Widow and Widower Mortality in India: A Research Note

Abstract

Widowhood is associated with elevated mortality risk in many social contexts. This research note is the first study to quantify and contextualize the mortality risk of widowhood for men (widowers) and women (widows) in India. We do so by using data from the first wave of the India Human Development Survey (2004–2005) on individuals whose survival status was observed seven years later in the second wave of the survey. We find no differences in mortality by widowhood status for adults aged 60 or older. However, we find higher mortality risks for widows and widowers aged 25–59 than for individuals who are married. Despite the unique vulnerabilities experienced by Indian widows, we find similar levels of elevated mortality for widows and widowers relative to married individuals aged 25–59. In this age group, we also document higher mortality for widows exposed to conservative and less egalitarian gender norms. These findings suggest that despite India’s similarity to other contexts with elevated mortality for both widows and widowers, unequal gender norms still shape life chances for Indian widows.

Introduction

A large body of research has documented the elevated risk of mortality among widowed men (widowers) and women (widows), a phenomenon known as the “widowhood effect” (Dabergott 2022; Moon et al. 2011; Shor et al. 2012). Studies have documented that widowhood influences mortality through the stress associated with bereavement, end-of-life care work, and changes in social relationships, health, and economic status (Anusic and Lucas 2014; Carr and Bodnar-Deren 2009; Ding et al. 2021; Jin and Chrisatakis 2009). Yet, almost all evidence on the associations between widowhood and mortality comes from high-income countries. The dearth of longitudinal data in many less developed countries has prevented the study of this phenomenon across a wider range of social and cultural contexts, where the implications of widowhood may differ. This research note fills this gap as the first study to directly estimate widow and widower mortality using longitudinal data in the Indian context. India, home to more than 56 million widowed individuals (Government of India 2015), is an important setting to study widowhood as a social determinant of mortality because the widowhood disadvantage there is uniquely gendered.

One of the most significant and widespread findings in the global literature on widowhood effects is that the excess mortality risk associated with widowhood is more pronounced for men than for women (Moon et al. 2011; Shor et al. 2012; Wang et al. 2020). With a few exceptions (Sullivan and Fenelon 2014), studies across the United States (Bulanda et al. 2016; Moon et al. 2014) and Europe (Brenn and Ytterstad 2016; Dabergott 2022; Kalediene et al. 2007; Nystedt 2002) have consistently found larger widowhood effects on mortality for widowers than for widows. Research suggests that widowers are more likely to adopt unhealthy coping behaviors, such as smoking, drinking, poor diet, and drug use (Lin and Brown 2020; Utz et al. 2012). Widowers might also struggle to manage the household labor and care work their wives previously provided (Kalish 1971; Smith and Zick 1996; Utz et al. 2004). The literature has explained the gender gap in widowhood effects on mortality as resulting from factors such as socioeconomic status, health behaviors, social relationships, anticipation, and coping strategies (Angel et al. 2007; Dabergott 2022; Shor et al. 2012).

We examine how these patterns differ, if at all, in the Indian context, where widows often face discrimination. Widowhood takes on uniquely gendered cultural meanings in India because of a prevalent cultural norm of “perpetual mourning” for widows, which can lead to social exclusion and lifelong restrictions on dress, movement, diet, and sex (Azeez E. P. et al. 2023; Chen 2000; Lamb 2000). Widow remarriage is generally seen as taboo, even for women widowed at younger ages (Chen 2000). Indian widows are often financially reliant on their adult sons (Agarwal 1998; Vlassofft 1990), and widow-headed households have the highest poverty rate (Drèze and Srinivasan 1997). Studies show that Indian widows suffer from worse health than married women when measured by nutritional status, mental health, cognitive ability, and the presence of chronic disease (Agrawal et al. 2021; Chen and Dreze 1992; Jensen 2005; Lloyd-Sherlock et al. 2015; Perkins et al. 2016). Few studies in the Indian context have looked at widowers. Of those that have, some found that widowers face fewer health disadvantages than widows (Agrawal and Keshri 2014; Perkins et al. 2016), whereas others found similar health risks regardless of gender (Prasad et al. 2023).

To our knowledge, no previous study has measured both widow and widower mortality in the Indian context. Bhat (1998) examined mortality among widows, but not widowers, using indirect estimation techniques owing to the absence of longitudinal data at the time of the study. Anderson and Ray (2019) used multiple datasets to model mortality risk and quantify missing unmarried women but did not directly measure the mortality of widowed individuals. This research note contributes to our understanding of the gendered nature of widow mortality by directly estimating relative mortality risk by widowhood status for Indian women and men. Given widows’ uniquely disadvantaged position in India, we hypothesize that Indian widows will have a higher relative mortality risk than widowers, deviating from the international pattern of higher widower mortality.

We then examine whether the relative mortality risk by widowhood status is associated with local gender norms. An emerging literature suggests that gender norms might be an important determinant of health inequalities and poor health outcomes. Studies in India, China, and Bangladesh have linked conservative gender norms with a higher prevalence of hypertension, inflammation, obesity, and poor self-rated health among women (Ahmed and Sen 2018; Alvarez-Saavedra et al. 2023; Reynolds et al. 2020; Stroope et al. 2021). Health inequalities by gender might result from lifestyle factors exacerbated by female seclusion norms, which limit women’s physical mobility (Stroope 2015). Physical and emotional violence and unequal distribution of household resources are other mechanisms through which conservative gender norms contribute to negative health outcomes for women (Dhakad and Saikia 2019; Dupas and Jain 2021; Sedlander et al. 2021). In the Indian context, conservative gender norms are linked to the stigmatization and deprivation of widows because patriarchal social structures devalue women’s identity outside of marriage. This marginalization might result in higher mortality risk for widows exposed to these norms. We hypothesize that Indian widows, but not widowers, will face higher mortality than their married peers in places with more conservative gender norms.

Material and Methods

Data

We use data from the India Human Development Survey (IHDS), one of the first large-scale, nationally representative panel studies in India with a sufficiently large sample to estimate disaggregated demographic rates (Desai et al. 2005; Desai and Vanneman 2012). To date, the IHDS has completed two waves: IHDS 1 in 2004–2005 and IHDS 2 in 2011–2012. IHDS 1 collected information on 41,554 households across 1,503 villages and 971 urban areas from 33 states. IHDS 2 was able to reinterview 83% of those households and complete a tracking sheet for all individuals observed in IHDS 1. For this analysis, we use the individual data file from IHDS 1, which contains information on all members of surveyed households using data collected in the household roster. We then match individuals observed in IHDS 1 with their survival status in the tracking data file in IHDS 2.

From the 215,754 individuals observed in the household rosters of IHDS 1, we limit our analysis to the 158,639 individuals who appeared on the IHDS 2 rosters or were reported dead in the tracking file, dropping approximately 26.5% of the Wave 1 sample that was lost to follow-up. Next, we restrict our analysis to the 84,305 individuals who were aged 25 or older in IHDS 1, given that this is the age by which most Indians have married (Desai and Andrist 2010; Singh et al. 2023). We exclude 4,187 individuals with a marital status other than married or widowed in IHDS 1 and further exclude 4,579 individuals whose marital status changed between the two survey waves. We cannot account for marital status changes between the survey waves in this analysis because we do not have full marital histories. Therefore, our analysis can examine mortality differences only by baseline marital status. Finally, we use casewise deletion to drop cases with missing values on key variables and remove 625 cases (0.8% of the sample), yielding a final sample of 74,914 individuals. Table 1 provides summary statistics for the analytic sample.

Table 1.

Summary statistics, IHDS 1 and IHDS 2

	Widowed		Married
	Women	Men	Women	Men
Age Group (%)
25–59	39.7	27.0	92.9	84.3
60+	60.3	73.0	7.1	15.7
Residence Type (%)
Rural	74.3	82.6	76.3	75.2
Urban	25.7	17.4	23.7	24.8
Monthly per Capita Consumption (rupees, median)	785	796	808	806
Count of Household Assets (mean, of 30)	11.0	10.7	11.6	11.6
Years of Education (mean, capped at 15)	1.3	2.9	3.4	5.8
Dominant Crop of State (%)
Historically rice-dominant	48.6	37.1	43.6	45.8
Not rice-dominant	51.4	62.9	56.4	54.2
Household Practices Purdah/Ghunghat/Pallu (%)
Yes	53.4	65.9	59.2	57.7
No	46.6	34.1	40.8	42.3
Cases (of 5) When Domestic Violence Is Common in Community (%)
0	15.4	11.9	13.4	14.1
1	34.6	33.0	35.1	34.4
2 or more	50.0	55.2	51.5	51.5
Total Numbers	6,369	1,576	31,241	35,728
% of Overall Sample	8.5	2.1	41.7	47.7

Notes: All estimates, except those in the bottom two rows, are weighted using survey weights. Information from 10,826 and 11,124 individuals was missing for the measures of purdah and cases of domestic violence, respectively.

Setup

Using the data on the 74,914 sample individuals, we create a dataset in which each observation is a person-month. We observe an individual’s survival status and exact age (starting from their age in IHDS 1) in each person-month. We do this by creating individual lifelines from the month and year of observation during IHDS 1 until the month and year of interview in IHDS 2 (if the individual survived) or the month and year of death (if the individual died).¹ Mortality between the two IHDS waves is measured in the tracking sheet. The IHDS observes the year of death, and we assign individuals a random calendar month of death in the year they died. This method is equivalent to assuming that these individuals, on average, lived halfway through their year of death. This assumption is common in mortality analyses, especially for mortality in adulthood (Preston et al. 2001).

Measures

The outcome variable in this study is mortality. We observe whether the individual was alive or dead for each month between the month and year of interviews in IHDS 1 and IHDS 2. We then model the binary probability of their death in a given month. Our robustness checks explore alternative mortality modeling approaches.

Our main independent variable is whether the person was married or widowed when observed in IHDS 1. As mentioned earlier, we restrict the sample to those individuals whose marital status remained the same between IHDS 1 and 2, assuming that these individuals did not experience any transitions between the two waves.

Because mortality is highly age-dependent, we control for age in all models. Our person-month dataset has a measure of exact age for each calendar month and year we observe an individual, allowing us to control for time-varying age. Supplementary analyses provided in the online appendix show the age patterns of mortality by widowhood status. We also find broadly similar results when modeling age patterns differently.² We stratify all results by broad age groups (25–59 and 60+) and gender.

We also include controls for region through state dummy variables and an indicator of rural/urban residence. Drawing on research suggesting that socioeconomic status (SES) is an important determinant of mortality among the widowed (Choi and Marks 2011; Kalediene et al. 2007; Kung 2020; Martikainen and Valkonen 1998), we include several controls for SES. These controls include a measure of monthly per capita household consumption, education (in years), a count of household assets, and social group (caste and religion).³ All region and SES controls are measured in IHDS 1.

We next seek to understand how widow and widower mortality varies by gender norms. We use three measures of gendered social norms that capture its manifestation across different domains of life, including gender performance, gender-based violence, and egalitarianism. These interaction variables are captured in IHDS 1. First, we use a measure of whether women in the household practice any form of head covering known as purdah, ghunghat, or pallu. This measure captures a heterogeneous set of gender practices often associated with female seclusion and the performance of gender differences (Desai and Andrist 2010). The other two measures of gender norms are at the community level rather than the household level. One indicator comes from questions asked to a woman in each household in the IHDS 1 regarding the prevalence of domestic violence in their community across five hypothetical situations.⁴ We calculate a sum across these five variables and create three categories to measure the prevalence of domestic violence in the community, ranging from not prevalent to very prevalent. The other indicator is whether the family lives in an area where rice was historically the dominant crop.⁵ Previous research has found that India’s predominantly rice-growing regions have more egalitarian gender norms, which are speculated to be partly due to women’s important role in rice cultivation (Bardhan 1974; Boserup 1970; Kishor 1993; Miller 1982). This indicator can be considered a rough proxy for regional historical norms of gender egalitarianism.

Methodology

Our main empirical strategy relies on discrete-time survival analysis of the probability of death for widowed and married individuals. On the whole, our approach is similar to the estimation of period demographic rates and probabilities, albeit in a regression framework.⁶ We define a hazard function for an individual $i$ living in sampling unit $j$ as the probability of $y$ (dying) in a discrete interval $t$ (a month in our case), given that the individual was alive before $t:h_{ij}(t)=Pr\left(y_{ij}(t)=1\mid y_{ij}(t-1)=0\right)$.⁷ As Allison (1982) showed, this function is equivalent to the conditional probability that a death occurs at time $t$, given that it has not already occurred. We fit a discrete-time logistic regression model of the following form:

$$\mathrm{l}\mathrm{o}\mathrm{g}\mathrm{i}\mathrm{t}\left[h_{ijt}\right]=\mathrm{l}\mathrm{o}\mathrm{g}\left(\frac{h_{ijt}}{1-h_{ijt}}\right)=\alpha +{\mathit{Widowed}}_{ij,Wave1}\mathrm{\beta }+{\mathit{Age}}_{ijt}\theta +\sum _{k=1}^K{X}_{ij}^{(k)}{\gamma }_k,$$

where $h_{ijt}$ is the hazard of death in month $t$ for respondent $i$ in primary sampling unit ${\mathit{Age}}_{ijt}$ is a control for time-varying age with individuals starting at their age in IHDS 1 and aging every month until their final month and year of observation (death or IHDS 2 interview month and year); and $X_{ij}^{(k)}$ for $k=1,...,K$ is a set of time-invariant controls. The coefficient of interest is $\mathrm{\beta }$. For ease of interpretation, we report odds ratios from our main models and predicted probabilities from interactions.

We stratify all results by age group (25–59 and 60+) and gender. In addition to controlling for age, we examine the extent to which differences in socioeconomic status and residence can explain differences in mortality between widowed and married individuals. We also use interactions to explore how the probability of death varies by gender norms. All our estimates are weighted using IHDS 1 individual national weights. We cluster our standard errors at the level of the primary sampling unit.

Results

Mortality Differences by Widowhood Status

The first set of results, depicted in Figure 1, shows the odds ratios of mortality from discrete-time logistic regressions modeling death in a month while controlling for time-varying age, stratified by gender and broad age groups.⁸ The vertical lines in Figure 1 reflect 95% confidence intervals (CIs), which account for clustering at the level of the primary sampling unit. Results for the first models, which include only the time-varying age control (blue diamonds), show substantially higher odds of mortality for both widowers and widows aged 25–59. The second model (red circles) shows that controlling further for region has little impact on the results. The final model (green squares) also includes SES controls. The results from this fully controlled model indicate that widows aged 25–59 had 40% higher odds of death between the two IHDS waves than married women (95% CI = 1.11–1.77). Widowers aged 25–59 had 50% higher odds of dying than married men (95% CI = 1.06–2.12). Our findings do not support our original hypothesis that widows would face a greater survival disadvantage than widowers. Instead, we find that among those aged 25–59, the higher mortality associated with widowhood was slightly larger for men than for women, fitting with the international pattern. Among those aged 60 or older, however, mortality did not differ between widowed and married individuals in any of the models. Consequently, we focus only on the 25–59 age group in the subsequent analysis. In the online appendix (Figure A2), we demonstrate that the results are identical when alternative modeling strategies are used.⁹

Fig. 1.

Odds ratios from separate discrete-time logistic regressions modeling the mortality of widowed individuals relative to married individuals, by gender and age group. Survey weights were used, and standard errors are clustered at the level of the primary sampling unit. The y-axes use a log scale. The first set of regressions (blue diamonds) controls only for time-varying age with a continuous measure. The second set of regressions (red circles) adds controls for state dummy variables and rural residence. The third set of regressions (green squares) additionally includes controls for years of education, social group (dummy variables for caste group and religion), a count of household assets, and monthly per capita consumption expenditure.

Heterogeneity by Gender Norms

Next, we test whether widow and widower mortality varies by local gender norms. Given the history of social exclusion of widows in India, we expect that widow mortality will be elevated in places with more conservative gender norms. Figure 2 depicts the predicted probabilities of death from discrete-time logistic regressions in which widowhood is interacted with the three gender norm measures.¹⁰ The models adjust for age.

Fig. 2.

Predicted probability of death in a month by gender, measures of gender norms, and widowhood status for younger adults (aged 25–59), based on discrete-time logistic regressions interacting widowhood status with measures of gender norms. National survey weights were used, and standard errors are clustered at the level of the primary sampling unit. Regressions adjust for time-varying age. The y-axis displays the predicted probability of death in a month.

The first measure of gender norms captures whether women in the household practice purdah, ghunghat, or pallu. The predicted probabilities of death calculated by interacting this dichotomous measure with the indicator of widowhood are shown in the leftmost area of Figure 2 (panel a). Among households that do not practice purdah, we find no statistically significant differences in mortality for women aged 25–59 by widowhood status. However, in households where purdah was practiced, the probability of death was roughly two times as high for widows compared with married women. This result is statistically significant, as evidenced by the nonoverlapping CIs and a statistical test of difference. We find no significant differences in mortality by purdah practices for men aged 25–59, as shown in the lower portion of Figure 2 (panel a).

The next two measures of gender norms capture community-level gender practices. The second column of analyses in Figure 2 (panel b) uses an interaction between widowhood and the measure of perceptions of domestic violence prevalence in the community. We find a statistically higher probability of death for widows than for married women only in households where women reported that domestic violence in their community is common (in at least two of five hypothetical scenarios). The differences between widows and married women are not statistically significant in areas where domestic violence is less prevalent or not prevalent. In addition, mortality does not meaningfully differ between widowers and married men by community prevalence of domestic violence.

The final column of analyses in Figure 2 depicts the predicted probabilities when widowhood is interacted with a state-level indicator for rice-dominant states, a proxy for regions with a historical pattern of greater gender egalitarianism (panel c). Widows aged 25–59 have a statistically higher probability of death than married women only in regions where rice was not dominant. We find no meaningful widowhood status differences in mortality for widowers by this measure. Figure 2 provides clear evidence that in places with less egalitarian or more conservative gender practices, widows faced higher relative mortality than married women. Gender norms, however, did not play a significant role in stratifying mortality differences between widowers and married men.

Discussion

Using survey data that observed individuals in 2004–2005 and measured their survival seven years later, we document the mortality risk for widowers and widows relative to married men and women, respectively, and explore how those relative risks are associated with local gender norms. We find no statistically higher mortality risk for widows or widowers aged 60 or older. This finding fits with the literature in other contexts, which has documented that the relative mortality risk associated with widowhood narrows with age (Brenn and Ytterstad 2016; Martikainen and Valkonen 1998). A spouse’s death at a younger age is more likely to be unexpected and, therefore, might cause more socioemotional and financial stress, often resulting in a higher relative mortality risk for the surviving spouse (Moon et al. 2014; Moon et al. 2011; Zick and Smith 1991). In the Indian context, SES selection into survival to old age also likely influences the diminished role of widowhood status for those aged 60 or older (Goli et al. 2014).

We also show that among those aged 25–59, the odds of death were 50% higher among widowers and 40% higher among widows than among married men and women, respectively, even after controlling for region, household SES, and time-varying age. From a comparative perspective, this relative mortality risk is similar in magnitude to that observed in high-income countries (Liu et al. 2020). Furthermore, as observed in other settings, we find that widowhood poses a slightly greater mortality risk to men than to women. This result for India is surprising, given that the literature has documented that Indian widows face unique vulnerabilities associated with the stigma of widowhood. The larger impact of widowhood on men’s mortality might be due to gendered bereavement practices, such as men’s greater reliance on unhealthy coping strategies (Lin and Brown 2020; Stroebe 1993; Utz et al. 2012), or to household gender roles that make the loss of their wife’s essential domestic labor especially consequential for men (Kalish 1971; Smith and Zick 1996; Utz et al. 2004). In addition, these results could suggest that marriage might not be as protective of women’s health in India. This surprising finding contradicts our hypothesis and points to a greater universality in the gender pattern of widowhood effects than expected.

More research is needed on widowers in India and the causes of their higher mortality risk. A contributing factor to the gender differences in relative mortality risk might be the unobserved selection processes into or out of widowhood. In India, remarriage is common and socially accepted for young widowers but is considered taboo for widows (Chen 2000). More advantaged widowers, especially those in the prime ages for family formation, might select out of widowhood through remarriage, resulting in the widower population being negatively selected on traits that determine marriageability, such as wealth and education. Because so few widows remarry, the Indian widow population is both larger (3.5 times larger in the 2011 census) and less selected than the population of widowers, potentially contributing to the gender differences in relative mortality risk. Future research could enumerate how large a role selection plays in the elevated mortality of widowers.

This research note extends the growing literature on the health implications of conservative gender norms by documenting how these norms exacerbate gaps in mortality risk by marital status for women. For widows in India aged 25–59, the higher odds of mortality associated with widowhood are observed only in households where women practice purdah, in communities where domestic violence is most prevalent, and in non-rice-dominant regions with historically less egalitarian gender relations. These results fit with qualitative evidence documenting the extreme deprivations and stigma that some young widows in India face (Azeez E. P. et al. 2023; Chen 2000; Lamb 2000). Despite widowers having higher relative mortality risks, we find no meaningful variation in widower mortality by gender norms. These findings highlight the importance of community-level gender norms in determining mortality risk for women.

This study has important limitations. Studying mortality with survey data poses challenges, but supplemental analyses presented in the online appendix suggest that the IHDS data can be used to estimate mortality reliably, at least relative to UN estimates from the same period.¹¹ Although relatively modest in the IHDS, attrition could also impact our results if mortality patterns were different in the households lost to follow-up. Previous research suggests that this form of bias is limited in the IHDS (Tiwari et al. 2022). In addition, mortality is measured only retrospectively at the year level in the IHDS, forcing us to impute the month of death by assuming that deaths occur, on average, halfway through the year. Finally, we lack marriage history data. Thus, although the IHDS is a panel survey, we examine the mortality risk associated with widowhood status as reported at baseline in IHDS 1. Given that IHDS has only two survey waves, we cannot observe the shock of widowhood in our dataset and track subsequent mortality risk for the surviving spouse. As a result, our findings are associational and cannot be interpreted as the causal effect of widowhood on mortality. Further, we cannot measure the role of remarriage in men’s greater selectivity into (or out of) widowhood and how mortality risk changes with time since widowhood.

Nevertheless, this research is the first to directly estimate widow and widower mortality using longitudinal data in the Indian context. In doing so, we help extend the widowhood effects literature to new social and cultural contexts where widowhood takes on different social meanings. By focusing on India, where widows face unique vulnerabilities, our study sheds light on the important role of gendered social norms in mediating the impact of widowhood on mortality for women. This study’s findings are especially relevant in the wake of the COVID-19 pandemic, which left millions of Indians widowed (Banaji and Gupta 2022). Both in India and globally, there is a need for researchers to measure the mortality implications of the pandemic-induced increase in widowhood. ■

Supplementary Material

ELECTRONIC SUPPLEMENTARY MATERIAL The online version of this article (https://doi.org/10.1215/00703370-12177893) contains supplementary material.