Social Interactions and Loneliness in Daily Life: A Study of Younger Adults and Cognitively Diverse Older Adults

Abstract

Feeling lonely is a common experience across the lifespan, and people’s feelings of loneliness often do not correspond with their levels of social interaction in expected ways (i.e., social asymmetry). It is unclear, however, whether loneliness in daily life differs by age or cognitive status, and how loneliness varies as a function of social interaction across age and cognitive status. The present research used experience sampling to investigate group differences in loneliness and social interaction in the daily lives of individuals (N = 219; Range_age = 21–84; 57% women; 67% White), including younger adults (M_age = 27), cognitively unimpaired older adults (M_age = 75), and older adults with mild cognitive impairment (MCI) (M_age = 77). Compared to older adults, younger adults reported being lonelier and their loneliness was more strongly tied to recent social interactions (i.e., greater reductions in loneliness versus when no recent interaction). In contrast, relative to younger adults, cognitively unimpaired older adults demonstrated an attenuated negative relation between their social interactions and loneliness levels, and no such association was present for older adults with MCI. Across groups loneliness was lower after social interactions that occurred face-to-face, but partner closeness primarily mattered for reduced loneliness in older adults with MCI. The findings suggest younger adults are particularly vulnerable to experiencing loneliness in their daily lives, and frequent face-to-face social interactions may serve as a buffer against loneliness. Although older adults may feel less lonely on average, older adults with MCI may be especially likely to experience social asymmetries.

Introduction

Loneliness has been defined as an aversive emotional experience that arises from a discrepancy between a person’s desired and achieved level of social activity (Perlman & Peplau, 1981; Russell et al., 1984). This consensus conceptual definition has been implemented to measure loneliness and establish psychometric scales (e.g., UCLA Loneliness Scale, Russell, 1996; de Jong Gierveld loneliness short scales, de Jong Gierveld & Van Tillburg, 2010) that have been extensively employed in empirical investigation (Mund et al., 2023). Feeling lonely is a common experience across the lifespan that has been studied in childhood and adolescence (Maes et al., 2019; Doane & Thurston, 2014), young adulthood (von Soest, Luhmann, & Gerstorf, 2020), middle adulthood (Luhmann & Hawkley, 2016), and older adulthood (Hawkley et al., 2019; Compernolle et al., 2021). It is associated with negative health consequences (for review, see Hawkley & Cacioppo, 2010), mortality risk (for review, see Holt-Lunstad et al., 2015), and risk for dementia (Holwerda et al., 2014).

Because of the relation between loneliness and a number of mental and physical health outcomes, loneliness has been a topic of extensive theoretical (e.g., Cacioppo, Cacioppo, & Boomsma, 2014; Cacioppo et al., 2015) and empirical (e.g., Jones, Hobbs, & Hockenberry, 1982; Watson & Nesdale, 2012) investigation for the past several decades. Researchers have often tried to identify the predictors and correlates of loneliness (Rokach, 1989; Theeke, 2009; Qualter et al., 2013). However, a limited but growing body of research has examined lifespan age differences (e.g., Green et al., 2001; Graham et al., 2024). Instead, past research has primarily focused on age-specific samples (e.g., Cohen-Mansfield et al., 2016; Steptoe et al., 2004; Doane & Adam, 2010). It is important to consider loneliness from a developmental perspective because precipitating events and predisposing factors that make people vulnerable to loneliness vary in relevance across the life course, which can provide insight into both the social standards or goals of people in different life stages and the conditions that contribute to loneliness across the lifespan (Luhmann & Hawkley, 2016).

Loneliness and Age

Older adulthood in particular has been the focus of substantial loneliness research (e.g., Cohen-Mansfield & Papura-Gill, 2007) because normative experiences of advanced age (e.g., decline in functional status, living alone) expose people to greater risk of loneliness, and the effects of loneliness compound over time (Victor et al., 2005; Hawkley & Cacioppo, 2010). Additionally, although involuntary changes to one’s social network (e.g., divorce, widowhood) may occur at any life stage, these changes may feel more irreversible or insurmountable in advanced age due to changes in functional ability, motivation, and cognition (Böger & Huxhold, 2018; Carstensen, 1992; Hess, 1994). This idea is supported by evidence that while older adults are not more likely than middle aged adults to become lonely, they are more likely to remain lonely (Huxhold & Henning, 2023). Despite this intuition, findings from the modest lifespan developmental literature empirically testing associations between age and loneliness are mixed. Some evidence suggests loneliness spikes in adolescence or early adulthood and among the oldest old (i.e., U-shaped curve; Victor & Yang, 2012, Graham et al., 2024), decreases as people age (i.e., negative linear association; Schnittker, 2007), peaks in adolescence and old age but fluctuates throughout the lifespan (i.e., complex non-linear association; Luhmann & Hawkley, 2016, Hawkley et al., 2022), or is unrelated to age (Mund et al., 2020). Moreover, the majority of this research has focused on global reports of overall loneliness or loneliness over a long period of time. Thus, we know very little about age differences in moment-to-moment experiences of loneliness in daily life. Examining how feelings of loneliness vary momentarily, rather than treating loneliness as a static individual difference, allows developmental researchers to discern whether observed age differences in loneliness stem from differential exposure to contexts that precipitate loneliness (e.g., experiencing fewer social interactions) or differential reactivity to those contexts (e.g., different levels of loneliness in response to social interactions) (Compernolle et al., 2021). Additionally, measuring momentary variations of loneliness permits precise identification of the contexts that are linked to people feeling more or less lonely than they typically feel (Van Bogart et al., 2024). Determining the contextual factors related to momentary loneliness across the lifespan can provide opportunities for intervention and facilitate understanding of how daily experiences of loneliness relate to longer term trait-like patterns. In sum, although there has been a vast literature on the prevalence and health consequences of loneliness for specific age groups, evidence for lifespan age differences in loneliness are mixed. Further, age differences in momentary experiences of loneliness in daily life have been largely unexamined.

Loneliness and Cognition

While the association between age and loneliness is uncertain, it is clear that cognition and loneliness are interdependent (Cardona & Andrés, 2023). Greater loneliness has been associated with lower general cognitive ability (Boss, Kang, & Branson, 2015; Gow et al., 2013), and cognitive decline can make it hard to maintain relationships. Difficulty maintaining relationships may predispose a person to loneliness, as it is predicted by diminished social contact and lack of stimulation (Aartsen et al., 2004; Pinquart, 2003). Cross-sectional research (O’Luanaigh et al., 2011) shows that impaired cognition is significantly associated with loneliness independent of social network size, and longitudinal findings (Zhong et al., 2017) indicate that loneliness and cognitive functioning bi-directionally influence one another in older adults, such that impaired cognition exacerbates loneliness and loneliness negatively affects cognitive functioning. In fact, one reason why loneliness in older age in particular has been the target of many interventions (for review, see Fakoya, McCorry, & Donnelly, 2020) is because of its link to social withdrawal, reduced regional brain activity, and cognitive decline longitudinally (Cacioppo et al., 2009; Wilson et al., 2007). Further, there is evidence that momentary loneliness has important consequences for cognition and contributes to maladaptive thinking that may expose people to risk for chronic loneliness (Van Bogart et al., 2024). However, it remains unclear the role that cognition plays in the momentary experience of loneliness and how it may relate to social interactions in daily life.

Loneliness and Social Interactions

Age

Research has emphasized the distinction between loneliness and the objective condition of being alone (Perlman & Peplua, 1981). Importantly, social isolation is not sufficient to elicit loneliness; instead, these feelings arise when there is a perceived difference between the desired and actual quantity or quality of one’s social interactions or network. This conceptual distinction recognizes that, despite lack of social interactions, a socially isolated person may be satisfied with solitude (i.e., not lonely), whereas, despite frequent social interactions, a social butterfly may be discontented with some aspect of their social activity (i.e., lonely). Further, there is reason to believe the relationship between loneliness and social interaction varies with age. The life course is characterized by distinctive motivations and goals, and therefore social experiences, needs, and expectations change across the lifespan (Heckhausen, Wrosch, Schulz, 2010; Qualter et al., 2015). For example, due to age-related motivational changes, older adults tend to derive more emotional satisfaction from smaller social networks characterized by more meaningful social interactions, whereas young adults are concerned with expanding their social network and interacting with new social partners (Carstensen 1992; English & Carstensen 2014; English & Growney, 2021). For this reason, the nature of the discrepancy between expected and achieved social interactions may differ based on age-related differences in social needs and expectations. This idea is supported by research from global self-report measures that shows contact frequency is more correlated with loneliness in young and middle adulthood (Victor & Yang, 2012), and contact closeness is more tied to loneliness levels in older adulthood, such that the average closeness of overall social networks was a predictor of lower loneliness for older adults but not young adults (Green et al., 2001).

Cognitive Status

In addition to the association between loneliness and social interaction varying with age, the relationship between loneliness and social interaction may vary across cognitive status. Cognitive processing (e.g., perception, attention, memory) plays a prominent role during social interactions, and cognitive dysfunction may negatively impact a person’s social interactions and interpersonal functioning (Kessels, Waanders-Oude Elferink, & van Tilborg, 2021). For loneliness to arise, one must first make a cognitive appraisal of a discrepancy between the desired and actual quantity or quality of one’s social activity. Even when a discrepancy is perceived, it does not inevitably lead to loneliness because individuals can engage in emotion regulation, which includes cognitive change, to modulate their response to the discrepancy (Burholt, et al., 2016). In fact, a meta-analysis of interventions targeting loneliness showed that interventions aimed at increasing social contact had no significant effect on loneliness, while addressing maladaptive social cognitions was most effective (Masi et al., 2011). However, the role of impaired cognition in daily experiences of loneliness is poorly understood. Changes in social cognition accompany declines in fluid cognition that may occur in pathological age-related cognitive decline (Bora & Yener, 2017). These changes have relevance for loneliness in daily life, because social cognition is an important aspect of social interaction (Hess, 1994; Zhaoyang et al., 2021) that has been identified as a critical target for reducing loneliness in interventions for older adults (Ong, Uchino, & Wethington, 2015). Mild cognitive impairment (MCI) is a clinical condition in older age that involves greater than average cognitive decline and mostly maintained functional abilities in daily life. It often but not always develops into a form of dementia (Bora & Yener, 2017). People with MCI may be at a particular disadvantage in managing loneliness because fluid cognitive ability is an important resource for social cognition and effective emotion regulation (Kessels, Waanders-Oude Elferink, & van Tilborg, 2021; Growney, Springstein, & English, 2023). Understanding how MCI is related to felt loneliness is crucial for informing loneliness prevention and intervention in this at-risk group, as well as for better isolating normative age-related shifts in loneliness.

Types of Social Interaction

Personal characteristics, like age and cognitive ability, are not the only relevant considerations to account for when evaluating variations in felt loneliness. Situational characteristics, such as the method of the social interaction (e.g., face-to-face, over the phone) may also influence the association between social interaction and loneliness. Past research has emphasized the benefits of face-to-face interaction for positive emotion, through pathways like eye contact and affective synchrony (Hietanen, 2018; Vacharkulksemsuk & Fredrickson, 2012). Nevertheless, technology-mediated communication is becoming more common in the 21^st century, and there is evidence that phone calls and messaging have the functional equivalence of face-to-face contact in some contexts (Burholt et al., 2020). Understanding how interaction method relates to loneliness is essential for informing interventions targeted at reducing loneliness. Importantly, the effects of the communication mode on loneliness may be age-differentiated or generationally differentiated based on the prevalence of certain communication technologies in historical time (Petersen et al., 2016). Additionally, there is reason to believe that cognitive ability may affect the association between communication mode and loneliness. On the one hand, fluid intelligence is related to technology adoption, and thus technology-mediated communication may be less preferable than face-to-face communication for older adults with MCI (Czaja et al., 2006). On the other hand, there is evidence that older adults with cognitive impairments may use communication technologies such as email to make up for communicative deficits, such as impaired verbal fluency or sentence comprehension, that can accompany MCI (Ruppell et al., 2016; Johnson & Lin, 2014).

Beyond the method of interaction, the person one interacts with may also affect their experience of loneliness. While social contact is important for protecting against loneliness (Pinquart, 2003), the meaningfulness of one’s interaction has been shown to be more important for predicting how lonely one feels (Wheeler, Reis, & Nezlek, 1983). Past research (Jones, 1981) has noted the relevance of the relationship type of the partner one interacts with for how this interaction influences loneliness. Additional research has found age effects in the association between social partners and loneliness, such that greater closeness of social network members is related to lower loneliness in older adults, whereas the size but not closeness of the social network is related to lower loneliness in young adults (Green et al., 2001). This finding supports the idea that relationships differentially confer benefits based on characteristics such as intimacy, and, as people age, intimacy is more closely related to emotional satisfaction than the quantity of relationships one has (Wheeler, Reis, & Nezlek, 1983; Carstensen, 1995). Finally, some research suggests that the specific types of partners one interacts with in daily social interactions has discriminate power in identifying people with MCI, such that older adults with MCI are less likely than non-MCI peers to interact with acquaintances and strangers (Zhaoyang et al., 2021). In sum, the social partner one interacts with may be associated with felt loneliness, and this relation may vary based on age and cognitive status.

Present Research

We examined the daily dynamics of social interaction and loneliness in an age and cognitively diverse sample to consider loneliness from a lifespan developmental perspective and to shed light on the effects of impaired cognition on emotional experience and social interactions in daily life. Building on prior research, we derived four preregistered research questions and corresponding hypotheses (see Table 1) to investigate how loneliness in everyday life is related to age, cognitive status, and recent social interaction. Within recent social interactions, we examined whether interaction method and relationship to one’s recent social interaction partner are related to loneliness. We extend cross-sectional studies of loneliness, age, and cognitive status by collecting repeated measures of respondents’ loneliness and context in daily life. The experience sampling method permits analyses of intra-individual (i.e., within-person) variations in experience, as well as inter-individual (i.e., between-person) group comparisons of mean levels of state-level loneliness based on trait-level characteristics (i.e., age and cognitive status). For example, we can compare an individual’s loneliness in the moments following a recent social interaction compared to moments when they had not recently interacted (i.e., within-person), and we can compare loneliness in people who have more social interactions versus people who have fewer social interactions on average (i.e., between-person). This intensive repeated measures approach affords higher temporal resolution than daily diary methods (Buecker, Horstmann, & Luhmann, 2024; Broen et al., 2023), but longer-term repeated measures would be necessary in order to distinguish age and cohort effects.

Table 1.

Research Questions and Hypotheses

Research Questions	Hypotheses
1. How are age (i.e., younger adults versus cognitively unimpaired older adults) and cognitive status (i.e., cognitively unimpaired older adults versus older adults with mild cognitive impairment) associated with mean levels of loneliness in daily life?	1. Cognitively unimpaired older adults will report lower mean levels of loneliness compared to older adults with mild cognitive impairment.
2.1. How is having had a recent social interaction related to loneliness?	2.1. Recent social interactions will be associated with lower loneliness on average, at the between- and within-person levels.
2.2. How does the relation between having had a recent social interaction and loneliness differ by age (i.e., younger adults versus cognitively unimpaired older adults) and/or cognitive status (i.e., cognitively unimpaired older adults versus older adults with mild cognitive impairment)?	2.2. The between- and within-person effects of recent social interactions on loneliness will be attenuated for older adults with mild cognitive impairment relative to younger adults and cognitively unimpaired older adults.
3.1. Within recent social interactions, how is the method of social interaction (i.e., face-to-face vs other methods) related to loneliness?	3.1. Face-to-face interactions will be associated with lower loneliness compared to other methods of interaction.
3.2. Does the relation between method of social interaction and loneliness differ by age (i.e., younger adults versus cognitively unimpaired older adults) and/or cognitive status (i.e., cognitively unimpaired older adults versus older adults with mild cognitive impairment)?	3.2. The reduction in loneliness for face-to-face interactions compared to other methods of interaction will be more pronounced for older adults relative to younger adults and for older adults with mild cognitive impairment relative to cognitively unimpaired older adults.
4.1. Within recent social interactions, how is one’s relationship to their social interaction partner (i.e., close other versus non-close other) related to loneliness?	4.1. Interactions with close others will be associated with lower loneliness compared to interactions with non-close others.
4.2. Does the relation between one’s relationship to their social interaction partner and loneliness differ by age (i.e., younger adults versus cognitively unimpaired older adults) and/or cognitive status (i.e., cognitively unimpaired older adults versus older adults with mild cognitive impairment)?	4.2. The reduction in loneliness for interactions with close others compared to interactions with only non-close others will be more pronounced for older adults relative to younger adults and for older adults with mild cognitive impairment relative to cognitively unimpaired older adults.

Specifically, we examined how age and cognitive status are associated with mean levels of loneliness in everyday life. Based on findings linking cognitive decline to loneliness (Aarsten et al., 2004; Zhong et al., 2017; O’Luanaigh et al., 2011), we predicted cognitively unimpaired older adults would report less loneliness compared to older adults with MCI. Throughout, we use the term “cognitively unimpaired” to refer to people who do not indicate cognitive impairment relative to normative shifts in cognition expected based on their age. Then, we examined the association between recent social interaction and loneliness and whether this association differs by age and cognitive status. Because frequent social exchanges are related to lower loneliness (Okun & Keith, 1998; Burholt et al., 2020), we predicted recent social interactions would be related to lower loneliness at the within and between-person levels. Additionally, regulating loneliness or engaging in social interaction may be cognitively demanding for individuals with cognitive impairment; thus, we predicted that the between- and within-person effects of recent social interactions on loneliness would be attenuated for older adults with mild cognitive impairment relative to younger adults and cognitively unimpaired older adults. Finally, within recent social interactions, we investigated differences in the association between loneliness, interaction method, and recent social interaction partner, and whether the associations differ across age and cognitive status. Phone or internet-mediated communication may be more cognitively demanding for individuals with MCI compared to cognitively unimpaired older adults (Czaja et al., 2006), and for older adults compared to younger “digital natives” (Petersen et al., 2016). Therefore, we expected the reduction in loneliness for face-to-face interactions compared to other methods of interaction to be more pronounced for older adults relative to younger adults and for older adults with mild cognitive impairment relative to cognitively unimpaired older adults. With respect to interaction social partners, we predicted loneliness would be lower after interactions with close others (Wheeler, Reis, & Nezlek, 1983), and that this effect with be more pronounced for older adults (Carstensen, 1995), particularly those with MCI (Zhaoyang et al., 2021). Research questions and corresponding hypotheses are presented in greater detail in Table 1.

Method

Transparency and Openness

This study received approval from Washington University in St. Louis’ Institutional Review Board (IRB). Questions, hypotheses, and analytic plan were preregistered (https://osf.io/n7j8v). Data were analyzed using R, and materials and analysis code are available (https://osf.io/a8ry7/?view_only=4bb80419e1d44887ac37b7fca9418e7b). The data which study conclusions are based are not publicly available because the authors do not have IRB permission to publicly post the data. However, the data can be shared with a data use agreement. Researchers interested in accessing the deidentified data should contact the corresponding author.

Participants and Procedures

The present research is part of a larger study on emotion regulation in cognitively diverse younger and older adults. We recruited 219 participants into three groups: 70 younger adults (20–34 years old, M = 27.4; SD = 4), 89 cognitively unimpaired older adults (70–84 years old, M = 75.2; SD = 3.79), and 60 older adults with MCI (70–84 years old; M = 77; SD = 4.4). The sample is 56.6% women, 42% men, and 1.4% other gender, and the racial/ethnic composition is reflective of the St. Louis, MO area (White or European American = 67%; Black or African American = 27.4%; Asian, Asian American, Or Pacific Islander = 2.3%; Hispanic or Latino = 2.3%; American Indian or Alaska Native = 0.5%; Middle Eastern or Arab American = 0.5%; other race or ethnicity = 2.3%) (participants could endorse more than one race or ethnicity). For demographic breakdowns by group, see Table 2. Participants were recruited through phone calls and letters sent to community members in the St. Louis, MO area, community flyering, and databases of research participants including Washington University’s Older Adult Participant Pool and Volunteer for Health. Individuals were not eligible to participate if they had possible dementia as indicated by a score of 21 or below on the Mini-Mental State Exam (MMSE).

Table 2.

Demographic statistics by group

	Young	CU	MCI	Group differences
White	58.60%	74%	65%	χ2= 4.39
Female	48.50%	67.40%	50%	χ2= 6.21*
Four-year college degree or more	71.40%	79.80%	63%	χ2=4.94
Physical health (mean, SD)	57.3 (7.12)	48.6 (8.61)	44.5 (10.9)	F =35.5**
Live alone	20.30%	49.40%	35%	χ2=14.35**
Married or in committed relationship	44.90%	52.80%	61.70%	χ2 =1.17

Note. Young = Younger adults. CU = Cognitively unimpaired older adults. MCI = Older adults with mild cognitive impairment. Physical Health was measured with the 36-item Short Form Health Survey Version 2 (SF-36v2).

^*p<.05.

^**p<.01.

After providing informed consent, participants completed the following procedures (1) cognitive assessments, self-report surveys, and instructions for the experience sampling procedure, (2) 9 days of experience sampling (seven prompts per day) completed on a smart phone, and (3) a laboratory emotion regulation task. For the current research, we will only use data from the cognitive assessments and the experiencing sampling. To comply with health and safety guidelines during the COVID-19 pandemic, some participants had only one laboratory session with both parts 1 and 3, with the initial self-report surveys completed from home.

For the cognitive assessment portion of the study, participants completed the NIH Toolbox Cognitive Battery (Dimensional Change Card Sort Task; Flanker Attention and Inhibitory Control Task; Pattern Comparison Processing Speed Task; Picture Sequence Memory Task; List Sorting Working Memory Task; Mungas et al., 2014) and the Montreal Cognitive Assessment (MoCA; Nasreddine et al., 2005). Additionally, a researcher certified in clinical dementia rating conducted the Structured Interview and Scoring Tool (SIST; Okereke et al., 2011) with both the participant and an informant.

For the experience sampling procedure, participants used their smart phone device. Participants without a smart phone (n=21) were provided with one. Fisher’s tests revealed that participants who borrowed a phone were significantly more likely to be older (n=20 out of 21; p=.003), but not more likely to be cognitively impaired (n=12 out of 21; p=.08). Participants received an in-depth tutorial covering each item on the survey and describing how to use the smart phone application, and prompts started the following day. Participants were prompted to fill out seven surveys per day during a 14-hour waking period of their choice. They were prompted randomly within each two-hour window (seven prompts per day) for nine consecutive days. At each prompt, participants were asked to respond to several items, including rating their current level of loneliness and reporting on recent social interactions. After receiving a notification for a prompt, participants had 15 minutes to complete the survey. If they did not open the survey within 5 minutes, they received a reminder prompt. In total, there were 13,521 prompts sent. Of these, participants completed 8,746 surveys. Some participants (n=16) did not receive all 63 prompts due to technical problems with the software application. The sample size for variables measured at all observed occasions ranged from 8,525 to 8,746, indicating relatively little item-level missingness. Of observed measurement occasions, 5,856 included a recent social interaction. The sample size for variables measured only after a recent social interaction ranged from 5,740 to 5,831, also indicating relatively little item-level missingness.

Measures

Cognitive Assessments

The NIH Toolbox Cognitive Battery (Mungas et al., 2014) is a battery of cognitive tests administered on an iPad. We examined age-corrected standardized scores and identified scores which were 1.5 standard deviations below the national average for the participant’s age group. Across all tests, the mean age-corrected standardized score is 100 with a standard deviation of 15, so we identified scores less than 77.5 (i.e., 77 or less). We examined scores on the following tests of fluid cognitive ability: Dimensional Change Card Sort Task (an assessment of cognitive flexibility), Flanker Attention and Inhibitory Control Task, Pattern Comparison, Processing Speed, Task Picture Sequence Memory Task (an assessment of episodic memory), and List Sorting Task (an assessment of working memory).

The Montreal Cognitive Assessment (MoCA; Nasreddine et al., 2005) is a 10–15 minute test that assesses cognition in the following areas: visuospatial/executive, naming, memory, attention, language, abstraction, delayed recall, and orientation. In the present study, we examined the overall score, which is a sum of all points awarded. We followed the scoring guidelines and awarded an additional point to participants who had less than or equal to 12 years of education A score of 26 or higher was considered cognitively unimpaired.

The Self and Informant Structured Interview and Scoring Tool (SIST; Okereke et al., 2011) includes questions about the extent to which problems with cognition impair the participant’s functioning in the following domains: memory, orientation, judgment and problem-solving, community affairs, home and hobbies, and personal care. For each interview (self and informant), an overall score of 0 (Normal), 0.5 (MCI or very mild dementia), 1 (mild dementia), 2 (moderate dementia), or 3 (severe dementia) was assigned by a researcher trained in clinical dementia rating (Hughes et al., 1982). We examined these overall scores.

Age and Cognitive Status (i.e., Group)

Participants between the ages of 20 and 34 were classified as younger adults. Participants between the ages of 70 and 84 were classified as older adults. Among the older adults, participants were further classified as cognitively unimpaired or as having MCI based on a protocol set in accordance with recommendations from Petersen et al (2018). Participants were classified as having MCI if one of the following criteria were met: (1) a score of 0.5 on both the self and informant SIST.; (2) One score of 0.5 on either the self or informant SIST, PLUS at least one of the following: a MoCA score below 26 OR a score below 1.5 SDs from the age-adjusted US population mean on one or more of the fluid cognitive ability tests from the NIH toolbox; (3) a MoCA score below 26 AND a score below 1.5 SDs from the age-adjusted US population mean on one or more of the fluid cognitive ability tests from the NIH toolbox.

Demographics

Race/ethnicity.

Participants provided information on their race/ethnicity. We recoded these responses to compute a binary variable: “White/European American” was recoded as 1 and all other were recoded as 0.

Gender.

Participants provided their gender identity as male, female, or non-binary/transgender/other (n=3). We recoded these responses to a compute binary variable: “Male” was recoded as 0, “Female” was recoded as 1, and “Non-binary/Transgender/Other” was recoded as NA.

Education level.

Participants provided the highest level of education they completed. We recoded these responses to compute a binary variable: “Less than 12th grade/GED”, “12th grade (high school grad/GED”, and “Some college or technical school” were recoded as 0 and “Completed 4 year college (BA, BS)”, “Completed Master s degree”, “Completed doctoral degree”, “Completed medical degree”, “Completed other professional degree”, “Completed some graduate school” were recoded as 1.

Physical health.

Participants completed the 36-item Short Form Health Survey Version 2 (SF-36v2). Using the guidelines from Ware et al. (2000), we computed the Physical Component Score as a measure of physical health. The normative mean on this measure is 50 (SD = 10), and higher scores indicate better physical health.

Experience Sampling Items

Loneliness.

Participants responded to the item: “At the time of the prompt, I felt lonely.” Response options ranged from 1 (Not at all) to 7 (Extremely).

Recent social interaction.

Participants responded to the item: “When was the most recent time you interacted with someone?” Response options included: at the time of the prompt; 1–30 minutes before the prompt; 31–60 minutes before the prompt; 1–1.5 hours before the prompt; more than 1.5 hours before the prompt. Although it is possible that participants occasionally interacted through multiple modalities within a single interaction, they were asked to select only one interaction modality per interaction. We used this item to compute a binary recent social interaction variable: “At the time of the prompt” and “1–30 minutes before the prompt” was recoded as 1 (i.e., recent social interaction), and all other responses were recoded as 0 (i.e., no recent social interaction).

Method of recent social interaction.

Participants responded to the item: “How did you interact with this person (or people)?” Response options included: face-to-face; phone call; video call (Skype or FaceTime); text or online message. We computed a binary variable in which face-to-face interactions were coded as 1 and all other interactions were coded as 0.

Relationship to one’s recent social interaction partner(s).

Participants responded to the item: “During your most recent interaction, who did you interact with? (check all that apply, but only one per person).” Response options included: romantic partner; family member; friend; co-worker/volunteer; acquaintance; health professional; stranger; pet (e.g., dog). We used this item to compute a binary variable. If the participant only selected “pet”, the variable was recoded to missing. If the participant selected romantic partner, family member, or friend, the variable was recoded as 1 (i.e., close other). If the participant did not select any of these close other relationships and selected at least one of the remaining options (besides pet), the variable was recoded as 0 (i.e., non-close other).

Statistical Analyses

We conducted a series of multilevel models to test our research questions. In all models, experience sampling surveys were nested within participants. We included fixed and random intercepts, fixed and random slopes for level 1 (i.e., occasion-level) predictors, fixed effects for level 2 (i.e., person-level) predictors, and used unstructured covariance matrices.

We used a series of dummy codes to compare each group to one another, and we report all possible comparisons. However, all our hypotheses concern differences between (a) younger adults and cognitively unimpaired older adults, or (b) cognitively unimpaired older adults and older adults with MCI.

Preliminary Analyses

Before running our primary hypothesis tests, we computed descriptive statistics for all study variables.

Research Question 1

To examine associations of age and cognitive status with loneliness in daily life, we used a multilevel model to predict loneliness from dummy-coded Group (three levels: young adults, cognitively unimpaired older adults, and older adults with MCI). In initial models, we treated cognitively unimpaired older adults as the reference level. This allowed us to compare cognitively unimpaired older adults to younger adults and to older adults with MCI. Although none of our hypotheses concerned a difference between younger adults and older adults with MCI, we also reran the models with younger adults as the reference group to allow for a complete set of statistical comparisons between all three groups.

Research Question 2

To examine the relation between social interactions and loneliness in daily life, we used a multilevel model to predict loneliness from the presence or absence of a recent social interaction. To disaggregate between- and within-person effects, we included person-mean frequency of recent social interactions (i.e., proportion of observed measurement occasions spent interacting; between-person effect) and fixed and random effects of occasion-specific recent social interaction (i.e., within-person effect).

To test whether the relation between social interactions and loneliness differed across groups, we ran two additional models. First, we added Group, and the interaction between frequency of recent social interactions (between-person interaction model) and Group (i.e., level 2 interaction) as predictors of loneliness. In a separate model, we added Group, and the interaction between occasion-specific recent social interactions (within-person interaction model) and Group (i.e., cross-level interaction) as predictors of loneliness.

Research Question 3

We tested Research Question 3 using only the measurement occasions in which a recent social interaction was reported. To examine the relation between method of social interaction and loneliness, we used a multilevel model to predict loneliness from the method of social interaction (i.e., face-to-face compared to other methods). To disaggregate between- and within-person effects, we included person-mean frequency of recent face-to-face social interactions (i.e., proportion of reported interactions that were face-to-face; between-person effect) and fixed and random effects of occasion-specific face-to-face social interactions (i.e., within-person effect).

To test whether the relation between the method of social interaction and loneliness differed between groups, we ran two additional models. Again, we added Group and the interaction between frequency of recent face-to-face social interactions and Group (i.e., level 2 interaction) as predictors of loneliness. In a separate model, we added Group, and the interaction between occasion-specific recent face-to-face social interactions and Group (i.e., cross-level interaction) as predictors of loneliness.

Research Question 4

We tested Research Question 4 using only the measurement occasions in which a recent social interaction was reported. To examine the relation between one’s relationship to their social interaction partner and loneliness, we used a multilevel model to predict loneliness from one’s relationship to their social interaction partner (i.e., close others compared to more distant others).

To disaggregate between- and within-person effects, we included person-mean frequency of recent social interactions with close others (i.e., proportion of reported interactions that were with a close other; between-person effect) and fixed and random effects of occasion-specific recent social interactions with close others (i.e., within-person effect).

To test whether the association between one’s relationship to their social interaction partner and loneliness differed between groups, we ran two additional models. Again, we added Group and the interaction between frequency of recent social interactions with a close other and Group (i.e., level 2 interaction) as predictors of loneliness. In a separate model, we added Group, and the interaction between occasion-specific recent social interactions with a close other and Group (i.e., cross-level interaction) as predictors of loneliness.

Results

Data were analyzed using R Version 4.1.2 and R Studio Version 2023.12.1+402. Descriptive statistics were computed using the package psych version 2.2.3 and multilevel models were formulated using the package nlme version 3.1–153. Figures were created using the packages ggplot2 version 3.4.4 and emmeans version 1.8.4–1.

To compute standardized coefficients for slopes in our hypothesis tests, we transformed the t-statistics for coefficients into r values (Page-Gould et al., 2019) using the packages parameters version 0.24.2 and effectsize version 0.6.0.1. We interpret the transformed r values in relation to average effect sizes in psychology (Funder & Ozer, 2019).

Preliminary Analyses

As pre-registered, participants were included for analyses if they completed at least one prompt. On average, participants completed 38.9 prompts. The majority of participants (n=160 out of 219) completed more than 50% of prompts, and 104 participants completed more than 70% of prompts. One-way ANOVA revealed a significant effect of Group on number of completed prompts (F(2,216)=3.78, p=.025). Pairwise comparisons conducted using Tukey’s Honest Significant Difference test revealed no differences in completed prompts between young adults (M=40.7, SD=14.5) and cognitively unimpaired older adults (M=40.7, SD=15.6), but older adults with MCI completed significantly fewer prompts (M=34.2, SD=17.4) than both young adults (p=.049) and cognitively unimpaired older adults (p=.036). Participants who borrowed a smart phone also completed significantly fewer prompts (F(1,217)=4.63; p=.038). We retain all data in analyses because those with lower compliance rates may systematically differ from those with high compliance rates, and engaging in list-wise deletion could bias the sample (Newman, 2014).

Participants reported relatively low levels of loneliness in daily life (M = 1.65; SD = 0.86; Range = 1.00–4.60 out of possible 1–7). The ICC for loneliness indicates that 47%f of the variance in loneliness was explained by between-person differences in average levels of loneliness while the remaining 53% was within-person variance and error.

Participants reported frequent recent social interactions (M = 66% of occasions; SD = 23% of occasions). The ICC for social interactions indicates that 21% of the variance in social interactions was explained by between-person differences in average frequency of social interactions while the remaining 79% was within-person variance and error. Most social interactions were face-to-face (M = 73%; SD = 27%) and with close others (M = 78%; SD = 20%). The ICC for social interaction method indicates that 28% of the variance in social interaction method was explained by between-person differences in average frequency of face-to-face social interactions while the remaining 72% was within-person variance and error. The ICC for social interaction partner indicates that 17% of the variance in social interaction partner was explained by between-person differences in average frequency of social interactions with close others while the remaining 83% was within-person variance and error.

Primary Research Questions

As shown in Table 3¹, younger adults reported greater levels of loneliness compared to cognitively unimpaired older adults (β=0.56, r=.28, p<.001) and compared to older adults with MCI (β=0.40, r=.19, p=.006). This small-to-medium effect size corresponds to younger adults feeling approximately one-half point more lonely on a 7-point scale compared to older adults. Contrary to Hypothesis 1, however, mean levels of loneliness did not differ between cognitively unimpaired older adults and older adults with MCI (β=0.16, r =.08, p=.261).

Table 3.

Results from Multilevel Models with Group Predicting Loneliness

Main Effect Model	B	SE	95% CI LB	95% CI UB	p
Intercept	1.42	0.09	1.25	1.59	< .001
Young (vs. CU)	0.56	0.13	0.30	0.82	< .001
MCI (vs. CU)	0.16	0.15	−0.12	0.43	.261
Pairwise-Comparison Model	B	SE	95% CI LB	95% CI UB	p
MCI (vs. Young)	−0.40	0.14	−0.69	−0.12	.006

Note. Young = Younger adults. CU = Cognitively unimpaired older adults. MCI = Older adults with mild cognitive impairment. SE = Standard error. CI LB = Confidence interval lower bound. CI UB = Confidence interval upper bound. In the initial model, we treated CU older adults as the reference level. To allow for a complete set of statistical comparisons between all three groups, we reran pairwise comparison models with younger adults as the reference group. Only the effects for MCI vs. Young group comparisons are reported from the pairwise comparison models.

Results from Research Question 2 are presented in Table 4. Partially consistent with Hypothesis 2.1, recent social interactions were associated with lower loneliness at the within-person level, but not at the between-person level. Participants reported small reductions in loneliness after a recent social interaction compared to occasions when they had not had a recent social interaction (β=−0.17, r=−.06, p<.001). Contrary to Hypothesis 2.2, the within-person association between recent social interaction and loneliness was stronger for younger adults relative to cognitively unimpaired older adults and older adults with MCI (see Figure 1). For young adults, recent social interactions had a small, stronger negative effect on loneliness compared to cognitively unimpaired older adults (β=−0.23, r=−.03, p=.003) and older adults with MCI (β=−0.27, r=−.03, p=.002). When examining the simple effects for each group, the negative association between social interaction and loneliness was statistically significant but attenuated for cognitively unimpaired older adults and was statistically non-significant for older adults with MCI. The between-person interaction between the person-mean frequency of recent social interactions and Group was not statistically significant. Results from Research Question 3 are presented in Table 5. Consistent with Hypothesis 3.1, loneliness levels were lower following recent face-to-face interactions compared to recent interactions that did not occur face-to-face. Compared to interactions that were not face-to-face, face-to-face interactions had a small negative effect on loneliness (β=−0.13, r=−.06, p=.001). Contrary to Hypothesis 3.2, group did not moderate the association between interaction method and loneliness.

Table 4.

Results from Multilevel Models with Social Interaction and Social Interaction by Group Interactions Predicting Loneliness

Main Effect Model	B	SE	95% CI LB	95% CI UB	p
Intercept	1.77	0.18	1.42	2.11	< .001
Recent Social Interaction (person-mean)	−0.01	0.25	−.50	.48	.959
Recent Social Interaction (occasion-specific)	−0.17	0.03	−.24	−.11	< .001
Between-person Interaction Model	B	SE	95% CI LB	95% CI UB	p
Intercept	1.67	0.23	1.21	2.13	< .001
Recent Social Interaction (person-mean)	−0.21	0.35	−0.90	0.48	.554
Recent Social Interaction (occasion-specific)	−0.17	0.03	−0.24	−0.11	< .001
Group (Young vs. CU)	1.11	0.49	0.15	2.07	.023
Group (MCI vs. CU)	0.12	0.39	−0.66	0.89	.768
Recent Social Interaction (person-mean)* Group (Young vs. CU)	−0.72	0.64	−1.99	0.55	.265
Recent Social Interaction (person-mean)* Group (MCI vs. CU)	0.09	0.59	−1.07	1.24	.883
Pairwise Comparison Model – Between-person	B	SE	95% CI LB	95% CI UB	p
Group (MCI vs. Young)	−0.99	0.53	−2.04	0.05	.062
Recent Social Interaction (person-mean)* Group (MCI vs. Young)	0.81	0.72	−0.60	2.22	.261
Within-person Interaction Model	B	SE	95% CI LB	95% CI UB	p
Intercept	1.71	0.18	1.35	2.06	< .001
Recent Social Interaction (person-mean)	−0.36	0.25	−0.85	0.13	.152
Recent Social Interaction (occasion-specific)	−0.12	0.05	−0.21	−0.02	.017
Group (Young vs. CU)	0.81	0.15	0.51	1.12	< .001
Group (MCI vs. CU)	0.13	0.15	−0.17	0.44	.387
Recent Social Interaction (occasion-specific)* Group (Young vs. CU)	−0.23	0.08	−0.38	−0.08	.003
Recent Social Interaction (occasion-specific)* Group (MCI vs. CU)	0.04	0.08	−0.11	0.20	.593
Pairwise Comparison Model – Within-person	B	SE	95% CI LB	95% CI UB	p
Group (MCI vs. Young)	−0.68	0.17	−1.01	−0.35	<.001
Recent Social Interaction (occasion-specific)* Group (MCI vs. Young)	0.27	0.09	0.10	0.44	.002

Note. Young = Younger adults. CU = Cognitively unimpaired older adults. MCI = Older adults with mild cognitive impairment. SE = Standard error. CI LB = Confidence interval lower bound. CI UB = Confidence interval upper bound. In initial models, we treated CU older adults as the reference level. To allow for a complete set of statistical comparisons between all three groups, we reran pairwise comparison models with younger adults as the reference group. Only the effects for MCI vs. Young group comparisons are reported from the pairwise comparison models. The main effect model reports fixed effects for person-mean and occasion-specific recent social interactions for loneliness, while the interaction models report the simple effects for the models with interaction terms.

Figure 1. Group differences in the within-person associations between social interactions and loneliness.

Note. Young = Younger adults. CU = Cognitively unimpaired older adults. MCI = Older adults with mild cognitive impairment. 0 = no recent social interaction, 1 = social interaction <31 minutes before survey prompt. The Y axis values represent scores on the 7-point Likert scale. Scores 2 standard deviations above and below the mean are depicted in the figure.

Table 5.

Results from Multilevel Models with Social Interaction Method and Social Interaction Method by Group Interactions Predicting Loneliness

Main Effect Model	B	SE	95% CI LB	95% CI UB	p
Intercept	1.90	0.17	1.56	2.24	< .001
Face-to-Face Interaction (person-mean)	−0.28	0.22	−0.71	0.15	.206
Face-to-Face Interaction (occasion-specific)	−0.13	0.04	−0.21	−0.05	.001
Between-person Interaction Model	B	SE	95% CI LB	95% CI UB	p
Intercept	1.47	0.26	0.96	1.98	< .001
Face-to-Face Interaction (person-mean)	−0.01	0.33	−0.66	0.64	.969
Face-to-Face Interaction (occasion-specific)	−0.13	0.04	−0.21	−0.05	.001
Group (Young vs. CU)	1.19	0.38	0.43	1.95	.002
Group (MCI vs. CU)	−0.03	0.43	−0.86	0.81	.953
Face-to-Face Interaction (person-mean)* Group (Young vs. CU)	−0.93	0.50	−1.91	0.05	.063
Face-to-Face Interaction (person-mean)* Group (MCI vs. CU)	0.30	0.52	−0.73	1.34	.561
Pairwise Comparison Model – Between-person	B	SE	95% CI LB	95% CI UB	p
Group (MCI vs. Young)	−1.21	0.44	−2.08	−0.35	.006
Face-to-Face Interaction (person-mean)* Group (MCI vs. Young)	1.23	0.55	0.15	2.32	.026
Within-person Interaction Model	B	SE	95% CI LB	95% CI UB	p
Intercept	1.60	0.19	1.24	1.97	< .001
Face-to-Face Interaction (person-mean)	−0.24	0.21	−0.66	0.18	.269
Face-to-Face Interaction (occasion-specific)	−0.10	0.07	−0.23	0.03	.122
Group (Young vs. CU)	0.62	0.15	0.32	0.92	< .001
Group (MCI vs. CU)	0.18	0.17	−0.15	0.51	.294
Face-to-Face Interaction (occasion-specific)* Group (Young vs. CU)	−0.11	0.09	−0.29	0.07	.239
Face-to-Face Interaction (occasion-specific)* Group (MCI vs. CU)	0.05	0.11	−0.16	0.27	.630
Pairwise Comparison Model – Within-person	B	SE	95% CI LB	95% CI UB	p
Group (MCI vs. Young)	−0.44	0.17	−0.78	−0.10	.012
Face-to-Face Interaction (occasion-specific)* Group (MCI vs. Young)	0.16	0.11	−0.05	0.37	.144

Note. Young = Younger adults. CU = Cognitively unimpaired older adults. MCI = Older adults with mild cognitive impairment. CI LB = Confidence interval lower bound. CI UB = Confidence interval upper bound. In initial models, we treated CU older adults as the reference level. To allow for a complete set of statistical comparisons between all three groups, we reran pairwise comparison models with younger adults as the reference group. Only the effects for MCI vs. Young group comparisons are reported from the pairwise comparison models. The main effect model reports fixed effects for person-mean and occasion-specific face-to-face social interactions for loneliness, while the interaction models report the simple effects for the models with interaction terms. These models were conducted in the subset of measurement occasions in which a recent social interaction occurred. Face-to-face interactions were compared to all other methods of interaction.

Results from Research Question 4 are presented in Table 6. Contrary to Hypothesis 4.1 and 4.2, participants’ relationship to their social interaction partner (i.e., close versus non-close) was not associated with loneliness at the within or between-person level, and there were no between-person level interactions with group. At the within-person level, there were significant pairwise interactions between occasion-specific interaction partner and group such that older adults with MCI were less lonely after interacting with a close partner compared to cognitively unimpaired older adults (β=−0.23, r=−.06, p = .030) and compared to younger adults (β=−0.20, r=−.03, p = .043) (see Figure 2).

Table 6.

Results from Multilevel Models with Relationship to One’s Social Interaction Partner and Relationship to One’s Social Interaction Partner by Group Interactions Predicting Loneliness

Main Effect Model	B	SE	95% CI LB	95% CI UB	p
Intercept	2.09	0.24	1.62	2.55	< .001
Close Other Interaction (person-mean)	−0.57	0.29	−1.14	0.003	.051
Close Other Interaction (occasion-specific)	−0.07	0.04	−0.14	0.003	.062
Between-person Interaction Model	B	SE	95% CI LB	95% CI UB	p
Intercept	1.50	0.44	0.64	2.36	< .001
Close Other Interaction (person-mean)	−0.09	0.52	−1.13	0.95	.865
Close Other Interaction (occasion-specific)	−0.07	0.04	−0.15	0.001	.052
Group (Young vs. CU)	0.70	0.55	−0.38	1.79	.204
Group (MCI vs. CU)	0.42	0.67	−0.91	1.74	.537
Close Other Interaction (person-mean)* Group (Young vs. CU)	−0.26	0.70	−1.63	1.11	.711
Close Other Interaction (person-mean)* Group (MCI vs. CU)	−0.31	0.79	−1.86	1.24	.695
Pairwise Comparison Model – Between-person	B	SE	95% CI LB	95% CI UB	p
Group (MCI vs. Young)	−0.29	0.61	−1.49	0.91	.638
Close Other Interaction (person-mean)* Group (MCI vs. Young)	−0.05	0.74	−1.51	1.41	.945
Within-person Interaction Model	B	SE	95% CI LB	95% CI UB	p
Intercept	1.59	0.26	1.08	2.10	< .001
Close Other Interaction (person-mean)	−0.27	0.30	−0.85	0.31	.364
Close Other Interaction (occasion-specific)	−0.02	0.06	−0.14	0.10	.739
Group (Young vs. CU)	0.52	0.16	0.22	0.83	< .001
Group (MCI vs. CU)	0.39	0.17	0.05	0.72	.024
Close Other Interaction (occasion-specific)* Group (Young vs. CU)	−0.02	0.08	−0.18	0.14	.802
Close Other Interaction (occasion-specific)* Group (MCI vs. CU)	−0.23	0.10	−0.43	−0.02	.030
Pairwise Comparison Model – Within-person	B	SE	95% CI LB	95% CI UB	p
Group (MCI vs. Young)	−0.13	0.18	−0.49	0.22	.455
Close Other Interaction (occasion-specific)* Group (MCI vs. Young)	−0.20	0.10	−0.40	−0.01	.043

Note. Young = Younger adults. CU = Cognitively unimpaired older adults. MCI = Older adults with mild cognitive impairment. SE = Standard error. CI LB = Confidence interval lower bound. CI UB = Confidence interval upper bound. In initial models, we treated CU older adults as the reference level. To allow for a complete set of statistical comparisons between all three groups, we reran pairwise comparison models with younger adults as the reference group. Only the effects for MCI vs. Young group comparisons are reported from the pairwise comparison models. The main effect model reports fixed effects for person-mean and occasion-specific close-other interactions for loneliness, while the interaction models report the simple effects for the models with interaction terms. These models were conducted in the subset of measurement occasions in which a recent social interaction occurred. Interactions that involved a close other were compared to all other interactions.

Figure 2. Group differences in the within-person associations between social interaction partner closeness and loneliness.

Note. Young = Younger adults. CU = Cognitively unimpaired older adults. MCI = Older adults with mild cognitive impairment. 0 = recent interaction with non-close partner 1 = recent interaction with close partner. The Y axis values represent scores on the 7-point Likert scale. Scores 2 standard deviations above and below the mean are depicted in the figure.

Robustness Checks

As a robustness check, we tested for group differences in gender, race, education, and physical health and found significant differences for gender and physical health (see Table 2). We re-ran our models controlling for these covariates, as well as and the number of prompts completed (this covariate was not pre-registered, for full table of preregistration deviations, see supplemental materials Table S1), and the pattern of results was unchanged. Results from these analyses can be found in the supplemental materials Table S2–Table S5. We also re-tested Research Question 2 with recent social interaction coded as 1 only for interactions that happened “at the time of the prompt.” The pattern of results was unchanged (see supplemental materials Table S6). Additionally, we re-ran all models omitting participants who borrowed a smart phone device, and the pattern of results remained consistent (see supplemental materials Table S7 for results from main effects models). Finally, we re-ran main effects models treating loneliness as an ordered outcome and using a probit link function, rather than treating it as a normally distributed continuous variable (Bürkner & Vuorre, 2019). The direction, significance, and magnitude of effects were consistent with the primary models.

Discussion

We examined loneliness and social interaction in daily life in an age and cognitively diverse sample. Our inclusion of older adults with MCI allowed us to examine differences between age groups as well as between cognitive status groups irrespective of age. Young adults reported the greatest levels of loneliness in daily life compared to cognitively unimpaired older adults and older adults with MCI. This finding supports other work that suggests young people (<30 years old) are the loneliest age group (Qualter et al., 2015) but it does not entirely support previous findings that loneliness follows a U-shaped curve throughout the lifespan, with loneliness decreasing after adolescence into young adulthood and middle age until peaking again in older age (Luhmann & Hawkley; Victor & Yang, 2012; Graham et al., 2024). Contrary to our expectations, there was no difference in reported loneliness between cognitively unimpaired older adults and older adults with MCI. This is inconsistent with past findings from global trait-level loneliness measures that demonstrate people with MCI are significantly lonelier than age-matched cognitively unimpaired counterparts (Yu, Lam, & Lee, 2018) and that greater loneliness is related to lower cognitive ability (Boss et al., 2015). One explanation for this discrepancy could be that we assessed loneliness multiple times in daily life rather than relying on a one-time global measure. Although trait and state level assessments tend to broadly measure the same construct, they differ in meaningful ways (Augustine & Larsen, 2012), especially in daily life (McMahon & Naragon-Gainey, 2020). Trait loneliness commonly measures the frequency with which a person feels lonely in general (e.g., UCLA Loneliness Scale; “How often do you feel you lack companionship?”) or the extent to which they agree with global statements about their social network (e.g., De Jong Gierveld Loneliness Scale; “There are plenty of people that I can lean on in case of trouble”) and is mostly stable throughout the lifespan (Mund, Lüdtke, & Neyer, 2019), whereas state loneliness measures the momentary intensity of loneliness and fluctuates across time (van Roekel et al., 2018; Halvorson & Kuczynski, 2024). It is possible that older adults endorse trait loneliness even if they are not actually experiencing state loneliness in their everyday lives. There are widespread societal prejudices directed toward older adults, and experiences of ageist discrimination and internalized age-stereotypes can lead to global reports of loneliness (Sutin et al., 2015; Levy et al., 2009).

Having had a recent social interaction was related to lower loneliness at the within but not between-person level. That is, individuals reported lower loneliness when they had a recent social interaction compared to when they had not, but individuals who tended to have more social interactions were not necessarily less lonely. This corresponds to findings from another experience sampling study, which found that, compared to no social interaction, having a social interaction predicted lower momentary loneliness hours later (Zhaoyang et al., 2022). Additionally, our results support findings from a daily diary study, which found that on days when people had more social interactions than usual, they reported feeling less lonely, but the overall quantity of social interactions was not related to loneliness at the between-person level of analysis (Kuczynski et al., 2022). The reduction in loneliness after a social interaction was strongest for younger adults compared to older adults, and it was not significant for older adults with MCI. This pattern may suggest that older adults with MCI are more likely to experience social asymmetry, that is, experiencing a mismatch between subjective loneliness and objective social isolation (Ward et al., 2021). Social asymmetry can be characterized by both experiencing less loneliness than one would expect relative to one’s social interactions (i.e., resilience) or more loneliness than one would expect relative to one’s social interactions (i.e., vulnerability). Thus, social asymmetry is not uniformly adverse, however, some evidence suggests a discrepancy between objective isolation and subjective loneliness may be predictive of longitudinal cognitive decline (McHugh et al., 2017).

As expected from previous research on trait loneliness among younger samples (Twenge Spitzberg, & Campbell, 2019; Costa, Patrão, & Machado, 2019), loneliness was lowest after face-to-face interactions compared to other forms of interaction. Surprisingly, this effect was not moderated by group, which suggests face-to-face interactions could be an important protection against loneliness across adulthood, even when experiencing accelerated cognitive decline. Finally, contrary to our expectations, loneliness did not generally differ depending on whether interaction partners were close versus non-close. However, there was support for our prediction that older adults with MCI were less lonely after interacting with close others compared to cognitively unimpaired older adults. This finding extends previous research (Zhaoyang et al., 2021) suggesting that interacting with close or familiar others is less cognitively demanding and thus preferable for older adults with MCI by providing preliminary evidence that interactions with close social others may be particularly important for reducing loneliness in the context of MCI. For older adults with MCI, loneliness was not generally reduced after social interactions but rather only after interactions with close others.

Strengths and Limitations

A particular strength of this study is the use of repeated, momentary measures to investigate loneliness and social interaction in daily life across age and cognitive status. The state of being lonely fluctuates during and across days (van Roekel et al., 2018; Tam & Chan, 2019; Buecker, Horstmann, & Luhmann, 2024), but loneliness is often operationalized as a static, trait-level measure (e.g., Victor & Yang, 2012; Doane & Thurston, 2014; Masi et al., 2011). Because loneliness is time-variant, studying the intra-individual (i.e., within-person) temporal dynamics of loneliness can provide insights into the construct, especially for understanding inter-individual (i.e., between-person) differences (i.e., age and cognitive differences) in intra-individual processes (e.g., association between social interactions and loneliness). Repeated, momentary measures are especially important for capturing the concordance between social interaction and loneliness as there is evidence that within-person designs are more sensitive than between-person designs in detecting synchrony (Mauss et al., 2005). However, to our knowledge, social asymmetry has not been tested in everyday life using daily or momentary measures.

Furthermore, there is evidence that older adults with and without MCI differ in characteristics of social interactions at the daily but not global level (Zhaoyang et al., 2021), which suggests that daily measures of loneliness and social interactions have particular pertinence in clarifying the role of cognition in the experience of loneliness and its association to social interaction. Additionally, as older adults with MCI were most likely to experience social asymmetry, future research could determine the discriminate power of social asymmetry for detecting MCI. Moreover, establishing the interaction-loneliness association could help identify points of entry for real-world prevention and intervention.

The use of experience sampling also contributes to the ecological validity of our findings because respondents reported on current emotion and social experience while engaged in naturalistic routines and contexts. Asking participants to report on current emotions contributes to measurement validity, as retrospective ratings of emotion may be biased due to episodic memory decay or belief-consistent reconstruction (Robinson & Clore, 2002). This is especially consequential for comparing emotion in an age and cognitively diverse sample, as these biases may be age-differentiated (Robinson & Clore, 2002), and those with MCI may suffer from greater recall bias (Zhaoyang et al., 2021). Current best practices for ecological momentary assessment, as supported by the literature (deVries, Baselmans, & Bartels, 2021), emphasize important advantages to using smart phone-based data collection. Although participants who borrowed smart phones had lower compliance on average, accommodating individuals who did not own a smart phone allowed us to include a more diverse range of participants, thus strengthening the generalizability of the study. Additionally, our shorter prompt response window relative to other studies could be related to lower compliance rates, however this method has the advantage of reducing memory biases, which is particularly important in a sample with cognitively diverse older adults.

In addition to these strengths, there are limitations to our study. Although smartphone-based experience sampling improves upon the accuracy and ecological validity of traditional methods, it is still influenced by recall and self-report biases. Future research could employ passive mobile sensing methods to gather more objective data on participants’ contexts and emotions (Springstein & English, 2024). Further, not all participants completed a survey every single time they were prompted, and our mixed effects models use list-wise deletion for occasion-level missingness. This method could impact our findings, as we cannot know what participants are doing when they did not respond. For example, it is possible that a non-response is more likely when participants are engaged in social interactions.

There is evidence for generational differences in loneliness (Hülür et al., 2016; van Tilburg, Aarsten, van der Pas, 2015; Suanet & van Tilburg, 2019), and although our use of momentary assessment allows for within-person inferences, our study cannot account for potential cohort effects. Future research with longer-scale cohort-sequential longitudinal measures is necessary to account for cohort or historical time effects. Age is confounded with generation and time of measurement in a majority of loneliness studies, and more research is needed to account for cohort and historical time differences. Future studies could also consider additional variables that vary with age (e.g., mental health) to examine potential mechanisms that explain the age differences we found. Further, one reason we may not have replicated previous findings (Victor & Yang, 2012; Luhmann & Hawkley, 2016) that indicate loneliness is elevated similarly in young adults and older adults is that our older adult sample does not include members of the so-called “oldest-old” (those 85 years or older). Members of this group have been distinguished from those in other phases of older adulthood and have links to well-being and loneliness that are distinct from older adults aged 65–84 (Suzman, Willis, & Manton, 1995; Luhmann & Hawkley, 2016). Because the oldest participants in our sample are 84, it is possible that our sample does not extend far enough into the lifespan to capture the entire trend. In addition, our sample limits age to an older adult group (70–84 years old) and a younger adult group (20–34 years old). Future research could compare momentary loneliness and social interactions in people spanning a broader age range, including middle-aged people and those who are “young-old” and “oldest-old.” Further, although our sample reflects the racial and ethnic diversity of St. Louis, where data collection took place, it may not accurately represent the general U.S. population, and the extent to which our findings can be generalized to other populations is limited.

Another limitation is our use of a proxy measure of closeness in assessing the association between interaction partner and loneliness, which could potentially explain why we did not observe an effect of interaction partner. We categorized partners as close vs. non-close based on the type of relationship to the participant, which is less sensitive than measuring the closeness to recent interaction partner directly. Additionally, the extent to which our conclusions can be applied to clinical populations in an intervention context is unknown due to our use of a community sample. Future research could explore loneliness in daily life in a clinical setting.

Conclusion

This study joins the growing body of literature examining age differences in momentary loneliness. Taken together, these findings advance the developmental literature by comparing loneliness of younger adults, cognitively unimpaired older adults, and older adults with MCI. Results from a nine-day experience sampling procedure suggest there are cross-sectional age differences in loneliness and its association to recent social interactions. These findings suggest that younger adults may be particularly vulnerable to experiencing loneliness in their daily lives, and that frequent face-to-face social interactions may serve as an important buffer against loneliness for younger adults. Additionally, differences in emotional responses to social interactions and partners between cognitively unimpaired older adults and older adults with MCI support research that emphasizes the role of cognition in emotion (Lazarus & Smith, 1988; Gross, 1999). Based on these findings, interventions aimed at reducing loneliness should increase opportunities for regular face-to-face social interactions. For older adults with MCI, interventions designed to maintain or improve close relationships may be particularly effective at preventing loneliness.

Public Significance Statement.

People across the lifespan experience loneliness, which is linked to negative physical and mental health outcomes. This study suggests that young adults are particularly at risk for experiencing loneliness in everyday life, and that frequent face-to-face interactions may protect against loneliness for this group. Additionally, results suggest that older adults may feel less lonely on average, but those with mild cognitive impairment may be especially likely to experience a discrepancy between objective isolation and subjective loneliness.