Life Events and Loneliness Transitions Among Middle-Aged and Older Adults Around the World

Abstract

Objectives

Adult loneliness is a substantial social problem and a growing point of concern for policymakers around the world. We assess whether the predictors of loneliness onset among middle-aged and older adults vary from country to country in a large array of settings across world regions. Taking a life course perspective, we focus on common life events in our focal age range, including changes in partnership, coresidence, work, and health, and we test whether changes in them have comparable prospective associations with loneliness onset in different countries.

Methods

We draw on respondent-level data from a diversity of world regions surveyed in 7 harmonized cross-national studies in 20 countries, representing 47% of the global population over the age of 50. Our innovative longitudinal approach estimates prospective transition probability models that examine how each life event predicts the transition into loneliness.

Results

Despite substantial variation in the prevalence of loneliness and life events across the range of countries in our sample, our results highlight consistency in the predictors of loneliness transitions. Family and household changes like divorce, coresidence, and especially widowhood are paramount predictors of loneliness transition across settings, with changes in work and health playing more minor and less universal roles.

Discussion

The results demonstrate the importance that family and household connections play in determining loneliness at these ages. These findings suggest that addressing late-life loneliness may require a focus on key life events, especially those concerning changes in families and households.

A considerable segment of the population experiences loneliness in countries across the world (Surkalim et al., 2022). Loneliness is a public health focus because in middle and older adulthood it is strongly associated with adverse individual and population health, influencing mortality (Holt-Lunstad et al., 2010) and physical (Valtorta et al., 2016), cognitive (Wilson et al., 2007), and mental (Cacioppo et al., 2006) aspects of health. Loneliness is defined as “an individual’s subjective perception of deficiencies in his or her network of social relationships” (Russell et al., 1984, p. 1313). As emphasized in classic works defining the concept (Baumeister & Leary, 1995; Perlman & Peplau, 1981; Sermat, 1978; Weiss, 1973), it has three key features: it (a) is subjective, (b) results from perceived deficiencies in social relations, and (c) is “unpleasant and distressing” (Perlman & Peplau, 1981, p. 32). After age 50, loneliness tends to increase with age (Mund et al., 2020), and there is evidence that this age pattern of loneliness can be attributed to worsening health and changes in social integration (Jylhä, 2004). With global population aging, late-life loneliness is poised to become an even more pressing issue in coming decades. Recent projections suggest that globally, the number of lonely adults 50 and older can be expected to triple between 2020 and 2050, solely because of population aging and population growth (Newmyer et al., 2022). For these reasons, national governments around the world have sought to tackle late-life loneliness as a social problem (Newmyer et al., 2021).

Policies and interventions targeting loneliness must know critical onset points. Among middle age and older adults, a central question is whether people arrive in this age range lonely and stay there, or whether they develop loneliness within this period. Although there is extensive research on correlates of loneliness, most of it is cross-sectional, yielding an incomplete understanding of loneliness dynamics in middle age and older adulthood. These issues even complicate the language used to refer to loneliness dynamics; for example, in this article, we discuss the onset of loneliness and transitions into it, but we explicitly note that loneliness can be transient or chronic. Therefore, transitioning “into loneliness” does not imply a permanent state. Further, whether correlates of loneliness are precursors or consequences remains unknown. This open question extends to some of its most frequently noted correlates of loneliness, including life events like losing a partner or widowhood, increased physical disability and mobility reduction, retirement, and reduced social activities (Aartsen & Jylhä, 2011; Dahlberg et al., 2015; Domingue et al., 2021; Nicolaisen & Thorsen, 2014b). In addition, research addressing predictors of loneliness onset is largely from single-country studies focused on North America and Europe (Aartsen & Jylhä, 2011; Domingue et al., 2021; Dykstra et al., 2005; Nicolaisen & Thorsen, 2014a; Victor & Bowling, 2012; Wenger & Burholt, 2004). This gap is noteworthy because studies of psychosocial phenomena increasingly recognize that findings from Western, educated, industrialized, rich, democracies generally do not apply to other contexts. Instead, across a range of core psychological measures, there is substantial cross-national variation (Henrich et al., 2010). Given massive differences in cultural emphases, economic organization, religiosity, levels of urbanization, and other such features between countries, we might suspect that whether, and the extent to which, a given life event predicts transitions into loneliness could vary from place to place. Despite growing diversity of the aging population worldwide, this question has not been assessed in prior work.

To address these challenges, we analyze harmonized data from seven surveys fielded in diverse regions and covering 20 countries that contain 47% of the global population over the age of 50. We first assess how common it is for loneliness transitions to occur in middle age and older adulthood. Upon finding that transitions are a key facet of loneliness in these age ranges, we then ask to what extent life events common in older adulthood (changes in family and household structure, work, and health) predict the onset of loneliness and, whether these events are consistent predictors of loneliness across countries. In supplemental results, we look at the same questions in reverse, asking about the cessation of loneliness. This analysis considerably broadens current understandings of middle- and late-life loneliness, offering new insight into who, at these ages, is at greatest risk of becoming lonely, and providing policymakers with a critical understanding of potential intervention points.

Cross-Sectional Correlates of Loneliness

There is considerable research on the individual-level correlates of loneliness in midlife and older adulthood (see Dahlberg et al., 2022, for a recent systematic review); this work can be grouped into three categories. First, some ascriptive or demographic characteristics and early-life experiences are associated with a higher likelihood of reporting loneliness. Loneliness tends to be curvilinear by age, with a higher prevalence among young adults (ages 18–29) and older adults (ages 65 and older) than among middle-aged adults (Nicolaisen & Thorsen, 2014b). Self-reported loneliness also tends to be higher among women than men (Borys & Perlman, 1985) and among racial and ethnic minority groups in the United States (Raymo & Wang, 2022). There is also evidence of an inverse relationship between educational attainment and older adult loneliness (Fokkema et al., 2012). Second, attributes of family and social networks are associated with individual loneliness. Those who live alone, are single, or have small kin networks are more likely to report being lonely than those who coreside with others, are married and report positive marital quality, or have large families (de Jong Gierveld et al., 2012; Dykstra & Fokkema, 2007; Margolis et al., 2022; Stephens et al., 2011). Conversely, grandparenthood and caregiving for grandchildren may reduce loneliness (Quirke et al., 2019). Third, health, wealth, and employment status are associated with loneliness in middle age and older adulthood. Poor health, lower socioeconomic status, and retirement are all associated with a higher likelihood of reporting loneliness (Barlow et al., 2015; Niedzwiedz et al., 2016; Shin et al., 2020; Stickley et al., 2013; Theeke, 2010). Taken together, this research, which is overwhelmingly based on cross-sectional data, paints a clear picture of the correlates of loneliness in middle age and older adulthood, but does not address what increases the risk of becoming lonely or address its possible variation across countries.

Life Events and the Onset of Loneliness

Limited longitudinal analyses of loneliness transitions mean that there is no clear evidence about whether the correlates of loneliness are preexisting or dynamic. For example, those in poor health are more likely to report being lonely. However, it is unclear whether people who are lonelier are also more likely to develop poor health or whether the transition to poor health increases the risk of loneliness. Theorizing does not offer clear predictions either. For instance, in terms of the association between losing a spouse and loneliness, caregiving for a sick spouse could be a lonely experience that predates widowhood, leading to unchanged levels or even potential reductions in loneliness upon the death of a spouse. Alternatively, it is also possible that the loss of a spouse could bring about new feelings of loneliness.

A small set of studies have directly investigated dynamic predictors of loneliness in middle age and older adulthood, finding that the loss of a spouse through death or divorce, physical health decline, and changes in social networks are experiences that increase the risk of becoming lonely. However, these works do not compare across contexts. For example, the death of a spouse predicted a subsequent increase in loneliness in the United States (Domingue et al., 2021), the Netherlands (Dykstra et al., 2005), Finland (Aartsen & Jylhä, 2011), and the United Kingdom (Wenger & Burholt, 2004) and broader changes in marital status like divorce lead to loneliness in Norway (Nicolaisen & Thorsen, 2014b) and England (Victor & Bowling, 2012). In addition, increased physical disability and deteriorating health led to increases in loneliness in England (Victor and Bowling, 2012), the United Kingdom (Wenger & Burholt, 2004), and Finland (Aartsen & Jylhä, 2011). These are single-country studies that do not explicitly compare the association between life events and the onset of loneliness across contexts. Additionally, evidence of these relationships is concentrated in the United States and Europe, and it is unknown whether they hold in other contexts.

Cross-National Comparisons of Life Events and Loneliness

There are several reasons to think that there may be cross-national variation in how life events are associated with loneliness. First, the prevalence of loneliness varies across ­countries—perhaps counterintuitively, people in societies that emphasize social cohesion report higher levels of loneliness than those in more individualistic societies (Dykstra, 2009; Johnson & Mullins, 1987). Therefore, how the experience of something like widowhood or divorce affects transitions into loneliness may also vary based on context-specific perceptions of social cohesion. Second, because the prevalence of life events may differ widely across countries, the experience of a life event and its effect on loneliness may vary. For example, divorce may be more likely to lead to loneliness in a context where it is rarer, or the transition to retirement may be less likely to cause loneliness in a context with a mandatory retirement age policy where the experience is shared by peers. Hagestad (1986) refers to life events that are experienced by a small subset of the population and thus “neither shared nor fully understood by peers” as “lonely transitions” (p. 120), and these transitions may be important predictors of loneliness. Third, there is considerable variation across contexts in other factors, such as religiosity or welfare states, that might affect variation in loneliness in response to life events. For instance, people in more religious societies or ones with more advanced welfare states may find greater community support when faced with individual health challenges, have an easier time navigating the financial challenges of divorce, or gain greater meaning through grandparenting.

On the other hand, certain life events, no matter the context, may have a universal association with loneliness. Evidence for this supposition can be found in the cross-sectional correlates of loneliness. For example, across contexts there is evidence that married individuals are less likely to report being lonely (Fokkema et al., 2012), and therefore experiencing widowhood or divorce may engender loneliness cross-nationally. Similarly, across contexts poor health is associated with higher likelihood of loneliness (Fokkema et al., 2012), and therefore the onset of chronic disease limiting an individual’s ability to engage in social activities may also have a universal association with becoming lonely. The fact that experiencing life events like widowhood, health decrements, and retirement are more common in older adulthood may underpin the universal association between older age and loneliness (Mund et al., 2020). However, without cross-national research comparing the association between life events and the onset of loneliness, the consistency or variation in these relationships across countries remains a puzzle.

The Present Study

To fill this gap in the existing research on loneliness, we draw on the Health and Retirement Study family of harmonized surveys to analyze longitudinal data on loneliness across 20 countries covering almost half of the global population over the age of 50. We focus on two research questions. First, we ask how common it is for loneliness transitions to occur in middle age and older adulthood and how this differs cross-nationally. To answer this, we estimate the predicted county-specific prevalence of panel respondents consistently ­reporting they are not lonely, consistently reporting they are lonely, and those who experience loneliness transitions (into loneliness, out of loneliness, and multiple transitions). Note that when examining loneliness transitions, we are constrained by left and right censoring and thus only able to examine loneliness during the observed window; some respondents who we classify as transitioning into loneliness may, for instance, transition out of loneliness after the observation window, a fact we elaborate on below. Second, we ask to what extent life events common in older adulthood predict loneliness transitions and whether these events are consistent predictors of loneliness across countries. We chose six life events that are both common in middle age and older adulthood and that previous research has shown are correlated with loneliness, as outlined in the background section: becoming a widow, getting divorced, ceasing coresidence (living alone after living with others), becoming a grandparent, the onset or increase in chronic conditions, and exiting employment. To answer this question, we estimate prospective transition probability models, stratified by country, assessing the association between each life event and the probability of becoming lonely. By answering these questions, we expand our knowledge of loneliness transitions and the life events in middle and older adulthood that leave one most vulnerable to loneliness.

Throughout the article when we look at the transition from not reporting loneliness in one wave to reporting loneliness in the next wave, we refer to this as an “onset of loneliness,” “becoming lonely,” or a “transition into/out of loneliness.” This is not meant to imply that once an individual experiences loneliness, it is a constant state. In fact, as the following results document, a portion of middle age and older adults who report loneliness in one wave do not report loneliness in subsequent waves, and vice versa. We use these terms, instead, for discursive simplicity to describe the transition from not reporting loneliness in one interview to reporting loneliness in a subsequent interview.

Data

We analyze data from seven longitudinal surveys representing 20 countries. Table 1 contains country and survey descriptive statistics. These studies were chosen because they are all nationally representative surveys of populations ages 50 and older, are part of the Health and Retirement (HRS) family of surveys, and have at least three waves of data harmonized by the Gateway to Global Aging with loneliness measures so that we can examine loneliness trajectories. Supplementary Appendix A provides details on each data set.

Table 1.

Summary of Countries Analyzed—Surveys, Waves, Percent of World Population, and Sample Size

Country	Survey	Waves analyzed	% Worlda population age 50+ in 2020	Sample size
				Loneliness trajectoriesb		Transition probability modelsc
				Respondents	Person-years	Respondents	Person-years
Austria	SHARE	5	0.2%	630	2,454	3,240	7,674
Belgium	SHARE	5	0.2%	1,839	7,183	4,889	13,283
China	CHARLS	3	24.9%	13,643	50,082	17,598	57,992
Czech Republic	SHARE	4	0.2%	891	3,069	4,359	10,005
Denmark	SHARE	5	0.1%	1,462	5,411	3,659	9,805
Englandd	ELSA	9	1.4%	12,280	74,914	14,452	79,258
France	SHARE	5	1.4%	1,236	4,453	3,682	9,345
Germany	SHARE	5	2.0%	984	3,717	4,548	10,845
Greece	SHARE	4	0.2%	1,317	4,350	2,377	6,470
Israel	SHARE	4	0.1%	855	2,737	2,108	5,243
Italy	SHARE	5	1.4%	1,658	6,193	3,961	10,799
Japan	JSTAR	3	3.2%	1,815	5,445	3,309	8,433
Korea	KLoSA	7	1.1%	8,849	50,756	9,670	52,398
Mexico	MHAS	4	1.4%	8,244	30,178	16,879	47,448
Netherlands	SHARE	3	0.4%	335	1,005	1,487	3,309
Poland	SHARE	3	0.8%	485	1,455	1,218	2,921
Spain	SHARE	5	1.0%	1,415	5,222	5,073	12,538
Sweden	SHARE	5	0.2%	1,490	5,597	4,159	10,935
Switzerland	SHARE	5	0.2%	808	3,035	2,753	6,925
United States	HRS	14	6.3%	29,187	218,893	35,681	231,881
Total			46.8%	89,423	486,149	145,102	597,507

Notes: CHARLS = China Health and Retirement Longitudinal Study; ELSA = English Longitudinal Study of Aging; HRS = Health and Retirement Study; JSTAR = Japanese Study of Aging and Retirement; KLoSA = Korean Longitudinal Study of Aging; MHAS = Mexican Health and Aging Study; SHARE = Survey of Health, Aging and Retirement in Europe.

^aUN World Population Prospects—source for percent of the global population age 50 and older.

^bAnalytic sample for loneliness trajectories limited to respondents with three or more loneliness observations.

^cAnalytic sample for transition probability models limited to respondents with two or more loneliness observations.

^dPercent world population age 50 and older in 2020 in the United Kingdom.

In our primary results, we use a direct single-item measure of loneliness because it is available across the most surveys and in the most waves, allowing us to take advantage of the rich longitudinal data and estimate loneliness trajectories and transitions in the largest possible sample. As documented in Supplementary Appendix B, in most surveys, the question asks respondents “Did you feel lonely in the past week?” No/Yes. For surveys with different question wording, we harmonize the measure into a binary indicator of loneliness (Newmyer et al., 2021). In a recent analysis of convergent validity between loneliness measures, Mund et al. (2023) conclude that the various versions of the single-item measures are highly correlated with each other and with the various multi-item scales; earlier work showed similar results across countries in our sample (Newmyer et al., 2021). See our sensitivity analysis section for results using alternative measures of loneliness.

To assess the extent to which and how transitions play a role in loneliness prevalence in middle age and older adulthood, we code cross-wave “loneliness profiles” for each respondent with a categorical variable for: consistently not lonely (respondents reporting they are not lonely in all observed waves), consistently lonely (respondents reporting they are lonely in all observed waves), and transitions (respondents who transition into or out of loneliness or make multiple transitions over observation). The purpose of this analysis is to assess the extent to which loneliness onset is a key issue in middle age and older adulthood, compared to those who arrive at these ages lonely and stay there and those who never become lonely. Next, drawing on previous research on life course experiences that may be risk factors or protective of loneliness, we study six life events as predictors of loneliness onset, coding a binary variable for experiencing each event in each person-year of data. Life events include: becoming widowed, getting divorced, ceasing coresidence (going from living with others to living alone), lost employment, onset or increase in chronic conditions (cancer, lung disease, heart problems, stroke), and becoming a grandparent.

Method

Our analysis is stratified by country, weighted using survey-provided individual weights, and accounts for survey design by using the SVY suite of commands in Stata MP 17. To gauge the extent to which transitions contribute to loneliness prevalence across countries, we use logistic regression to predict the cross-wave measure of loneliness (consistently not lonely, consistently lonely, loneliness transitions), on an analytic sample with one person-year per respondent among those with at least three observations, adjusting for a categorical measure of the number of observations. We report the predicted probability of each loneliness trajectory at three observations of loneliness by country in Figure 1.

Figure 1.

Distributions of loneliness trajectories across countries (weighted and adjusted for number of waves). Estimated using logistic regression predicting binary variables of cross-wave loneliness trajectory (consistently not lonely, consistently lonely, transition into loneliness, transition out of loneliness, multiple transitions), adjusting for number of observations (top coded at four). Predicted probabilities estimated at three observations reported here. Analytic sample restricted to respondents with three or more loneliness observations. Detailed transition estimates provided in Supplementary Appendix C1.

To estimate the association between each life event and the transition into loneliness, we estimate prospective transition probability models using a long file with a person-wave for every observation of loneliness. We regress each of the six life events separately on the transition into loneliness from one wave to the next using logistic regression, adjusting for respondents’ age at survey, number of sample observations, a binary indicator for death over observation period, and a categorical variable for educational attainment (less than high school, high school/vocational training/some college, college or more—which are harmonized for cross-national comparison by the Gateway for Global Aging). Our sample for this analysis includes respondents with at least two loneliness observations. We account for multiple observations per respondent by adjusting standard errors to account for survey design, including observations nested within individuals. Following current best practices for testing cross-model effect size differences in nonlinear models (Mize et al., 2019), we estimate and plot the average marginal effect (AME) of each life course transition on the probability of transitioning into loneliness. In supplemental results we provide statistical tests of the difference between each life event using seemingly unrelated estimation procedures (Mize et al., 2019).

Results

First, to examine loneliness across countries, Table 2 presents cross-sectional estimates (range and mean across the waves analyzed) and longitudinal estimates (predicted prevalence of reporting loneliness at least once across the observation period) of loneliness. As would be expected, longitudinal estimates are higher than the maximum observed in any wave. However, the extent to which this is the case varies substantially across countries (range: 3% higher in Denmark to 24% higher in Mexico), which suggests that transitions into and out of loneliness are a key factor driving older adult loneliness, albeit a more important factor in some countries than others.

Table 2.

Loneliness and Life Event Prevalence Estimates by country

Table 2 also shows that the overall predicted prevalence of each life event varies across countries. The most common life course transition is worsening health, ranging from 16% to 29% in the majority of countries. Similarly, leaving employment (range: 9%–39%) and the transition to grandparenthood (range: 7%–21%) are also quite common during the time frame we observe respondents. Becoming widowed, getting divorced, and ceasing coresidence are less common than changes in employment, health, and grandparenthood. For example, the predicted percent of respondents becoming widowed ranges from 2% in Korea to 16% in Mexico, with 4%–8% of respondents experiencing this life event in the majority of sample countries.

To better assess the role of transitions in the experience of loneliness in middle age and older adulthood, Figure 1 presents the distribution of loneliness trajectories across countries. On the right side is Denmark, with the highest predicted percentage of respondents never reporting loneliness (78%) and the lowest predicted percentage of respondents consistently lonely (4%). On the other end of the spectrum, Greece has the lowest predicted percentage of respondents never reporting loneliness (34%) and the highest predicted percent always reporting loneliness (15%). The United States is among the countries with a low predicted prevalence of consistent loneliness (7%) and a high predicted prevalence of respondents consistently not lonely (61%).

Importantly, Figure 1 illustrates that loneliness transitions are very common across countries. The gray portion of the bars in the middle represents transitions into and out of loneliness, and multiple transitions of loneliness. We estimate that transitions into loneliness (reporting not lonely in one wave, and lonely in all subsequent waves) comprise an average of 39% of the total transitions within each country, ranging from just less than a quarter in Austria and the Netherlands and a half in the Czech Republic. Our estimates show that another 29% of individuals experiencing loneliness transitions across countries are experiencing multiple transitions. That is, they may report no loneliness, subsequently report loneliness, and then report no loneliness again or vice versa (and so on, depending on the number of waves). Transition-specific data are included in Supplementary Appendix C1. Because transitions make up a considerable portion of all those reporting loneliness, to understand loneliness levels across countries, it is critical to investigate what predicts transitions into loneliness.

To investigate what predicts becoming lonely and how this might vary across countries, we examine how experiencing each life event prospectively predicts becoming lonely across countries. Figure 2 plots the AMEs, with solid black dots denoting statistical significance and empty dots denoting associations that fall below conventional statistical significance thresholds (p < .05). Our findings point to cross-national consistency in the predictors of becoming lonely. Across sample countries, becoming a widow has the strongest and most consistent association with becoming lonely in middle age and older adulthood of all the life events we analyze. The top left panel in Figure 2 shows that compared to those who are not widowed, widowhood increases the probability of transitioning into loneliness by 0.22 in China at the lowest end, and by 0.74 in Greece at the highest end. Of the 20 countries in our analysis, only in three (Israel, Japan, and Korea) is widowhood not a statistically significant predictor of the transition into loneliness.

Figure 2.

Average marginal effect (AME) of life course transitions on the transition into loneliness. AME for divorce cannot be estimated in Poland or Japan, AME for coresidence cannot be estimated for Korea or Japan, and AME for grandparenthood cannot be estimated in China or Japan. Supplementary Appendix C2a presents significance tests of the difference between the effect of widowhood and each other life event on becoming lonely. Supplementary Appendix C8a presents corresponding odds ratios.

We compare the other five life events and how they predict the transition into loneliness with the strong association for widowhood. We plot the AMEs for each life event using the same hollow and filled circles we used for widowhood, and to facilitate comparison to the widowhood associations, we include a line showing where each country’s widowhood association stood. The second panel in Figure 2 shows that becoming divorced has a relatively large association with the probability of becoming lonely across most contexts, associations that are positive (those who divorce are more likely to become lonely than those who do not divorce) and statistically distinguishable from zero in most countries. Divorce associations are similar to widowhood ones, emphasizing the importance of transitioning out of marriage on loneliness. In Switzerland and Italy, the divorce AMEs are statistically significantly greater than widowhood AMEs (see Supplementary Appendix C2). In most other countries, AMEs for divorce are lower than or not distinguishable from the widowhood AMEs at a statistically significant level. The Czech Republic and Japan stand out, however; getting divorced in these two countries decreases the probability of becoming lonely at a statistically significant level. The third panel of Figure 2 shows that ceasing coresidence is a significant predictor of loneliness across all contexts, ranging from 0.06 in Mexico to 0.65 in Greece, highlighting the great risks of newly living alone for the onset of loneliness.

The next three factors we examine—exiting the workforce, increasing chronic illness, and becoming a grandparent—are less consistently associated with transitioning into loneliness than widowhood, divorce, and newly living alone. In three-quarters of countries, exiting the workforce is not associated with becoming lonely. The exceptions are the United States, France, Japan, Greece, and Poland, where exiting the workforce statistically significantly associates with a higher probability of transitioning into loneliness compared to remaining employed or already being unemployed. However, these are comparatively small associations in magnitude. As visualized with dots lower than the widowhood line, the magnitude of the AME of exiting the workforce is considerably lower than the AME of becoming a widow in all countries, even the five where workforce transitions are statistically significant predictors.

Although cross-sectional research implicates poor health as a risk factor for being lonely in middle and older adulthood (Fokkema et al., 2012), we find that in 12 out of 20 countries worsening health, as measured by the onset of or an increase in chronic conditions, is not associated with becoming lonely. Like exiting the workforce, in the minority of countries where worsening health is a risk factor for loneliness (England, United States, Sweden, China, Mexico, Greece, Poland, and Korea), AMEs for worsening health are of comparatively low magnitude.

Despite some prior research finding that grandparenthood is protective of physical and mental health (Arpino et al., 2018; Zhang et al., 2022), in most countries we find no statistically significant association between becoming a grandparent and loneliness. Only in five countries (Czech Republic, Austria, Spain, Korea, and United States) is becoming a grandparent protective of transitioning into loneliness. Poland and Italy stand out as contexts where becoming a grandparent increases the probability of transitioning into loneliness. Note that even when significant, the absolute magnitude of AMEs for grandparenthood is modest compared to widowhood.

Overall, although the strength of the association varies from country to country, life events that involve changes in partnership and living arrangements (widowhood, divorce, coresidence) are unfailingly the biggest risk factors for becoming lonely across countries. Life events involving employment, health, and grandparenthood do not have as strong or as consistent an association with becoming lonely.

Sensitivity Analyses

We conducted several sensitivity analyses. First, we conducted two analyses to assess the choice of using a direct single-item measure of loneliness on results. Supplementary Appendix B details how the measurement of loneliness varies across surveys and waves, and Supplementary Appendix C3 presents a sensitivity and specificity analysis of the two types of measures that are included in both HRS and English Longitudinal Study of Aging. The high level of specificity shows that the single-item measure does not misclassify many individuals who are not lonely as lonely, but the moderate sensitivity levels suggest that using the single-item measure does not classify all lonely people correctly as lonely. Therefore, our estimates of loneliness prevalence by country (Figure 1) can be considered conservative. Additionally, Supplementary Appendix C4 presents results from prospective transition probability models using each life event to predict both the direct single-item loneliness measure and a dichotomized version of the University of California, Los Angeles (UCLA) three-item loneliness scale. A comparison of results shows striking similarity in the association between each life event and loneliness, regardless of the loneliness measure used.

Second, to better gauge the stability of our results, we assess how experiencing each life event prospectively predicts exiting loneliness across countries (i.e., people who are lonely and experience a life event then report not being lonely in the subsequent interview). As visualized in Supplementary Appendix C5, life events that involve changes in partnership and living arrangements have the strongest associations with transitions out of loneliness across almost all countries, like they do for the transition into loneliness, even though the associations with exiting loneliness are weaker. These results align with our general conclusions.

Third, because of evidence of higher rates of self-reported loneliness among women than men and gendered variation in experiencing key life events like widowhood, we stratified the analysis by gender. As the results presented in Supplementary Appendix C6 show, across all countries, women are more likely to report being consistently lonely than men. Likewise, women are also less likely to report being consistently not lonely. For both men and women, widowhood has the strongest and most consistent relationship of all the life events with transitioning into loneliness. In about half the countries, this relationship is approximately equivalent by gender, and in the other half of the countries, the association with widowhood is greater for men (Supplementary Appendix C7), a pattern that also holds for divorce and coresidence. For both men and women, lost employment, increased chronic conditions, and becoming a grandparent have lower and mostly statistically insignificant associations with the probability of transitioning into loneliness.

Discussion

Exploring loneliness transitions among middle-aged and older adults across 20 countries, representing almost half of the adults 50 and older globally, our findings offer new understandings of loneliness in middle and late life in light of current global diversity in aging. Our analysis extends beyond the more commonly studied contexts of the United States, Europe, and China to include Korea, Japan, and Mexico in a comparative framework. Of note, Mexico and Korea are both among the five countries with the highest proportion of individuals reporting loneliness over the course of observation: 63% in Mexico and 57% in Korea. The high prevalence of loneliness in these places underscores the need to extend research on older adult loneliness to these and other rapidly aging countries.

Our cross-national longitudinal analysis of loneliness extends prior understandings of loneliness in several significant ways. These results highlight that far greater numbers of older adults are experiencing loneliness than is evident from cross-sectional estimates, either those previously published or as estimated here. For example, our estimates show that the cross-sectional prevalence of loneliness in the United States is between 14% and 23% but that a predicted 40% of midlife and older adults experience loneliness when analyzed longitudinally. While our findings of higher rates are not surprising since loneliness is often transient, these results suggest it is critical to understand what produces transitions into loneliness in this age range, not simply whether some types of people are at greater (perhaps lifelong) risk of being lonely. Because these data include varying waves of observation across surveys, and between individuals within surveys, our results do not allow estimations of loneliness expectancies. This is an active area for future research and could be accomplished by following the exemplary calculations of loneliness using a Sullivan life table approach (Raymo & Wang, 2022). Doing so would allow more nuance on current estimates and projections of middle age and older adult loneliness (Newmyer et al., 2022).

Of even more significance, the fact that our findings document substantial and surprising consistency in the association between life events and transitions into loneliness, across 20 countries globally, extends our understanding of the centrality of family and household transitions to loneliness risk. In 85% of the countries analyzed here (17 out of 20), becoming widowed is associated with a significantly higher likelihood of becoming lonely compared to those who were not widowed. Becoming widowed has a positive but not statistically significant association with becoming lonely in the remaining three countries. Of all the life events we examine, widowhood has the strongest association with loneliness transitions in all but five countries. Overall, life events that involve changes in partnership and living arrangements (widowhood, divorce, coresidence) have the strongest associations with the onset of loneliness across almost all countries despite variation in loneliness trajectories and prevalence of life events by country.

That we do not find a strong and consistent relationship between the onset/increase in chronic illness and loneliness is noteworthy considering the sizable research attention paid to the health–loneliness relationship. Future research should consider the association between the onset of disability and becoming lonely, in light of the potentially isolating effect of activity limitations (Korporaal et al., 2008). Likewise, the limited association between exiting employment and transitioning into loneliness is surprising. This may be explained by the fact that our measure of exiting employment includes both voluntary (retirement) and involuntary exit (laid off or fired; Shin et al., 2020). The conflation of these different types of exit from employment may explain the weak association. Future research, with data that can disentangle voluntary and involuntary exit from employment, should separately investigate their association with loneliness. These results draw into focus the prominence of family and household connections, and life events involving changes in these factors, in older adult trajectories of social well-being.

Some life events are much more common than others. Leaving employment, worsening health, and becoming a grandparent are quite common at ages 50 and older, but other life events are much less common (Table 2) and, therefore, may be more isolating. For example, between 9% and 39% of our respondents left the labor force in the countries we analyzed, while less than 2% experienced a divorce. Those experiencing the transition out of employment are likely to have friends and acquaintances going through the same thing, much more so than divorce. It is possible that the low prevalence of widowhood, divorce, and ceasing coresidence make them “lonely transitions” (Hagestad, 1986) and contributes to the stronger associations of these life course experiences vis-à-vis loneliness. Hagestad’s “lonely transitions” occupy a particularly prominent place in life course theories of adolescent well-being (Patterson et al., 2020; Shanahan, 2000), but their relevance to middle age and older adult health remains undertheorized.

Comparing the association of widowhood and divorce on loneliness contributes additional depth to our findings. Whereas widowhood is consistently positively associated with becoming lonely, the relationship between divorce and loneliness is inconsistent across contexts. For example, in the Netherlands and France, divorce is associated with a higher likelihood of transitioning out of loneliness—divorce is protective against loneliness in these contexts. These results highlight the importance of studying both determinants of marital dissolution in older adulthood, including “gray divorce” (Lin et al., 2019; Wright et al., 2020) and widowhood (Dahlberg et al., 2015).

Further, our gender-stratified results leave opportunities for future research. On the one hand, women are more likely to consistently report loneliness than men across countries (Supplementary Appendix C6). On the other hand, in half of the countries we investigate, the relationship between becoming a widow, getting divorced, and ceasing coresidence is greater for men than for women (Supplementary Appendix C7). This may mean that men are more vulnerable loneliness as a result of experiencing these life events, even if their experience of loneliness is more transient.

The harmonized data we use represent the best cross-national data available to investigate loneliness in a life course framework among middle age and older adults. However, these data are not without limitations. Because the interviews occur approximately every 2 years, we are not able to capture short-term changes in loneliness between surveys. We are also unable to account for loneliness before or after the observation period. Additionally, while we analyze some of the most common life events in midlife and older adulthood known to be associated with loneliness, our exploration of life events is not exhaustive. Future research should look at other events, particularly those that may be protective against loneliness like repartnering, becoming a caregiver, and starting to volunteer, in prospective analyses of loneliness transitions. Despite limitations, these results provide the most comprehensive cross-national study of middle age and older adult loneliness trajectories to date.

Our results indicate that around the world, middle age and older adults experiencing widowhood, divorce, and changes in living arrangements are at the greatest risk of detrimental social health. Considering decreased fertility characterizing the Second Demographic Transition (Lesthaeghe, 2014), which led to smaller families and increasing numbers of older adults without children (Verdery et al., 2019), experiencing widowhood and divorce in absence of other close kin at older ages may become an even greater risk factor for loneliness in the future. In light of these findings, interventions targeting middle age and older adults who experience changes in family structure and residence are critical tools in cross-national efforts to decrease the risk of loneliness worldwide.

Funding

Research reported in this publication was supported by the National Institute on Aging of the National Institutes of Health (1R01AG060949). We also acknowledge the Government of Canada—Social Sciences and Humanities Research Council (435-2017-0618 and 435-2022-0764).

Conflict of Interest

None.