How common are loneliness and isolation?

Background

Loneliness and social isolation have emerged as significant public health concerns, and some have even described them as an ‘epidemic’ (Holt-Lunstad, 2017). However, despite clear evidence of the harms that loneliness and social isolation have on mental and physical health, there have been insufficient public health investments in social and community programs (Barreto et al., 2024). One reason for this might be that the scale of the problem is difficult to conceptualize.

Purpose

The purpose of this evidence brief is to describe the prevalence of loneliness and social isolation. In doing so, we recognize that loneliness and social isolation are diverse subjective experiences and thus it is difficult to differentiate between different types and severities thereof. Additionally, we acknowledge that estimating a true prevalence of any social factor can be difficult for a variety of reasons, including (a) geographic, temporal, or sub-population variations in the occurrence of loneliness; (b) cross-cultural or interpersonal differences in how loneliness is conceived; and (c) stigma and other sources of bias in the reporting of loneliness. As such, this review, rather than providing a definitive point-estimate for the prevalence of these conditions, aims to describe the range of likely prevalence estimates and the major factors contributing to heterogeneity in prevalence estimates.

Evidence from Existing Studies

Prevalence estimates of loneliness and social isolation vary widely according to a range of factors (Stegen et al., 2024). Summarizing across studies, time periods, and regions, somewhere between 5% and 50% (i.e., ~35%) of people likely experience moderate to severe levels of loneliness and between 5% and 35% (i.e., ~20%) experience social isolation (Teo et al., 2023; Chawla et al., 2021; Zhenrong et al., 2024; Lin, 2023; Lee et al., 2023; Cuesta-Lozano et al., 2020; Zhong et al., 2016, 2018; Routasalo et al., 2006; Saito et al., 2010; Simon et al., 2014; Cheng et al., 2015; O’Shea, 2021; Shimada et al., 2014; Losada et al., 2012; Kotwal et al., 2021; Rohr et al., 2022; Ran et al., 2024; Chokkanathan, 2020; Shimada et al., 2014). Of course, many of the estimates included in these ranges come from specific sub-populations with elevated risk for loneliness and social isolation (Ruan et al., 2023; Alexandra et al., 2018).

Among the populations at increased risk for loneliness and isolation include older adults, people living in long-term care facilities, and those with co-morbid psychiatric or physical health conditions (Jansson et al., 2020). For example, in some clinical samples, the prevalence of loneliness and social isolation are found to be as high as 60-90% (Galiatsatos et al., 2022; Gardiner et al., 2020; Susanty et al., 2022), whereas prevalence estimates of loneliness in non-clinical samples might be as low as 5-20% (Surkalim et al., 2022; Pengpid & Peltzer, 2021; Huang et al., 2021; Tomstad et al., 2017; Srivastava & Srivastava, 2023; Antunes, 2015; Na et al., 2022; Igami et al., 2023; Peng et al., 2022; Peltzer & Pengpid, 2019; Shevlin et al., 2013; Theeke, 2009; Beutel et al., 2017). Though of course, this is not universally true: some clinical samples report relatively low rates of loneliness as well and vice versa (Mullen et al., 2019; Hajek & Konig, 2022; Fortuna et al., 2023).

The magnitude of differences in experiences of loneliness and isolation across various sub-groups varies considerably. For example, Emerson et al. (2023) reports that the prevalence of loneliness among citizens in the United Kingdom was 46% among those with disabilities compared to 22% in those without disability; while Lampinen et al., (2022) report that in a representative sample of people, aged 85 or older, the prevalence of loneliness did not differ considerably between those with (50.9%) and without (46.0%) dementia. Such variation highlights the complexity of estimating the prevalence of loneliness and social isolation across sub-groups. Differences in loneliness may exist across any number of factors ranging from demographic and personality traits to situational and life course factors (Barjakova et al., 2023).

Furthermore, the prevalence estimates for loneliness and social isolation are highly contingent on how they are measured. Sometimes a specific cut-off score from a validated loneliness scale is used (e.g., scores of >6 on the three-item UCLA loneliness scale; Stickley & Ueda, 2022; Victor & Pikhartova, 2020). Other times, a single item is used to assess loneliness and participants are classified as lonely based on their response to that item (e.g., for frequency-based measures, it is not uncommon to see “never,” “rarely,” or “sometimes” classified as “not lonely” and “often,” “most”, or “all of the time” classified as “lonely”; Srivastava & Srivastava, 2023; Reinwarth et al., 2023). There are also considerable measurement challenges when trying to estimate the prevalence of social isolation. For example, living alone is sometimes used, but Shimada et al., (2014) demonstrated that social isolation (measured using the Lubben Social Network Scale) differs only slightly between adults living alone (31%) and those living with others (24.1%) – highlighting challenges in defining what constitutes “social isolation.” Notably, Stegen et al., (2024) and others report that loneliness prevalence tends to be higher when scales are used (compared to single-item questions; Victor et al., 2020). The same may also be true for measuring social isolation (Hawthorne et al., 2006; Zavalea et al., 2017). Similarly, some studies suggest that measurement choices may influence which individuals are classified as lonely or not, giving rise to potentially artificial differences in loneliness across groups (Nicolaisen & Thorsen, 2014).

In considering the various measures used to define loneliness, it is important to note that there is variation in prevalence estimates between “moderate” loneliness and isolation (30% [~25-40%]) and “severe” loneliness or isolation (~10% [5-15%]), with the former typically being much more prevalent than the latter (Victor et al., 2021 [2.5x]; Hajek et al., 2024 [1.4x]’ Stickley & Ueda, 2022 [2.5x]; Reinwarth et al., 2023 [6.8x]; Luo & Li, 2022 [2.4x]; Cudjoe et al., 2020 [6x]; Chawla et al., 2021 [3.3x]). Of course, definitions for moderate and severe loneliness vary across studies. As well, loneliness has multiple dimensions that may vary differently and affect different populations (e.g., emotional vs. social dimensions; Manoli et al., 2022; Fierloos et al., 2021; Valle et al., 2022; Diehl et al., 2018). Recently, researchers have also increasingly attended to differences between the “acute/episodic/transient” and “chronic” phenotypes of loneliness and isolation (Roddick & Chen, 2021; Hosozawa et al., 2022; Wolska & Creaven, 2023; Martin-Maria et al., 2020; Shiovitz-Ezra & Ayalon, 2010; Zhong et al., 2016). For example, Lim et al., (2023) report that the prevalence of episodic loneliness is about 1.6x greater compared to chronic loneliness (21% vs. 13%). Similarly, they found that episodic social isolation was about 3x more common compared to chronic isolation (13% vs. 4%). Along these lines, Martin-Maria et al., (2021) found that transient loneliness was 3.4x more common than chronic loneliness (23% vs. 6%). Generally speaking, most estimates of chronic or severe phenotypes are lower compared to episodic or moderate phenotypes (Sheftel et al., 2024; Amendola et al., 2023; Pengpid & Peltzer, 2023; Qi et al., 2023).

As well, there are differences in the estimated prevalence of loneliness and social isolation across time and space. Several studies have noted increasing prevalence of loneliness over time (Madsen et al., 2019; Dagnew & Dagne, 2019; Lin et al., 2024). For example, among Norwegian adolescents, Parlikar et al. (2023) estimate that loneliness doubled (5.9% to 10.2%) between 1995/97 and 2017/19. Similarly, Girault et al. (2023) showed that loneliness increased among adults between 2000 (13.2%) and 2021 (27.4%). However, other studies suggest that the prevalence of loneliness has been unchanged or decreased (Surkalim et al., 2023; Qualter et al., 2021) – particularly among populations with an already high burden of loneliness (Nyqvist et al., 2017; Aunsmo et al., 2023; Timmermans et al., 2019; Victor & Bowling, 2012). Geographically speaking, Stegen et al., (2024) report that cultural factors that vary regionally— including power distance, uncertainty avoidance, indulgence, and collectivism—are associated with higher loneliness. Similarly, Rapoliene & Aartsen (2021) report that low-trust societies (e.g., post-totalitarian countries) have greater loneliness. Other authors have also noted considerable geographic variation in levels of loneliness (Chawla et al., 2021; Surkalim et al., 2022; Schroyen et al., 2023), including at the sub-regional level (Menec et al., 2019).

Altogether, the existing evidence on the prevalence of loneliness demonstrates very high heterogeneity. Further, a wide variety of factors are known to influence the estimated prevalence and it is likely that studies may be influenced by multiples of these factors, making it difficult to isolate the direction of bias. Unfortunately, no agreed upon framework for loneliness measurement exists and there is therefore considerable unexplained heterogeneity and a wide variety of hypotheses could potentially explain differences in existing estimates (Hajek et al., 2023).

Analyses from The Canadian Alliance for Social Connection and Health

To characterize the prevalence of loneliness in Canada and highlight challenges in doing so, we conducted analyses of two datasets: The Canadian Social Connection Survey and The Canadian Social Survey.

In the Canadian Social Connection Survey, an online survey by the Canadian Alliance for Social Connection and Health, participants were asked “During the PAST WEEK, have you felt lonely…” “All of the time (e.g. 5-7 days)”, “Occasionally or a moderate amount of time (e.g. 3-4 days), “Some or a little of the time (e.g. 1-2 days)”, “Rarely (e.g. less than 1 day)” “None of the time (e.g., 0 days)”.

In the Canadian Social Survey, a population mail-based household survey by Statistics Canada, participants were asked, “How often do you feel lonely?”, and could choose from the options “Always”, “Often”, “Sometimes”, “Rarely”, and “Never.” Table 1 provides prevalence estimates from each survey, conducted in 2021, by using the top two levels of each item as an indicator for “loneliness.”

As shown above, the prevalence estimates are higher in the 2021 CSCS – particularly for older adults. Of course, these differences across surveys may be due to several factors: First, the differences could be due to sampling design, with online surveys resulting in recruitment of a higher proportion of people with loneliness. Similarly, the CSCS is not representative and tends to over-recruit some important demographic groups. Second, the categories of “always” or “often” may be more restrictive compared to “All of the time (e.g., 5-7 days)” or “Occasionally or a moderate amount of time (e.g., 3-4 days). For example, someone who completed the CSS might say they experience loneliness “often,” but when answering the CSCS they might indicate that they only experience loneliness 1 or 2 days a week (which would be classified as “not lonely.”) Given these multiple differences between the surveys, it is hard to say why the prevalence estimates differ. As such, we sought to further explore the distribution of responses across these two surveys using data from the 2023 CSCS and 2024 CSS (See Table 2).

As shown in the table, the prevalence of loneliness grew significantly in the CSCS (from 2021 to 2023), but not the CSS – suggesting that the pool of participants responding to the CSCS likely shifted towards recruiting a greater proportion of individuals who were lonely.

Figure 1 illustrates the shifting distribution in reported loneliness across 2021, 2022, and 2023.

We also show within-person changes in loneliness over time using data from the Canadian Social Connection Cohort (n = 255; See Figure 2).

These results highlight frequent transition within individuals across the different frequency levels of loneliness. These changes are underscored by calculating an Average Cohen Kappa coefficients for the data of 0.35, indicating only fair agreement across years. Of course, this might be expected given that the CSCS measures loneliness by asking participants to think about the number of days they felt lonely over the past seven days. Loneliness for many may vary from week to week, not to mention for different weeks across different years. As such, it is possible that recall period may be an important factor shaping response patterns to our one-item measure of loneliness.

Given potential challenges with the one-item measure, we also assessed the within-person variance in loneliness over time, by examining changes in DeJong Loneliness Scale Scores. The results showed a statistically significant decrease in loneliness across the years (β coefficient of -0.255, p < 0.0001). However, a relatively low marginal R² (1.16%) suggests that changes in loneliness scores over the years, as captured by the survey collection year, contribute only modestly to the overall variation in loneliness. Furthermore, we found that the Intraclass Correlation Coefficient (ICC) was approximately 0.74, indicating substantial agreement in ratings over time. Taken together, these analyses suggest that scales might be more stable than one-item measures in capturing a global experience of loneliness. Of course they are still imperfect.

Speaking further to the scale and one-item direct measures, we calculated average Bhattacharyya coefficient for the DeJong Loneliness Scale Scores across all levels of the direct measure of loneliness. The coefficient value was 0.71, indicating a high degree of overlap between the distributions (See Figure 3). This substantial overlap suggests that the DeJong Loneliness Scale scores are relatively consistent across different categories of the direct measure, suggesting that while the one-item direct measure of loneliness correlates with the DeJong Loneliness Scale Score (p < 0.0001), there is a fair degree of variation in loneliness within each of the levels of the direct measure. In fact, the direct item measure only explains 33.2% of the variance in the DeJong Loneliness Scale Score.

Given variations in the frequency and severity of loneliness, it is also important to recognize that experiences of loneliness can be either chronic or acute. As noted in our evidence review, a great many people experience loneliness only on a fleeting basis, others experience a constant state of lonely agony. Differentiating between these types of loneliness is difficult and there is no clear standard for doing so. In the CSCS, we asked a subset of participants whether they “continually experience loneliness or social isolation on an ongoing basis?” and 42.9% of respondents indicated they did. Figure 4 shows the distribution of DeJong Loneliness Scores for those who are chronically lonely and those who are not, highlighting, once again, considerable overlap across groups. Considering this, we also looked at the relationship between chronic loneliness and the one-item direct measure of loneliness. These analyses suggested that chronic loneliness was common among those who reported feeling lonely all of the time (93.8%), occasionally or a moderate amount of the time (76.0%), some or a little of the time (52.8%), but less common among those who reported feeling lonely only rarely (23.5%) or none of the time (4.3%). This indicates a clear dose response in the relationship between frequency and chronicity of loneliness, which is further supported by a McFadden’s R² value of 0.29 – indicating that approximately 29% of the variation in chronicity of loneliness can be explained by the past week’s experience of loneliness. This value is not greatly increased by the inclusion of the DeJong Loneliness scores – which further underscores the challenge of using measures such as these to characterize chronic loneliness.

As noted in the literature review, another important issue to consider is that there are different scales used to measure loneliness. We tested the relationship between two commonly used scales using data from the 2021 Canadian Social Connection Survey: The De Jong Emotional and Social Loneliness Scale and the 3-item UCLA Loneliness Scale. These analyses show that the correlation between the scales is weak. The De Jong Emotional Loneliness Subscale explains only 21.58% of the variance in UCLA Loneliness Scale Scores, while The De Jong Emotional Loneliness Subscale explains only 13.3% of variation in the same measure. Furthermore, Emotional Loneliness Subscale scores explains only 8.6% of the Social Loneliness Subscale scores. These variations underscore significant differences in the underlying constructs being measured across these scales within our survey. Even the overall De Jong Loneliness Scale Score explains only 27.1% of the variation in UCLA Loneliness Scale Scores.

Similarly, loneliness scores correlate only weakly with measures of social isolation. For example, the number of “close friends” a participant reports explains only 5.0% of the variation in De Jong Loneliness Scores and the number of hours someone spends socializing with others during the course of a week explains only 1% of the variation. This may be because participants report generally low levels of social isolation: only 7.5% said they had no close friends and a similar proportion indicated that they spent less than 5 hours in the past week socializing (7.2%). These analyses highlight the difficulty of conceptualizing loneliness and isolation and determining what specific measures should be used when trying to assess the prevalence of these conditions.

Taken together, all of the analyses above highlight the considerable challenges of measuring loneliness. Indeed, different survey methods, sample compositions, measurement tools, and the general difficulty of conceptualizing what you are measuring each make it difficult to understand the prevalence of loneliness and social isolation.

Furthermore, there is one more complicating factor: Loneliness differs across different sub-groups. To demonstrate this, we define loneliness using a cut off score of 3 or more on the DeJong Loneliness Scale. For reference, Table 3 shows the prevalence of loneliness across a pooled sample from 2021, 2022, and 2023.

Across age groups, there was a statistically significant difference in the prevalence of loneliness (φ_c = 0.22, p < 0.0001) with loneliness highest among younger individuals (See Figure 5).

For gender, there was a statistically significant difference in the prevalence of loneliness (φ_c = 0.10, p < 0.0001) with loneliness higher among men and non-binary people (See Figure 6). For ethnicity, there was a statistically significant difference in the prevalence of loneliness (φ_c = 0.22, p < 0.0001) with most ethno-racial groups having higher rates of loneliness that participants identifying as white (See Figure 7). For household income, there was a statistically significant difference in the prevalence of loneliness (φ_c = 0.13, p < 0.0001) with rates of loneliness decreasing with increasing household income (See Figure 8).

For relationship status, there was a statistically significant difference in the prevalence of loneliness (φ_c = 0.19, p < 0.0001) with people in a relationship having the lowest levels of loneliness (See Figure 9). For disability status, there was a statistically significant difference in the prevalence of loneliness (φ_c = 0.06, p < 0.0001) with people living with a disability having higher loneliness (See Figure 10).

Discussion

The evidence reviewed above indicates a wide range of prevalence estimates for loneliness and social isolation, influenced by various factors such as geographic location, population demographics, and measurement tools. Studies show that between 5% and 50% of people reasonably experience moderate to severe loneliness, and between 5% and 35% experience social isolation. Generally speaking, individuals experiencing severe and chronic forms of loneliness and isolation make up a relatively small proportion of those who experience loneliness. Specific sub-populations, such as older adults and those with co-morbid conditions, are at higher risk, with prevalence rates reaching as high as 60-90% in some samples. Wide heterogeneity in results is linked to variations in methodology.

Despite extensive research, several gaps and challenges remain. Future studies should focus on standardizing measurement tools, as there is a need for a consensus on the best practices for measuring loneliness and social isolation. This includes developing standardized scales that can be used across different studies to allow for better comparison and aggregation of data. More longitudinal research is required to understand the dynamics of loneliness and social isolation over time. This can help differentiate between transient and chronic loneliness and identify the long-term effects of these conditions. Additionally, more research is needed to understand the unique experiences of different sub-populations, such as people with disabilities, various age groups, and ethnic minorities. This includes identifying specific risk factors and protective factors for each group. There is also a critical need for research on effective interventions to reduce loneliness and social isolation. This includes evaluating the impact of social and community programs and developing new strategies tailored to different population groups. Finally, understanding how cultural factors influence the prevalence and experience of loneliness and social isolation can provide insights into global patterns and inform culturally sensitive interventions.

Regardless of the need for more research, it is clear from the prevalence estimates reviewed above that loneliness and social isolation are widespread. As such, they merit substantial investments to address – particularly given their significant impact on morbidity and mortality (Wang et al., 2023).

Conclusion

Based on the evidence reviewed above, we recommend the inclusion of appropriate measures for loneliness and social isolation in local, regional, national, and international surveillance frameworks. Such inclusion should be based on widely agreed upon measures and tools and leverage appropriate representative sampling designs. Such efforts will support the monitoring of loneliness and isolation allowing us to understand the scale of the problem and whether ongoing efforts to reduce loneliness and social isolation are making sufficient impact so as to bring rates of these social conditions down.