Negative Financial Shock Increases Loneliness in Older Adults, 2006–2016: Reduced Effect during the Great Recession (2008–2010)

1. Introduction

Social relationships and a sense of social connectedness are critical for health and wellbeing, as is evident in the associations between social relationship deficits and morbidity and mortality (Hawkley & Cacioppo, 2010; Holt-Lunstad et al., 2010, 2015). Although social relationships are complex and multidimensional, research has shown that simply feeling alone—commonly known as loneliness—is associated with a range of adverse physical and mental health outcomes (Cacioppo & Cacioppo, 2014). Loneliness is defined as a distressing feeling that accompanies a perceived discrepancy between actual and desired social relationships and is more highly related to the quality than the number of relationships (Peplau & Perlman, 1982). Thus, loneliness is correlated, but not synonymous, with actual social contacts and connections.

A number of factors increase the risk for loneliness, and in older age, these factors tend to revolve around losses—of spouse, friends and family members to death or geographic relocation; health; mobility; and independence, among other losses (Pinquart & Sorensen, 2003). In cross-sectional analyses, loneliness is more prevalent among those with low income and low education (Cohen-Mansfield, Shmotkin, & Goldberg, 2009; Niedzwiedz et al., 2016; Pinquart & Sorensen, 2003), and in prospective analyses, incident loneliness 3.5 years later was more likely to occur among older adults who reported insufficient financial resources (Cohen-Mansfield et al., 2009). However, very little research has explored the effect of income and wealth losses on loneliness. This study examines the impact of significant financial losses on loneliness in older adults, and whether the impact is exaggerated during the Great Recession (2008–2010) when financial and property losses increased in prevalence relative to other periods. Results may reveal a possible reason for changes in loneliness prevalence across historical periods and may indicate that financial assistance is a relevant target for loneliness interventions.

1.1. Background

A growing body of research demonstrates that loneliness is a prevalent problem with adverse health consequences, including premature mortality (Cacioppo & Cacioppo, 2014, 2018; Valtorta et al., 2016). Such findings have prompted attention to the remediation of high rates of loneliness (Holt-Lunstad, Robles, & Sbarra, 2017). Typical candidate intervention targets have included increased social contact and social support (Masi, Chen, Hawkley, & Cacioppo, 2011). To date, personal financial shocks have not been considered relevant intervention targets. Loneliness theory and intervention research would benefit from data showing that life experiences that are only indirectly related to social relationships - such as financial shocks - can have an impact on feelings of isolation.

Financial shock, an unexpected and sizeable loss of income or wealth, is a stressful experience. In cross-sectional data, stress is associated with higher levels of loneliness, depression, and related psychosocial variables (Cacioppo, Hawkley, & Thisted, 2010; Niedzwiedz et al., 2016; Pool, Needham, Burgard, Elliott, & de Leon, 2017; Yilmazer, Babiarz, & Liu, 2015). In longitudinal data, the stress of the loss of wealth after the crash of 2008, largely triggered by home foreclosures, increased depression and the use of antidepressant medications (McInerney, Mellor & Nicholas, 2013). Financial shock may also increase loneliness (De Jong Gierveld, Keating, & Fast, 2015). Support for this conjecture comes from evidence, first, that individuals who experience financial loss are prone to feeling shame about what are perceived as personal failings (Starrin, Åslund, & Nilsson, 2009). Second, the effects of shame on one’s identity and self-esteem can negatively affect relationship quality, and can cause people to actively withdraw from others, or to engage in self-protective emotional distancing in social relationships (Heretick, 2013; Starrin et al., 2009), thus increasing the risk of loneliness (Hawkley & Cacioppo, 2010).

Does this pattern of effects differ during a general economic downturn such as the Great Recession? Previous research has provided mixed evidence regarding this question. On the one hand, financial shock was prevalent during the 2008–2010 economic decline (Burgard, Seefeldt, & Zelner, 2012; Catalano et al., 2011; Tsai, 2015), and created a unique social context that could have altered the way people perceive their economic status. When an external explanation (i.e., a national economic crisis) can be provided for personal financial downturns, there may be a greater tendency to feel a sense of belonging to a group that shares a common experience. We posit that the impact of financial shock on loneliness may be ameliorated by a sense of shared experience during a recession relative to non-recessionary periods. This finding is consistent with research on period differences in the risk for depressive symptoms and suicide, where risk increases during periods when financial shocks are rare relative to periods when they are more common and more likely to be shared among individuals and groups (Corcoran & Arensman, 2010; McInerney & Mellor, 2012; Neeleman, 2002).

On the other hand, there is also theoretical justification to posit that financial shock has a greater impact on loneliness during a recession than otherwise. This hypothesis is based on evidence that perceptions are more closely related to negative outcomes than are actual experiences of disadvantage and loss (Ferraro & Shippee, 2009; Wilkinson, 2016). We posit that negative perceptions of loss, and the magnitude of the loss, may have been exacerbated during the Recession when media bombarded the public with news about the Recession’s widespread and adverse financial, material, and health consequences. The result would be a greater increase in loneliness during the Recession if individuals overestimated the direness of their futures during this relative to other periods. Wilkinson (2016) found that decreased financial resources during the 2006–2010 period were associated with perceptions of financial strain, which, in turn, were robustly associated with worsening mental health. However, Wilkinson’s study did not examine whether the association was stronger in Recession versus non-Recession periods, nor did his examination of mental health include loneliness.

What might account for a link between financial shock and loneliness? Although not the primary focus of the current study, the lack of research in this area motivated an exploratory examination of plausible mechanisms. For instance, poor physical health and physiological functioning are associated with financial shocks (Boen & Yang, 2016; Erixson, 2017), and poor health is itself a risk factor for loneliness, particularly when it involves functional limitations that affect the ability to maintain social activities and interactions (Luo, Hawkley, Waite, & Cacioppo, 2012). We posit that poor health and physical disabilities may mediate the association between financial shock and loneliness.

In addition, people may alter their social activities in response to financial losses, including the frequency with which they attend religious services. Religious and related charitable organizations play important roles in providing financial and material assistance to those who suffer economic loss (Allard, Wathen, & Danziger, 2015). Religious organizations may also provide a safety net and, under the best of circumstances, a nonjudgmental environment to form and maintain social connections that prevent or alleviate loneliness. More frequent attendance at religious services thus increases the likelihood of receiving social support (Rote, Hill, & Ellison, 2013), which, in turn, has been shown to reduce loneliness (Masi, et al., 2011).

Third, the quality of social relationships may change as a result of experiencing financial shock. In general, better quality social relationships can moderate the negative consequences of stressful events on individuals’ health (Cohen, 2004). However, economic hardship does not necessarily elicit positive support from others. Financial strain has been linked with strain in social relationships (Conger et al., 1990; Gudmunson, Beutler, Israelsen, McCoy, & Hill, 2007; Stephens, Alpass, & Towers, 2010). Poor quality relationships, in turn, are closely associated with loneliness (Hawkley et al., 2008; Peplau & Perlman, 1982).

1.2. The Current Study

We use longitudinal data from the Health and Retirement Study to address two primary research questions: (1) Are financial shocks associated with increases in loneliness? And (2), are the effects of financial shock on loneliness ameliorated or exacerbated during the Great Recession? A secondary research goal is to explore three types of possible mediators: health and physical functioning, social activity, and relationship quality. As illustrated in our theoretical framework (Figure 1), our approach isolates the causal effect of financial shock by controlling for multiple sources of confounders and ensures a time-lag between financial shock and the mediator and outcome to account for possible reverse causality. The existing literature tends to use either loss of income or loss of wealth to denote financial shock, and the loss of either resource has been found to affect mental health (Margerison-Zilko, Goldman-Mellor, Falconi, & Downing, 2016; Pool et al., 2017). However, losses of income and wealth are often interrelated. If only one is measured without considering the other, the effect of wealth loss may be confounded by that of income loss. Thus, we distinguish between negative income shocks and negative wealth shocks, measuring their effects in the same model and assessing their net, independent effects on individuals’ loneliness.

Fig. 1.

A conceptual model illustrating temporal causal relationships among financial shock, outcome (loneliness), mediators and covariates.

2. Method

2.1. Sample

The Health and Retirement Study (HRS) is a longitudinal panel study that began in 1992 with a nationally representative sample of U.S. adults over age 50 and interviews roughly 20,000 respondents biennially on subjects related to health care, housing, assets, employment, and pensions. The current study relies on data from the RAND HRS Longitudinal File 1992–2016 and the HRS Psychosocial and Lifestyle Questionnaire (PLQ). The PLQ is a self-administered module that is assigned to a random 50% sample of respondents in every other wave; thus observations for a given individual from two consecutive waves are separated by a four-year gap. Table 1 shows the sampling design of our analysis. The UCLA-R loneliness scale questions were piloted in a subset of respondents in 2004 during the in-person interview, and these questions were formally moved to the PLQ in 2006. For our outcome measure, we use loneliness data from 2006–2016. The definition of the treatment variables, outcome variables, and covariates will be described in the next section. The sample is restricted to respondents aged 50 years and older in 2004. Our choice of analytic strategy (described in Section 2.3) requires that the analytic sample be restricted to respondents who have multiple observations across waves and whose loneliness score varies over time. The final sample consists of 6,532 respondents with non-missing data on the financial shock variables and other covariates.

Table 1.

Design of the analytic sample and measurements, with columns indicating years at which different variables are measured.

	Wave 7	Wave 8	Wave 9	Wave 10	Wave 11	Wave 12	Wave 13
	2004	2006	2008	2010	2012	2014	2016
Main Survey	All Sample	All Sample	All Sample	All Sample	All Sample	All Sample	All Sample
Leave Behind Questionnaire		Subsample1	Subsample 2	Subsample 1	Subsample 2	Subsample 1	Subsample 2
Measurement
Outcome		Y	Y	Y	Y	Y	Y
Treatment	A	A	A	A	A	A
Time-Lagged Covariates	Lt-1	Lt-1	Lt-1	Lt-1	Lt-1	Lt-1

Data source: The Health and Retirement Study, 2004–2016.

Note: Y = UCLA-R loneliness scale, A = financial shock, L_t-1 = time-lagged covariates, including marital status, insurance status, labor force status, household size, and health status, L = time-invariant demographic covariates, include age, gender, race, and education.

2.2. Measures

Treatment variables

Adjusted Annual Household Wealth and Household Income.

Total household wealth is calculated as the sum of the value of the primary residence (secondary residences are excluded), real estate, vehicles, business, individual retirement account (IRA), stocks, mutual funds, investment trusts, checking, savings, money market accounts, certificates of deposits (CDs), government saving bonds, treasury bills, bonds, bond funds, and all other savings. Net household wealth is calculated by subtracting from total household wealth the sum of the value of mortgages, land contracts, home loans, and all other debts. Household income is defined as the sum of individual earnings, household capital income, individual earnings from employer pension or annuity, and individual income from social security DI (Disability Insurance) or SSI (Supplementary Security Income), social security retirement, unemployment or workers’ compensation, and other governmental transfers, plus all other household income. Missing data in income and wealth were both replaced with imputed values provided by RAND HRS. We adjusted all net wealth and income values using the 2013 Consumer Price Index (CPI) and transformed the variables using an inverse hyperbolic sine function to deal with distributional skewness and negative values (Burbidge, Magee, & Robb, 1988). We corrected CPI-adjusted net wealth and income variables for household size by dividing values by the square root of household size.

Financial shocks.

The treatment variables are negative wealth shock and negative income shock. We operationalized negative wealth shock as has been done in prior research (Pool et al. 2017, 2018). Specifically, the loss of 75% or more of the net household wealth between two consecutive waves was defined as a negative wealth shock. The person-year observations with net household wealth less than or equal to zero were included in the analytic sample and were treated as no negative wealth shock. The same threshold was used for negative income shock. A loss of 75% or more of the household income between two consecutive waves was regarded as a shock, and the person-wave observations without any income were treated as no negative income shock.

Outcome variables

UCLA-R Loneliness.

This study used the validated 3-item UCLA Loneliness Scale (Hughes et al., 2004). The three items are: “How often do you feel that you lack companionship?”, “How often do you feel left out?”, and “How often do you feel isolated from others?” Each item has three response options (1 = often, 2 = some of the time, 3 = hardly ever) and loneliness scores were calculated by summing all items and rescaling the totals to generate values that ranged from 0–6, with higher values indicating more frequent feelings of loneliness. This scale has good reliability, with a Cronbach’s α of 0.76 at baseline (2006). Observations with missing loneliness measures are dropped from our analysis (< 3%).

Covariates

Because our analytic model implicitly controls for any time-invariant variables, such as gender, race/ethnicity, education or birth cohort, our analysis includes only time-varying covariates that may confound the relationship between household income/wealth and loneliness (Cohen-Mansfield et al., 2009; Pinquart & Sorensen, 2003; Pool et al., 2017; Margerison-Zilko et al., 2016). Because the time-varying covariates are measured at wave t-1, prior to the wave when loneliness is measured at t, we refer to these variables as time-lagged covariates. Figure 1 provides a conceptual model of relationships among the treatment variables, the time-lagged covariates, and the outcome variable.

Time-lagged covariates include measures of labor force status (employed, retired, or other status including being disabled and not being in labor force), annual net household wealth, annual household income, marital status (partnered or not), health insurance (have any health insurance or not), and household size (single-person household, two-person household, or household with more than two people) at the previous wave. We also include three measures of health status at the previous wave: comorbidity index, a variable that sums the respondent’s number of chronic conditions and ranges from 0 to 8; difficulties with activities of daily life (ADL), a variable that sums across five activity types (dressing, bathing, eating, getting in and or of bed, and using the toilet), the number with which the respondent reports difficulties and thus ranges from 0 to 5; and difficulties with instrumental activities of daily life (IADL), a variable that sums across three activity types (using the phone, managing money, and taking medications), the number with which the respondent reports difficulties and thus ranges from 0 to 3. All these covariates contain very few missing cases (<1%), and we removed all observations that have one or more missing values.

We also created two other time-varying covariates: Recession, a dummy variable denoting 2010 (Wave 10 of the HRS data), which marked the end of the Great Recession (2008–2010); and Wave, a chronological indicator of the data collection wave from which an observation was drawn. The inclusion of these variables allowed us to control for temporal changes in loneliness and evaluate whether the Recession moderates the effects of financial shocks (see Figure 1).

Mediators

We examined three possible mediators. The comorbidity index described above served as a measure of health, and ADLs and IADLs, also described above, served as measures of functional limitations. Religious service attendance assessed whether or not the respondent attended any religious services in the past year (0 = not at all; 1 = once a year or more). Spousal strain and family strain assessed the degree to which respondents experience stress in their social relationships with the spouse and other family members, respectively. For each relationship type, strain is assessed with four items asking how often/how much [family members/spouse]…(1) “make too many demands on you?”, (2) “criticize you?”, “let you down when you are counting on them?”, and “get on your nerves?”. Each item has four response options (1 = a lot, 2 = some, 3 = a little, and 4 = not at all). Following Smith, Ryan, Sonnega, and Weir’s (2017) recommendations, the four items are averaged. After reversing the scale, the composite spousal strain and family strain variables range from 1 to 4, with higher values indicating a higher level of strain. The mediators are concurrent with the outcome variable and lagged by one wave relative to the independent variables.

2.3. Analytic strategy

Loneliness values are negatively skewed, and we therefore use ordinal logistic fixed-effect models to estimate the effect of negative financial shock on loneliness. The model includes controls for individual differences in the baseline level of loneliness and the potentially confounding effects of time-invariant unobserved factors (see Figure 1). Compared to regular fixed-effects models for continuous outcome variables, our model assumes that the dependent variable is discrete but rank-ordered. The model is specified as follows:

$$Y_{it}^{*}={\beta }_0+{\beta }_{1}W{S}_{it}+{\beta }_{2}I{S}_{it}+{\beta }_{3}Wav{e}_{it}+{\beta }_4{\mathit{\text{Recession}}}_{it}+\mathit{\gamma }{\mathit{x}}_{it-1}+{\alpha }_i+{\epsilon }_{it}$$

$$Y_{it}=\{\begin{array}{l}0\text{ if \hspace{1em}}{Y}_{it}^{*}<{\delta }_0,\hfill \\ 1\text{ if }{\delta }_0\le Y_{it}^{*}<{\delta }_1,\hfill \\ \hfill ⋮\hfill \\ J\text{ if }{\delta }_{J-1}\le Y_{it}^{*}\hfill \end{array}$$

where Y_it denotes the level of loneliness for individual i at wave t with possible values ranging from j = 0 to 6. We derive the ordinal logistic model from a measurement model in which a latent variable $Y_{it}^{*}$ ranging from −∞ and ∞ is mapped to the observed variable Y_it. The thresholds or cut-points δs show the mapping rules—for example, if $Y_{it}^{*}$ falls between the latent cut-points δ₀ and δ₁, the observe Y_it is 1. The treatment variables WS_it and IS_it denote negative wealth and income shocks that happened between wave t-1 and t, respectively; Wave_it denotes the wave or year of data collection; Recession_it refers to whether the data were collected right after the end of the Great Recession; X_it-1 refers to a set of time-lagged covariates measured at wave t − 1 ; α_i denotes the individual-level fixed-effects; and ε_it is assumed to follow the standard logistic distribution.

In order to estimate unbiased coefficients, we use Baetschmann et al.’s (2015) conditional ordinal logistic fixed-effects model. The model controls for individual characteristics that do not vary by time, such as gender, race, birth cohort, levels of education, and variables that are not explicitly measured in the data. Note that α_i will be dropped in the fixed-effects models and thus can help rule out omitted variables at the individual level, leaving variation within each individual as the focus of the analysis. Note that our estimates of financial shocks may not be causal or may be biased if (1) time-varying variables that are omitted from the model are correlated with both financial shocks and loneliness; (2) there is a dynamic relationship between the treatment and outcome variables; and (3) the functional form of the model is misspecified.

To examine mediating effects, we first regressed the mediators on the negative income and negative wealth shock variables and their interactions with recession period, as well as time-varying covariates. Linear fixed-effects models were used for comorbidity, ADLs, IADLs, spousal strain, and strain in friendships. Ordinal logistic fixed-effects models were used for religious service attendance. Sample sizes for these models varied dependent on the availability of data for each mediator. Only mediators that were significantly predicted by negative financial shock were considered plausible for the second step in which the mediators were entered into the primary ordinal logistic fixed-effects model predicting loneliness. We did not conduct a formal mediation analysis because financial shocks did not predict any of these mediators (see Mediation section below), suggesting that the mediation effects of these variables are negligible.

3. Results

3.1. Sample characteristics

Table 2 presents the summary statistics of the analytic sample at baseline. As shown in Figure 1, the baseline is 2004 for pre-shock (time-lagged) covariates, 2006 for the outcome variable, and 2004 to 2006 for the financial shock. The mean loneliness level was low at 1.77, and this was reflected in a low POMP score (percent of maximum possible score) of 29.7 (Cohen, Cohen, Aiken, & West, 2010). The proportion of respondents who experienced financial shock was 6.3% for wealth shock and 4.2% for income shock between 2004 and 2006. During the Recession period, the proportions increased, reaching 9.1% for negative wealth shock and 5.2% for negative income shock (not reported in Table 2). Most have retired (51.4%) and have a spouse or partner (70.5%). Although time-invariant demographic variables are not included in the model, we note that our analytic sample is roughly evenly divided among 50–64 year-olds (52.0%) and those 65 and older (48.0%). In addition, more than half of the analytic sample is female (60.3%), and 13.2% is Black.

Table 2.

Summary statistics of the baseline sample.

Variable	Mean (SD)/%
Outcome
UCLA-R Loneliness Scale (mean)	1.77 (1.61)
UCLA-R Loneliness POMP Score (mean)	29.65 (26.85)
Feel lonely (CES-D Loneliness Item)(%)	16.8
Treatment
Negative Wealth Shock (%)	6.3
Negative Income Shock (%)	4.2
Lagged-Time Covariates
Median Annual Household Income	51109.9
Median Total Annual Net Household Wealth	228806.3
Labor Force Status (%)
Employed	37.0
Retired	51.4
Other	11.7
Marital Status (%)
not Partnered	29.6
Partnered	70.4
Insurance (%)
Insured	27.3
not Insured	72.7
Household Size (%)
One (single)	19.1
Two	57.0
More than two	23.9
Comorbidity Index	1.69 (1.27)
Difficulties with ADLs	0.16 (0.57)
Difficulties with IADLs	0.06 (0.28)

Data source: The Health and Retirement Study, 2004–2016.

Note: The descriptive statistics are calculated based on all persons from the baseline sample. The baseline sample is comprised of outcome variables from 2006 and treatment variables and covariates from 2004. Numbers in parentheses are standard deviations for continuous variables. ADL = Activities of Daily Living. IADL = Instrumental Activities of Daily Living. POMP Score = Percent of Maximum Possible score (range = 0–100%).

3.2. Negative financial shock, the Great Recession, and loneliness

Table 3 presents the coefficients from two ordinal logistic fixed-effects models. Model 1 simultaneously assesses the effects of negative wealth shock, negative income shock, and their interactions with the recession period on loneliness. Model 2 adds time-varying covariates to Model 1. Time-invariant demographic covariates, including gender, education, ethnicity, and age, are implicitly controlled in all the models. Additional models (shown in Supplementary Table 1) were conducted to sequentially assess the bivariate impact of each independent variable without shock-by-recession interaction terms, and with and without the addition of time-varying covariates. The pattern of results did not alter our interpretation of Models 1 and 2, presented here.

Table 3.

Coefficients (SEs) from ordinal logistic fixed-effect model regressions of UCLA-R loneliness scale on negative wealth shock and negative income shock, recession period, and their interactions, adjusting for time-lagged covariates.

	Model 1	Model 2
Negative Wealth Shock	0.098	0.105
	(0.081)	(0.084)
Negative Income Shock	0.264*	0.323**
	(0.108)	(0.110)
Recession	0.001	0.003
	(0.041)	(0.041)
Negative Wealth Shock × Recession	0.135	0.137
	(0.143)	(0.143)
Negative Income Shock × Recession	−0.652**	−0.645**
	(0.204)	(0.205)
Wave	0.001	−0.011
	(0.009)	(0.013)
Time-lagged Covariates
Finance
Annual Household Income		−0.046*
		(0.018)
Annual Net Household Wealth		−0.001
		(0.005)
Labor Status (ref = employed)
Retired		−0.060
		(0.066)
Other		0.017
		(0.095)
Marital Status (ref = not partnered)
Partnered		−0.266**
		(0.095)
Insurance (ref = not insured)
Insured		−0.049
		(0.051)
Household Size (ref = 2)
Single Household		0.011
		(0.086)
More than two		0.195**
		(0.064)
Health Status
Comorbidity Index		0.011
		(0.035)
Difficulties with ADLs		0.011
		(0.037)
Difficulties with IADLs		0.100
		(0.065)
Individual Fixed-Effects	Yes	Yes
Log Likelihood	−11852.428	−11807.624
# Person-Year Observations	17,060	17,060
# Individuals	6,532	6,532

Data source: Health and Retirement Study, 2004–2016.

Note: ADL = Activities of Daily Living. IADL = Instrumental Activities of Daily Living.

^†p < 0.1,

^*p < 0.05,

^**p < 0.01,

^***p < 0.001, standard errors are presented in parentheses.

As shown in Model 1, negative income shock has a significant positive effect on loneliness (B = 0.264, p < .05), but this effect is moderated by recession period (B = −0.652, p < .01). This effect is illustrated in Figure 2 (left panel). The effect of negative wealth shock on loneliness is not significant, nor is there a significant wealth shock-by-recession period interaction (Figure 2, right panel). After adjusting time-varying covariates (Model 2), the pattern of effects is unchanged. These findings show that a loss of 75 percent or more of one’s income, but not wealth, is associated with increased loneliness, and that this effect is not confounded by other critical life changes, such as the dissolution of marriage, which may have happened contemporaneously with the negative income shock. However, the significant income shock-by-recession interaction indicates that this effect was reversed during the Great Recession. Older adults who experienced a loss of 75 percent or more of their income during the Great Recession were much less lonely than those who experienced such a loss during other periods between 2006 and 2016, consistent with our amelioration hypothesis.

Fig. 2.

Average levels of loneliness (unadjusted) among those who experienced or did not experience income shock (left panel) or wealth shock (right panel) as a function of the Recession period (Yes = 2008–2010; No = all other 2-year periods from 2006 to 2016). Experiencing income shock during the Great Recession significantly attenuated loneliness relative to experiencing the same kind of shock during non-Recession periods.

There is no evidence of a main effect of the Great Recession in any of the models. However, the effect of negative wealth shock increases in size in a more parsimonious model that removes the non-significant wealth shock-by-recession period interaction from the model (see Supplementary Table 1, Model 7). Although this effect (B = 0.139, SE = 0.74, p < .1) does not achieve statistical significance, additional research is warranted. We return to this issue in our robustness analyses to evaluate possible mortality bias.

The effects of the lagged time-varying covariates are also worth noting. For instance, respondents who have more income are less likely to feel lonely, all else being equal. This finding is consistent with the fact that financial resources enable people to engage in social activities (Niedzwiedz et al., 2016), including paying for transportation to social events and membership or participation fees (Heretick, 2013). Additionally, consistent with prior research (Stack, 1999), results show that married or partnered respondents are less likely to feel lonely than their unmarried peers. In contrast, older adults who live in households with more than two members are lonelier than those who live in a 2-person household. Older adults may find themselves in larger households because they are dependent on others (e.g., mobility issues, financial constraints), circumstances that involve diminishing autonomy and self-efficacy, factors that are known to increase feelings of loneliness (Suanet et al., 2019).

3.3. Mediation

Supplementary Table S2 shows coefficients for associations between negative financial shocks (income and wealth) and candidate mediators. Regarding the health mediators, neither comorbidity nor ADLS were predicted by negative income shock or its interaction with recession period (ps > .05). Negative income shock was modestly but non-significantly associated with IADLs (B = 0.031, SE = 0.015, p < .1). Similarly, negative wealth shock was unrelated to each of the health variables. On the other hand, recession period was related to comorbidity and IADLs such that the number of chronic conditions and functional limitations in instrumental activities deteriorated during the Great Recession relative to other periods, independently of personal financial losses. Religious service attendance, spousal strain, and strain in friendships were not predicted by negative income shock, negative wealth shock, or their interactions with recession period (ps > .05). Yet, recession period was related to religious service attendance, such that attendance was significantly less frequent during the Great Recession (B = −0.119, SE = 0.051, p < .05) than other periods. These results ruled out a role for these possible mediators in explaining the effect of negative income shock on loneliness.

3.4. Robustness checks

We conducted four sensitivity analyses to test the robustness of our findings. First, to test whether the observed effects are robust to alternative loneliness measures, we replicated our analysis in Table 3 using the loneliness item from the CES-D scale, which asks whether the respondent felt lonely in the past year (Radloff, 1977). The response is dummy-coded (1 = yes and 0 = no). We used a logistic fixed-effects model to analyze the effect of financial shock on the CES-D loneliness item. The Results shown in Supplementary Table S3 confirm our main finding. In all model specifications, negative income shock significantly increases loneliness; however, the effect does not vary by the recession period when using the single loneliness item. In addition, the effect of negative wealth shock, which only approached significance with the UCLA-R loneliness scale, achieves statistical significance when the CES-D loneliness item is the outcome.

Second, we tested the robustness of the results to the 75% cut-point for financial shock. Specifically, we created an ordinal financial shock measure with five categories: (1) no loss, (2) less than 25% loss, (3) 25% to less than 50% loss, (4) 50% to less than 75% loss, and (5) 75% or more loss. We analyzed the effect of the ordinal income and wealth shock variables on the UCLA-R loneliness score (Supplementary Table S4). Only an income loss of 75% or more significantly increased loneliness. Overall, we conclude that our choice of threshold for negative financial shock (i.e., >75%) is justified; only at this level of loss does income shock elicit significant increases in loneliness. Moreover, ancillary analyses that included interaction terms between negative financial shock (income and wealth) and baseline economic factors including baseline income, baseline wealth, and poverty status showed no effect on loneliness (available from the first author upon request).

Third, selective mortality may bias our results. To address this issue, we stratified the analyses by age group: those who were 50–64 years old at study onset (“younger group”), and those 65 and older (“older group”). A significant income shock effect (B = 0.464, SE = 0.144, p < .01) and an income shock-by-Recession interaction (B = −0.362, SE = 0.130, p < .01) was evident in the younger group, consistent with what was found in the sample as a whole. The income shock effect for the older subsample was attenuated (B = 0.082, SE = 0.173, ns), but the income shock-by-Recession interaction remains significant and sizeable (B = −0.315, SE = 0.157, p < .01). Wealth shock, on the other hand, had a non-significant but nominally larger effect in older adults (B = 0.135, SE = 0.117, ns) than in younger adults (B = 0.097, SE = 0.097, ns).

Finally, marital dissolution may be confounded with financial shocks and thus may bias results. To address this issue, we stratified the sample on those who remained married throughout the study versus changed marital status during the study. Our main findings were replicated in both subsamples. Income shock was associated with loneliness in the consistently married subsample (B = 0.417, SE = 0.154, p < .01) and in the sample whose status changed during the study (B = 0.374, SE = 0.159, p < .05). Moreover, a significant income shock-by-Recession interaction was evident in both groups (B = −0.878, SE = 0.286, p < .01, and B = −1.052, SE = 0.290, p < .01). We conclude that marital dissolution is inadequate to explain the effects of financial shocks on loneliness.

4. Discussion

This study shows for the first time that a negative income shock affects feelings of loneliness. Specifically, American older adults who experienced a loss of 75% or more of their income over a two-year period showed increased levels of loneliness (see Table 3 and Figure 2, left panel), a potentially consequential effect given the health consequences that have been attributed to loneliness (Cacioppo & Cacioppo, 2014). This effect was independent of baseline levels of household income and wealth, each of which has been associated with a lower risk of loneliness (Cohen-Mansfield et al., 2009; Pinquart & Sorenson, 2003). The effect was also independent of life changes that were contemporaneous with the negative income shock, including the effect of changes in marital status.

We hypothesized that the impact of negative income shock on loneliness could be exacerbated or ameliorated during the Recession. The data support an amelioration hypothesis: negative income shock had an adverse effect during non-Recession periods, but reduced loneliness during the Great Recession (see Table 3). This effect is consistent with research that has shown that risk for other adverse psychosocial outcomes (e.g., depression) is greater during periods when financial shocks are rare than when they are more common (Corcoran & Arensman, 2010; McInerney & Mellor, 2012; Neeleman, 2002). We posit that attenuation of loneliness during the Great Recession may be attributable to the sense of camaraderie that comes from a shared experience. A sense of being part of an entire segment of the population that is experiencing similar financial stress may foster a general sense of belonging, whereas financial stresses experienced during other periods may be perceived as solitary experiences that result in feelings of isolation from the rest of society. By way of comparison, the shared experience of financial losses during the Great Recession may have bolstered solidarity among its “victims” much as tragedies are known to boost social solidarity (Ryan & Hawdon, 2008). Conversely, the “deaths of despair” (suicides, overdoses) that have become particularly prevalent among middle-aged white Americans may be attributable to people feeling disenfranchised from society, unable to reap the economic benefits that others, by all appearances, seem to be obtaining (Case and Deaton, 2015). A general principle underlying these hypotheses is that perceptions of disparity matter as much as real disparity, and this has implications not only for feelings of isolation and loneliness but also for downstream health consequences. Research is needed to examine the hypothesis that perceptions of disparity account for the differential effects of negative income shocks during recessionary versus non-recessionary periods.

Our negative income shock findings were robust to our choice of loneliness measure: both the UCLA-R loneliness scale and the CES-D loneliness item were significantly impacted by income losses of 75 percent or more. However, loneliness assessed using the single loneliness item did not vary across recession periods. Prior research has shown that these two measures of loneliness do not produce consistent effects (Shiovitz-Ezra & Ayalon, 2012), so this discrepancy is not entirely surprising. Nevertheless, our results warrant replication with independent data. If perceptions of disparity help to explain period differences in the effects of income shocks, then data from other western countries that also experienced the Great Recession should produce comparable period differences in the effect of income shock on loneliness.

We tested whether income shock effects were moderated by baseline economic status. In a study that relied on HRS data (McInerney et al., 2013), the effects of the Recession on depressive symptoms were greatest among those with the highest levels of stock holdings prior to the crash. However, we found no evidence for a moderating effect of income, wealth, or poverty status on the effects of income shock. Income shock may be less vulnerable than wealth shock to baseline economic status because income shock has a real impact on everyday life in the here and now, regardless of one’s general economic position. Wealth shock may be vulnerable to baseline wealth levels if, as was done by McInerney et al. (2013), wealth is defined primarily as non-housing wealth (e.g., stocks) because, in later life, those with more initial financial assets have more to lose and may not have time to recover from those losses. In our case, we assessed wealth from both non-financial assets (e.g., property) and financial assets, less any debts. For most older adults in the U.S., home ownership constitutes the largest portion of their wealth and this type of wealth serves to balance out vacillations in income and wealth from financial assets. Indeed, higher levels of non-financial assets, but not financial wealth, have been shown to improve men’s mental health after being widowed (Kung, 2020), highlighting the importance of considering sources of wealth when examining effects on outcomes such as loneliness.

We selected a “shock” cut-point of a loss of at least 75 percent of income or wealth, a cut-point that has been used in prior research (Pool et al., 2017, 2018), but that arguably signifies something different for those at the low versus the high end of the socioeconomic spectrum. Theoretically, even a small percentage loss might be expected to impact low-income individuals, whereas a higher percentage loss might be needed to detect an impact on high-income individuals. Nevertheless, our comparison of varying degrees of loss failed to identify any impact on loneliness of income losses less than 75 percent. Losses of this magnitude may be necessary to detect effects on loneliness (and depression, as per prior research), but we have no reason to expect that these shock thresholds are applicable to other outcomes. In future research, the choice of cut-point will require empirical testing for each unique outcome. For instance, as little as 10 percent loss of wealth contributed to a significant decrement in health in the research reported by Schwandt (2018; see also Erixson, 2017).

Mortality bias is an issue in studies of older adults. We addressed this issue by conducting age-stratified analyses. Results were replicated in the younger subsample of 50–64 year-olds, a group that is less subject to mortality than those 65 years and older. Negative income shock effects were attenuated in older adults, an effect that could be expected based on prior research, which has shown that the loneliest people die earlier than their age-matched less lonely peers (Luo et al., 2012). Alternatively, the younger respondents may be more vulnerable to income shocks than the older respondents because they are unlikely to have social security income to compensate for income loss caused by unemployment, failed investment, or health problems. Yet, the respondents in the group 65 years and older are not invulnerable, either. Income shock is not likely to be a “shock” when individuals first retire since they likely expected and planned for such a shock, but significant income shocks after having been retired for some years may be very unexpected (i.e., a shock). Thus, older adults’ vulnerability may be associated with their retirement status (recent vs. long-standing retirement) and sources of income. In sum, the difference between the two age groups may be explained by selective mortality, the lower vulnerability of older people to income shocks, the transition to retirement, and birth cohort differences, or a combination of these factors. Given that the date of retirement is unavailable in the data, we cannot test detailed mechanisms that explain the observed age differences in the effects of income shock.

The discovery of an effect of financial shock on changes in loneliness inevitably raises the question of mechanism. Possible mechanisms for the effect of negative income shock on loneliness include a wide range of behaviors and social relationships that threaten the capacity of individuals to good quality social relationships. In cross-sectional research, the association between income or wealth and loneliness is typically confounded with marital status; for instance, married individuals tend to have higher incomes and tend to be less lonely (Hawkley et al., 2008). In the present study, we observed significant effects of income shock on loneliness, holding constant concurrent changes in marital status. Moreover, the effects continued to be significant when analyses were limited to those who remained continuously married throughout the study period. We conclude that changes in marital status are insufficient to explain the effects of income shock on loneliness. Findings from cross-sectional research also suggest a confounding of health status and financial status; poorer individuals tend to be sicker and lonelier. In the present study, we found no effect of income shock on comorbidity and physical functioning, and holding these time-lagged factors constant did not alter the influence of income shock on loneliness. We conclude that changes in health and functional status are also insufficient to explain the effects of income shock on loneliness. Similarly, religious service attendance and social strain in relationships with a spouse and family members were not affected by negative income shock and therefore, could not account for the effect of income shock on loneliness.

4.1. Limitations.

Our amelioration hypothesis was premised on the assumption that people’s perceptions of a shared economic shock dampens the impact of their personal financial shocks. That assumption requires additional research since perceptions were not measured in the current study, much less whether perceptions differed during recessionary versus non-recessionary periods.

Our results were robust to the potentially biasing effects of selective mortality and marital dissolution. However, these robustness tests are not sufficient to permit broader generalization of our results. Our findings warrant replication in other countries that also experienced the Great Recession to determine whether comparable period differences are observed in the effect of income shock on loneliness.

Our choice of cut-point for financial shock, namely a loss of 75% or more of income or wealth, was supported by the data showing that smaller shocks did not exhibit an association with loneliness. In the case of wealth shock, the absence of an association with loneliness warrants further research to determine whether the loss of wealth in some classes (e.g., property assets) is more or less impactful than loss in other classes (e.g., financial assets, including stock holdings). Moreover, research is need to examine whether effects on loneliness differ among those with different levels of initial wealth in each class.

The data allowed a limited exploration of mechanisms for the impact of income shock, namely changes in marital, health, and functional status, and changes in social activity and social support. None of these variables were found to explain the main effect. Future research with richer social, behavioral, and psychological health data is needed to explore other mechanisms that may mediate the relationship between financial shock and loneliness.

5. Conclusions

Loneliness has been repeatedly observed to be correlated with low income and wealth (Pinquart & Sörensen, 2003), but this study is the first to use longitudinal data to show that income losses lead to increased loneliness. Using panel data from adults 50 years and older in the nationally representative Health & Retirement Study, we showed that individuals who lost 75% of more of household income over a two-year period were significantly lonelier than those who did not experience such a loss. The impact of income shock differed during the Great Recession of 2008–10 relative to other two-year periods between 2006 and 2016. Specifically, during the Great Recession, income shock reduced loneliness. This ameliorative effect may be attributable to perceptions of shared experience that generate a sense of camaraderie and belonging. Conversely, perceptions of disparity are more likely during non-recession periods when conversations around personal finances are “taboo,” and thus foster greater feelings of exclusion and isolation.

We posited several mechanisms that might account for the effect of income shock on loneliness. Loss of a spouse and changes in health and mobility are associated with greater loneliness but did not account for the effect of income shock on loneliness. Similarly, changes in the frequency of religious service attendance and in levels of social support did not explain the effect. Other plausible mechanisms include relationship quality with close others and frequency of community activity, including volunteerism, each of which are associated with loneliness and are likely adversely affected by significant losses of income.

Financial shocks become more frequent with age (Sass, 2018). The current study shows that targeting older adults’ financial difficulties could directly or indirectly help to reduce loneliness in this vulnerable population. This suggestion is plausible given evidence that a recently introduced pension program in China produced significant decreases in depressive symptoms, especially among those with greater financial constraints (Chen, Wang, & Busch, 2019). Perceptions of and actual inequity may each be a critical component of a successful intervention.

Highlights.

• In older adults, a large loss of income leads to greater loneliness net of covariates.

• Loneliness decreased in those who experienced income shock during the Great Recession.

• Changes in health and social experiences did not explain the income shock effect.

• Reducing financial difficulties may reduce loneliness in older adults.