Examining the Bidirectional Nature of Loneliness and Anxiety Among Older Adults in Daily Life

Abstract

Objectives

Loneliness in later life increases the risk for adverse health outcomes; however, less is known about how loneliness is maintained. Anxiety may play an important role in maintaining loneliness, but little is known about how this connection plays out over time in daily life. This study thus focused on the within-person associations between momentary loneliness and anxiety among older adults.

Methods

Participants were 317 diverse older adults (40% Black; 13% Hispanic, mean age = 77.45 years, 67% women) systematically recruited from the Bronx, NY, who completed ecological momentary assessments 5 times daily for 14 consecutive days. Multilevel models tested bidirectional contemporaneous, momentary cross-lagged (t − 1), day-level cross-lagged (average day to end of day), and day-to-day cross-lagged associations between loneliness and anxiety. Separate sensitivity analyses controlled for concurrent overall mood valence or depressed state. Gender and mild cognitive impairment (MCI) status were tested as moderators at all timescales.

Results

Significant bidirectional associations between loneliness and anxiety were found at the contemporaneous and momentary cross-lagged (t − 1) timescales. Higher average daily loneliness predicted higher end-of-day anxiety, but not vice versa. Loneliness and anxiety were not significantly associated from day to day. Sensitivity analyses revealed some associations varied depending on inclusion of either concurrent mood valence or depressed state. Neither gender nor MCI status moderated associations at any timescale.

Discussion

Findings shed light on the complex temporal ordering of loneliness and anxiety in daily life and extend contemporary theoretical notions of loneliness, including the possibility of interventions that target key moments in daily life.

Ecological momentary assessment, Mental health, Social connection

Loneliness is a distressing feeling that can negatively affect health during older adulthood (Courtin & Knapp, 2017; Ong et al., 2016a). The subjective feeling of loneliness is characterized by the perception of a discrepancy between desired and actual social relationships (Peplau & Perlman, 1982) and is distinct from objective social isolation (Victor et al., 2000). Loneliness is not only an unpleasant emotional experience but is also linked with many negative physical health outcomes, including cardiovascular disease (Holt-Lunstad & Smith, 2016; Valtorta et al., 2016), dementia (Lara et al., 2019), Alzheimer’s disease (Sundström et al., 2020), and higher rates of mortality (Holt-Lunstad et al., 2015; Luo & Waite, 2014). A large-scale meta-analysis found that one in four older adults (aged 60 years and older) is lonely (Chawla et al., 2021). This high prevalence reflects recent urgent calls for more research to understand the impact of loneliness on older-adult health (National Academies of Sciences, Engineering, and Medicine, 2020). Importantly, because there are many different sources of loneliness across the life span (which often vary by developmental stage, sociocultural context, and life circumstances; Qualter et al., 2015), it can be challenging to develop and implement effective intervention strategies that apply to a wide range of individuals. Therefore, understanding factors that might maintain loneliness may inform future interventions to help mitigate loneliness among older adults.

Anxiety may maintain or perpetuate the experience of loneliness. According to the regulatory loop model of loneliness proposed by Cacioppo and Hawkley (2009), loneliness elicits hypervigilance to social threats, setting off a cycle of negatively biased cognition and behavior that is accompanied by feelings of anxiety, which can in turn contribute to or extend feelings of loneliness (Cacioppo et al., 2006). Thus, there are many potential socioemotional and cognitive factors that may be bidirectionally related to loneliness over time, and anxiety is one important factor to consider. Anxiety may be an indicator of hypervigilance to threat, as it is a prominent feature of anxiety (Weierich et al., 2008), a state that includes excessive tension and worries. Higher loneliness is both cross-sectionally (Hoffman et al., 2021) and longitudinally (Domènech-Abella et al., 2019) associated with higher anxiety among a wide range of adults (~18–87 years); this association appears to be bidirectional (Lim et al., 2016; Maes, Nelemans, et al., 2019), with each feeling influencing the other. For example, cross-sectional research linking anxiety to loneliness among older adults suggests that this association may in part be due to the role that anxiety can play in impairing personal relationships and decreasing a sense of connection to others (Hoffman et al., 2021). Other empirical work linking loneliness to anxiety suggests that fear of rejection from others and a negative bias (both of which tend to be present in the context of loneliness) may contribute to heightened anxiety (Lim et al., 2016).Therefore, it may be valuable to assess loneliness in relation to anxiety, broadly. Although previous studies show that higher trait loneliness is associated higher trait anxiety, little is known about how loneliness and anxiety relate to one another at shorter timeframes in daily life. This is important because the cumulative effects of loneliness and anxiety experienced in daily life (and over time) may be a potential pathway by which broader, longer-term health effects are observed. Because ecological momentary assessment (EMA) data are collected in natural environments, such data allow for the detection of dynamic changes in real-life psychosocial and behavioral processes over time (Smyth et al., 2017). Although it appears that momentary loneliness varies as a function of psychological distress in daily life among young adult college students (Meng et al., 2020; Yung et al., 2021), it is unknown whether such momentary and daily associations apply to older adulthood, a distinctly different developmental stage of life that presents unique risks for health and well-being.

Momentary Loneliness and Anxiety

Although associations between loneliness and anxiety have been established among older adults, studies have typically relied on between-person comparisons (i.e., people higher in loneliness versus people lower in loneliness). This is important because associations between loneliness, affect, and cognitions (including anxiety) are typically described as processes occurring within individuals (Cacioppo & Hawkley, 2009; i.e., moments higher in loneliness compared to moments lower in loneliness, for a given person). Studies that assess between-person associations provide important information comparing lonely to nonlonely (or less lonely) individuals, which tells us who is at risk for negative outcomes related to loneliness. However, within-person examination is needed to know what happens when a person feels lonely. As between-person associations do not always exhibit the same pattern of results as within-person associations, it is critical to investigate the correct level of inquiry particularly when informing or designing interventions. Few EMA studies that assess within-person processes of loneliness in older adults exist, and most focus on contextual daily life factors, such as being home alone or around others (Compernolle et al., 2021, 2022) and leisure activities (Fingerman et al., 2021). Such studies provide important information regarding social and environmental context surrounding loneliness in daily life; however, they do not provide adequate insight into the psychosocial processes (such as anxiety) that may help maintain loneliness over time.

Most work linking loneliness and anxiety in older adults’ daily life stems from research spurred by the COVID-19 pandemic. For example, one study used a network analysis approach and found that loneliness was positively associated with negative affect among older adults during the pandemic (Badal et al., 2022). More work is needed outside the context of COVID-19 to determine whether such processes are distinct to such a uniquely stressful context, such as the pandemic. One study that took place prior to the COVID-19 pandemic found that EMA-aggregated levels of loneliness and distressed affect (a composite of anxiety, worry, and fear) were related in a single day of repeated (four) assessments (Steptoe et al., 2011); however, this study focused on between-person- not within-person-associations, and data were constrained to a single day, which did not allow for substantial variability in loneliness.

It can also be important to examine for whom within-person associations are strongest, as such work may help inform ways to target interventions. Prior research linking gender with trait loneliness and well-being suggests that gender may be a particularly important sociodemographic factor to examine in the context of the loneliness and anxiety link. Previous research demonstrates inconsistent findings, with some studies showing no gender differences in the prevalence of loneliness across the lifespan (for meta-analysis; see Maes, Qualter, et al., 2019), other studies suggesting a stronger impact of loneliness among men (e.g., Ernst et al., 2021), and other studies showing more loneliness and worse emotional and overall well-being among women (Pinquart & Sörensen, 2001; Shiovitz-Ezra et al., 2009). Another particularly important factor to consider in the present sample of older adults is mild cognitive impairment (MCI). There are emotional and behavioral changes that accompany cognitive decline prior to dementia onset among older adults, and anxiety specifically has been highlighted as one factor that is associated with accelerated cognitive decline prior to dementia onset (Pietrzak et al., 2014, 2015). Individuals with MCI also report less enjoyment of social activities (Zhaoyang et al., 2021) which could place them at higher risk of loneliness. Therefore, a better understanding of how gender and MCI status relate to the timing of associations between loneliness and anxiety might help identify groups most at risk for factors related to sustained loneliness.

The Current Study

The current research aims to address the gap in the literature regarding how momentary loneliness and anxiety relate on a within-person level among older adults. To shed light on a potentially vicious cycle between loneliness and anxiety, it is not only helpful to determine concurrent and moment-to-moment associations between loneliness and anxiety but also how loneliness relates to anxiety on other timescales in daily life. The present research asks whether loneliness felt at one point in time relates to anxiety levels at the same time, hours later, over the course of a full day, or into the next day. It is also important to ask if this association is bidirectional. Given that the timeframe between feeling lonely and feeling anxious is not specified in theoretical models of loneliness—and that loneliness and anxiety are expected to reinforce one another—we examined bidirectional associations between loneliness and anxiety at four different timescales during daily life: contemporaneous moments, moment level-lagged (t − 1), day level-lagged (average to end-of-day), and day to day. We hypothesized that loneliness and anxiety would be positively associated bidirectionally at all timescales. Lastly, we examined whether associations varied by gender and MCI status, on an exploratory basis (i.e., without directional hypotheses).

Method

Participants and Procedure

Participants were recruited as part of the ongoing Einstein Aging Study using systematic random sampling of New York Registered Voter Lists in Bronx County. Data for the current study were collected between May 2017 and March 2020 (i.e., prior to the announcement by the World Health Organization of the COVID-19 pandemic [March 11, 2020] and the U.S. government-issued stay-at-home orders). Recruitment letters describing the study were mailed to potential participants and interested participants were screened for eligibility over the phone. To be included, participants had to be 70 years or older and ambulatory, be fluent in English, and reside in the community (participants diagnosed with dementia were excluded from participating; see Author Note 1). The final study sample consisted of 317 older adults who provided valid data during the 14-day EMA session from burst 1 (see Table 1 for participant characteristics).

Table 1.

Descriptive Information on Participant Characteristics (N = 317)

	Mean or %	SD	Range
Sex/gender	67.5% (n = 214) women
Age	77.45	4.83	70.4–90.6
Race/ethnicity
White (non-Hispanic)	45.9% (n = 145)
Black	40.2% (n = 127)
Hispanic, White	9.8% (n = 31)
Hispanic, Black	2.8% (n = 9)
Asian	1.3% (n = 4)
Income	42.6% (n = 132) <$30,000
Education (years)	14.97	3.55
Married	33.4% (n = 106) married
Live alone	53.6% (n = 170) live alone
MCI status (yes)	31.9% (n = 101)
Loneliness (trait)	4.02	1.37	3–9
Momentary loneliness	13.21	16.58	0–100
Momentary anxiety	19.90	16.85	0–100
Momentary depressed state	12.36	13.83	0–100
Momentary mood valence	82.05	13.99	0–100

Notes: SD = standard deviation; one participant did not report race/ethnicity; seven participants did not report income data; mood measured as 0 (bad) to 100 (good); MCI = mild cognitive impairment status (see Author Note 4).

Eligible participants provided consent and visited the research clinic to complete demographic and psychosocial questionnaires. At this visit, participants also completed 1.5 hr of training on the EMA study protocol, including use of the study smartphones. The next day, participants completed a 2-day practice EMA session, immediately followed by the 14-day formal EMA protocol. The EMA protocol included participant-initiated wake-up and end-of-day surveys and four quasi-random beeped surveys (approximately 3.5 hr apart). A more detailed description of the study procedure can be found elsewhere (Zhaoyang et al., 2021, 2022).

The current study used data from baseline assessments of demographic and psychosocial variables and EMA-beeped surveys across the 14 days (i.e., only the four quasi-random beeped surveys and the end-of-day survey; morning surveys did not assess loneliness or anxiety). Participants were generally highly compliant, with 83.38% of all beeped surveys completed and 80.26% of all end-of-day surveys completed. Missing data analyses suggested that compliance was not significantly associated with demographic factors (such as race/ethnicity or gender), except for a positive association between education and completion of end-of-day surveys (see Zhaoyang et al., 2022, for more details). The compliance rates of any surveys were not associated with baseline trait loneliness. The 317 participants in the present research completed 18,364 momentary assessments across the EMA period (14,801 beeped surveys and 3,563 end-of-day surveys).

Measures

Momentary loneliness and anxiety

At each EMA (and the end-of-day survey), participants answered, “Right now, do you feel lonely?” by using a slider scale beginning with “not at all” (0) and ending with “extremely” (100). They also answered, “Right now, do you feel tense/anxious?” by using a slider scale beginning with “not at all” (0) and ending with “extremely” (100).

Other variables of interest

Because each person was compared to themselves, the analytical plan of the present research naturally adjusted for between-person differences (Curran & Bauer, 2011). However, because loneliness and anxiety are often related to depression and overall negative mood (Cacioppo et al., 2006; Leitenberg, 1990; Lim et al., 2016), levels of EMA self-reported depressed state and overall mood valance were added individually to models for sensitivity analyses. Including these variables in follow-up analyses allowed us to test the association between loneliness and anxiety above and beyond other momentary aspects of negative affect. For momentary depressed state, participants answered, “Right now, do you feel depressed/blue?” by using a slider scale beginning with “not at all” (0) and ending with “extremely” (100). For overall mood valence, participants answered, “Right now, how is your overall mood?” by using a slider scale beginning with “bad” (0) and ending with “good” (100).

Analytic Plan

All models were estimated in a multilevel modeling (MLM) framework to examine bidirectional within-person associations between loneliness and anxiety at four different time scales (i.e., contemporaneous, momentary-level cross-lagged, day-level cross-lagged, and day-to-day cross-lagged). This analytic framework was chosen because it is a commonly recommended approach for analyzing nested data (i.e., repeated assessments within individuals). All models were estimated using the nlme package in R (Pinheiro et al., 2017; R Core Team, 2022). All models included random intercepts and slopes. For model convergence, correlations among the random effects were fixed to zero in the multivariate multilevel models. Three models were estimated at each timescale: unadjusted, adjusted for concurrent mood valence, and adjusted for concurrent depressed state.

Contemporaneous associations between momentary loneliness and anxiety were estimated via univariate MLMs with autocorrelated residuals (AR1) to account for the possibility that each outcome is influenced by levels measured at the previous assessment. We used within-person centered predictor values to test how a higher momentary score than is typical for a given person (i.e., loneliness or anxiety) is associated with the outcome at the same assessment (i.e., loneliness or anxiety). The within-person centered value was created by subtracting each person’s mean value from each of their own momentary values.

Our second set of analyses tested the momentary-level-lagged (t − 1) association between loneliness and anxiety to test potential “carry-over” effects at the next EMA (3–4 hr later in the same day). Bidirectional associations were evaluated using a multivariate multilevel model for intraindividual coupling (Bolger & Laurenceau, 2013; Hox et al., 2017; MacCallum et al., 1997), which is equivalent to a random-intercept cross-lagged panel model but adapted for EMA data. This model simultaneously evaluates the autocorrelation amongst momentary loneliness and anxiety, and the cross-lagged associations between loneliness and anxiety to shed insight into their temporal ordering or precedence (i.e., a test for Granger causality).

The third set of analyses examined whether average levels of loneliness (aggregated across the day) predicted anxiety assessed at end-of-day (and vice versa) using multivariate multilevel models for intraindividual coupling. For both loneliness and anxiety, we created a day-level aggregated score for each person using the four quasi-random EMA-beeped assessments throughout the day (i.e., not including the end-of-day score). We used these day-level average scores to predict loneliness or anxiety assessed at the evening (end-of-day) EMA on the same day.

For the final set of analyses examining whether average levels of loneliness and anxiety within a day had “carry-over” effects into the following day, multivariate multilevel models for intraindividual coupling were estimated. The same day-level aggregated scores used in the day-level cross-lagged models described above (i.e., average scores using the four quasi-random EMA-beep assessments) were used. As in the model estimated for momentary cross-lagged associations, we could not estimate the correlations among the random effects. Lastly, moderation by gender and MCI at all timescales was conducted on an exploratory basis (see Author Note 2).

Results

Descriptive Statistics

Across the 317 participants, average levels of momentary loneliness during the 14-day study period were relatively low (M = 13.21 [SD = 16.58] on a 100-point scale). The intraclass correlation for momentary loneliness (ICC = 0.75) indicated that 25% of the variation was at the within-person level. This suggests that although momentary loneliness was generally stable, individuals varied from their typical levels from moment to moment. Overall, participants had relatively low levels of momentary anxiety (M = 19.90 [SD = 16.85] on a 100-point scale), low levels of momentary depressed state (M = 12.36 [SD = 13.83] on a 100-point scale) and were overall in a moderately good mood across the EMA period (M = 82.05 [SD = 13.99] on a 100-point scale). See Table 1 for additional descriptive information.

Contemporaneous Models

Contemporaneous models revealed significant bidirectional associations between loneliness and anxiety, even when controlling for concurrent mood valence or depressed state (see Table 2). The unadjusted model showed that EMAs characterized as lonelier than typical were associated with higher anxiety at the same EMA (b = 0.295, SE = 0.028, p < .001) and vice versa (b = 0.108, SE = 0.011, p < .001).

Table 2.

Contemporaneous Models (Univariate MLMs)

		Unadjusted			Adjusted for mood valence			Adjusted for depressed state
Outcome	Fixed effects	Est	SE	p	Est	SE	p	Est	SE	p
Anxiety	Intercept	19.898*	0.945	< .001	19.970	0.945*	<.001	19.899*	0.946	<.001
	Loneliness	0.295*	0.028	<.001	0.225	0.025*	<.001	0.169*	0.024	<.001
	Mood valence	—	—	—	–0.323	0.008*	<.001	—	—	—
	Depressed state	—	—	—	—	—	—	0.400*	.011	<.001
	Random effects	Est	CI lower	CI upper	Est	CI lower	CI Upper	Est	CI lower	CI Upper
	Intercept	16.641	15.366	18.021	16.655	15.381	18.033	16.674	15.399	18.055
	Loneliness	0.333	0.282	.393	0.290	0.243	.346	0.268	0.221	.323
	Phi	0.193	0.178	.208	0.169	0.154	.184	0.157	0.142	.172
	Residual	15.241	15.076	15.408	14.533	14.378	14.690	14.640	14.484	14.798
	Fixed effects	Est	SE	p	Est	SE	P	Est	SE	p
Loneliness	Intercept	13.212*	0.931	< .001	13.228*	0.931	<.001	13.211*	0.931	<.001
	Anxiety	0.108*	0.011	< .001	0.089*	0.011	<.001	0.067*	0.010	<.001
	Mood valence	—	—	—	–0.066*	0.005	<.001	—	—	—
	Depressed state	—	—	—	—	—	—
	Random effects	Est	CI lower	CI upper	Est	CI lower	CI upper	Est	CI lower	CI upper
	Intercept	16.508	15.260	17.858	16.504	15.257	17.853	16.515	15.252	17.882
	Anxiety	0.139	0.139	0.177	0.155	0.138	0.175	0.140	0.123	0.159
	Phi	0.179	0.164	0.194	0.175	0.160	0.190	0.159	0.143	0.175
	Residual	9.238	9.138	9.338	9.197	9.098	9.296	9.013	8.917	9.111

Notes: Phi = autocorrelation; residual = within-group standard error.

^*p < .05.

Momentary Cross-Lagged (t − 1) Models

Unadjusted models showed that moments characterized by higher than typical levels of loneliness for a given person were associated with higher levels of anxiety at the next EMA, approximately 3–4 hr later (b = 0.055, SE = 0.020, p < .001). This cross-lagged association was also bidirectional, with higher than typical levels of anxiety at one EMA associated with higher levels of loneliness during the next EMA (b = 0.026, SE = 0.007, p < .001). Effects of cross-lagged anxiety on loneliness were robust to the inclusion of concurrent momentary mood valence as a covariate (b = 0.019, SE = 0.007, p = .007), but not significant for cross-lagged loneliness on anxiety (b = 0.033, SE = 0.019, p = .076). Neither cross-lagged associations were significant when controlling for concurrent momentary depressed state (ps > .321). However, autoregressive terms for loneliness and anxiety were statistically significant regardless of the inclusion of covariates. Specifically, levels of loneliness at one EMA were associated with levels of loneliness approximately 3-4 hr later (b = 0.159, SE = 0.018, p < .001) and levels of anxiety at one EMA were associated with levels of anxiety approximately 3–4 hr later (b = 0.188, SE = 0.016, p < .001). Full model results are displayed in Table 3.

Table 3.

Momentary Cross-Lagged (t − 1) Models (Multivariate MLM)

	Unadjusted			Adjusted for mood valence			Adjusted for depressed state
Fixed effects	Est.	SE	p	Est.	SE	p	Est.	SE	p
Loneliness intercept	13.073*	0.935	<.001	13.121*	0.935	<.001	13.104*	0.783	<.001
Loneliness autocorr	0.159*	0.018	<.001	0.156*	0.018	<.001	0.146*	0.016	<.001
Anxiety cross	0.026*	0.007	<.001	0.019	0.007	.007	0.007	0.007	.321
Mood valencet	—	—	—	-0.073	0.007	<.001	—	—	—
Depressed statet	—	—	—	—	—	—	0.225*	0.010	<.001
Anxiety Intercept	19.790*	0.967	<.001	20.023*	0.973	<.001	19.844*	0.819	<.001
Anxiety autocorr	0.188*	0.016	<.001	0.161*	0.015	<.001	0.160*	0.013	< 0.001
Loneliness cross	0.055*	0.020	.006	0.033	0.019	.076	0.015	0.020	.451
Mood valencet	—	—	—	-0.340*	0.010	<.001	—	—	—
Depressed statet	—	—	—	—	—	—	0.380*	0.013	<.001
Random effects	Est.	CI lower	CI upper	Est.	CI lower	CI upper	Est.	CI lower	CI upper
Loneliness intercept	16.511	15.256	17.868	16.517	15.261	17.875	13.775	12.885	14.726
Anxiety intercept	16.928	15.619	18.354	17.065	15.747	18.493	14.311	13.366	15.324
Loneliness autocorr	0.179	0.152	0.210	0.179	0.152	.210	0.126	0.103	0.155
Anxiety autocorr	0.181	0.157	0.209	0.158	0.136	.185	0.146	0.125	0.170
Loneliness cross	0.117	0.077	0.180	0.097	0.059	.159	0.132	0.095	0.183
Anxiety cross	0.059	0.044	0.079	0.056	0.040	.076	0.040	0.024	0.066
Rho	0.118	0.099	0.138	0.091	0.071	.111	0.008	−0.004	0.021
Residual	9.109	8.983	9.237	9.061	8.935	9.188	9.915	9.761	10.071

Notes: t = at time of assessed outcome. Models would not converge with correlated random effects. Statistically significant estimates are bolded; italics indicates p < .10.

^*p < .05.

Day-Level Cross-Lagged (Average Day to End-of-Day) Models

Unadjusted day-level cross-lagged analyses showed that days characterized as lonelier than typical (indicated by day-level aggregated scores) were associated with higher levels of anxiety at the end of the day (b = 0.318, SE = 0.037, p < .001). Further, effects of loneliness on anxiety held after controlling for concurrent (end-of-day) associations of mood valence or depressed state (see Table 4). However, day-level cross-lagged models were not bidirectional; days with higher levels of anxiety than typical (indicated by day-level aggregated scores) were not significantly associated with higher levels of loneliness at the end of the day (b = 0.008, SE = 0.017, p = .626).

Table 4.

Day-Level Cross-Lagged Models (Average Day to End-of-Day; Multivariate MLM)

	Unadjusted			Adjusted for mood valence			Adjusted for depressed state
Fixed effects	Est.	SE	p	Est.	SE	p	Est.	SE	p
Loneliness intercept	12.692*	0.320	<.001	20.988*	0.915	<.001	10.128*	0.339	<.001
Loneliness autocorr	0.802*	0.260	<.001	45.846*	1.507	<.001	0.697*	0.025	<.001
Anxiety cross	0.008	0.017	.626	−0.005	0.017	.757	−0.012	0.017	.495
Mood statet	—	—	—	−0.346*	0.017	<.001	—	—	—
Depressed statet	—	—	—	—	—	—	0.199*	0.013	<.001
Anxiety intercept	17.773*	0.838	<.001	45.846*	1.507	<.001	12.168*	0.730	<.001
Anxiety autocorr	0.276*	0.033	<.001	0.222*	0.031	<.001	0.224*	0.029	< .001
Loneliness cross	0.318*	0.037	<.001	0.237*	0.033	<.001	0.143*	0.029	<.001
Mood statet	—	—	—	−0.346*	0.017	<.001	—	—	—
Depressed statet	—	—	—	—	—	—	0.453*	0.020	<.001
Random effects	Est.	CI Lower	CI Upper	Est.	CI Lower	CI Upper	Est.	CI Lower	CI Upper
Loneliness i	3.865	3.175	4.705	20.988	19.196	22.781	3.431	2.845	4.138
anxiety intercept	13.571	12.317	14.953	45.846	42.892	48.800	11.227	10.221	12.331
Loneliness autocorr	0.271	0.230	0.321	0.749	0.697	0.801	0.254	0.217	0.297
Anxiety autocorr	0.319	0.265	0.385	0.222	0.162	0.283	0.254	0.205	0.314
Loneliness cross	0.228	0.151	0.343	0.237	0.172	0.302	0.085	0.024	0.310
Anxiety cross	0.120	0.084	0.171	−0.005	−0.039	0.028	0.121	0.087	0.168
Rho	0.145	0.109	0.181	0.100	0.064	0.136	0.074	0.038	0.111
Residual	7.923	7.706	8.146	7.786	7.575	8.003	1.478	1.424	1.533

Note: t = at time of assessed outcome. Models would not converge with correlated random effects. Statistically significant estimates are bolded. SE = standard error.

^*p < .05.

Day-to-Day Cross-Lagged Models

In all models (unadjusted and adjusted), day-to-day analyses revealed no statistically significant cross-lagged associations between average daily loneliness and average daily anxiety (all ps > 0.194). However, autoregressive associations were statistically significant for both average daily loneliness and average daily anxiety. Specifically, days where loneliness levels were higher than typical for a given person were associated with higher levels of loneliness on the following day (b = 0.337, SE = 0.027, p < .001), an effect that was robust to the inclusion of average daily mood valence (b = 0.374, SE = 0.027, p < .001) and average daily depressed state (b = 0.350, SE = 0.027, p < .001). Similarly, days when participants reported higher levels of anxiety than typical were associated with higher levels of anxiety on the following day (b = 0.066, SE = 0.023, p = .005). This effect was also robust to the inclusion of average daily mood valence (b = 0.059, SE = 0.022, p = .008) and average daily depressive state (b = 0.047, SE = 0.022, p = .029). Full model results are displayed in Table 5.

Table 5.

Day-to-Day Cross-Lagged Models (Multivariate MLM)

	Unadjusted			Adjusted for mood valence			Adjusted for depressed state
Fixed effects	Est.	SE	p	Est.	SE	p	Est.	SE	p
Loneliness intercept	–0.597	0.506	.238	–0.571	0.508	.261	–0.607	0.519	.243
Loneliness autocorr	0.337*	0.027	<.001	0.374*	0.027	<.001	0.350*	0.027	<.001
Anxiety cross	–0.011	0.015	.469	–0.012	0.015	.412	–0.017	0.013	.194
Mood valencet	—	—	—	–0.046*	0.008	<.001	—	—	—
Depressed statet	—	—	—	—	—	—	0.154*	0.010	<.001
Anxiety intercept	0.365*	0.167	.029	0.421*	0.163	.010	0.347*	0.160	.030
Anxiety autocorr	0.066*	0.023	.005	0.059*	0.022	.008	0.047*	0.022	.029
Loneliness cross	–0.002	0.010	.855	–0.004	0.009	.671	–0.008	0.009	.386
Mood valencet	—	—	—	–0.173*	0.012	<.001	—	—	—
Depressed statet	—	—	—	—	—	—	0.289*	0.015	<.001
Random effects	Est.	CI lower	CI upper	Est.	CI lower	CI upper	Est.	CI lower	CI upper
Loneliness intercept	8.166	6.968	9.570	8.224	7.020	9.635	8.848	7.287	9.861
Loneliness autocorr	0.276	0.235	0.325	0.275	0.234	0.324	0.267	0.226	0.315
Anxiety autocorr	0.209	0.173	0.252	0.195	0.160	0.237	0.197	0.162	0.239
Anxiety cross	0.125	0.096	0.161	0.121	0.093	0.158	0.102	0.075	0.139
Rho	0.204	0.172	0.235	0.189	0.157	0.221	0.145	0.112	0.177
Residual	6.445	6.283	6.611	6.421	6.259	6.587	6.255	6.098	6.415

Notes: t = at time of assessed outcome. Models would not converge with correlated random effects or the random intercept for anxiety and the random slope for loneliness cross. Statistically significant estimates are bolded. CI = confidence interval; SE = standard error.

^*p < .05.

Exploratory Moderation Analyses

Neither gender nor MCI status significantly moderated the association between loneliness and anxiety at any timescale (see Supplementary Material for details).

Discussion

Loneliness is a serious concern during older adulthood because it is associated with poorer overall health and well-being (Holt-Lunstad & Smith, 2016; Lara et al., 2019; National Academies of Sciences, Engineering, and Medicine, 2020; Ong et al., 2016b). Past research suggests that trait anxiety may contribute to the maintenance of trait loneliness over time, with some studies showing a significant bidirectional association (Lim et al., 2016; Maes, Nelemans, et al., 2019). The current study provides an important extension of such work by examining within-person associations of momentary loneliness and anxiety assessed via EMA among older adults during daily life. Our findings suggest that when an older person is feeling lonelier than is typical for them, they likely also feel more anxious at the same time, 3–4 hr later, and at the end of the same day. Further, the reciprocal nature of the momentary cross-lagged findings supports the notion that the vicious cycle of momentary loneliness and anxiety may go in both directions. Our findings align with past research on older adults that has shown that momentary loneliness assessed throughout a single day is associated with distressed affect that day (Steptoe et al., 2011).

Given that loneliness, depression, and anxiety are related and commonly conflated dysphoric constructs (Cacioppo et al., 2006; Leitenberg, 1990; Lim et al., 2016), separate sensitivity analyses were conducted to include levels of concurrent overall mood valence or depressed state to determine the distinct effect of loneliness and anxiety. For the contemporaneous models, a significant bidirectional association between loneliness and anxiety remained even when controlling for either mood valence or depressed state. For the momentary cross-lagged model, the effect of anxiety on loneliness at the next EMA remained significant when concurrent mood valence was controlled, whereas the effect of loneliness on anxiety at the next EMA was null. The effect was also null in both directions when adjusting for concurrent depressed state. In the current study, stronger correlations for anxiety (compared to loneliness) with mood valence and depressed state may explain why the bidirectional associations were not robust to their inclusion. Past research has found that although loneliness, anxiety, and depression are related, they are distinct constructs (Fung et al., 2017). Moreover, depressive state or mood valence may serve as a moderator or mediator in certain contexts. Given that including such variables in models in various ways often leads to different results, it is an ongoing challenge for researchers to disentangle these concepts.

The day-to-day cross-lagged analysis was not significant with or without the inclusion of either concurrent mood valence or depressed state. The differences in model results by time frame (specifically, the null effects at longer time frames) may suggest that daily loneliness and anxiety (calculated as the aggregate of momentary states) do not have as strong of a lingering effect on each other as trait-like tendencies of loneliness and anxiety seem to do over a much longer time interval (e.g., Domènech-Abella et al., 2019). Trait loneliness may reflect more of a chronic perception of loneliness, whereas a momentary state or daily level is understood to be transient and perhaps circumstantial (Peplau & Perlman, 1982). Therefore, it may also be that momentary loneliness and anxiety have an effect on each other the next day, but that the effect is not detectable until accumulation reaches a certain threshold.

It is also possible that the link between loneliness and anxiety may indicate the presence of hypervigilance for social threats, a key component of the loneliness loop model (Cacioppo and Hawkley, 2009) and a prominent cognitive feature of social anxiety (Lim et al., 2016). More research is needed that specifically tests various aspects of social hypervigilance in order to make the connection with loneliness more explicit. The current study did not include an explicit measure of social hypervigilance. It would be interesting for future research to collect EMA data that measures social monitoring, negative attentional biases, and interpretations of ambiguous social information throughout daily life to examine specific patterns of such behavior in relation to loneliness. Although past research has observed a link between hypervigilance and anxiety regardless of clinical status (Bar-Haim et al., 2007), the way in which anxiety that is part of a clinical condition (e.g., generalized anxiety disorder, PTSD) may relate to loneliness may vary from subclinical levels of anxiety. It is further possible that different subtypes of anxious states (e.g., social anxiety specifically versus general anxiety) might be differentially linked with different subtypes of lonely states in daily life. For example, it may be important to distinguish among loneliness experiences in different contexts, such as among peers, family (Lasgaard et al., 2011), or romantic relationships (DiTommaso et al., 2004). Future research could specifically test these different dimensions of loneliness and anxiety in daily life.

Our findings emphasize the importance of considering anxiety when evaluating momentary loneliness for older adults. Future interventions that aim to disrupt sustained loneliness may aim to use methods to deliver interventions during key moments in daily life (e.g., anxious moments), such as just-in-time adaptive interventions (JITAI; Nahum-Shani et al., 2018). This aligns with a recently proposed pyramid model of vulnerability for loneliness that could help guide policy and interventions (Holt-Lunstad, 2021). The pyramid model depicts three levels of vulnerability: primary (at risk for isolation), secondary (beginning to disconnect and isolate), and tertiary (highly isolated). Implementing a JITAI may offer the opportunity to influence individuals at any of these levels and offset loneliness over time. However, future research would need to replicate the present findings and contextualize the associations observed in the present research by examining components of daily life that are unique to older adulthood or various moderators before testing the efficacy of specific interventions. Given the complexity of the multivariate models and temporal focus in the current research, our analyses prioritized the internal, psychological associations and did not include the role of environmental factors. Future work might examine how social/environmental factors in daily life (such as being outside the home, or living with others) may strengthen or attenuate the associations between momentary loneliness and anxiety the current study found for older adults.

Lastly, gender and MCI status did not moderate the association between loneliness and anxiety at any timescale. Although past studies have generally shown no gender differences in the prevalence of loneliness across the lifespan (for meta-analysis, see Maes, Qualter, et al., 2019), research on the impact of loneliness on well-being indicators have been inconsistent, with some work observing stronger associations among men (e.g., Ernst et al., 2021), and others showing associations among women (Pinquart & Sörensen, 2001; Shiovitz-Ezra et al., 2009). It thus may be valuable for future work to continue to examine gender and MCI as moderators (as well as other sociodemographic variables) to help identify groups most at risk for factors related to sustained loneliness.

Strengths, Limitations, and Future Directions

A prominent strength of the current study is the use of frequent momentary assessments, which ensured that the data had high ecological validity and allowed us to model within-person processes over time. The statistical approach we used was another strength in this study because it allowed us to test the bidirectional associations between loneliness and anxiety simultaneously within the same models while also controlling for the autoregressive effects of each. Multivariate multilevel models have not been commonly used in the context of loneliness, making this among the first investigations of bidirectional associations between loneliness and anxious states in daily life. Another important contribution of the current research is the inclusion of sensitivity analyses that include mood valence or depressed state in the models. These results add to growing literature that aims to better understand related mood states in daily life. Lastly, while past research among older adults has primarily been in majority White samples, the current study consisted of a diverse sample (40% Black; 13% Hispanic). However, there were some limitations of the present work.

Theoretical models of loneliness typically depict processes and associations that apply to individuals who have higher levels of (and chronic) loneliness. However, the sample in the current study reported low levels of trait and momentary loneliness overall (see Table 1 for sample means). Although this is in line with findings which suggest that, on average, older adults report less negative emotion than younger adults (Mather, 2012), it is possible that more pronounced or prolonged bidirectional associations may be evident in samples who report higher levels of loneliness and anxiety. It is also possible that gender differences may be more evident in populations with higher levels of momentary loneliness and anxiety. Another limitation of the current study is that results may not be generalizable to older adults in other environmental contexts beyond the Bronx, New York, such as those living in rural areas or areas with less racial/ethnic diversity.

It would be valuable for future work to extend the current findings by testing how the current micro-level processes might accumulate and translate to longer-term health issues over time. Such analyses are possible by using “slice of life” methods, such as EMA methodology (Smyth et al., 2017). Bridging the fine-grained temporal effects of momentary states with longer-term effects might help shed light on the amount of time it might take for momentary states to become problematic (see Author Note 3).

Conclusion

This study provides an important extension of prior work that demonstrates bidirectional associations between trait loneliness and anxiety over longer periods of time. Findings from the current study are consistent with the possibility that there may be a vicious cycle by which loneliness may be influenced and maintained by anxiety in daily life among older adults. This study was among the first to establish bidirectional associations contemporaneously and 3–4 hr later among older adults in daily life. The detailed analyses presented here add nuanced temporal information to contemporary theoretical models of loneliness that propose that sustained loneliness operates in a vicious cycle over time.

Author Notes

1. Dementia diagnosis was based on the DSM-IV standardized clinical criteria. For more information on the larger study methods, please see Katz et al. (2021).

2. Moderation by gender was tested with a variable coded as 0 (women) or 1 (men) and MCI with a variable coded as (0) No or (1) Yes.

3. For more information on the concept of accumulation and aging, see Ferraro and Morton (2018).

4. MCI status of each participant was determined using Jack/Bondi criteria (Bondi et al., 2014; Jack et al., 2009). For more details on use of this criteria in other published work, see Zhaoyang et al. (2021).

Funding

This work was supported in part by National Institute on Aging grants RF1 AG056487 (Engeland and Graham-Engeland), P01AG003949 (Lipton and Sliwinski), and T32 AG049676 to The Pennsylvania State University.

Conflict of Interest

None.

Data Availability

The data, analytic methods, and study materials on which the manuscript is based will be made available per request. This study is not preregistered.

Acknowledgments

The authors would like to thank the participants of the Einstein Aging Study for sharing their personal daily life experiences.