Would the Real Loneliness Please Stand Up? The Validity of Loneliness Scores and the Reliability of Single-Item Scores

1Humans are social beings, and they rely and depend on each other. Interpersonal relationships are among the most powerful sources for well-being and health. When interpersonal relationships are perceived as deficient regarding quantitative (e.g., network size, number of friends) and/or qualitative aspects (e.g., relationship closeness, trust, intimacy), individuals experience loneliness.

2The large body of literature on loneliness that has accumulated is very consistent in demonstrating the negative effects of loneliness on health-related outcomes (for reviews, see;;;;; ) and interpersonal outcomes. This consistency is to be highlighted given the multitude of loneliness measures that have been developed over the last decades, including unidimensional and multidimensional multi-item scales, various abbreviated versions thereof, and a variety of single-item measures. Although issues of validity have been investigated in prior research, these studies are limited in that they (a) have examined either convergent validity between item scores, but not their correlations with external variables (e.g.,;;;; ), (b) examined the nomological net of scores of only one or two measures at once, and, when doing so, (c) mostly only considered demographic variables such as age, gender, and education as external correlates of loneliness. Thus, it is unclear whether item scores of different loneliness measures for adults tap into different aspects of loneliness or capture a common core.

3In this study, we extend prior work in four ways. First, we simultaneously examined the convergent validity among and the nomological net of scores of the most widely used measures of loneliness in adulthood: the Rasch-Type Loneliness Scale (RTLS) and its facets emotional and social loneliness, the University of California Los Angeles Loneliness Scale (UCLA-LS) in its full and two abbreviated versions, and three single-item measures of loneliness. Second, we examined the convergence between scores of self-ratings and informant-ratings of loneliness to extend the assessment of the validity beyond self-reports. Third, in addition to demographic correlates of loneliness, we investigated the nomological net of scores of the various measures using a wide variety of constructs (e.g., Big Five personality traits, domain and life satisfaction, social support, depressiveness, and shyness) to pinpoint similarities and differences between item scores of the selected measures. Fourth, we examined the reliability of scores of three single-item measures to investigate the often-raised criticism that such measures lack reliability.

4To achieve these goals, we conducted three independent studies. Studies 1 and 2 were designed to investigate the validity of item scores and agreement between self-ratings and ratings by close others and to examine the nomological nets of scores of different loneliness measures. In Study 3, we examined the reliability of scores of three single-item measures of loneliness using a procedure developed by in a study extending over three measurement occasions, each 2 weeks apart.

5Given its subjective nature, loneliness has most often been assessed using self-reports. The various self-report measures that have been developed in recent decades primarily differ regarding three aspects: (a) the dimensionality of the instrument (unidimensional vs. multidimensional), (b) the number of items (i.e., single-item measures vs. multi-item scale), and (c) whether the instrument assesses loneliness in a direct or an indirect way.

6Measurement instruments to assess loneliness can be differentiated according to whether they were developed based on the conceptualization of loneliness as a unidimensional construct or as a construct with multiple interrelated facets. Instruments such as the UCLA-LS have been designed to assess global loneliness. This notion has been challenged by several studies showing the UCLA-LS scores to be multidimensional and to be accounted for by two, three, or even five factors. By contrast, other studies support the notion of the UCLA-LS scores being essentially unidimensional. Despite this controversy, we decided to use the UCLA-LS scores as unidimensional in the present work for two main reasons. First, there is no consensual multidimensional solution of UCLA-LS item scores in the literature, and even studies reporting the same number of underlying factors differ markedly in which items are allocated to which factor and how those factors are interpreted. Second, in most research contexts, scores on the UCLA-LS seem to be treated as unidimensional. Specifically, several recent meta-analyses refrained from considering specific aspects of loneliness because there were not enough studies using the UCLA-LS reporting on those facets.

7Other scales such as the RTLS have been designed to capture multiple aspects of loneliness such as emotional and social loneliness. Multidimensional approaches offer the opportunity to measure loneliness in a more nuanced fashion and to better understand differential predictors and consequences of those facets. However, when taking the RTLS as an example, surprisingly few studies seem to build on its multidimensionality (for meta-analyses, see; ). This might be due to relatively high correlations between scores of the two facets of the RTLS, possibly indicating that the two facets capture actually the same aspects of loneliness. In this study, we will examine both global loneliness as measured by scores of the RTLS and its facets emotional and social loneliness.

8Single-item measures of loneliness have been used frequently and have been attested substantial content validity. However, their usage has been discouraged for several reasons. For example, the validity of scores of single-item measures has been questioned due to possible socially desirable responding and lack of standardization. That is, there is a plethora of single-item measures of loneliness that differ in wording and response format so that it is not fully clear whether scores of all these measures tap into the same construct. Studies comparing scores of single-item with multi-item measures have found a lower prevalence of loneliness when single items are used (;, but see, for a contrasting finding). Moreover, it seems that the individuals who are classified as “lonely” based on cut-off scores on multi-item scales are not the same individuals who are classified as “lonely” based on a cut-off score applied to single-item measures. Thus, it might be questioned whether scores of single items capture the complexity of the construct in a way scores of multi-item scales do. Apart from prevalence rates, however, the results obtained with scores of single-item measures converge very well with those obtained with scores of multi-item scales regarding demographic aspects, aspects of mean-level and rank-order stability, personality correlates, and health outcomes, including early mortality.

9It has also been argued that scores of single items are unreliable and capture more noise than substantial interindividual differences. In line with this reasoning, a study employing STARTS models found that scores of single-item measures capture more (potentially unreliable) state variance (≈50% state variance) than scores of longer scales such as a six-item version of the RTLS (≈25% state variance; ). However, it has been shown for scores of single-item measures of subjective well-being that state variance can contain a substantial amount of reliable variance. Similarly, demonstrated high reliability of scores of a single-item measure of self-esteem by applying a procedure introduced by. In this approach, the reliability of scores of a single-item measure is estimated based on its pattern of autocorrelation across three measurement points. In this study, we will adopt this approach to estimate the reliability of scores of three single-item measures of loneliness.

10In direct measures of loneliness, the respondents are confronted with the target construct insofar as words such as “lonely” are contained in the questions (e.g., “I am lonely”; “How often have you felt very lonely?”). Indirect measures, by contrast, avoid such references to loneliness. One item from the UCLA-LS, for example, asks participants to indicate “How often do you feel left out?,” and an item from the RTLS asks participants to rate to what extent they agree with the statement “I experience a general sense of emptiness” (indicative of emotional loneliness).

11The reasoning for using indirect measures is to disguise the researchers’ interest in loneliness, thereby avoiding the activation of negative stereotypes and circumventing socially desirable responding. Previous studies have shown that gender differences in the prevalence of loneliness can be observed when scores of direct measures are used, with men reporting lower levels of loneliness on average than women. Those gender differences disappear or even reverse when scores of indirect measures are used. It should be noted that direct measures of loneliness are often single items, which, necessarily, are unidimensional. By contrast, most multi-item measures are indirect measures of loneliness (for an exception, see ), and some of them are multidimensional.

12Previous studies comparing measures of loneliness for adults usually included no more than two measures at once and primarily considered their association with demographic variables. Regarding aspects of validity, we extend prior research by (a) considering scores of multiple measures, and their facets, of loneliness for adults, (b) investigating correlations among self-reports and the agreement between self-ratings and informant-ratings on item scores of those measures, and (c) considering a broad range of demographic and psychological variables to investigate similarities and differences in the nomological nets of scores of those measures.

13Although several studies have investigated the psychometric properties of scores of multiple loneliness measures for children and adolescents, studies on scores of loneliness measures for adults have only compared scores of two measures at once, with a particular focus on the UCLA-LS. However, examining the convergent validity of item scores of multiple measures is important to ensure that scores of all those measures capture interindividual differences in the same construct.

14The correlation between self-reports and informant-reports is an important aspect of construct validity. If scores of an instrument merely capture noise instead of reliable interindividual differences, the correlation with informant-ratings would be low. Thus, self-informant agreement is indicative of the substance of the measured construct. Using a nine-item version of the UCLA-LS, found self-informant agreements for partnered individuals of and, when the informants were friends, parents, and partners, respectively. For individuals without a partner, the self-informant agreement with friends amounted to and to with parents. Using the 20-item UCLA-LS, reported a correlation of between self-ratings and an aggregated informant-rating of up to three friends in a study with 118 self-(aggregated) informant pairs. Finally, also using the 20-item UCLA-LS, reported a correlation of between self-ratings and ratings by a friend, roommate, or parent in a study with 74 self-informant pairs. Studies investigating the self-informant agreement in adult samples for scores of other measures than observed using the UCLA-LS are lacking to date.

15As another facet of self-informant agreement, it is also possible to examine mean-level differences in self-ratings and informant-ratings of loneliness. While the correlation between self-ratings and informant-ratings pertains to the detectability of interindividual differences, comparing mean levels provides an evaluation of the exact agreement between self-reports and informant-reports. Previous studies on mean-level differences in loneliness measured using the UCLA-LS have found that parents and friends, but not partners, tend to underestimate the degree of target’s loneliness. No studies have examined mean-level differences between self-ratings and informant-ratings in scores of measures other than the UCLA-LS so far.

16The construct validity of the scores of measures investigated in the present work is further examined using a comprehensive nomological net. Previous studies comparing the nomological nets of scores of loneliness measures for adults were restricted to a few, mostly demographic aspects. In the present work, we investigated broad nomological nets of scores of the selected measures including demographic aspects, personality characteristics, domain-specific and life satisfaction, and aspects of social networks. Large differences in the nomological nets across item scores would indicate that researchers and practitioners need to choose their measures carefully to make sure that the obtained item scores measure the intended aspects.

17In this article, we present a series of three studies. In Study 1, we investigated the validity (convergence, self-other agreement, and nomological nets) of scores of six measures of loneliness in a sample of young adults. In Study 2, we examined the validity of scores of four partly different measures of loneliness than those used in Study 1. Furthermore, data in Study 2 were collected as part of a larger project on partner relationships; accordingly, informants in Study 2 were romantic partners. In Study 3, we investigated the reliability of scores of three single-item measures of loneliness by designing a short-term prospective study with three measurement occasions each 2 weeks apart. The single studies were not preregistered. Analysis scripts and data used in the analyses are accessible at https://osf.io/7gsfw/.

18The study was conducted in May 2018 as a cross-sectional online study using the formr survey framework. The study was announced on several university mailing lists and on dedicated websites for people interested in psychological studies. The study protocol was approved by the Institutional Review Board of the Friedrich-Schiller-Universität Jena, Germany (FSV 18/03). A total of 697 German-speaking individuals participated in the self-rating part of the study. Among those, 72.17% were female. The majority of the participants were students (73.39%), 15.88% of the participants were employed or self-employed, 2.43% were enrolled in nonacademic education (e.g., vocational learning, traineeship), and 1% were fulfilling a civil service; 7.30% of the participants were unemployed, unable to work because of ill-health, or not working for other reasons (e.g., retired, parental leave). On average, participants were 26.06 years (SD = 9.84, Mdn = 23) old, ranging from 18 to 99 years. After completion of the survey, participants received personalized feedback on their personality and well-being, and could sign up for a voucher lottery.

19Very early in the survey, participants were asked to provide contact details of up to six individuals who could provide ratings of the targets’ personality and well-being. Those contacts were then sent an email and asked to fill in a brief questionnaire. In this way, we obtained 282 informant-ratings on 160 targets. On average, each of the 160 participants was rated by 1.68 informants (SD = 1.03). Most of the targets (60.3%) were rated by one, 21.8% were rated by two, 9.9% by three, 6.2% by four, 1.2% by five, and 0.6% by six informants, respectively. On average, the informants were 28.25 years old (SD = 13.58, Mdn = 22), ranging from 18 to 70 years. Among the informants, 52% were friends of the targets, 23% were kin (e.g., parents, siblings), 18% were the target’s romantic partner, 3% were acquaintances, 3% were colleagues, and 1% indicated to be in another relationship with the target. Given the uneven distribution of types of informants, we aggregated the ratings of all informants in the analyses.

20Informants were asked to indicate how well they knew the target using a scale ranging from 0 (not well) to 10 (very well). On average, informants stated to know the targets very well (8.43, SD = 1.54, Mdn = 9). Upon completion of the survey, informants were offered the option to sign up for a voucher lottery. Informants were also invited to take part in the self-rating. In this way, it was possible that self-raters also served as informants for other targets. Furthermore, it is possible that some individuals provided informant-ratings for more than one person. The primary data used for the analysis are available at https://osf.io/7gsfw/.

21Details on internal consistency, sample items, response formats, descriptive statistics, and zero-order correlations are provided in Supplemental Tables S1 to S3.

22The loneliness measures were presented in the self-rating and informant-rating parts of the study.

23We used the 11-item version of the RTLS to assess global loneliness and the facets emotional and social loneliness. Internal consistency (coefficient ) was high for self-ratings and informant-ratings. The scores of RTLS Emotional and RTLS Social were highly correlated in self-ratings and informant-ratings.

24We used the German version of the 20-item UCLA-LS. Internal consistency was high for self-ratings and informant-ratings.

25We employed a three-item version of the UCLA-LS that has been developed for use in large panel studies. Internal consistency was satisfactory in self-ratings and informant-ratings.

26As a direct single-item measure of loneliness, we asked the participants to what extent the statement “I feel lonely” applies to them.

27As an indirect single-item measure of loneliness, we asked participants to indicate their agreement with the statement “I feel alone.”

28Adapting the SI Direct, we also asked participants about the frequency of loneliness (i.e., “How often do you feel lonely?”).

29All correlates were assessed only in the self-rating part of the study. We examined correlates from the domains of (a) demography, (b) personality, (c) satisfaction, and (d) network characteristics.

30Participants were asked to indicate their age, gender, educational status, and whether they currently have a partner or not.

31The Big Five personality traits were assessed using a 15-item version of the Big Five Inventory. Self-esteem was assessed using a three-item version adapted from the Rosenberg Self-Esteem Scale that has been used previously in a large panel study. Depressiveness was measured using the 10-item State-Trait Depression Scale. The explicit affiliation motive was assessed using four items from the Unified Motive Scales. We also assessed need satisfaction of the affiliation motive by two items from the Basic Need Satisfaction and Frustration Scale. A product interaction between both variables was calculated to create an index of frustration of the affiliation motive. Social desirability was assessed using the 17-item Social Desirability Scale. Shyness and sociability were assessed using the Shyness and Sociability Scales for adults.

32Participants were asked to indicate their overall satisfaction with (a) life, (b) education, (c) leisure, (d) friends and social contacts, (e) family, and (f) partner relationship using items taken from larger panel studies.

33Using single-item measures, participants were asked to indicate the number of (a) close friends, (b) friends on Facebook, (c) persons with whom the participant would be ready to discuss personal or occupational problems (i.e., emotional support), (d) persons who could be contacted for practical help (e.g., when relocating; i.e., instrumental support), and (e) persons the participant would ask for advice. Furthermore, participants were asked to indicate the frequency of (f) contact with their closest friend(s) and (g) joint activities with their friends.

34Convergent validity, operationalized as the correlation between scores of the six loneliness measures, was assessed separately for self-ratings and informant-ratings. Furthermore, for scores of each measure, we examined the correlation between self-ratings and informant-ratings. Nomological nets of the item scores were examined by calculating the correlations with the external variables (e.g., demography, personality).

35To investigate whether scores of the selected loneliness measures differ in their associations with external variables, we conducted pairwise model comparisons. In the first step of this approach, we entered all item scores and allowed unrestricted correlations between the scores of the loneliness measures and the external variables, resulting in a saturated model. In the next step, we constrained the correlation between scores of two loneliness measures (e.g., RTLS Total and UCLA-LS20Items) and a correlate (e.g., shyness) to be equal. We then compared the fit of this constrained model with the unconstrained model via a χ2-difference test with 1 degree of freedom. A significant decrease in model fit in this setup indicated that scores of the two loneliness measures differed significantly in their association with the external variable. Due to the large number of pairwise comparisons, we only consider differences significant when the p-value of the χ2-difference test was lower than.001. To facilitate the interpretation of the findings, we computed the average absolute correlation for scores of each loneliness measure and each domain (i.e., demography, personality, satisfaction, network characteristics).

36To estimate differences in the correlation patterns between self-ratings and informant-ratings as well as between Study 1 and Study 2, we calculated mean absolute deviations between correlation matrices. Furthermore, we compared whether the two correlation matrices (e.g., convergent validity in Study 1 vs. convergent validity in Study 2) are equal to each other. To this end, we used a procedure described by. Specifically, noted that the sum of squared z-transformed correlation coefficients follows a χ2-distribution under the assumption that the coefficients in the two matrices are equal. For all comparisons between Study 1 and Study 2, we used disattenuated coefficients as input. This was done because the measures in Study 2 were shorter and less internally consistent. As a consequence, the correlations in Study 2 might be smaller than those found in Study 1, although they might follow a very similar pattern. Mean-level differences between self-ratings and informant-ratings were examined using paired-samples t-tests and by calculating a standardized mean difference to quantify the effect size of the differences.

37To examine the similarity of correlation profiles, we calculated the double-entry intraclass correlation, which is similar to a one-way random-effects model for exact agreement. Using, each element of the compared profiles is entered twice, but in reverse order. For instance, when comparing the correlation between RTLS Emotional, UCLA-LS3Items, and SI Direct across Study 1 and Study 2, the correlations observed in Study 1 are first entered as the first number, whereas the coefficients found in Study 2 are entered second. Then, the coefficients observed in Study 2 are entered first and the coefficients observed in Study 1 constitute the second entry. The correlation between those three measures across studies, thus, is based on four entries instead of two. is then calculated as the correlation between those four data points. In this way, captures differences in both elevation and shape of the compared profiles. In terms of interpretation, can range between −1.0 and +1.0, with high positive values indicating a strong similarity between profiles, high negative values indicating a strong dissimilarity between profiles, and an around 0 indicating no association between profiles. In all analyses, all variables were treated as manifest variables. All syntax files used in the study are available at https://osf.io/7gsfw/.

38Before turning to the key analyses, we conducted some preliminary analyses. First, we examined whether loneliness would be associated with the number of informant-ratings. Across all measures, loneliness was correlated with the number of informant-reports such that individuals with higher scores on loneliness were rated by fewer individuals. On average, this correlation amounted to, ranging from for the RTLS Total to for the SI Direct frequency. Second, we determined the consensus among the informants regarding the target’s loneliness and examined the reliability of the informant-ratings. We calculated ICC(1) and ICC(1, k) to determine the reliability of single informant-ratings and the ratings of all informants together, respectively. ICC(1) amounted to 0.53 for the RTLS Total, 0.55 for the RTLS Emotional, 0.19 for the RTLS Social, 0.53 for the UCLA20Items, 0.46 for the UCLA3Items, 0.54 for the SI Direct, 0.47 for the SI Indirect, and 0.40 for the SI Direct frequency, respectively. ICC(1, k) amounted to 0.87 for the RTLS Total, 0.88 for the RTLS Emotional, 0.58 for the RTLS Social, 0.87 for the UCLA20Items, 0.83 for the UCLA3Items, 0.88 for the SI Direct, 0.84 for the SI Indirect, and 0.80 for the SI Direct frequency, respectively. Thus, for all measures except RTLS Social, the absolute agreement between raters was high and even single informants provided a reliable estimate of the target’s loneliness.

39The intercorrelation between scores of the different measures of loneliness is displayed in Table 1 separately for the self-ratings and aggregated informant-ratings. Overall, the correlations were high for self-ratings and informant-ratings. Furthermore, the average difference between the two correlation matrices was small and the difference between the matrices was not statistically significant. Finally, the profile similarity between the coefficients in the two matrices was high.

40Within the self-ratings, the correlations between scores of RTLS Social and the three single items were lower than other associations and varied between and. The scores of the UCLA-LS20Items were more strongly correlated with the scores of the RTLS Total than the UCLA-LS3Items. Similarly, the scores of the UCLA-LS20Items were more strongly correlated with the scores of the RTLS Social than the UCLA-LS3Items. Among the informant-ratings, the scores of the SI Direct frequency were least strongly associated with scores of all other measures of loneliness (.34 ≤ r ≤.51).

41Taken together, the findings converge with prior research showing high intercorrelations between scores of different measures of loneliness for adults. Furthermore, we extended prior research by simultaneously examining the convergent validity of scores of multi-item and single-item instruments, and by demonstrating that the high convergence between item scores also extends to informant-ratings. Thus, we conclude that all the instruments included in Study 1 are appropriate to assess between-person differences in loneliness from an internal (i.e., self-ratings) and an external (i.e., informant-ratings) perspective.

42Table 2 displays the correlations between self-ratings and aggregated informant-ratings. Self-informant agreement on scores of the same scale (shown in the diagonal of Table 2) varied across instruments (.49 ≤ r ≤.61). The highest self-informant agreement was found for scores of the RTLS Total and the UCLA-LS20Items; the lowest agreement was observed for scores of the SI Direct. Across all item scores, the self-informant agreement amounted to. This average correlation converges with other studies in which self-informant agreement in loneliness was investigated. Paired-samples t-tests indicated no statistically significant differences between self-ratings and informant-ratings obtained with either measurement instrument (.32 ≤ p ≤.89) and only minor effect sizes of differences between self-ratings and informant-ratings ranging between for scores of the RTLS Social and 0.07 for scores of the RTLS Emotional. The only exception were scores of the UCLA-LS3Items, for which self-ratings were significantly higher than the informant-ratings.

43To our knowledge, this is the first study that investigated the self-informant agreement on scores of multiple measures of loneliness. The results show that the overall convergence between self-ratings and informant-ratings is high within and across measures. In fact, the amount of self-informant agreement found in the present and previous studies is similar to self-informant agreement in broader personality characteristics such as the Big Five traits (for a meta-analysis, see ). Furthermore, the mean differences between self-ratings and informant-ratings are, overall, negligible, indicating that the aggregated informant-ratings provide a relatively accurate description of the target’s loneliness (for similar findings regarding broader personality characteristics, see;; ).

44Table 3 displays the correlations between scores of the different loneliness measures and a wide variety of correlates pertaining to demography, personality, satisfaction, and aspects of social networks. All correlations were compared pairwise via a model comparison with 1 degree of freedom. We will not go into detail on these model comparisons but instead highlight the overall pattern of results. To facilitate the inter-pretation, we computed scale-wise average absolute correlations for each block of correlates (i.e., demography, personality, satisfaction, network characteristics). Table 4 shows scale-wise profile correlations derived from the correlations in Table 3.

45Item scores of all measures were only modestly related to demographic aspects. Average absolute correlations ranged from for scores of the UCLA-LS3Items to for scores of the UCLA-LS20Items and the RTLS Social.

46The average absolute correlations ranged from for scores of the three single-item measures to for scores of the UCLA-LS20Items. These differences held also when computing for disattenuated correlations (RTLS Total =.41, RTLS Emotional =.39, RTLS Social =.39, UCLA-LS20Items=.45, UCLA-LS3Items =.41, SI Direct =.29, SI Indirect =.27, and SI Direct frequency =.29—for the single items, we used the reliability estimated in Study 3 as correction), accounting for the possibility that differences in occurred due to differences in the reliability of the item scores rather than the true association with the correlates. These differences were probably due to substantially lower correlations between scores of the single items and personality aspects such as openness, agreeableness, conscientiousness, sociability, and need frustration. Furthermore, the correlations suggested that scores of the UCLA-LS20Items were more strongly related to aspects associated with extraversion (i.e., sociability, shyness, affiliation motive) than scores of other instruments.

47The average absolute correlations between loneliness scores and domains of satisfaction ranged from for scores of the SI Direct frequency to for scores of the RTLS Total and the UCLA-LS20Items. Scores of the RTLS Total, RTLS Emotional, and UCLA-LS20Items stood out in this regard with average absolute correlations between and, whereas all other item scores varied on a somewhat lower level with.38 ≤ ≤.40 The largest differences between item scores emerged for the association with satisfaction with friends and social contacts. Scores of single-item measures were less strongly associated with this aspect (.43 ≤ ≤.47) than scores of the multi-item instruments (.55 ≤ ≤.66).

48Across all item scores, loneliness was associated with smaller friendship and support networks. All scores were uncorrelated with the self-reported number of friends on Facebook. The average absolute correlations ranged from for scores of the SI Direct frequency to for scores of the RTLS Social and the UCLA-LS20Items. Thus, as with the personality correlates, scores of the UCLA-LS20Items seem to tap into aspects related to sociability and support. The average absolute correlations with network characteristics were particularly low for the single items ( for scores of the SI Direct frequency, SI Direct, and SI Indirect, respectively) and were somewhat higher for scores of the UCLA-LS3Items and the RTLS Emotional. Except for the number of friends providing instrumental support and advice, scores of the single items were markedly less strongly correlated with network characteristics than scores of the other instruments; these differences were particularly strong with regard to the frequency of contact with friends. In this case, scores of the single items were not significantly correlated with this variable, whereas scores of the multi-item scales were.

49The correlation between the profiles of the loneliness scores and the correlates examined in Study 1 are displayed in Table 4. On average, the profiles of the item scores were very similar to each other, indicating that their nomological nets overlap to a large extent. Nevertheless, it might be noted that the profile correlation between scores of the single items and scores of the RTLS Total (.87 ≤ ICCDE ≤.89), RTLS Social (.82 ≤ ICCDE ≤.87) and UCLA-LS20Items (.80 ≤ ICCDE ≤.84) was relatively low. By contrast, the profile correlations were substantially higher with the RTLS Emotional and the UCLA-LS3Items (.92 ≤ ICCDE ≤.93). Similarly, the profile of the UCLA-LS3Items was less strongly correlated with the profile of the RTLS Social and the UCLA-LS20Items than with the RTLS Emotional. By contrast, and resembling the findings presented in Table 3, the profile of the UCLA-LS20Items was closely related to the RTLS Social.

50Overall, the results of Study 1 show that (a) all scores of measures of loneliness are highly correlated in the self-reports and informant-reports, (b) self-informant agreement was high for scores of all measures, and (c) scores of all measures have a similar nomological net. These findings suggest that scores of all measures tap into the same underlying latent construct and that there are no strong qualitative differences between the measures. It should be noted, though, that the nomological net of the scores obtained using single-item measures and—to a lesser extent—scores of the UCLA-LS3Items overlapped less strongly with related phenomena pertaining to personality (e.g., extraversion, sociability) and social networks (e.g., contact frequency). It could be argued that scores of these measures provide purer assessments of loneliness. However, it could also be argued that these lower correlations are due to the lower reliability of scores of these measures. Indeed, the UCLA-LS3Items had a lower internal consistency in the present study than the UCLA-LS20Items and the RTLS Total.

51The data used in Study 2 were taken from a larger online study on personality and partner relationships (1 citation removed for masked review). The study protocol was approved by the Institutional Review Board of the Friedrich Schiller University Jena (FSV 18/47). The larger study combines features of longitudinal and diary methods in that couples are asked every 3 months about aspects of their own and their partner’s personality, evaluations of the partner relationship, and aspects related to partner perception and communication patterns. In addition, participants are invited in February and March each year to participate in a diary period extending over 28 days in total, separated by a 3-week break after the first 14 days. Participants were recruited via advertisements on social media platforms, several university mailing lists, and online forums visited by people interested in research participation.

52In this study, we use data from the first five longitudinal waves. These waves were conducted in January 2020, April 2020, July 2020, October 2020, and January 2021, respectively. We included data of couples when they participated for the first time. That is, for the purpose of this study, we did not build on the longitudinal nature of the study, but on its dyadic nature and the large sample size. Accordingly, when a couple participated more than once, we only used data from the first time both partners provided data. This first-time participation could have been anytime during the five waves. In this way, we included data from 1,216 individuals (608 mixed-sex couples). Of those 608 couples, 136 entered the study in January 2020, 48 new couples entered in April 2020, 141 new couples entered in July 2020, 219 new couples entered in October 2020, and 64 new couples entered in January 2021.

53Despite couples entering the study at different measurement occasions, analyses of variance indicated no mean-level differences across waves for scores of the UCLA-LS8Items,, UCLA-LS3Items,, or the SI Direct, F(3,753) = 1.87, p =.13. There was a difference, however, for the RTLS Total (. Post hoc Tukey tests revealed that the mean of the RTLS Total was significantly higher in October 2020 compared with April 2020 (∆ = 0.26, p =.013). All other pairwise comparisons, however, were not statistically significant (.075 ≤ p ≤.999). This difference in the RTLS Total is attributable to differences between waves in the RTLS Social. Tukey post hoc tests indicated that social loneliness was higher in October 2020 than in January 2020 (∆ = 0.18, p =.036) and April 2020 (∆ = 0.27, p =.030). All other pairwise comparisons were not statistically significant (.299 ≤ p ≤.998). No difference between waves was observed for RTLS Emotional,.

54The same pattern of results was observed when comparing data collected before the COVID-19 pandemic (i.e., Wave 1, January 2020, N = 272) with data collected during the pandemic (i.e., from Wave 2 conducted in April 2020 onward; N = 944). By using independent-samples Welch tests, we found no differences in the levels of loneliness before and during the pandemic for the UCLA-LS20Items,, the UCLA-LS3Items,, or the SI Direct, t(560.37) = −0.61, p =.541, Hedges"s g = −0.04. For the RTLS Total, however, there was a stati-stically significant difference in the item scores, t(429.27) = −2.00, p =.046, Hedges’s g = −0.14. As before, this difference can be attributed to scores of the RTLS Social being higher during the pandemic, t(420.87) = –2.18,, whereas no difference emerged for RTLS Emotional, t(426.03) = –1.24, p =.216,. In sum, the few observed differences were small in size and inconsistent across loneliness measures. Thus, we conclude that there are no strong differences in the mean levels of loneliness before and during the pandemic in the present sample (for similar findings, see;;; for a review, see ). Accordingly, we decided to proceed with the analyses using the pooled data set.

55On average, participants were 26.69 years old and had been engaged in their current relationship for 40.85 months on average, with a range from 2 weeks to 44 years. The majority of couples lived in a noninstitutionalized relationship (81.25%), and 11.35% were married. The remaining participants indicated other relationship forms (e.g., in registered partnership) or preferred not to answer the question. A minority of participants (5.83%) reported to have children. Of those, 81.25% reported to have one child, 11.35% reported to have two children, and 7.40% reported to have three or more children. As highest educational degree, 0.08% of participants reported to have finished primary school, 0.99% reported to have achieved a secondary school diploma, 6.71% have achieved a high school diploma, 6.13% have achieved an entrance qualification for a University of Applied Sciences (Fachabitur), 36.87% have earned a University entrance qualification (Abitur), 47.97% have earned a diploma from a University or a University of Applied Sciences, and 1.24% have completed a PhD. The remaining 0.01% of participants indicated “other” degrees or preferred not to answer this question. Almost half of the participants (48%) were students at the time of study participation. A subset of the data without any personal information (age, gender, etc.) is available from https://osf.io/7gsfw/.

56Details on internal consistency, sample items, response formats, descriptive statistics, and zero-order correlations are provided in Supplemental Tables S4 to S6. Differences between samples of Study 1 and Study 2 regarding their average levels of loneliness are discussed in Section 3 of the Supplement, including Supplemental Tables S7 and S8.

57For all loneliness measures, we obtained a self-rating and a partner-rating. That is, each participant rated and was rated by their partner on each loneliness measure.

58In Study 2, we used the six-item version of the RTLS, which also allows to differentiate between emotional and social loneliness. The correlation between scores of the two facets was less strong than in Study 1 for both the self-rating and the partner-rating. Internal consistency was high for scores of the full scale and its facets in the self-rating as well as in the partner-rating.

59We used the German version of the eight-item UCLA-LS. Internal consistency was high for self-rating and partner-rating.

60As in Study 1, we employed a three-item version of the UCLA-LS as proposed by. Internal consistency was satisfactory in the self-rating and the informant-rating.

61As a direct single-item measure of loneliness, we used the same measure as employed in Study 1. Self-ratings of this item were collected in January 2020, April 2020, July 2020, and January 2021, but not in October 2020. Partner-ratings were collected at all measurement occasions.

62As in Study 1, we used variables from several domains to examine the nomological net of scores of the loneliness measures. Specifically, we used correlates from the domains of (a) demography, (b) personality, and (c) satisfaction.

63Participants were asked to indicate their age, gender, educational status, and their relationship duration.

64The Big Five were assessed using a 15-item version of the Big Five Inventory Version 2. Self-esteem was assessed using the 10-item Rosenberg Self-Esteem Scale. Depressiveness was measured using the five negatively worded items from the State-Trait Depression Scale. The explicit affiliation motive was assessed using two items from the Unified Motive Scales. Shyness and sociability were assessed using the Shyness and Sociability Scales for adults. Sociability was only assessed at the first three measurement waves (January, April, and July 2020, respectively).

65Participants were asked to indicate their overall satisfaction with (a) life, (b) education, (c) leisure, (d) friends and social contacts, and (e) family using single items. Relationship satisfaction was assessed with the seven-item Relationship Assessment Scale.

66The intercorrelation between scores of the loneliness measures used in Study 2 is displayed in Table 5 separately for self-ratings (below the diagonal) and partner-ratings (above the diagonal). For self-ratings, intercorrelations were high and comparable to Study 1. The average correlation for the partner-ratings was lower than the average correlation in the self-ratings of Study 2 and the informant-ratings of Study 1. This relatively lower convergence seems to be due to the partner-rated SI Direct. For this item, the correlation with other measures was weak, ranging from with partner-rated RTLS Social to with partner-ratings on the UCLA-LS8Items. Without the SI Direct, the average correlation in the partner-ratings increased to —a value which is comparable to self-ratings in Study 2 and informant-ratings in Study 1.

67The average difference between the self-ratings and partner-ratings in Study 2 was =.19. Without the SI Direct, the two correlation matrices were nearly identical ( =.03). The same pattern emerges when testing whether the two correlation matrices are equal. Including the SI Direct, the two matrices were significantly different. By contrast, there was no statistically significant difference without SI Direct. The profile correlation between self-ratings and partner-ratings amounted to for scores of all measures and increased to when SI Direct was left out.

68Finally, we compared the average differences and profile correlations between Study 1 and Study 2 separately for the self-ratings and informant-ratings for the measures included in both studies (i.e., and facets,,, SI Direct). The absolute difference in the self-ratings between Study 1 and Study 2 amounted to =.16. Both with and without SI Direct, the difference in the correlation matrices was statistically significant. For the informant-ratings, amounted to.31 when SI Direct was included and to.22 when SI Direct was left out. The difference in the correlation matrices was again statistically significant with and without SI Direct. Regarding the profile correlations between Study 1 and Study 2, amounted to.86 for the self-ratings, indicating a very high similarity of the profiles across studies despite the fact that we used different measures in Study 2. For the informant-ratings, the profile similarity amounted to and was even higher when the SI Direct was excluded.

69Taken together, the results regarding the convergent validity of scores of different loneliness measures obtained in Study 2 converged very well with the results observed in Study 1. Across the two studies, the correlation profiles were very similar and the differences in the average correlations were small to moderate. Unlike Study 1, the SI Direct turned out to be somewhat problematic in the partner-rating of Study 2 in the sense that it showed only modest correlations with scores of other partner-rated loneliness measures. One possible explanation for this finding is that the SI Direct forced individuals to explicitly rate their partners as “lonely,” which might threaten the self-concept of the rating partner. In Study 1, such effects might have been removed through the aggregation of the informant-ratings across multiple raters.

70The agreement between self-ratings and partner-ratings are displayed in Table 6. The diagonal shows the agreement on the same instrument, ranging between for scores of the SI Direct and for scores of the. Apart from the SI Direct, however, the correlations in the diagonal are close to each other with.42 ≤ r ≤.55. With the exception of, the agreement between self-rated SI Direct and partner-rated measures was almost twice as high (.42 ≤ r ≤.44) as the correlation with partner-rated SI Direct. Across all item scores, self-informant agreement amounted to with and to without the SI Direct, respectively. These coefficients were lower than the average correlation observed in Study 1. It should be noted, though, that we employed shorter measures of the UCLA-LS and RTLS in Study 2 that evinced a lower internal consistency. When correcting for attenuation, the self-informant agreement for scores of the, for example, increased to, which is very similar to a corrected self-informant agreement for scores of the in Study 1.

71In contrast to Study 1, paired-samples t-tests indicated consistent statistically significant differences between self-ratings and informant-ratings (all ps <.001). These differences were moderate in size, ranging between for the and for the. For all measures, the scores of the informant-reports were larger than the self-report scores, indicating that individuals described their partners as lonelier than they described themselves. This finding is in contrast to a previous study reporting no difference between self-rated and partner-rated loneliness using a nine-item version of the UCLA-LS in a sample of partnered students. A meta-analysis on differences in self-reports and informant-reports of broader personality characteristics has found that closer informants typically see targets in a more favorable light (e.g., targets report higher scores on neuroticism than informants; ).

72Comparing self-informant agreement between Study 1 and Study 2, we found a weak profile correlation of. However, the average difference between the matrices was only small ( =.11). The difference between the matrices was statistically significant both with and without SI Direct. Thus, small shifts in the profiles might have led to low profile similarity, whereas the actual coefficients do not differ much between Study 1 and Study 2.

73The correlations between the different loneliness item scores and external correlates are displayed in Table 7. As in Study 1, we focus on the average absolute correlations when discussing the results. Table 7 additionally contains the results of pairwise model tests with 1 degree of freedom for each scale and each correlate. The similarity between the correlation profiles of the different loneliness scores with external correlates is displayed in Table 8.

74The absolute average correlations varied only slightly between the loneliness measures with (.04 ≤ ≤.10. In many cases, the correlations between the demographic aspects and the loneliness scores were not statistically significant.

75Compared to Study 1, the absolute average correlations in Study 2 were substantially lower, ranging from for to for. When calculating disattenuated correlations, the values were very close to those observed in Study 1 ( =.39 vs..41, =.39 vs..39, =.32 vs..39, =.45 vs..45, =.37 vs..41, and SI Direct =.37 vs..27), indicating that the lower correlations found in Study 2 might be attributed to the lower internal consistency of the abbreviated measures.

76As in Study 1, scores of the showed the highest associations with aspects related to sociability (i.e., extraversion, sociability, shyness, affiliation) and, thus, were similar in this regard to the. By contrast, scores of other scales such as the or the SI Direct were less strongly or even not significantly associated with aspects of sociability.

77On average, there were no strong differences between scores of the loneliness measures with.27 ≤ ≤.35. The strongest differences between the measures emerged for satisfaction with friends and social contacts. In this domain, scores of the,, and showed quite strong associations, whereas the association with scores of the remaining measures were lower.

78The profile similarities for Study 2 are displayed in Table 8. The profiles of the different loneliness scores varied between.77 ≤ ICCDE ≤.99. The highest similarity was found between the SI Direct and the and the lowest between and. Similarly, the profile correlations between and SI Direct (.84) and (.79) were lower than the profile similarities between other scores.

79Comparing the profiles of scores of the measures that were used in both studies (i.e., RTLS and facets,,, SI Direct) regarding the correlates that were used in both studies yielded a high profile similarity. Specifically, the profile similarity between the nomological nets in Study 1 and Study 2 was high for scores of the,,,,, and SI Direct. Thus, we conclude that the nomological nets of scores of these popular measures are largely invariant across samples and studies (for similar findings regarding the Big Five, see ).

80The results of Study 2 compared well with the results obtained in Study 1. We found a similar pattern of convergent validity among scores of different loneliness measures in the self-ratings and, with the exception of SI Direct, also in the partner-ratings. Moreover, we found a very high similarity between the profiles of different loneliness item scores and external correlates across the two studies. Although Study 2 cannot be considered a direct replication of Study 1, the similarity between the results indicates that the findings are robust. Thus, Study 2 further corroborates that scores of all included measures are valid and reliable measures of loneliness in adults.

81In Studies 1 and 2, we found that the nomological nets of scores of single-item measures of loneliness overlap less with phenomena related to extraversion and sociability. This finding might point either to purer nomological nets or to a lack of validity of the scores of single items—because of low reliability, their correlation with other constructs might be reduced.

82Two previous studies have used STARTS models on single-item scores of loneliness. In this way, estimated a reliability of.595 for scores of an SI Direct, whereas the results reported by imply a reliability of scores of different single-item measures in the range between and. An alternative to STARTS models is a procedure proposed by. In this approach, the reliability of single-item scores is estimated based on their autocorrelations over three evenly spaced measurement occasions. Specifically, assumed that the stability of a construct depends on the retest-reliability of the scale scores used to measure the construct and the decline in true stability of the construct over time. Using the observed autocorrelations of scale scores across three measurement occasions, it is possible to derive a reliability estimate that is free of transient or other effects of temporal change. In Study 3, we adopted this approach and examined the retest-reliability of the scores of three single-item measures of loneliness.

83In 2019, we recruited a sample of 411 individuals via the service provider Prolific (www.prolific.com; for an overview, see ). As the study only involved self-ratings, no ethical approval was necessary according to German regulations; however, all participants provided informed consent before entering the study. On average, participants were 30.06 years old, ranging from 18 to 63 years. Almost half of the sample (47.7%) were women. Regarding employment status, 46% and 25% of the sample reported to work full-time or part-time, respectively; 7.2% reported not to be in paid work (e.g., homemaker, disabled, retired). The remaining ≈12% of individuals reported to be currently job seeking, about to start a new job, or “other” status. Most participants were of German (76%) or Austrian (14%) nationality. The Prolific score (possible range from 0 to 100) of the participants, resembling the quality and validity of their responses based on previous studies they participated in, was high (M = 99.57,,, range from 87 to 100).

84Two weeks after the first assessment (T1), all participants were reinvited to participate a second time (T2). Of the initial 411 individuals, 278 (67%) participated again. Another 2 weeks later (i.e., 4 weeks after the initial assessment), all participants were invited to the third survey (T3) and 276 individuals (67% compared to T1) provided the necessary data. For estimating the reliability of scores of the single items, we conducted an analysis based on all available data and, separately, for the panel sample of 276 individuals who participated in all three assessment points. The data used for the key analyses are available from https://osf.io/7gsfw/.

85As measures of loneliness, we included the three single items also included in Study 1. That is, we included the SI Direct, SI Indirect, and the SI. Descriptive information for all measures is displayed in Supplemental Table S9.

86The three single-item measures of loneliness were interspersed with other measures. We included the three-item measure of self-esteem developed for large panel studies that was also used in the present Study 1. Furthermore, participants rated the frequency of being tense, joyful, sad, and happy. We will consider scores of self-esteem in the following analyses, but the emotion ratings were merely used as fillers and will not be considered further.

87The procedure described by, Formula 9) requires three autocorrelations among the items measuring loneliness: the correlation between scores of T1 and T2, T1 and T3, and T2 and T3. Based on these three correlations, reliability is estimated as. We applied this formula separately to scores of all three single-item measures. Moreover, we applied the formula to the point estimate of the observed correlations and additionally to the lower and upper bound of the 95% confidence interval of the observed correlations.

88Before estimating the reliability of scores of the single-item measures, we investigated their correlation with scores of the brief self-esteem measure. The correlation between self-esteem scores and scores of SI Direct (rT1 = –.55, rT2 = –.51, rT3 = –.51), SI Indirect, and converged well with the coefficients observed in the present Studies 1 and 2.

89Table 9 displays the reliability of the scores of single-item measures of loneliness for the full and panel sample. The highest reliability was estimated for scores of the SI Indirect, followed by scores of the. The reliability of scores of the SI Direct was somewhat lower, but still above typically used cut-offs of.70. Thus, scores of these three single-item measures have a reliability that can at least be considered adequate. The often-cited notion that scores of single items are unreliable appears questionable against the backdrop of the results of this analysis. Furthermore, the reliability estimated in this study is similar to reliability estimates for scores of single items of self-esteem ( using the Heise approach; ) and life satisfaction (average across four independent studies =.72 using bivariate STARTS models; ). Thus, it seems possible to reliably capture loneliness in adults using scores of single-item measures. This finding is particularly important when financial or time-related constraints do not allow researchers to include multi-item measures of loneliness.

90In the present studies, we examined the psychometric features of scores of several measures of loneliness—the perception of one’s relationships as deficient regarding the quantitative and/or qualitative aspects. We included versions of the RTLS, UCLA-LS, and different single-item measures because those are among the most popular and widely used measures of loneliness in adulthood. At the same time, these measures represent broader operationalizations of loneliness: The RTLS represents an indirect multidimensional approach, whereas the UCLA-LS represents an indirect unidimensional approach to loneliness. The included single-item measures are necessarily unidimensional, but represent both direct and indirect approaches. This study largely extends previous research by (a) focusing on several loneliness measures at once, (b) using self-ratings and informant-ratings to assess loneliness, (c) examining and comparing broad nomological nets, and (d) paying attention to the validity and reliability of different versions of single-item measures. The results presented in this article offer many vantage points for discussion, including (a) the similarities and differences between measures of loneliness, (b) the debate around the use of single items, and (c) recommendations for future research.

91In Study 1 and Study 2, we investigated the convergent validity of scores of different measures of loneliness, self-informant agreement, and the nomological nets of these item scores. The results indicate that scores of all measures included in this study are useful measures of loneliness. All item scores show high convergent validity in the self-reports and informant-reports. Furthermore, the nomological nets of scores of all measures were very similar to each other as indicated by high profile correlations. However, it should be noted that consistent across Studies 1 and 2, scores of the UCLA-LS, in its 20-item as well as in its 8-item version, were more strongly related to affiliative aspects like sociability and extraversion than scores of other measures, even after controlling for unreliability in other item scores. Similarly, scores of the were more strongly related to sociability, extraversion (Studies 1 and 2), and network characteristics such as the frequency of joint activities with friends (Study 1). By contrast, scores of the were more strongly related to variables such as neuroticism and self-esteem. Thus, the findings for scores of the RTLS support the idea that, despite the high intercorrelation between scores of its facets, it allows to differentiate between emotional and social aspects of loneliness. The correlation between scores of the RTLS and UCLA-LS with personality and network characteristics demonstrates that the indirect approach taken by both instruments might come along with a less clear separation of loneliness from related constructs. As we will discuss below, this approach might have consequences for specific research questions.

92Across all measures and both Study 1 and Study 2, depressiveness and life satisfaction emerged as key correlates of loneliness. These findings are in line with prior research suggesting that loneliness might be a risk factor for depression or co-occur due to a common origin. The association with lower life satisfaction further suggests that loneliness might have very broad influences on how individuals see and approach the world. More specifically, although loneliness is conceptualized to emerge from perceived deficiencies in one’s social relationships, its consequences seem to reach far beyond social relationships and to cloud individual’s perspective on life more generally.

93Consistent with prior research, self-informant agreement within and across item scores was generally high and similar to self-informant agreement in broader personality characteristics. This finding indicates that loneliness can be perceived by others. From a psychometric perspective, this supports the notion that scores of the measures selected in this article capture reliable interindividual differences that are, to a large extent, validated by informants. From an applied perspective, this finding might also indicate that close others can accurately evaluate a person’s loneliness. Thus, campaigns to tackle loneliness might also benefit from equipping close others with the knowledge of how to help and support lonely individuals in their environment in addition to encouraging lonely people to seek help proactively.

94It should be noted that the informant-ratings in the present and previous studies were obtained from close others such as partners, friends, or parents. Loneliness might be less well observable for less close informants, such as teachers or casual acquaintances. Unlike extraversion, for example, which can be judged accurately even after a few seconds of interaction, loneliness is a rather internal state that might not manifest readily in specific observable behaviors (see also ). Instead, loneliness might express itself in the content of interactions, for example, by a specific emotional tone present in repeated conversations, or specific language patterns (; for findings on neuroticism, see ) or interaction patterns.

95In Study 1 and Study 2, we demonstrated that scores of different versions of single-item measures correlate highly with each other and with scores of multi-item scales. Furthermore, we demonstrated that the nomological nets of scores of single items are consistent across different formulations and with the nomological nets of the established RTLS and UCLA-LS item scores. In Study 3, we also demonstrated that scores of single items are reliable. Thus, the results of the present studies indicate that scores of single-item measures are not deficient or invalid measures of loneliness per se. Consequently, we argue that single-item measures of loneliness have their place as robust and trustworthy measures in research on loneliness.

96The practice of measuring loneliness via single items has been controversial. However, single-item measures are highly prevalent in research on loneliness. This controversy seems to resemble similar discussions surrounding single-item measures of self-esteem and life satisfaction. Specifically results in those research fields have suggested that both self-esteem and life satisfaction are so highly schematized constructs that single-item scores are reliable and valid. The results of this study seem to suggest that loneliness is highly schematized as well and can be measured using single items in contexts where financial or time-related constraints prohibit the use of multi-item scales. However, it should also be noted that the use of single-item and other ultra-short measures of loneliness such as the comes with the disadvantage of losing bandwidth. That is, short measures are very focused measures that cannot capture a construct in all its complexity. Thus, although feasible in many research contexts, single-item and ultra-short measures of loneliness should not be used in clinical practice or in settings where a very nuanced perspective is required.

97Broadly speaking, when it comes to measuring the core of loneliness, we proclaim the well-known Dodo Verdict that “everybody has won, so all shall have prizes.” Put differently, we believe that the phenomenon of loneliness in adulthood can be measured robustly, validly, and reliably by scores of any of the measures investigated in this article.

98A more nuanced view is necessary within specific research contexts. For example, in studies designed to investigate the behavioral correlates and/or the observability of loneliness, researchers need to take care that any observed effects are not due to other factors. For example, most studies that have examined self-informant agreement in loneliness have used scores of the UCLA-LS. Given the results presented in the current studies, the high self-informant agreement found in these studies might be attributable, to some extent, to informants having actually observed the target’s extraversion or sociability, which are well observable even for raters who are unacquainted with the target. As another example, the similarity in loneliness between romantic partners has been found to be moderate using the, but to be near zero in a study using a single item. By contrast, the effects of loneliness on later relationship satisfaction were similar in both studies.

99The general recommendation that can be taken from the present findings is that loneliness measures should be selected with a particular regard to the research question. Some of the existing measures (e.g., RTLS, UCLA-LS) might need to be adjusted statistically to rule out confounding with overlapping constructs (e.g., extraversion). Furthermore, some measures might not work well when informant-ratings are to be collected, as was the case for the SI Direct in the present Study 2; in such cases, indirect measures might be better suited. Ideally, researchers should include multiple measures whenever possible to cross-check the results. The nomological nets presented in this study might help to spot potential sources of disagreement between measures in existing and future research.

100The findings presented in this article should be evaluated against the backdrop of some limitations. First, recruitment of the participants was conducted via the internet and, thus, was prone to self-selection. Overall, the samples were composed of mostly young and highly educated adults. Hence, it is unclear whether the results can be generalized to samples from other age groups, other educational or cultural backgrounds, or with chronic health conditions.

101Second, we investigated a selection of measurement instruments that are frequently used in contemporary research on loneliness in adulthood. However, other loneliness measures are also available. Some of those measures have been developed for samples of children or adolescents, and some have been developed with a broad multidimensional and relationship-specific perspective. We hope that the nomological nets presented in this study might serve as a foundation for further investigating the construct validity of scores of those measures.

102Third, although we investigated comprehensive nomological nets of scores of different loneliness measures, several important correlates were not considered. These include, first and foremost, health behaviors, but also other phenomena such as externalizing problems (e.g., hostility;; ) or social media usage.

103Across three studies, we investigated aspects of the validity and reliability of scores of different loneliness measures—including several versions of established scales such as the RTLA and the UCLA-LS, but also different versions of single-item measures. Overall, the results indicate that scores of all measures are highly correlated in self-ratings as well as informant-ratings. Furthermore, we found substantial self-informant agreement on scores of all measures. The nomological nets of the included item scores were shown to be consistent within and across studies. Finally, we have shown that scores of single-item measures of loneliness possess adequate reliability. Taken together, the findings of the present studies provide rich evidence for the validity of scores of popular loneliness measures, thereby providing a nuanced picture of the nomological net of each measure’s item scores. Thus, this study might guide further research in selecting an appropriate loneliness measure and at the same time provide a solid foundation for future research on the validity of scores of other loneliness measures.