−0.045**	0.016
Education	−0.040	0.024	−0.041	0.027	0.116**	0.042	−0.052**	0.019	−0.043*	0.017
Gender (male)	−0.072	0.047	0.055	0.055	0.229**	0.084	−0.122***	0.038	−0.082*	0.034
Self-rated health	−0.153***	0.022	−0.237***	0.025	0.216***	0.038	−0.116***	0.017	−0.181***	0.016
Married	−0.149***	0.022	−0.019	0.025	0.150***	0.038	−0.074***	0.017	−0.044**	0.016
Cognitive impairment	−0.021	0.027	−0.037	0.031	0.106*	0.048	−0.040	0.021	−0.033	0.019
ADL	0.188***	0.039	0.121**	0.046	0.115	0.069	−0.014	0.031	0.136***	0.028
Changes in intellectual activities between waves (ref.: no change)
Decrease	0.119***	0.026	0.097***	0.030	−0.045	0.046	0.079***	0.021	0.099***	0.019
Increase	0.011	0.029	−0.007	0.034	0.031	0.052	0.037	0.023	−0.047*	0.021

Notes: ADL = activities of daily living; CES-D = Center for Epidemiologic Studies Depression Scale.

* p < .05. ** p < .01. *** p < .001 (two tailed).

Effects of Changes in Social Activities on Dimensions of Depressive Symptoms

Table 6 summarizes the effects of changes in social activities on the four CES-D subscales and the full scale. Decreased social activities between waves were only associated with higher CES-D scores ( b = .055, p < .01), but were not significantly related to any of the four subscales at follow-up. In contrast, increased social activities were not only associated with higher positive affect scores ( b = .115, p < .05), but also with lower depressed affect ( b = −.055, p < .05), interpersonal difficulties ( b = −.056, p < .01), and total CES-D ( b = −.061, p < .01) scores in the subsequent wave.

Table 6.

Changes in Social Activities and Dimensions of Depressive Symptoms (Separate Model)

	Depressed affect		Somatic symptoms		Positive affect		Interpersonal difficulties		Total CES-D
	Estimate	SE	Estimate	SE	Estimate	SE	Estimate	SE	Estimate	SE
Prior dependent variable	0.258***	0.016	0.253***	0.022	0.043**	0.016	0.202***	0.016	0.467***	0.013
Prior social activities	−0.041***	0.009	−0.011	0.011	0.040*	0.017	−0.028***	0.007	−0.026***	0.007
Hazard of attrition	0.032	0.023	0.030	0.026	−0.099*	0.041	0.045*	0.018	0.031	0.016
Age	−0.036	0.022	−0.031	0.025	0.066	0.039	−0.050**	0.017	−0.035*	0.016
Education	−0.034	0.024	−0.038	0.027	0.102*	0.042	−0.048*	0.019	−0.036*	0.017
Gender (male)	−0.048	0.046	0.069	0.054	0.175*	0.082	−0.102**	0.037	−0.058	0.033
Self-rated health	−0.145***	0.021	−0.234***	0.025	0.199***	0.038	−.110***	0.017	−0.173***	0.016
Married	−0.154***	0.022	−0.023	0.025	0.157***	0.038	−0.079***	0.017	−0.049**	0.016
Cognitive impairment	−0.006	0.026	−0.028	0.030	0.081	0.047	−0.029	0.021	−0.018	0.019
ADL	0.204***	0.039	0.134**	0.045	0.092	0.068	−0.003	0.031	0.152***	0.028
Changes in social activities between waves (ref.: no change)
Decrease	0.035	0.026	0.006	0.030	−0.007	0.045	0.009	0.020	0.055**	0.019
Increase	−0.055*	0.027	−0.050	0.031	0.115*	0.047	−0.056**	0.021	−0.061**	0.019

Notes: ADL = activities of daily living; CES-D = Center for Epidemiologic Studies Depression Scale.

* p < .05. ** p < .01. *** p < .001 (two-tailed).

Effects of Changes in Physical Activities on Dimensions of Depressive Symptoms

Table 7 lists the effects of changes in physical activities between waves on the four CES-D subscales and the full scale. After adjusting for the prior value of the dependent variable, prior physical activity level and other control variables, declines in physical activities between waves were related to more somatic symptoms ( b = .061, p < .05), higher CES-D scores ( b = .051, p < .01) and lower positive affect scores ( b = −.153, p < 001) at follow-up. On the other hand, increased physical activities were associated with higher positive affect ( b = .109, p < .05) and lower CES-D ( b = −.072, p < .001) scores in the subsequent wave.

Table 7.

Changes in Physical Activities and Dimensions of Depressive Symptoms (Separate Model)

	Depressed affect		Somatic symptoms		Positive affect		Interpersonal difficulties		Total CES-D
	Estimate	SE	Estimate	SE	Estimate	SE	Estimate	SE	Estimate	SE
Prior dependent variable	0.260***	0.015	0.249***	0.023	0.046**	0.016	0.206***	0.016	0.470***	0.013
Prior physical activities	−0.078***	0.019	−0.042	0.023	0.101**	0.034	−0.031*	0.015	−0.046***	0.014
Hazard of attrition	0.042	0.023	0.035	0.027	−0.110**	0.041	0.049**	0.018	0.037*	0.016
Age	−0.045*	0.022	−0.036	0.025	0.077*	0.039	−0.053**	0.018	−0.040*	0.016
Education	−0.041	0.024	−0.042	0.027	0.111**	0.042	−0.051**	0.019	−0.040*	0.017
Gender (male)	−0.069	0.046	0.059	0.054	0.198*	0.082	−0.112**	0.037	−0.069*	0.033
Self-rated health	−0.149***	0.021	−0.236***	0.025	0.198***	0.098	−0.112***	0.017	−0.174***	0.016
Married	−0.153***	0.022	−0.022	0.025	0.154***	0.038	−0.078***	0.017	−0.046**	0.016
Cognitive impairment	−0.018	0.026	−0.034	0.030	0.095*	0.047	−0.033	0.021	−0.025	0.019
ADL	0.182***	0.039	0.122**	0.046	0.126	0.069	−0.011	0.031	0.139***	0.028
Changes in physical activities between waves (ref.: no change)
Decrease	0.043	0.026	0.061*	0.030	−0.153***	0.046	0.020	0.021	0.051**	0.019
Increase	−0.040	0.027	−0.007	0.031	0.109*	0.047	−0.029	0.021	−0.072***	0.019

Notes: ADL = activities of daily living; CES-D = Center for Epidemiologic Studies Depression Scale.

* p < .05. ** p < .01. *** p < .001 (two-tailed).

Effects of All Leisure Activities on Dimensions of Depressive Symptoms

In Table 8 , the three types of leisure activities were added to the full models simultaneously to test whether their effects on the four subscales and the total CES-D persisted. All the significant associations between changes in leisure activities and the subscales or total CES-D in the separate models remained true in the full models, except the following three relationships: (a) increased social activities and depressed affect; (b) decreased physical activities and somatic symptoms; (c) increased intellectual activities and total CES-D. Specifically, controlling for all presumed factors and changes in the other two types of leisure activities, declines in intellectual activities between waves remained closely related to higher subsequent depressed affect ( b = .113, p < .001), somatic symptoms ( b = .093, p < .01), interpersonal difficulties ( b = .078, p <.001), and total CES-D ( b = .095, p <.001) scores. In contrast, increases in intellectual activities between waves did not substantially influence any facet or the total scores of depressive symptomatology in the subsequent wave.

Table 8.

Changes in Leisure Activities and Dimensions of Depressive Symptoms (Full Model)

	Depressed affect		Somatic symptoms		Positive affect		Interpersonal difficulties		Total CES-D
	Estimate	SE	Estimate	SE	Estimate	SE	Estimate	SE	Estimate	SE
Prior dependent variable	0.251***	0.016	0.249***	0.023	0.040*	0.016	0.196***	0.016	0.464***	0.014
Prior sedentary activities	−0.062*	0.024	−0.053	0.029	0.098*	0.043	−0.057**	0.020	−0.061***	0.018
Prior social activities	−0.036***	0.010	−0.008	0.011	0.083*	0.035	−0.015	0.016	−0.032*	0.014
Prior physical activities	−0.058**	0.020	−0.033	0.023	0.031*	0.018	−0.026***	0.008	−0.023***	0.007
Hazard of attrition	0.043	0.023	0.040	0.027	−0.124**	0.041	0.051**	0.019	0.042*	0.017
Age	−0.047*	0.022	−0.041	0.026	0.091*	0.039	−0.056***	0.018	−0.045**	0.016
Education	−0.040	0.024	−0.044	0.028	0.121**	0.042	−0.051**	0.019	−0.043*	0.017
Gender (male)	−0.069	0.047	0.051	0.055	0.235**	0.084	−0.117**	0.038	−0.079*	0.034
Self-rated health	−0.146***	0.022	−0.237***	0.025	0.206***	0.038	−0.111***	0.017	−0.174***	0.016
Married	−0.152***	0.022	−0.020	0.025	0.150***	0.038	−0.075***	0.017	−0.045**	0.016
Cognitive impairment	−0.022	0.027	−0.042	0.031	0.114*	0.048	−0.039	0.021	−0.033	0.019
ADL	0.171***	0.040	0.109*	0.046	0.150*	0.070	−0.020	0.031	0.122***	0.028
Changes in intellectual activities between waves (ref.: no change)
Decrease	0.113***	0.026	0.093**	0.030	−0.035	0.046	0.078***	0.021	0.095***	0.019
Increase	0.014	0.029	−0.005	0.034	0.025	0.052	0.041	0.023	−0.041	0.021
Changes in social activities between waves (ref.: no change)
Decrease	0.028	0.026	−0.002	0.030	0.014	0.045	0.005	0.020	0.054*	0.019
Increase	−0.049	0.027	−0.045	0.031	0.096*	0.047	−0.053*	0.021	−0.049*	0.019
Changes in physical activities between waves (ref.: no change)
Decrease	0.032	0.026	0.055	0.030	−0.146**	0.046	0.012	0.021	0.039*	0.019
Increase	−0.033	0.027	0.001	0.031	0.099*	0.047	−0.024	0.021	−0.063***	0.019

Notes: ADL = activities of daily living. CES-D = Center for Epidemiologic Studies Depression Scale.

* p < .05. ** p < .01. *** p < .001 (two-tailed).

Decreased social activities were only associated with higher CES-D scores ( b = .054, p < .05), while increased social activities between waves were associated with higher positive affect ( b = .096, p < .05), lower interpersonal difficulties ( b = −.053, p < .05), and lower CES-D ( b = −.049, p < .05) scores in the next wave.

Decreased physical activities were related to lower positive affect ( b = −.146, p < .01) and higher CES-D ( b = .039, p < .05) scores in the subsequent wave. Similarly, increases in physical activities were related to higher positive affect ( b = .099, p < .05) and lower CES-D ( b = −.063, p <.001) scores at follow-up.

Discussion

This study examines how changes in different types of leisure activities are related to four dimensions of depressive symptomology. The use of longitudinal data collected over a 12-year period enables researchers to apply more rigorous tests of correlational effects among variables, and to extend current knowledge based on cross-sectional data. Recognition of the complex links among various leisure activities and multiple facets of depressive symptoms provides important insights into the development of leisure activity programs that can help improve mental health outcomes in later life.

The analytical results demonstrated that changes in the three types of leisure activities between waves all forecast overall depressive symptom scores in the subsequent wave. Older adults with increased intellectual, social and physical activities between waves reported lower CES-D scores at follow-up, compared to those who maintained the same activity level. Meanwhile, those who decreased participation in these three types of leisure activities all reported significantly higher CES-D scores at follow-up. These findings were generally consistent with those of previous longitudinal studies ( Chiao, Weng, & Botticello, 2011 ; Ku et al., 2012 ). The analytical results also supported the “successful aging model” assumption proposed by Rowe and Kahn (1997) , which suggested that active engagement in life promotes positive mental health, and thus should be considered integral to successful aging.

This investigation showed that changes in physical activities between waves significantly influenced the positive affect domain of depressive symptoms. Increased physical activities were related to more positive affect while decreased physical activities were associated with less positive affect. In contrast, changes in physical activities did not substantially affect any subscale assessing negative symptoms of depression, such as depressed affect, somatic symptoms, and interpersonal difficulties. While numerous previous studies have supported the importance of physical activity to depressive symptoms in later life (e.g., Ku et al., 2012 ; Lee, Lan, & Lee, 2012 ), the findings of this study further demonstrate that the benefits of physical activity largely underlie its effectiveness in reinforcing the positive affect facet of depressive symptomology. As noted by Stathi, Fox, and McKenna (2002) , engagement in physical activity is particularly effective in enhancing positive attitude toward life as well as feelings of happiness and enjoyment. The stimulation of positive emotional experience thus is the most notable outcome of participation in physical activities.

Older adults with increased social activities between waves tended to display higher positive affect and lower interpersonal difficulties scores at follow-up than those who maintained a constant activity level. These findings agree with those of previous research on social activity and depression (e.g., Hong et al., 2009 ). The current study provides more detailed information demonstrating that the positive affect and interpersonal difficulties domains of depressive symptoms were more likely than other dimensions to be affected by increasing social activity involvement. According to Simone and Haas (2013) , engagement in social activity provides opportunities for positive interpersonal interactions, also broadens the social lives of older individuals and increases their confidence that support will be available if necessary. The former can effectively reduce interpersonal problems, while the later can make elders less sensitive to negative life experiences, and ultimately contribute to positive affect.

Interestingly, increased intellectual activities between waves did not appear to affect any aspect of depressive symptoms. In contrast, reduced intellectual activities between waves emerged as a risk factor for depressed mood, somatic symptoms, and interpersonal difficulties. The analytical results were partially supported by previous studies of the oldest-old ( Hilleras et al., 1999 ). The average age of the study participants was over 80 years old, and people of advanced age tend to withdraw from more emotionally and physically demanding activities (e.g., friendship or outdoor-physical exercise) and take up more passive activities (e.g., watching TV, listening to radio, and reading) to compensate for their loss of physical and mental functioning ( Paillard-Borg, Hui-Xin, Winblad, & Fratiglioni, 2009 ; Silverstein & Parker, 2002 ). Therefore it is plausible to assume that intellectual activities are less demanding than other activities, and generally compensate for more demanding activities. Consequently, increasing intellectual activities do not significantly influence individual emotional state. However, when physical and mental limitations become so severe that an older individual must discontinue intellectual activities, this becomes especially detrimental because those affected lose their ability to sustain their active lifestyle.

Additionally, the research findings have indicated that increases in intellectual activities between waves were related to fewer interpersonal difficulties at follow-up. Based on the work by Jacobs, Hammerman-Rozenberg, Cohen, and Stessman (2008) , intellectual activities (i.e., watching TV, reading newspaper, and playing chess) may provide potential opportunities for positive social interactions such as going out and meeting people, buying newspapers from nearby stores or having good topics of conversation with others. This could be why engagement in such activities can reduce the subjective appraisals of older individuals in relation to interpersonal difficulties.

In response to the first research question, changes in intellectual, social, and physical activities between waves predicted subsequent levels of depressive symptoms. As for the second research question, changes in intellectual, social, and physical activity affected different dimensions of depressive symptoms at follow-up. For instance, reducing intellectual activities increased scores on the three subscales measuring the negative symptoms of depression (i.e., depressed mood, somatic symptoms, and interpersonal difficulties). Meanwhile, increases in physical and social activities were associated with higher scores on the positive affect items. The results held true in both the partial and complete models. Silverstein and Parker (2002) suggested that engagement in a wider variety of activities benefits quality of life more than participation in a single type of activity. Similarly, the analytic results in this study showed that the beneficial effects of various leisure activities on depressive symptoms appear cumulative. Involvement in more varied activities (i.e., intellectual, social, and physical activities) implies greater mental health benefits because each activity type makes its own contribution to different aspects of depressive symptoms. These effects remained unchanged even with the presence of other activity types.

Finally, in line with previous works on the “selection, optimization, and compensation” (SOC) assumption ( Baltes & Baltes, 1990 ; Menec, 2003 ; Silverstein & Parker, 2002 ), the study results supported that gains in certain activities may compensate for losses in others. Specifically, socializing with others can balance the detrimental effects of reducing physical activities on the positive affective features of depression. Having more opportunities to chat with others may also offset the growing difficulties in interpersonal relationships that result from reduced intellectual activities. This study thus has enhanced understanding of mental health in the elderly by clarifying which dimensions of depressive symptomology are affected by reducing involvement in certain leisure activities, and how these impacts can be mitigated by increased participation in other activities.

The current investigation has several limitations that require addressing. First, the leisure activities and depressive symptoms used for the analyses were based on a self-reported questionnaire, raising concerns regarding recall bias. Second, due to the limitation of the questionnaires, the analysis could only include 10-item shortened version of the CES-D scale and eight leisure activities. Because the multidimensional nature of leisure activities and depressive symptoms, some subscales for these two concepts are based on a small number of items. Although it may be possible to reduce the item number while still maintain a strong factor with a large data set ( Costello & Osborne, 2005 , p. 5), subscales with multiple and heterogeneous items are usually preferred in order to adequately reflect the construct of interest. Future analyses may expand this study by including more comprehensive measures of leisure activities and depressive symptoms. Third, the national survey did not include any psychological measures related to the appraisals of older adults regarding their leisure experiences, such as satisfaction with activities engaged in or their perceived value, which are considered important potential mechanisms that link participation in activities and mental well-being ( Adams et al., 2011 ; Lu, 2011 ).Fourth, future research will be needed to examine several intervening internal or external resources that may moderate or mediate the relationships between leisure activity type and different facets of depressive symptoms. For example, physical activities may increase self-esteem, and in turn contribute to increased positive affect. Similarly, social activities may increase social support, and thus eliminate interpersonal problems.

Acknowledgments

The data in this study were provided by Bureau of Health Promotion, Department of Health, Taiwan, Republic of China. The author would like to thank the National Science Council of the Republic of China, Taiwan for supporting this research under Contract No. NSC 101-2410-H-002-215-SS2.

Author notes

Decision Editor: Nicholas G. Castle, PhD