Data Source
The data for this study came from the Household Income and Labour Dynamics of Australia (HILDA) survey. HILDA is a major largescale population-based longitudinal survey of Australian households available over a period of 17 years (2001-2017). The survey follows the lives of more than 17,000 Australians each year. It collects information on economic and personal well-being, labour market dynamics, family life, household and family relationships, income and employment, health and education 24. We analysed wave 14 data, the latest available, as information on community participation, social connection, trust and reciprocity were not available in each wave. Our sample contained Australians aged 15 years and over who had completed the self-completion questionnaire (SCQ) as most of our dependent/independent variables come from the SCQ. For more information on HILDA survey sampling methodology please refer to HILDA User Manual and HILDA project discussion paper series #1/15 24 25. Our sample consisted of 12,767 individuals located over 5004 SA1s.
Measures
Statistical Area Level
The geographical units of analysis in this study were Australian Census of Population and Housing (ABS, 2011) Statistical Area Level 1 (SA1). SA1s have generally been designed as the smallest local or neighbourhood units for the release of census data. SA1s have a population of between 200 and 800 people with an average population size of approximately 400. Whole SA1s aggregate to form Statistical Areas Level 2 (SA2) in the Australian Statistical Geography Standard (ASGS) Main Structure 23.
Health Outcomes
Following Mohanty and Niyonsenga (2019) 1, the outcome variables included in this study were self-reported health, scores on the Short Form 36 Questionnaire (SF-36), a widely used multi-dimensional measure of health-related quality of life 26, using data available in HILDA surveys. We used the general health, physical functioning and mental health domains as our outcome variables. Scores on each of the SF-36 domains are standardised, and component scores range from 0 to 100, higher scores indicating better health 26. It is a generic measure, as opposed to one that targets a specific age, disease, or treatment group 1.
Socio-economic and Demographic Control variables
We have included a range of explanatory variables in the model that include a set of variables relating to the community and societal social capital realised at individual level, socio-economic and demographic variables, physical activity status, any major adverse health event that occurred in the previous year, that may have influenced the general, physical or mental health of an individual along with his/her care-giving status. Like in Mohanty and Niyonsenga (2019) the care-giving status of the individual, the main factor, is defined as a dichotomous variable taking the value 1 if the individual actively cares for a household member/non-resident due to a long-term health condition or elderly status and 0 otherwise. The study also controlled for the number of hours a person spends on care-giving duties in a week irrespective of their carer status. As seen below, Tables 1-3 present the indicators of community participation, social support and trust-reciprocity respectively, while Table 4 lists the predictors and outcome variables used in this analysis, along with their descriptive statistics. All these variables included in the model are fully detailed in the 2017 HILDA User Manual – Release 16 24.
Table 1: Community Participation: Rotated factor loadings (pattern matrix) and unique variances.
Variable (n= 37290)
|
Civic engagement, political participation and breadth of participation
|
Informal social connectedness
|
Unique Variation in Variables
|
Community participation: Attend events that bring people together such as fetes, shows, festivals or other community events
|
0.505
|
0.473
|
0.521
|
Community participation: Get involved in activities for a union, political party, or group that is for or against something
|
0.756
|
0.031
|
0.428
|
Community participation: Make time to attend services at a place of worship
|
0.590
|
0.078
|
0.645
|
Community participation: Encourage others to get involved with a group that’s trying to make a difference in the community
|
0.841
|
0.141
|
0.273
|
Community participation: Volunteer your spare time to work on boards or organising committees of clubs, community groups or other non-profit organisations
|
0.755
|
0.150
|
0.407
|
Community participation: Get in touch with a local politician or councillor about issues that concern me
|
0.762
|
0.021
|
0.419
|
Community participation: Give money to charity if asked
|
0.406
|
0.314
|
0.736
|
Community participation: Have telephone, email or mail contact with friends or relatives not living with you
|
-0.005
|
0.804
|
0.353
|
Community participation: Chat with your neighbours
|
0.296
|
0.462
|
0.699
|
Community participation: Talk about current affairs with friends, family or neighbours
|
0.348
|
0.515
|
0.613
|
Community participation: Make time to keep in touch with friends
|
0.087
|
0.794
|
0.362
|
Community participation: See members of my extended family (or relatives not living with me) in person
|
0.117
|
0.654
|
0.559
|
Table 2: Personal Social Cohesion: Rotated factor loadings (pattern matrix) and unique variances.
Variable (n= 184076)
|
Personal Social Exclusion
|
Personal Social Cohesion
|
Unique Variation in Variables
|
People don’t come to visit me as often as I would like
|
0.728
|
0.015
|
0.469
|
I don’t have anyone that I can confide in
|
0.750
|
-0.360
|
0.308
|
I have no one to lean on in times of trouble
|
0.760
|
-0.367
|
0.288
|
I often feel very lonely
|
0.729
|
-0.236
|
0.412
|
I often need help from other people but can’t get it
|
0.766
|
-0.182
|
0.380
|
When I need someone to help me out, I can usually find someone
|
-0.411
|
0.759
|
0.254
|
I enjoy the time I spend with the people who are important to me
|
-0.146
|
0.825
|
0.298
|
There is someone who can always cheer me up when I am down
|
-0.291
|
0.703
|
0.421
|
When somethings on my mind, just talking with the people I know can make me feel better
|
-0.102
|
0.861
|
0.248
|
I seem to have a lot of friends
|
-0.354
|
0.451
|
0.671
|
Table 3: Trust and Reciprocity: Rotated factor loadings (pattern matrix) and unique variances.
Variable
|
Trust
|
Distrust
|
Unique Variation in Variables
|
Most people you meet keep their word
|
0.79 3
|
-0.254
|
0.307
|
Most people you meet make agreements honestly
|
0.858
|
-0.204
|
0.222
|
Generally speaking, most people can be trusted
|
0.821
|
-0.215
|
0.280
|
Most of the time people try to be helpful
|
0.855
|
-0.168
|
0.241
|
Most people would try to take advantage of you if they got a chance
|
-0.294
|
0.755
|
0.343
|
Most people you meet succeed by stepping on other people
|
-0.356
|
0.749
|
0.312
|
People mostly look out for themselves
|
-0.079
|
0.803
|
0.349
|
Table 4: Descriptive Statistics
Variable
|
Mean/Percent (N=12,767)
|
Standard. Deviation
|
Minimum
Value
|
Maximum
Value
|
SF36 General Health Component Score
|
67.62
|
20.83
|
0.00
|
100
|
SF36 Mental Health Component Score
|
74.05
|
17.58
|
0.00
|
100
|
Actively cares for a household member/non-resident due to long-term health condition, elderly (%)
|
7.76%
|
0.27
|
|
|
Time spent in (hrs/mins) per week Caring for disabled/elderly relative (n=8,416)
|
1.69
|
9.87
|
0.00
|
168
|
Age
|
45.10
|
18.71
|
15.00
|
97
|
Gender: 1 for male
|
47%
|
0.50
|
|
|
Household financial year gross total income ($)
|
124708
|
111903
|
-74269
|
2827827
|
Life events in past year: Serious personal injury/illness
|
10%
|
0.30
|
|
|
Educational level
|
Tertiary University Education (Base Category)
|
26.44%
|
0.44
|
|
|
Avd diploma, diploma, Cert III or IV
|
31.71%
|
0.47
|
|
|
Year 12
|
15.15%
|
0.36
|
|
|
Year 11 and below
|
26.71%
|
0.44
|
|
|
Employment Status
|
Employed - works 35+ hours a week (Base Category)
|
40.89%
|
0.49
|
|
|
Employed - works less than 35 hours a week
|
21.23%
|
0.41
|
|
|
Unemployed, retired, home duties, students & others
|
37.87%
|
0.49
|
|
|
How often participate in Physical Activities?
|
Not at all (Base Category)
|
10.95%
|
0.31
|
|
|
Less than once a week
|
16.68%
|
0.37
|
|
|
1 to 2 times a week
|
23.54%
|
0.42
|
|
|
3 times a week
|
15.75%
|
0.36
|
|
|
More than 3 times a week
|
33.09%
|
0.47
|
|
|
Social Capital Summary Variables
|
Civic engagement, political participation and breadth of participation
|
1.93
|
1.13
|
-0.13
|
7.17
|
Informal social connectedness
|
5.69
|
1.20
|
1.20
|
8.87
|
Personal social exclusion
|
5.78
|
1.64
|
1.47
|
12.12
|
Personal social cohesion
|
8.97
|
1.29
|
1.79
|
12.56
|
Level of trust in the community
|
7.66
|
1.19
|
1.61
|
11.28
|
Level of distrust in the community
|
6.98
|
1.56
|
1.78
|
12.44
|
Note: The Table 4 presents summary statics of the variables used in the fully-adjusted multi-level mixed effects models for the SF36 mental and general health outcomes in wave 14 of HILDA data
Personal Social Capital Variables
Following Berry et al (2010) and Berry and Rickwood (2000) 20 21, we identified 12 variables in HILDA that assessed community participation, 10 variables that described informal social connectedness or exclusion and 7 variables that described level of trust/distrust in the community. We conducted principal component factor analysis to summarize and identify any common underlying theme in each category separately. Our analysis clearly identified two different underlying constructs or factors for each of the community participation, personal social cohesion and trust and reciprocity categories. Following Berry et al (2010) 20, we named Factor 1 as ‘Civic engagement, political participation and breadth of participation’ and Factor 2 as ‘Informal social connectedness’ in the community participation category. Likewise, in the personal social cohesion category, the two factors were named as ‘Personal social exclusion’ and ‘Personal social cohesion’. Further, in the trust and reciprocity category, the factors were named as ‘trust’ and ‘distrust’. The set of variables grouped under these factors in each category, their rotated factor loadings (pattern matrix), unique variances are presented in Tables 1-3 in the Additional file (Additional file 1.doc).
Statistical Methodology
The study used cross-sectional multilevel mixed effects regression analysis considering the hierarchical or clustered structure of the data: individuals/households are clustered within geographical areas or communities they live; and communities are also nested within other higher administrative units 27-29. Communities are more likely to share similar policy and health care infrastructure than those living in other communities.
Our explanatory variables were grouped into two levels to reflect the hierarchical nature of the data. Level 1 variables corresponded to individual level characteristics. At the individual level, the model included carer/non-carer status, time spent caring for disabled/elderly relatives, employment status, age, gender, any major injury/illness in past year, educational level, physical activity status, household gross total income and the sets of community participation, social connection, trust and reciprocity summary variables as main effects. To test the moderation effect of the community participation, social connection and trust and reciprocity on carer health, the model also included interaction-effects of carer’s status with these social capital summary variables. Level 2 variables corresponded to SA1 random effect (i.e., allowing for variations in average values of health scores across SA1s) and caregiving status (carer/non-carer) as a random slope (i.e., allowing for variations in the impact of caregiving on carers’ health across SA1s).
We fitted separate models for each of the SF-36 component scores namely, General Health, Mental Health and Physical Functioning, as outcome measures. We estimated a series of regression models on each health outcome measure (7 models each) and added the explanatory variables sequentially in groups into the successive models. Our analysis for the SF-36 physical functioning measure did not show any significant effects of carer/non-carer status and the results are not presented here. Analyses were performed using STATA 14 (XTMIXED FE || RE:) with robust standard errors estimation 30 31.
The model 1 in each health component score (Tables 5 and 6, first column), was the simplest multilevel model with one level 1 explanatory variable - carer/non-carer status (main effect), and SA1 level random effects (intercept) at level 2.
In model 2 for each health component score (Tables 5 and 6, second column), we additionally included carer/non-carer status as a random slope at level 2 - SA1 level. This means that we also allowed the nature of the relationship between the carer status and the health component scores to vary across SA1s (random slope) 32 33. In models 3-5 (Tables 5 and 6, 3rd-5th column), we added the summary variables for community participation, personal social connection and trust and reciprocity in sequence. We added the other individual level confounders such as: age, gender, employment, education, household income, any major personal injury/illness last year and the physical activity/exercise status in models 6. Finally, in models 7 (Tables 5 and 6, 7th column), we interacted the community participation, social connection and trust and reciprocity level summary variables with carer status.
Estimates of effects were reported with associated 95% confidence intervals as suggested by NEJM guidelines 34. The a priori level of significance was set at the usual 5% alpha and all p-values reported in Tables using the asterisk convention: ***: p < 0.01; **: p < 0.05; *: p < 0.10 35, with the last category meant to show that a “trend towards statistical significance” has to be noted 36-38.
Table 5: Multilevel mixed effects estimates for SF36 general health component
Variables (no. of observations=12767 and no. of SA1s=5004)
|
Model 1:
Coeff (95% CI)
|
Model 2:
Coeff (95% CI)
|
Model 3:
Coeff (95% CI)
|
Model 4: Coeff (95% CI)
|
Model 5: Coeff (95% CI)
|
Model 6: Coeff (95% CI)
|
Model 7:
Coeff (95% CI)
|
Variance (Individual-level random effects)
|
381.874***(370.388, 393.716)
|
376.868*** (365.081, 389.037)
|
363.916*** (352.640, 375.552)
|
340.609*** (330.159, 351.389)
|
335.010*** (324.729, 345.634)
|
281.753*** (273.349, 290.414)
|
281.744*** (273.339, 290.408)
|
Variance (SA1 level random effect)
|
51.784***
(42.917, 62.483)
|
54.451***
(44.969, 65.931)
|
43.959*** (35.467, 54.483)
|
31.729***
(24.525, 41.048)
|
32.350***
(25.171, 41.577)
|
13.973***
(9.359, 20.863)
|
13.960***
(9.346, 20.852)
|
Variance (Carer random slope)
|
|
67.163**
(32.154, 140.29)
|
54.575** (23.572 126.355)
|
36.886* (12.601, 107.967)
|
33.125* (10.240, 107.160)
|
37.124** (14.716, 93.652)
|
37.268**
(14.7694, 94.042)
|
Covariance (Carer and SA1)
|
|
-17.374
(-38.073, 3.324)
|
|
|
|
|
|
Carer
|
-5.045***
( -6.387,
-3.704)
|
-5.336 ***
(-6.773, -3.899)
|
-5.504***
(-6.88, -4.125)
|
-4.302***
(-5.614, -2.992)
|
-4.264*** ( -5.555, -2.973)
|
-1.135*
(-2.459, 0.189)
|
-2.565
( -14.750, 9.621)
|
Civic engagement, political participation and breadth of participation
|
|
|
0.071
( -0.242, 0.385)
|
-0.098
(-0.398, 0.202)
|
-0.302**
(-0.602, -0.003)
|
-.002
(-0.279, 0.276)
|
0.009
(-0.279, 0 .297)
|
Informal social connectedness
|
|
|
3.913***
(3.622, 4.205)
|
1.034***
(0.706, 1.362)
|
0.628***
(0.297, 0.959)
|
0.545***
(0.239, 0.851)
|
0.547***
(0.230, 0.865)
|
Personal social exclusion
|
|
|
|
-3.447***
(-3.671, -3.224)
|
-2.894***
(-3.130, -2.657)
|
-1.988***
( -2.205, -1.771)
|
-1.995***
(-2.221, -1.769)
|
Personal social cohesion
|
|
|
|
3.059***
(2.774, 3.343)
|
2.599***
(2.308, 2.890)
|
1.972***
(1.706, 2.237)
|
2.009***(1.732, 2.286)
|
Trust
|
|
|
|
|
1.753***
(1.442, 2.063)
|
2.416***
(2.125, 2.708)
|
2.348***
(2.045, 2.651)
|
Distrust
|
|
|
|
|
-1.170***
(-1.401, -0.940)
|
-1.421***
(-1.632, -1.211)
|
-1.409***
(-1.629, -1.191)
|
Carer # Civic engagement, political participation and breadth of participation
|
|
|
|
|
|
|
-0.181
(-1.177, 0.815)
|
Carer # Informal social connectedness
|
|
|
|
|
|
|
0.467
( -0.637, 1.570)
|
Carer # Personal social exclusion
|
|
|
|
|
|
|
-1.886***
(-2.647, -1.125)
|
Carer # Personal social cohesion
|
|
|
|
|
|
|
1.601***
(0 .680, 2.523)
|
Carer # Trust
|
|
|
|
|
|
|
3.207***
(2.197, 4.217)
|
Carer # Distrust
|
|
|
|
|
|
|
-1.589***
(-2.335, -0.843)
|
Notes. All the p-values have been replaced by stars and categorised as follows. ***: p < 0.01; **: p < 0.05; *: p < 0.10; However, confidence intervals use the usual 95% confidence level. Model 1: Null model with SA1 random effects and carer status at individual level; Model 2: Model 1 added with carer random slopes at SA1 level; Model 3: Model 2 added with community participation at individual level; Model 4: Model 3 added with personal social connection at individual level; Model 5: Model 4 added with trust at individual level; Model 6: Model 5 added with other individual level confounders; Model 7: Model 6 with community and social connection interactions with Carer.
Table 6: Multilevel mixed effects estimates for SF36 mental health component
Variables (no. of observations=12767 and no. of SA1s=5004)
|
Model 1
Coeff (95% CI)
|
Model 2
Coeff (95% CI)
|
Model 3
Coeff (95% CI)
|
Model 4
Coeff (95% CI)
|
Model 5
Coeff (95% CI)
|
Model 6
Coeff (95% CI)
|
Model 7
Coeff (95% CI)
|
Variance (Individual-level random effects)
|
270.740***
(262.516, 279.222)
|
267.462***
(259.074, 276.121)
|
245.992***
(238.372, 253.855)
|
202.326***
(196.193, 208.652)
|
335.019***
(324.729, 345.634)
|
183.028***
(177.458, 188.773)
|
183.003***
(177.434, 188.748)
|
Variance (SA1 level random effect)
|
39.162***
(32.587, 47.063)
|
38.244***
(31.498, 46.434)
|
25.019***
(19.585, 31.959)
|
15.656***
(11.755, 20.851)
|
32.350***
(25.171, 41.577)
|
12.569***
(9.127, 17.309)
|
12.556***
(9.116, 17.294)
|
Variance (Carer random slope)
|
|
47.688**
(23.183, 98.096)
|
37.779**
(16.590, 86.032)
|
44.053***
(24.043, 80.719)
|
33.125**
(10.240, 107.160)
|
28.850**
(12.680, 65.640)
|
28.255**
(12.210, 65.384)
|
Covariance (Carer and SA1)
|
|
4.101
(-10.91, 19.114)
|
|
|
|
|
|
Carer
|
-3.863***
(-4.996, -2.730)
|
-3.890***
(-5.145, -2.635)
|
-4.164***
(-5.335, -2.994)
|
-2.626***
( -3.712, -1.540)
|
-4.264***
( -5.555, -2.973)
|
-1.770***
( -2.879, -0.661)
|
1.381
( -8.751, 11.513)
|
Civic engagement, political participation and breadth of participation
|
|
|
0.827***
(0.570, 1.084)
|
0.630***
(0.399, 0.861)
|
-0.302**
( -0.602, -0.003)
|
.082
( -0.144, 0.308)
|
.097
(-.137, 0.331)
|
Informal social connectedness
|
|
|
4.691***
(4.452, 4.930)
|
1.279***
(1.026, 1.531)
|
0.628***
(0 .297, 0.959)
|
1.039***
(0 .790, 1.288)
|
1.035***
(0.777, 1.293)
|
Personal social exclusion
|
|
|
|
-4.538***
(-4.710, -4.366)
|
-2.894***
( -3.130, -2.657)
|
-3.656***
( -3.832, -3.479)
|
-3.669***
(-3.852, -3.486)
|
Personal social cohesion
|
|
|
|
3.063***
(2.844, 3.282)
|
2.599***
(2.308, 2.890)
|
2.600***
(2.384, 2.817)
|
2.659***
(2.434, 2.885)
|
Trust
|
|
|
|
|
1.753***
(1.442, 2.063)
|
1.849***
(1.612, 2.086)
|
1.823***
(1.577, 2.069)
|
Distrust
|
|
|
|
|
-1.170***
( -1.401, -.940)
|
-1.310***
( -1.482, -1.139)
|
-1.312***
(-1.489, -0.134)
|
Carer # Civic engagement, political participation and breadth of participation
|
|
|
|
|
|
|
-0.091
( -0.922, 0.740)
|
Carer # Informal social connectedness
|
|
|
|
|
|
|
1.126**
(0.207, 2.046)
|
Carer # Personal social exclusion
|
|
|
|
|
|
|
-3.456***
( -4.091, -2.820)
|
Carer # Personal social cohesion
|
|
|
|
|
|
|
1.912***
(1.143, 2.681)
|
Carer # Trust
|
|
|
|
|
|
|
2.133***
(1.291, 2.976)
|
Carer # Distrust
|
|
|
|
|
|
|
-1.343***
( -1.964, -0.721)
|
Notes. All the p-values have been replaced by stars and categorised as follows. ***: p < 0.01; **: p < 0.05; *: p < 0.10; However, confidence intervals use the usual 95% confidence level. Model 1: Null model with SA1 random effects and carer status at individual level; Model 2: Model 1 added with carer random slopes at SA1 level; Model 3: Model 2 added with community participation at individual level; Model 4: Model 3 added with personal social connection at individual level; Model 5: Model 4 added with trust at individual level; Model 6: Model 5 added with other individual level confounders; Model 7: Model 6 with community and social connection interactions with Carer.