Examining correlates of life satisfaction among Indian older adults using household fixed-effect approach

Life satisfaction is one of the most important components of a person’s overall quality of life and a vital element of successful ageing. Few studies have previously attempted to investigate the level of life satisfaction among Indian older adults; however, the majority of them were at risk of omitted variable bias. This study, while controlling for household-level unobserved confounders, aims to investigate the correlates of life satisfaction among Indian older adults using household fixed-effect approach. To achieve the objectives of the study, cross-sectional data from the first wave of the Longitudinal Ageing Study in India (LASI, 2017–18) were utilized. Households with only one study participant were excluded from the study because household fixed effect analysis requires at least two respondents per household. The total sample size of the study was 14,307 older adults (Nfemale = 7259; Nmale = 7048) aged 60 years or above. Simple OLS regression model, random effect model, and household fixed-effect model were employed to assess the factors associated with life satisfaction among older adults in India. According to the household fixed effect model, age, education, functional health, food insecurity, and self-rated health were found to be significantly associated with life satisfaction. On the other hand, sex, marital status, working status, and morbidity status were not found to be associated with life satisfaction. According to the random effect model, among the household-invariant factors, place of residence, caste, MPCE (monthly per capita expenditure) quantile, and region were significantly associated with life satisfaction, while religion was found insignificant. This study offers new insights on the predictors of life satisfaction among older adults in India. In order to improve the general well-being of the elderly, the study urges policymakers to address a number of issues, including functional health and food security.


Introduction
The world's population is ageing faster than ever. The global population of older adults aged 60 years and above has increased to 962 million in 2017 from 382 million in 1980 and is further expected to rise to 2.1 billion in 2050 [1]. Moreover, the global population of oldest old adults (aged 80 or above) is expected to rise to 425 million in 2050 from 137 million in 2017, a more than threefold increase [2]. Decreased fertility rates and increased life expectancy are the primary reasons for this demographic phenomenon of rapid population ageing. Although progress in medical research and public health can be seen as an achievement of the human race, the resultant population ageing comes with its own set of challenges. For the past few 1 3 survey to collect the data. The eventual unit of observation in LASI was a LASI-eligible household (LEH). LEH was defined as a household with at least one member aged 45 and above. From the selected households, all men and women aged 45 and above and their respective spouses were included in the LASI. The overall household and individual response rate in the survey was 95.8% and 87.3%, respectively. The survey agencies obtained prior consent from the respondents before data collection. To conduct the survey, the Indian Council of Medical Research (ICMR) provided the required guidelines and ethical permission. Survey design and data collection details, together with the detailed methodology, have been published elsewhere [8,16].

Study sample
The overall sample size of LASI is 72,250 individuals aged 45 and above. However, this study only focuses on older adults aged 60 years and above; hence 40,786 respondents aged less than 60 years were dropped from the study. In this study, we have applied a household fixed-effect regression model, which can be applied only on the sample of households with two or more survey participants. Thus, those older adults who were the single survey participants from a household (n = 16,740) were dropped from the study. After removing missing and invalid cases (n = 417), the final sample size of 14,307 older individuals was achieved.

Outcome variable
In LASI, life satisfaction was assessed using five items: The responses for all the five questions were collected on a 1 to 7 Likert scale in which 1 to 7 values were coded as "Strongly disagree", "Somewhat disagree", "Slightly disagree", "Neither agree nor disagree", "Slightly Agree", "Somewhat Agree", and "Strongly Agree" respectively. The internal consistency of the five items was checked using Cronbach' alpha. The Cronbach' alpha value was 0.90, indicating a very high internal consistency/reliability. The final life satisfaction variable was constructed by adding the above-mentioned five items. The life satisfaction variable was a continuous variable with range 5 to 35, mean 24.39, and a standard deviation of 7.14. Higher values of the variable reflect higher life satisfaction.

Individual factors
1. Age was coded into three categories as 60-69 years, 70-79 years, and 80 years and above. 2. Sex had two categories as male and female. 3. Education was categorized into five categories as (i) no education, (ii) primary or less than primary, (iii) middle, (iv) secondary, and (v) higher. 4. Marital status was coded into three categories as (i) currently married, (ii) widowed, and (iii) other. 5. Working status was coded into two categories as working and not working.

Household/community level factors
1. Place of residence had two categories as rural and urban. 2. Religion was categorized into three categories as Hindu, Muslim, and others. 3. Caste had four categories as (i) scheduled caste (SC), (ii) scheduled tribe (ST), (iii) other backward class, and (iv) none of the above three. 4. MPCE (monthly per capita expenditure) quantile was categorized into five categories as (i) poorest, (ii) poorer, (iii) middle, (iv) richer, and (v) richest. 5. Region represents geographical region of the country which have six categories as (i) north, (ii) east, (iii) central, (iv) west, (v) north-east, and (vi) south.

Health and wellbeing indicators
1. Morbidities represents number of morbidities an individual had at the time of survey. The variable was categorized into four categories as 0, 1, 2, 3 + . 2. Difficulty in activities of daily living (ADL) variable was categories into three categories as (i) no ADL, (ii) 1 ADL, and (iii) more than 1 ADL. ADLs represents basic day-to-day activities (sitting to standing, feeding, dressing, personal hygiene, grooming, bathing etc.). 3. Difficulty in instrumental activities of daily living (IADL) variable was categorized into three categories as (i) no IADL, (ii) one IADL, and (iii) more than one IADL. IADLs are not necessary for functional living of an individual; however, are very important for a person to live independently in a society. IADLs include cooking, cleaning, laundry, transportation, managing finance, etc. 4. Variable of frequent physical exercise was coded into two categories as yes and no. 5. In LASI, food insecurity status was assessed using three questions (1) In the last 12 months, did you ever reduce the size of your meals or skip meals because there was not enough food at your household? (2) In the last 12 months, were you hungry but didn't eat because there was not enough food at your household? (3) In the past 12 months did you ever not eat for a whole day because there was not enough food at your household? All the three questions were Yes/No type question. An individual was categories as food insecure if he/she answered 'yes' to any of the three question and was categorized as food secure if he/she answered 'no' to all the three questions. 6. Self-rated health (SRH) was categories into three categories as (i) poor, (ii) fair, and (iii) good.

Analytical approach
Descriptive statistics and bivariate analysis were employed to understand the socioeconomic and health profile of the study participants. The primary analysis explored the correlates of life satisfaction using three regression models: ordinary least square regression (OLS), household random-effect model, and household fixed-effect model. The OLS model was employed to get a basic understanding of the correlates of life satisfaction among older adults. The normality assumption of OLS regression was checked using the Shapiro-Wilk test of normality. The residual-verses-fitted plot was visually inspected to check the heteroscedasticity assumption of the OLS regression; however, no evidence of heteroscedasticity was observed. To check potential multicollinearity in the model, variance inflation factors (VIF) was estimated with a cut-off value of 10.
Household random-effect model was employed to explore the correlates of life satisfaction among older adults to control for between-household unobserved heterogeneity. The Breuch-Pagan Lagrange Multiplier (LM) test was utilised to determine whether using a random-effect model is advantageous to simple OLS regression. The null hypothesis of the LM test was that the variances of the relevant variables across the households were zero. The null hypothesis was rejected (p-value < 0.001), and it was concluded that the random effect model is advantageous over the simple OLS model. The random-effect model had the following structure: where Y ij is the outcome variable for the j th individual in the i th household, α is the intercept, the matrix of predictor variables is denoted by X, and the matrix of associated fixed regression coefficients is denoted by β. The total variance of the model is broken up into two parts: u i representing the between-household error, and within-household error represented by e ij .
Furthermore, household fixed-effects regression model was used to account for unobserved within-household heterogeneity. Unknown factors such as food behaviour, traditions, ethnic traits, household structure, and social environment are generally shared between family members within one household, not across households. These kinds of common unobserved characteristics shared by individuals from the same household can be controlled using a household fixed-effect model. This approach has been widely used in previous research [17][18][19][20]. One very important advantage of the fixed-effect model is that you do not have to worry about omitted variable bias at the household level. Hausman's specification test was performed to examine whether the preferred model is fixed-effect model or random-effect model. Basically, Hausman's test tests the (null) hypothesis that within-household errors (ui) are not correlated with the regressors. The p-value of the test was found to be less than 0.001, indicating that the fixed-effect model is appropriate to use. The household fixed-effect model had the following structure:

3
where α i is the unknown intercept for each household and u ij is the error term.
Although the fixed-effect model was found to be advantageous over the random-effect model, the random-effect model was included in the analyses to explore the household invariant correlates of life satisfaction. Variables like place of residence, MPCE quintile, region, caste, and religion were invariant within the household i.e. they cannot be included in the household fixed-effect model. All the analyses were conducted using STATA-16.

Results
Background characteristics, along with average life satisfaction score, have been presented in Table 1. The majority of the study participants were aged 60-69 years (61%), living in rural areas (67%), had no education (51%), were currently married (94%), belonged to the Hindu religion (74%), were not working (70%), and living in the southern region of the country (22%). The overall average life satisfaction score was 24.4 (SD = 7.1) among the study participants. According to the results of ANOVA and t-test, the average life satisfaction score was not significantly different among the categories of age, morbidity status, and physical exercise. Individuals living in urban areas (x̅ = 25.6; SD = 7.1) had a higher level of life satisfaction compared to their rural counterparts (x̅ = 23.8; SD = 7.1). The average life satisfaction level increased progressively with increase in education level. Older adults belonging to the richest MPCE quantile (x̅ = 25.2; SD = 7.2) had significantly higher life satisfaction than the older adults belonging to the poorest MPCE quantile (x̅ = 23.3; SD = 7.2). Among regions of the country, elderlies living in the western and eastern regions had the highest (x̅ = 27.4; SD = 6.6) and lowest (x̅ = 23.1; SD = 6.9) levels of life satisfaction, respectively. Older adults who reported food insecurity had significantly lower levels (x̅ = 21.0; SD = 7.6) of life satisfaction than their counterparts (x̅ = 24.3; SD = 7.1). Older adults with good self-reported health seemed to have higher life satisfaction compared to older adults with poor self-reported health.
The results of multivariable regression analyses are presented in Table 2. In the table, model-1 is a simple OLS model, model-2 is a household random effect model, and model-3 is a household-fixed effect model. According to all three models, being 80 + years old was significantly associated with increased life satisfaction. According to model-3, adults aged 80 and older had 1.11 units (β = 1.11; 95% CI 0.63-1.59) higher life satisfaction than those aged 60 to 69. Sex was not found to be significantly associated with life satisfaction in all three models. Place of residence is a household invariant factor; thus, it was not included in model-3. However, according to model-1(β = 0.49; 95% CI 0.21-0.77) and model-2 (β = 0.83; 95% CI 0.48-1.17), urban older adults had significantly higher levels of life satisfaction compared to rural older adults. Marital status was found insignificant in all three models. According to the random-effect model, religion was not significantly associated with life satisfaction. However, individuals of 'other' religions had a significantly higher level of life satisfaction than Hindu individuals at a 5% level of significance. Caste was found to be significantly associated with life satisfaction in both model-1 and model-2. MPCE quantile was positively associated with life satisfaction. According to the random effect model, older adults belonging to the richest MPCE quantile had 0.89 (β = 0.89; 95% CI 0.40-1.38) units higher level of life satisfaction than older adults belonging to the poorest MPCE quantile. Working status was significantly associated with life satisfaction in model-1 (β = 0.46; 95% CI 0.19-1.07) and model-2 (β = 0.23; 95% CI 0.01-0.45) at 5% level of significance, but the association became insignificant after controlling for unobserved heterogeneities in model-3. Surprisingly, having morbidities was not found to be associated with life satisfaction in all three models. According to all three models, older adults with difficulty in one ADL did not have significantly different life satisfaction levels than older adults without difficulty in any ADL. However, older adults with difficulty in more than one ADL had significantly lower life satisfaction levels than older adults with no difficulty in any ADL. According to the fixed-effect model, older adults with difficulty in more than one ADL had 0.47 (β = − 0.47; 95% CI − 0.86 to − 0.13) units lower level of life satisfaction than older adults with no difficulty in ADL. Similarly, compared to individuals without difficulty in any IADL, individuals with difficulty in more than one IADL had 0.57 (β = − 0.57; 95% CI − 0.88 to − 0.27) units lower level of life satisfaction. Frequent physical exercise was not found to be significantly associated with life satisfaction in all three models. According to all three models, older adults with food insecurity had significantly lower levels of life satisfaction compared to their counterparts. However, the beta coefficient decreased to − 1.81 (β = − 1.81; 95% CI − 2.23 to − 1.39) in the random effect model from -2.52 (β =− 2.52; 95% CI − 2.23 to − 1.39) in the OLS model and further decreased to − 0.91 (β = − 0.91; 95% CI − 1.44 to − 0.36) in the household fixed effect model. Self-rated health was found to be positively associated 1 3 with life satisfaction in all three models. Compared to older adults with poor SRH, older adults with fair and good SRH had significantly higher levels of life satisfaction. However, the beta coefficient was substantially lower in the fixed effect model than in the OLS and random effect models. 1 3

Discussion
The objective of this study was to investigate the demographic, socioeconomic, and health factors associated with life satisfaction among older adults in India using household fixed-effect regression model. Age, educational attainment, functional health (ADL and IADL), food security, and self-rated health were found to be significantly associated with life satisfaction after controlling for household-level unobserved heterogeneities. According to previous studies, to make a judgement about one's satisfaction with life, an individual might review and compare his/her objective life conditions with respective imagined ideal conditions, weight each circumstance according to its relative importance, and then add up all of these evaluations to form an overall judgement [21]. It is reasonable to predict that levels of pleasure and life satisfaction should decrease as people age since they are more vulnerable to poor health, financial insecurity, and diminished social capital [22]. Socioemotional selectivity theory, on the other hand, proposes that people get happier and more satisfied as they get older. According to the theory, as people enter into later years of life, they become increasingly conscious of the amount of remaining time in their life [23]. They become more mindful of savouring the present moments as a result of their growing mortality awareness. Our findings support the second proposition by demonstrating that older adults aged 80 and up had significantly higher levels of life satisfaction than adults aged 60 to 69. The comparison, however, is within older age groups and makes no comparison to younger age groups.
Sex, marital status, and religion were not found to be significantly associated with life satisfaction in any of the three models. Results from earlier studies on the relationship between gender and life satisfaction have been varying. In some studies, women were found to have higher levels of life satisfaction than males [24][25][26], whereas, in others, men were found to be more satisfied [27,28]. Apart from that, there is a third group of studies which show no gender differences MPCE Monthly per capita income   [29,30]. One thing to be noted here is that most of the studies which found a significant association between gender and life satisfaction had a small effect size [31]. The evidence of contradictory findings suggests that a number of variables that differ from community to community may mediate the association between gender and life satisfaction. Similar was the case with the association between marital status and life satisfaction; some studies found a significant association between the two variables, while others did not [32,33]. In both the OLS regression model and the random effect model, older adults belonging to OBC and general caste had significantly higher levels of life satisfaction than SC older adults. There has not been much research on life satisfaction among caste groups in India, but it is probable that the discrimination and denial of basic human rights that SC and ST groups experience in society might contribute to their lower levels of life satisfaction. Education and MPCE quantile were found to be positively associated with life satisfaction. Previous studies have shown that higher levels of education are associated with better socioeconomic status in society and a higher quality of life, both of which may increase a person's sense of fulfilment in life [34]. The results of this study show that in OLS and random effect models, the level of life satisfaction improved steadily with increase in educational attainment. But according to the household fixed-effect model, there was no statistically significant difference in the degree of life satisfaction among the first four categories of education; only those people with 'higher' levels of education had a significantly better level of life satisfaction compared to those with no education. Since MPCE quantile was a household invariant variable, it could not be included in the household fixed-effect model. Higher levels of MPCE quantile show better economic stability and fewer financial vows, which are vital aspects of life, especially in old age. Hence, it is plainly justified to obtain a positive association between the MPCE quantile and life satisfaction. Similar results were found in previous studies as well [34,35].
Multimorbidity has become a serious public health concern among older adults as it has been found to be strongly associated with poor functional health [36], increased burden of healthcare financing [37], and decreased life expectancy Model-1 is simple OLS model, Model-2 is random effect model, and Model-3 is household fixed effect model MPCE Monthly Per Capita Income a P-value < 0.001, b P-value < 0.01, c P-value < 0.05, ns: not significant  [38]. Considering the severe impact of multimorbidity on numerous dimensions of life, it was hypothesised that it would have a strong negative association with life satisfaction. Contrary to expectations, no significant association was found between life satisfaction and multimorbidity. Similar results were found in a study conducted on very old Japanese people [39]. These findings indirectly indicate that older people who participated in the study might not give weightage to morbidity status while making a judgement about life satisfaction. Furthermore, since the items used to calculate the life satisfaction score make implicit references to the past, as a result, even if they were morbid at the time of the interview, older people may have claimed high life satisfaction based on their earlier life experiences. There have been other studies as well which show that multimorbidity is associated with lower quality of life among older adults, although none of them was conducted in an Indian context [40,41]. To thoroughly evaluate the potential impact of morbidities on life satisfaction, longitudinal follow-up studies are required. Functional health is an important dimension of the overall well-being of an individual. Difficulty in performing ADLs and IADLs implies an increased dependency on others and demand for care. This study found that having difficulty in more than one ADL or IADL is significantly associated with lower life satisfaction. According to the household fixed-effect model, having trouble performing more than one ADL was weakly but significantly associated with lower life satisfaction. These findings are in line with the previous research [42]. On the other hand, having difficulty in more than 1 IADL was strongly associated with lower life satisfaction than having no difficulty. Frequent physical activity has been found to be associated with key health and well-being indicators such as good functional capacity, decreased risk of noncommunicable diseases, good mental health, better body composition, and better cognitive functioning [43]. Despite the numerous advantages, no significant association was found between physical activity and life satisfaction, which implicitly suggests that study participants did not consider physical activity to be an element of life satisfaction. Similar results were found in a study conducted in Spain [44].
As has already been widely established in previous studies, food insecurity was found to be strongly associated with lower levels of life satisfaction [35,45]. Considering the definition of food insecurity, which says 'limited or uncertain availability of nutritionally adequate and safe foods' [46], one might envisage the dire social and economic circumstances in which a person experiencing food insecurity finds themselves. Among older people, the phenomenon of food insecurity has been found to be associated with poverty, social inequality, and financial deprivation, all of which have a significant negative impact on one's health and well-being. In light of this, we can think how satisfied one might feel with life when he/she is not even able to get adequate food to meet his/her nutritional requirements. Therefore, it is recommended that policymakers focus on the food security of older adults in order to improve subjective as well as objective well-being.
In all three models used in the study, self-rated health (SRH) was found to be strongly associated with life satisfaction. The finding is consistent with past international studies conducted in numerous countries [47,48]. SRH indicates one's assessment of his/her current overall health. So, the findings of this study suggest that if a person feels good about his health, it is likely that he/she will feel comparatively more satisfied with his life, and vice versa.

Strengths, limitations, and implications for future studies
This study is unique in that it examined the correlates of life satisfaction using a household-fixed effect model, controlling for all unobserved confounding factors that are constant within a given household. This study is more robust than previous studies conducted on this subject as we have eliminated the risk of omitted variable bias, at least at the household level, using the household-fixed effect model. In addition to that, the study also reported results of OLS and random effect models, which will allow for comparative insights. However, the study has a few limitations to note. First of all, the results of this study are based on cross-sectional data; hence no causal inference can be drawn from the results. Second, to apply the household fixed model, only those households were included in the analysis that had two or more than two older adults, which resulted in a significant loss of study sample. Hence, the results of the study are not nationally representative.
Future studies can explore causal factors associated with the life satisfaction of older adults using a more robust study design, such as cohort study or longitudinal study. Due to a lack of data, this study could only include a limited number of factors as independent variables in the analyses. In previous studies, some interesting factors such as environmental quality (such as comfort in terms of noise, temperature, furniture, appliances etc.) were found to be associated with laterlife life satisfaction [44]. Future studies can investigate these types of extended factors to comprehensively understand life satisfaction among older adults in the country.