Introduction

A growing ageing population in any country carries enormous social, economic and public health implications, which include higher expenditure on pension and healthcare, need for social security reforms, shrinking of workforce, and shortage of active persons who are able to support dependent older adults (Singh et al. 2013; Husain and Ghosh 2011; Kalavar and Jamuna 2011; Irudaya Rajan and Kumar 2003; Irudaya Rajan et al. 1999). The older adult population (aged 60 years and above) of India is currently the second largest in the world, after China’s (Population Reference Bureau 2012) and is projected to grow from 8 % in 2010 (around 80 million) to 19 % (encompassing 323 million people) by 2050 (Bloom 2011; Population Reference Bureau 2012).

The process of ageing has resulted in the emergence of a new epidemiological scenario in low-and-middle income countries, with a high prevalence of degenerative diseases that act as a major cause of disability, lack of mobility and death (Chatterji et al. 2008). Depression (a poor state of mental health) is a common health problem among older adults and a growing major public health concern, particularly in low-and-middle income countries (Peltzer and Phaswana-Mafuya 2013; Fiske et al. 2009), including India (Kulkarni and Shinde 2015; Maulik and Dasgupta 2012). According to the World Health Organization, “Depression is a common mental disorder, characterized by sadness, loss of interest or pleasure, feelings of guilt or low self-worth, disturbed sleep or appetite, feelings of tiredness, and poor concentration” (WHO 2008). Decreased physical, cognitive and social functioning, greater self-neglect and increased risk of suicide, all of which are in turn associated with increased mortality (Blazer 2003; Hasin et al. 2005), make late life depression an important public health problem (Fiske et al. 2009).

In India, according to the WHO Survey on Global AGEing and Adult Health (SAGE) 2007–10, about one in five older adults (19.3 %) had depressive symptoms coupled with considerable diversity across different socioeconomic groups (IIPS 2013). The other studies in India on depression among older adults were region specific. For instance, a study conducted among the elderly in a rural South Indian community estimated 13 % prevalence of depression according to the International Classification of Diseases, 10threvision (ICD-10) criteria (Rajkumar et al. 2009). However, a cross-sectional study in the rural area of West Bengal among the elderly, (60 +), observed the prevalence of depression at 54 %, using a pretested semi-structured schedule and geriatric depression scale (Maulik and Dasgupta 2012).

According to WHO, factors increasing the risk of depression in older adults include genetic susceptibility, chronic disease and disability, pain, frustration with limitations in activities of daily living (ADL), personality traits (dependent, anxious or avoidant), adverse life events (separation, divorce, bereavement, poverty, social isolation) and lack of adequate social support (WHO 2001, 2008, 2013). These factors have been empirically examined across different settings in both developed (Moussavi et al. 2007; Murata et al. 2008; Zheng et al. 2012) as well as low-and-middle income countries (Chou and Chi 2005; Gureje et al. 2007; Peltzer and Phaswana-Mafuya 2013).

The importance of social network on the mental health of older adults in an Indian rural setting cannot be undermined. The traditional family system (joint or undivided) has been providing social security for the elderly, who expect to be treated with respect (Samanta et al. 2014). However, during the last two decades, India, like many other Asian countries, is experiencing globalization led changes including rapid urbanization and huge out-migration of younger adults in search of better education and employment opportunities (Knodel and Ofstedal 2003; Vera-Sanso 2004; Croll 2008; Samanta et al. 2014). This has led to an increasing trend of nuclear family set-ups and diminishing preference for intergenerational coresidence (Kulkarni and Shinde 2015; Singh et al. 2013; Adams 2010; Bongaarts and Zimmer 2002), due to which older people are likely to be exposed to emotional, physical, and financial insecurities. This may make the elderly more prone to depressive disorders (Pilania et al. 2013), especially in rural areas, where most of the aged population live with scarce social security support, health insurance and poor public healthcare system and infrastructure (Bloom et al. 2010; Irudaya Rajan 2007).

Although, studies in the Indian setting have demonstrated the relationship between depression and various socioeconomic factors such as advanced age, low education, poverty and manual occupation (Kulkarni and Shinde 2015; Sengupta and Benjamin 2015; Sati et al. 2013; Reddy et al. 2012; Sharma and Sharma 2012; Barua et al. 2010; Ganguli et al. 1999), little attention has been paid to examine the role of social network on depression among older adults, particularly in rural areas.

This study assessed the effect of socioeconomic and demographic characteristics on depression and investigated the effect of specific and total social network on depression, especially among the elderly. The study was conducted in the rural areas of Varanasi district located in eastern Uttar Pradesh, the most populous state of India, which reported a higher prevalence of depression among older adults (30 %) than the national average (IIPS 2013).

Data and Methodology

Conceptual Framework

The study has used the conceptual framework proposed by Berkman et al. (2000), to examine the effect of social network on depression in the Indian context. The Berkman model proposed multiple pathways through which social network has potential influence on well-being. The Berkman’s model begins with the ‘upstream forces’, that is, the macro-social context in which the social network operates. These larger social and cultural factors may condition the extent, shape, nature and structure of the social network, and may include socioeconomic status, culture and processes of societal change. The structure of the social network and interaction between the members of the social network are described as the ‘mezzo’ level in Berkman’s model. The structural component, such as size (the number of network member), proximity of network members along with aspects of network that relate to interaction, such as frequency of contact, shape the definition of the social network. This study focuses on the effects of the mezzo level of the social network on depression among the older population in rural areas. The ‘micro’ level in the model includes the ways in which the social network may function, through the provision of social support, social influence, social engagement and access to material goods and resources.

This ‘micro’ level of social network function is then hypothesised to influence depression through a number of more proximate pathways. These may include psychological stress responses, health-damaging behaviour such as tobacco consumption, health-promoting behaviour such as appropriate health service utilization, medical adherence and exercise, and finally, exposure to infectious and non-communicable diseases.

Study Area and Participants

The data for the present study has been obtained from a cross-sectional survey conducted in the rural areas of Varanasi district of Uttar Pradesh, a northern Indian state during 2011. Though the proportion of elderly population in Uttar Pradesh is smaller than that of Kerala (state with the highest share of older adults), the absolute number of elderly in Uttar Pradesh is three times more (ORG and CC 2011). The old-age dependency ratio in Uttar Pradesh (136) is almost equal to the national average (CSO 2011). Health issues are a major concern in Uttar Pradesh because many older persons suffer from health problems, particularly chronic illnesses such as hypertension, diabetes mellitus, chronic obstructive pulmonary disease, heart disease, osteoarthritis, and lower back pain (Agrawal and Arokiasamy 2010). Both male and female aged 60 years and above in rural areas were included in the survey. The sample was selected by multi-stage sampling and the inclusion criteria were - aged 60 years and above, living in the village for more than one year, no evidence of severe mental disease or cognitive disorders and no hearing or speech impairment. On an average, one elderly in each village was having one of these conditions, and was therefore not included in the study.

Sampling Design

A multi-stage stratified random sampling was used to select households for the interview. Varanasi district was divided into two Tehsils (or sub-districts), namely Varanasi and Pindra. Pindra tehsil was purposively selected for this study. According to Census 2001, over 90 % of the population in Pindra tehsil live in rural areas (ORG and CC 2001). Out of four blocks in Pindra tehsil, Harhua block was selected as it was convenient for the researcher in terms of distance, time and costs. Twelve villages from Harhua block were picked by Probability Proportional to Size (PPS) sampling procedure. From a total of 2046 older persons in the twelve selected villages, the number of subjects was calculated regarding proportional to size of each village. The proportion of this study is 0.31(equal 631/2046). Selecting households in each village was conducted by systematic random sampling using household lists. One elderly per household, who met the inclusion criteria, was interviewed. The non-response was around 5 %.

Ethical Statement

The purpose of the study and procedure of data collection were described to the Gram Pradhans (elected head of each village), and the district social welfare officer to attain their permission and cooperation. Potential respondents were approached and interviewed for their willingness to participate in the study, and the overall purpose, protocol of the study, and time required to complete questionnaires were explained to them. Confidentiality and also any inconvenience that may be caused by the interview were explained to potential participants so that they could refuse to answer any question, and withdraw from the study at any time. The study did not offer any incentives either in cash or kind for participation. The entire interview, including the questionnaire and scales used in this study, was conducted in the local language, which is Hindi.

Measuring Social Network and Social Support

Glass et al. (1997) proposed multidimensional models of social network that reflected social network with children, relatives, friends, and confidant as well as total social network for each study participant using confirmatory analysis (CFA). They highlighted the need to consider the strength of each specific network separately because of the different social roles potentially fulfilled by different relationships with various people. In this model, both the number and frequency of contact with people in each type of network were used to construct summary scores for each specific type of network. A total social network measure was calculated as the sum of the four component network variables. To measure each of these constructs, the elderly respondents were asked:

  1. i.

    How many network members do you have?

  2. ii.

    How often do you have visual contact with them?

  3. iii.

    How often do you have non-visual contact with them?

These questions were asked for each network component, that is, children, siblings or relatives, friends or neighbours, and confidant. A confidant is defined as “one special person whom you feel very close and intimate with, someone you share confidences and feeling with, someone you feel you can depend on”. For measuring contact with the network component, each item was scored to a range of

  1. i.

    daily

  2. ii.

    three or more times a week

  3. iii.

    once or twice a week

  4. iv.

    once or twice a month

  5. v.

    every few months

  6. vi.

    once or twice a year

  7. vii.

    less than once a year

  8. viii.

    never

  9. ix.

    not applicable

The composite reliabilities of the latent variables in the present study were 0.83 for children social network, 0.78 for relative’s social network, 0.69 for friend’s social network and 0.68 for confidant social network.

The Perceived Support Scale (PSS) was developed by Krause and Markides (1990) to measure the receipt of support from children, relatives, friends and confidant. The original scale consisted of forty-one items measuring four support dimensions both support received (informational support, tangible help, and emotional support) and support given (integration). Of these forty-one items, Krause and colleagues used ten to measure three kinds of support received from significant others: informational support, tangible help, and emotional support (Krause and Shaw 2002; Krause 1999). To be consistent with the Indian context, these ten items were modified in this study to measure perceived support from network members in three dimensions: informational, emotional, and instrumental support. This modification was based on interviews with older persons about the support they received. The social support scale of this study was performed by deleting one item of information assistance of Krause and Makides (1990): “How often has someone told you what they did in a stressful situation that was similar to one you were experiencing”. This item was deleted because the older persons in the village stated that they rarely suffered from stressful situations and that they seldom discussed them with others.

The tangible support dimension was modified by changing the items, “How often has someone helped you with shopping?” into “How often has someone provided you clothes or groceries?” and “How often has someone pitched in to help you do something that needed to get done like household chores or yard work?” into three items of tangible support:

  1. i.

    How often has someone helped you do something that needed to get done inside the house like household chores?

  2. ii.

    How often has someone helped you with bathing, eating, dressing, toileting, etc.?

  3. iii.

    How often has someone helped you do something outside the house, like yard work, harvesting, or any income generating activity?

As financial support is important for the elderly, an item was added to the tangible support dimension: “How often has someone lent or given you money?” The details of each item were considered as appropriate and consistent with the situation in the Indian context.

Thus, the social support scale of this study consisted of twelve items: informational support - two items, emotional support - four items, and instrumental support - six items. The participants of the study were asked to indicate on a four-point scale rated from 1 (never) to 4 (routinely) the support they received. For example,

  1. i.

    Information Support: “How often have Relatives Suggested Some Action that you should take in order to Deal with the Problem you were Having?”

  2. ii.

    Emotional Support – “How often have Relatives taken Care and been Right there with you in a Stressful Situation or when you were sick?”

  3. iii.

    Instrumental Support - “How often have Relatives taken you to a Health Service, Provided Transport to get to the Doctor urgently, or taken you to some place?”

A score of social support for each network domain was created by calculating a mean of sub domain support received. A high score indicates that support was received from network members more frequently. Social support scale resulted in acceptable reliability (Cronbach’s alpha 0.83 for children support, 0.81 for relatives support, 0.80 for friends support and 0.82 for confidant support).

Pretesting was done on a small sample of respondents from the target population. Both the interviewer(s) and the respondents were asked a series of questions regarding the survey and the process of data collection during the debriefing session. Such debriefing sessions helped in detecting any problem with the questionnaire design and the process of administering the survey leading to ambiguity of words, misinterpretation of questions, inability to answer a question, sensitive questions.

Defining Depression

The most widely used method of depression assessment is self-administered questionnaires, for example, Beck Depression Inventory, the Center for Epidemiological Studies Depression (CESD) scale, and the Patient Health Questionnaire–9. These scales are easy to use, inexpensive, and user-friendly (Fisher et al. 2007). However, most of them do not address clinical diagnostic criteria directly; rather, they consist of a list of emotional symptoms that are endorsed by the respondent as present or absent during a specified time period (Fisher et al. 2007). This study used the Composite International Diagnostic Interview (CIDI), depression module to identify persons with one or more psychiatric disorders. The CIDI is a diagnostic instrument used internationally to generate prevalence estimates of major mental illnesses (Kessler and Ustun 2004).

Analytical Strategies

Social network construct was performed by using Confirmatory Factor Analysis (CFA) with Weighted Least Square (WLS) method to confirm whether the four dimensions were valid and reliable. To illustrate, CFA tests a proposed measurement model that describes the relationships between the observed and latent variables. In the present case, the latent variables are the social networks. To understand the effects of social network on depression, bivariate and multivariate analyses were performed. Bivariate analyses were performed to examine the nature of association between depression by selected socioeconomic and demographic background characteristics. Multivariate analyses used logistic regression to investigate the net effect of social network on depression. Three multivariate models were applied to examine the effect of four networks, namely children, relatives, friends, and confidant while adjusting for the potential background socioeconomic and demographic characteristics like, age, sex, education, current work status, caste, wealth index and living arrangement. All the variables identified as significant in the bivariate analyses using the chi-squared test were included in the binary logistic regression model. The results are presented by estimated odds ratio with 95 % confidence intervals (CIs). The analyses were conducted using SPSS version 15.0 (SPSS Inc 2007).

Results

Sample Characteristics

Table 1 shows that majority of the household had a family size of 7 (SD = 3.9) and about 75 % of the households in the study had a family size of five and above, and 14 % with a family size of six persons. In the study sample, on an average, an older women had 5.7 (SD = 2.4) children ever born. However, there was not much difference between average sons (3.0) and daughters born to older women. About 10 % of the older adults reported that they did not have any daughter compared to 4 % in case of son. Sample distribution further suggests that the average number of children survival was nearly 4.3 (SD = 1.9). The average number of children currently living with older adults was estimated at 1.4 (SD = 1.2).

Table 1 Mean and (Standard Deviation) of the Selected Background Characteristicsof Older Adults, Rural Varanasi, Uttar Pradesh, India
Full size table

The median age of the respondents was found to be 65 years (Table 2). Since mean and median ages of the respondent’s were fairly close, one could conclude that data on age of the respondents was quite close to a normal distribution and that there were few chances of recall bias. The sample distribution of older adults by broad age groups shows that a higher proportion (58 %) belonged to the 60–69 age group. Majority of surveyed older adults were female (53 %) and about two out of five, widowed. However, the study also found that majority of the women was widowed (52 %) as compared to the men (19 %) in the sample (not shown in Table). The sample distribution shows that majority of the elderly (67 %) had never attended any formal school. Among the elderly who were working, more than two-thirds (71 %) were unskilled workers followed by one–fourth (23 %) whose livelihoods were based on agriculture. The findings show that 50 % of the elderly were residing with their spouse and at least with one of their children. Only 5 % of the older adults were living alone and nearly 15 % were residing with their spouse at the time of survey. Nearly 43 % of the older adults were living with spouse and son (either married or unmarried) and grandchildren at the time of survey (not shown in table). Over 95 % of the older adults belonged to the Hindu religion and nearly 50 % of them were from Other Backward Castes (OBCs) – a collective term used by the Government of India to classify castes who are socially and educationally disadvantaged.

Table 2 Percentage Distribution of the Older Adultsby their Background Characteristics, RuralVaranasi, Uttar Pradesh, India
Full size table

Measurement of Social Networks: Confirmatory Factor Analyses (CFA)

Results suggest that while the overall fit of our model is good, a more exact fit to the data could have been achieved by allowing single indicators to load on multiple factors. In particular, an examination of modification indices indicated that a single improvement of fit could have been achieved by allowing the indicator for number of children to load on the factor for other relatives and friends. While fitting these paths may have made substantive sense, our goal was to permit items to load on one factor only in order to combine indicators into meaningful subscale in the most parsimonious manner.

The factor loadings (both standardized and standardized) for the measurement model are presented in Table 3. The results suggest that the coefficients of determination of the individual observed variables were moderate. The λ s ij values for the children, relatives, friends, and confidant latent variables were, however, comparable in size to those reported by Glass et al., where the λ s ij ranged from 0.31 to 0.99.The composite reliabilities of the latent variables in the present study were 0.83 for children social network, 0.78 for relatives social network, 0.69 for friends social network and 0.68 for confidant social network. The analogous figures reported by Glass et al., were 0.88 for children social network 0.67 for relatives social network, 0.80 for friends social network and 0.97 for confidant social network.

Table 3 Factor Loadings and Reliabilities for Four Latent Variable Measurement Model, Rural Varanasi, Uttar Pradesh, India
Full size table

Glass et al. (1997) calculated their specific network variables by summing the observed variables that made up each latent variable, arguing that their social network variables were identical whether or not the λ ij were used as weights. These authors appealed to work by Liang et al. (2005); Liang et al. 1990), suggested that when a composite variable is created from variables with similar variances and factor loading, then the difference in terms of reliability that arises from equal weighing versus differential weighing by λ ij is very small. However, summing of the observed variables by Glass et al. (1997) resulted in specific social network variables with different ranges, and these specific social network variables could not be compared easily with each other.

In this study, the average of the standardized observed variables was used to calculate the four specific network variables. Averaging was carried out in preference to summing because there were four observed variables for the children social network and three observed variables for the relatives, friends and confidant social network. A total social network variable was also calculated as the sum of the four specific social networks’ variables. Each of the social network variables was also categorized according to its tertiles, resulting in variables with categories of lower, middle and higher for each social network type and the total social network variable.

Bivariate Analysis

Table 4 presents the bivariate results of depression with socioeconomic and demographic variables. A significant difference was observed by marital status with a higher percentage of currently widowed/single/divorced reporting depression (34 %) than respondents who were currently married (23 %). Depression was highest among those elderly who did not attend any level of schooling (30 %). As far as current working status is concerned, about three in ten reported depression among those elderly who were not working, while the corresponding figure for depression among the elderly who were working was lower (22 %). About one out of five elderly who belonged to the richest wealth quintile were depressed, which was much lower than older adults who were from the poorer group (38 %). The percentage of depressed was significantly higher among those who were living alone (38 %), followed by those elderly who were living without spouse and at least one child (33 %).

Table 4 Percentage of Older Adults with Depression by Socio economic and Demographic Characteristics, Rural Varanasi, Uttar Pradesh, India
Full size table

The results presented in Table 5 suggest that sub-network with relatives, friends and confidant were inversely associated with depression among the elderly in rural areas. There was no significant association between children network and depression. Prevalence of depression by relatives network showed that about 33 % of the elderly who had lower relatives network reported depression. However, the corresponding figure was lowest (17 %) among the elderly who had a higher relatives network. In keeping with patterns obtained in the case of relatives network, depression among older adults was higher among those with lower friends network (37 %) and considerably lower at 14 % among those who had a higher network of friends. Although, the strength of association was not as significant as seen in the case of relatives and friends network, the findings suggest lower depression among those older adults who had a higher confident network (20 %) as compared with a lower one (32 %). In the case of total social network, the gap in depression between lower total social network (34 %) and higher (17 %) was nearly double.

Table 5 Percentage of Older Adults with Depression by Specific and Total Social Network, Rural Varanasi, Uttar Pradesh, India
Full size table

The percentage of elderly with depression by social support shows significant difference by children, relatives, friends and total support (Table 6). About 27 % of the older adults who had lower social support reported depression, the corresponding figure was 20 % among those who had a higher children network. Just over one in five elderly from higher and middle relatives support, reported depression, much lower than those who had lower relative support (37 %). The difference in depression was almost twice between the elderly with the higher and middle friends support than those with lower friends support. For example, two-fifths of the elderly with lower friends support were depressed. However, the percentage of depression was lowest among the elderly who had higher friends support (15 %), followed by those who had middle friends social support (26 %). Results further suggest a negative association between total support and depression among the elderly.

Table 6 Percentage of Older Adults with Depression by Specific and Total Social Support, Rural Varanasi, Uttar Pradesh, India
Full size table

Multivariate Analysis

To get the adjusted effects of social network on depression, multivariate analysis was carried out. The results presented in Table 7 suggest that specific network with children, relatives and confidant did not have any significant effect on depression. However, there is evidence of protective effect of network with friends on depression among older adults. In Model 1, the results show that the friends and relatives network significantly affects the odds of depression among the elderly. The odds of depression were lower among those elderly with middle friends network compared to the elderly with lower friends network (OR = 0.353; 95%CI = 0.176–0.071). Similarly, the likelihood of reporting depression was lower (OR = 0.304; 95%CI = 0.185–0.498) among the elderly who had higher friends network than among the elderly who had lower friends network. The odds of depression were less likely among those elderly (OR = 0.600; 95 % CI = 0.374–0.962) who had higher relatives network than among the elderly with lower relatives network.

Table 7 Binary Logistic Regression Model Estimates the Odds Ratios and Confidence Intervals for the Effect of Specific Social Networks on Depression among Older Adults
Full size table

Model 2 shows the adjusted effect of four networks on depression after including children, relatives, friends and confidant support in the regression analysis. A significant association between friends network and depression was evident, while the effect of relatives network on depression did not show any significant effect. The odds of depression was lower among the elderly who had middle friends network compared to the elderly who had lower friends network (OR = 0.406; 95%CI = 0.190–0.868). Similarly, the likelihood of depression was 66 % (OR = 0.342; 95%CI = 0.342–0.617) lower among the elderly who had higher friends network than among the elderly who had lower friends network. After adjusting selected socioeconomic and demographic predictors in Model 3, the effect of friends network remained significant. The odds of depression was lower (OR = 0.391; 95 % CI = 0.179–0.855) among the elderly with middle friends network than among the elderly who had lower friends network. The likelihood of depression was about 67 % (95 % CI = 0.182–0.613) lower among the elderly who had higher friends network compared to the elderly with lower friends network.

Discussion and Conclusion

The results of this study highlight the higher prevalence of depression among older adults in rural areas. The pattern of depression observed by this study matches with the overall prevalence of depression in Uttar Pradesh (30 %), estimated by the nationally representative survey conducted during 2007–10 (IIPS 2013). Results suggest considerable variations in depression among older adults by selected socioeconomic characteristics such as marital status, level of education, work status, household economic status and across living arrangements. These socioeconomic variations in depression have been well documented in previous studies from India (Chadda 2014; Reddy et al. 2012; Jain and Aras 2007; Sengupta and Agree 2002; Tiwari 1999). For instance, a recent study documented the significant effect of wealth status on the depression symptomatology of older adults, which linked poor wealth status to the household’s inability to provide adequate nutrition, access to healthcare and poor quality of life that ultimately lead to depressive disorders (Kulkarni and Shinde 2015). Global literature also suggests that having more income or wealth, more years of education, and a prestigious job, as well as living in stable and healthy neighbourhoods are factors associated with better mental and physical health status of older adults (Chiao et al. 2011; Hughes et al. 2007; Greenglass et al. 2006; Beckfield 2004).

This study did not find significant gender differences in depression among older adults in the rural sample. This result could be supported by other studies conducted in Northern India and elsewhere, where gender difference in depression among older adults was not found. For instance, a study conducted in the rural community of Ballabgarh in the state of Haryana (a state in North India) did not observe any gender difference in depressive symptoms, cognitive and functional impairment of the elderly (Ganguli et al. 1999). Similarly, no gender difference was documented in a recent study conducted in the rural areas of Vellore (Tamil Nadu), which suggested that depressed individuals in low-income communities rarely subscribe to biomedical causal models and hold more to psychosocial as well as interpersonal explanatory models for depression (Rajkumar et al. 2009; Pereira et al. 2007).

The main objective of this paper is to examine the role of specific and total social networks of older adults on depression in a rural setting, for which it is imperative to apply social network theories and measures. Research on India has consistently demonstrated that the family still remains the central source of support for older persons (Samanta 2014; Irudaya Rajan 2006). Although, the debates over dissolution of the extended family forms and the rise in nuclear families in India are inconclusive (Sathyanarayana et al. 2012; Irudaya Rajan and Kumar 2003; Shah 1996) there is considerable evidence that socio-demographic changes along with rapid urbanization led migration from rural to urban areas has started affecting the traditional support system of older adults, particularly in rural areas (Bloom et al. 2010; Chokkanathan 2009). India does have a tradition of providing pension to the retired elderly but this is available only to around 10 % of the Indian population who had been part of the formal workforce, thus majority of older persons depend on filial piety and intergenerational support during their old age (Sathyanarayana et al. 2012; Gupta and Pillai 2002). The growing importance of social network on the health and well-being of older adults in a resource poor setting is justifiable and timely.

Findings of this study suggest greater social networks with friends had statistically significant protective effects against depression among older adults in rural areas and that networks with children, relatives and confidant were not statistically significant predictors of depression. This pattern highlights the importance of sub-network disaggregating kin and non-kin networks, rather than relying on measures of total social networks. The strength of this finding lies with regard to the approach of the study that applied specific social networks of older adults for the first time including number, proximity and frequency of contacts in a rural Indian setting. The role of living arrangement on older adults health and well-being including disability, chronic morbidity and healthcare utilization have been highlighted by many studies (Samanta et al. 2014; Singh et al. 2013; Agarwal 2012; Gupta. 2009; Irudaya Rajan 2006; Irudaya Rajan and Kumar 2003), whereby the most common type of living arrangement and support for the elderly in India is found to be living with married sons and their families (Singh et al. 2013; Prakash 2001). However, little has been explored about, the role of friends or neighbours on the health of the elderly, particularly depression.

It is imperative to understand that the requirements of older adults could be different under varying circumstances. For instance, in the Indian context family or son and his family in particular provide instrumental support that includes ensuring healthcare during any illness or morbidity, ensuring adequate and timely provision of diet for older parents, and ensuring basic living conditions to the older adults in the household (Das et al. 2014). However, recent studies reveal a high prevalence of elder abuse in India (Sebastian and Sekher 2010; Siva Raju 2011). A recent study based on content analysis of reports published in two leading newspapers between 2004 and 2008 shows that most of the crimes against the elderly remain unreported (Patel et al. 2010). Surprisingly, about 60 % of the crimes were committed indoors and 25 % of the perpetrators were their own family members (Patel, 2010). In a study conducted in Pune (Maharashtra state, India) around 50 % of the elderly felt they were abused and 40 % felt totally neglected by the family (Bambawale 1997). Similarly, in Andhra Pradesh a study found that the elderly had been neglected by denying them proper medical care and food, and they were also exploited financially (Ushasree and Basha 1999). The prevalent patterns of elder abuse include mainly psychological abuse in terms of verbal assaults, threats and fear of isolation, physical violence and financial exploitation (Siva Raju 2002; Rao 1995). Moreover, studies found that either the son and daughter-in-law together (Kumar 1991; Rao 1995) or other family members were the main abusers (Sebastian and Sekher 2010).

Only a few studies have recognised the growing importance of friends on older adult’s health and well-being (Samanta 2014; National Research Council 2012). Friendships are voluntary relationships and being chosen as a friend and receiving emotional support from friends may build self-esteem and contribute effectively in reducing mental stress more than family relationships, especially because friends are typically members of the same age group and often share personal characteristics, cohort experiences, and are a source of enjoyment, socializing, and talking about ‘good old times’. Some studies argue that interaction with friends may improve well-being among older adults through multiple channels, such as through diffusion of health information, mutual assistance and economic support, which help people to deal with distress related to age and sickness (Webster et al. 2015; Lindström et al. 2003; Berkman et al. 2000; Kawachi et al. 1999).

The cross-sectional nature of the study limits the ability to draw causal inferences. For instance, it could be that people who foster positive and supportive relationships or who are less depressed are also more likely than others to have a varied, diverse network. In fact, research has shown that distressed individuals may withdraw from or alienate social relationships (Matt and Dean 1993) and that a high level of depression or physical limitations may actually lead to a decline in emotional support over time (Gurung et al. 2003). In other words, it is difficult to determine from the present study whether certain network types or relationships ‘cause’ good health, or whether good health promotes certain types of interactions, relationships or networks. Although longitudinal research could better assess two different possibilities, the association is likely to be bi-directional.

In summary, this study has shown that good social networks with friends protected rural older adults from depression after controlling for other socioeconomic and demographic confounders. The findings suggest the need for programmes that enhance collaborations within communities and develop network centres for older persons to facilitate interaction with friends within or between communities, provide opportunities for elders to participate in group activities, and promote exercise, recreation and knowledge towards better health and health practices. The paucity of research on social network and its effect on the well-being of the elderly in the Indian context provides huge scope to explore other dimensions of health and well-being.