A Temporal Perspective on Loneliness and Trustworthiness Judgments in Everyday Life

Participants

The study protocol and analysis plan were preregistered on the Open Science Framework (https://doi.org/10.17605/OSF.IO/R46A2). The final study deviates from the preregistered plan for reasons detailed in the “Deviations from Preregistration” section below. Data were collected from 127 individuals (age > 18, German fluency) between December 2024 and February 2025. We included only those participants in our analyses who indicated in more than one-third (> 9) of the 28 possible prompts that they had a social interaction since the previous prompt. This threshold was set to ensure sufficient within-person variability in social interaction experiences, a prerequisite for modeling how momentary evaluations of such interactions relate to loneliness and interpersonal trustworthiness beliefs. An independent samples t-test revealed no significant difference in trait loneliness (UCLA-Loneliness Scale, Russell et al., 1980) between excluded and included participants, t(8.81) = 0.17, p = .87, 95% CI [−0.51, 0.60]. Including participants with very few or no reported interactions would have limited the interpretability of within-person effects and may have introduced noise rather than informative data. Four participants reported no social interactions at all and nine participants had fewer than nine prompts following a social interaction, which resulted in a final data set of 114 individuals (mean age = 22.59, SD = 5.37 years; 98 women, 15 men, one did not indicate their gender). These participants responded on average to 25.8/28 prompts (i.e., 92% compliance; SD = 3.04, range = 14–28) and 19 prompts on average (SD = 4.37) targeted social interactions (n = 2169 total social prompts). We preregistered 100 participants and calculated the power using the ema.powercurvefunction from the R-package EMATools (Kleiman, 2017). With our compliance rate, 114 participants, two surveys per day over 14 days and an intra-class correlation (ICC) of.19–.34 (see results), our sample was well powered to detect medium-to-large effects targeted at the main effects as hypothesized in H1 and H2. Participants earned €0.50 for each prompt answered, a bonus of €1 for answering all prompts, and an additional €7 for completing the intake survey, which included demographic information, trait questionnaires, and an explanation of the study procedure, resulting in a maximum total of €22 (or the equivalent in course credits for psychology students). The study was carried out by the recommendations of the “Ethical Guidelines” of “The Association of German Professional Psychologists” (“Berufsethische Richtlinien, Berufsverband Deutscher Psychologinnen und Psychologen”). All participants provided written informed consent by the Declaration of Helsinki before participating in the studies.

Procedure

First, participants completed a series of online questionnaires, collecting detailed data on socio-demographic characteristics, as well as several trait measures, including trait loneliness (University of California Los Angeles [UCLA] loneliness scale; Russell et al., 1980), generalized trust (inclusive general trust scale [IGTS]; Yamagishi et al., 2015). Following this, participants started a 14-day EMA phase. During this period, participants received two daily prompts at randomized times between 11 a.m. and 1 p.m. and between 5 p.m. and 7 p.m. These prompts were administered via the m-Path smartphone application (m-path.io), directing them to complete a short questionnaire tailored to their recent experiences. In case a social interaction has occurred since the previous prompt, participants were asked to evaluate the interaction, reflecting on its quality, their self-esteem during the encounter (not relevant for the current analyses), and their momentary feelings of loneliness and interpersonal trustworthiness beliefs. If no interaction has taken place, they completed an alternative questionnaire of equal length to avoid response biases. This version inquired about the activity they were engaged in, whether their solitude was voluntary and enjoyable, and their current general affect, and feelings of loneliness and trustworthiness beliefs.

Measures

EMA Survey

In the EMA survey, the following items were rated on a 7-point scale (1) not at all to (7) very much: For the assessment of state loneliness, participants indicated how much they agree with the statement “I feel lonely” (van Roekel et al., 2014). The items on trust and the evaluation of social interaction were answered in relation to the most recent social interaction since the last prompt. Interpersonal trust was assessed by the question “To what extent did you trust the person(s)?” (Weiss et al., 2021). The evaluation of the most recent social interaction was measured with the following items: “I feel threatened by this person/people”, “I feel judged by this person/people”, “I feel the other person(s) dislike me”, “I feel accepted by this company”, “I feel excluded” (van Roekel et al., 2014), “I feel valued” (Kirtley et al., 2021), “I feel comfortable in this company” (van Roekel et al., 2014). For exploratory reasons, we also assessed momentary self-esteem using the items “I feel good about myself” (Unger, 2023) and “I feel self-confident” (Kirtley et al., 2021). As contextual variables, we assessed whether the last social interaction took place at the moment, within the last 30 minutes or whether there was no social interaction since the last prompt. We asked participants about the context in which the interaction took place (e.g., private, work or study-related), whether it was face-to-face or virtual, whether they interacted with one or multiple individuals, whether their interaction partner(s) were romantic partner, colleague(s), family, friend(s), or stranger(s), and which gender they attributed to the other person.

Analyses

All analyses were conducted in R version 4.5.2. Predictors were decomposed into within-person (WP; person-mean centred) and between-person (BP; grand-mean centred) components following the recommendations of Bolger and Laurenceau (2013). Within-person deviations were computed prior to lagging, such that lagged predictors reflect deviations from each person’s own mean at the preceding prompt. All models were estimated using maximum likelihood to allow likelihood-ratio comparisons.

Deviation from preregistration

Our manuscript deviates from the preregistered plan in three main ways, primarily stemming from a necessary shift in the study’s primary focus following recruitment challenges. First, the primary research question was changed. The study was initially designed to test hypotheses concerning the relationship between socio-political disaffection, loneliness, and interpersonal trustworthiness beliefs. Following recruitment challenges that yielded insufficient variance in the socio-political variables, we shifted the analytical focus to disentangle the state and trait effects of loneliness on social appraisals and interpersonal trust beliefs. Second, the primary statistical models were revised. The preregistered plan specified concurrent multilevel models (outcome_t ~ predictor_t). Consistent with the refined theoretical focus on loneliness dynamics, we revised the primary analyses to use a one-prompt lagged structure (outcome_t ~ predictor_t–1), which better supports prospective inference. Third, several preregistered variables were handled differently. As a consequence of the shifted focus, analyses pertaining to the collected socio-political variables, satisfaction with democracy (Neu et al., 2023), populism (Neu et al., 2023; Vehrkamp & Merkel, 2020), perceived political influence (Langenkamp & Stepanova, 2024), party affiliation, and self-assessed social status (MacArthur Scale; Adler et al., 2000) were removed from the main manuscript. Furthermore, one additional baseline measure was not included in the current study: trait self-esteem (Rosenberg Self-Esteem Scale; von Collani & Herzberg, 2003), which was part of a separate research project. For full transparency, all preregistered analyses for the current research project are provided in the Extended Data (https://doi.org/10.17605/OSF.IO/VP2EJ).

Preliminary analyses

To examine the factor structure of the seven social evaluation items, we conducted a parallel analysis to determine the optimal number of factors, followed by an exploratory factor analysis (EFA; minres extraction, varimax rotation). We then compared a one-factor and a two-factor multilevel confirmatory factor analysis (CFA) fitted to the nested data structure (observations within persons). Given the high interfactor correlation observed in the two-factor solution, we formed a single composite social appraisal score (mean of all seven items after reverse-coding) for use in all main analyses. Intraclass correlation coefficients (ICCs) were computed for each relevant EMA variable. Paired-samples t-tests compared momentary loneliness and trust between social-interaction and non-social prompts.

H1 and H2: Lagged effects of loneliness on social appraisals and trust beliefs

To test H1, we estimated a linear mixed-effects model predicting social appraisal quality at time point t from the WP deviation of momentary loneliness at t–1, controlling for autoregressive appraisal at t–1 and BP mean loneliness. To test H2, the analogous lagged model predicted trust at t from WP loneliness deviation at t–1, controlling for autoregressive trust and BP mean loneliness. All main models included random intercepts for participants. In the H1 model, the random-intercept variance converged to zero (singular fit), such that the BP mean is estimated but effectively absorbs all between-person variance; in H2, the BP mean loneliness coefficient reflects stable between-person differences in loneliness beyond the random intercept. To aid replication, the models were specified as follows:

H3: Trait loneliness as moderator

To test H3, we examined whether trait loneliness (grand-mean centred UCLA score) moderated the within-person loneliness–appraisal (H3a) and loneliness–trust (H3b) associations. For both models, a random slope for WP loneliness at t–1 was included when feasible.

Preliminary analyses: Factor structure of social evaluation items

Parallel analysis suggested retaining two factors. The EFA (varimax rotation) yielded a social belonging factor (MR1: accepted = .74, valued = .80, comfortable = .71) and a social safety factor (MR2: not threatened = .64, not judged = .57, not disliked = .65, not excluded = .61), together accounting for 59% of total variance. Multilevel CFA confirmed superior fit for the two-factor model (CFI = .976, TLI = .962, RMSEA = .053 [.046, .060], SRMR_within = .026, SRMR_between = .051) compared to the one-factor solution (CFI = .932, TLI = .898, RMSEA = .087), Δχ²(2) = 298.7, p < .001. However, the two factors were strongly correlated at both the within-person (r = .845) and between-person level (r = .789), indicating substantial shared variance. In line with the principle of parsimony, all main analyses therefore used a single social appraisal composite score (mean of all seven items after reverse-coding; M = 6.05, SD = 0.51, ICC = .27).

Descriptive statistics

For a detailed overview of the study’s main variables, see Table 1. State (r = −.28) as well as trait loneliness (r = −.50) correlated significantly and negatively with momentary trustworthiness beliefs, and the appraisal composite score (r = −.52 ≤ r ≤ −.49, see Figure 1). Momentary trustworthiness beliefs and the appraisal composite score were positively correlated with each other (r = .62). 87% of the interactions reported in our study were face-to-face, 21% with partners, 36% with family, 26% with friends and the rest with colleagues or known persons. Accordingly, 83% of the interactions were private and 14% were work-related.

Figure 1. Pearson correlations between the focal baseline (trait loneliness, generalized trust) and EMA variables (momentary loneliness, momentary trust, Social appraisal). Momentary trust refers to interpersonal trustworthiness beliefs.

The paired t-tests showed that momentary loneliness was lower after social interactions compared to when no interaction took place (t(54) = −5.94, p < .001). Momentary trustworthiness beliefs were likewise significantly higher at social compared to non-social prompts (t(54) = 2.90, p = .005). Note that only 55 participants reported that no interaction has taken place since the previous prompt. ICCs indicated that approximately one third of variance in momentary loneliness was attributable to stable between-person differences (ICC = .34), with the remaining two thirds reflecting within-person fluctuation. Momentary trustworthiness beliefs showed somewhat lower between-person stability (ICC = .19), and the social appraisal composite fell in between (ICC = .27). These ICC values confirm sufficient within-person variability to support the lagged multilevel analyses.

H1: Loneliness Predicting Subsequent Social Appraisals

H1 was not supported: WP deviations in momentary loneliness at t–1 did not significantly predict social appraisal quality at t (B = −.002, SE = .02, p = .932, see Figure 2). Autoregressive stability of social appraisals was significant (B = .07, SE = .02, p = .002). The random-intercept variance was estimated at zero (singular fit), so the model effectively reduces to a fixed-effects specification. The marginal R² was .091; the conditional R² was.247; this value primarily reflects the between-person loneliness predictor (B = −.32, SE = .05, p < .001) rather than the focal loneliness predictor.

Figure 2. Fixed-effect estimates (unstandardized coefficients B) and 95% confidence intervals for within-person (WP) lagged predictors across the three main multilevel models.

Panel H1 shows predictors of social appraisal quality at t; panels H2 and Joint show predictors of interpersonal trustworthiness beliefs at t (trust(t)). Red diamonds indicate effects significant at p < .05; grey diamonds indicate non-significant effects. AR = autoregressive control (outcome at t–1). BP = between-person. All models estimated with maximum likelihood; H2 included random intercepts (N = 114), while the H1 random-intercept variance converged to zero (singular fit).

H2: Momentary Loneliness Predicting Subsequent Trustworthiness Beliefs

H2 was similarly not supported: WP loneliness deviation at t–1 did not significantly predict momentary trust at t (B = .01, SE = .03, p = .727, see Figure 2). Trust showed modest autoregressive stability (B = .07, SE = .02, p = .003). The BP mean loneliness was significantly associated with trust (B = −.24, SE = .08, p = .003), indicating that more persistently lonelier individuals reported lower momentary trust overall. The marginal R² of the WP-loneliness model was small (.023); the conditional R² was.174, reflecting substantial between-person variance captured by the random intercept and BP term.

H3: Trait Loneliness as Moderator

H3a was only partially supported: the interaction between WP loneliness deviation at t–1 and trait loneliness marginally predicted subsequent social appraisal quality (B = .07, SE = .04, p = .057, see Figure 3A), but in the opposite direction of what we expected. The interaction indicates that the negative within-person association between momentary loneliness and subsequent social appraisals was moderated by trait loneliness: individuals lower in trait loneliness showed a stronger negative coupling, whereas this association was attenuated and reversed among those higher in trait loneliness. A random slope for WP loneliness was included and supported (σ² = .011, intercept–slope r = .47). Trait and BP loneliness exerted a significant negative main effect on subsequent social appraisals (trait: B = −.32, SE = .07, p < .001, BP: B = −.25, SE = .05, p < .001). The within-person deviation of loneliness at t–1 did not significantly predict subsequent social appraisals as a main effect (B = −.01, SE = .02, p = .732).

Figure 3. Predicted values from the trait loneliness moderation models at −1 SD, mean, and + 1 SD of trait loneliness (grand-mean centred UCLA score).

Shaded bands represent 95% confidence intervals. Panel A displays predicted social appraisal quality as a function of within-person loneliness deviation at t–1 (H3a; interaction p = .057). Panel B displays predicted momentary trustworthiness beliefs as a function of within-person loneliness deviation at t–1 (H3b; interaction p = .070). Predictions were computed using marginal means holding all other covariates at their observed values.

H3b was similarly only partially supported: the interaction between WP loneliness at t–1 and trait loneliness did not reach significance (B = .08, SE = .05, p = .070, see Figure 3B), though the direction mirrored H3a: the negative WP association was attenuated and reversed among individuals higher in trait loneliness. Due to singularity of the random slope model, a random intercept-only structure was retained. However, trait loneliness showed a pronounced negative main effect on subsequent trustworthiness beliefs (B = −.55, SE = .10, p < .001), indicating that individuals higher in trait loneliness reported substantially lower momentary trustworthiness beliefs overall. The BP loneliness control was non-significant in this model (B = −.12, SE = .07, p = .119). Prior trustworthiness beliefs positively predicted subsequent ones (B = .07, SE = .02, p = .002), indicating moderate temporal stability. The within-person deviation of loneliness at t–1 did not predict subsequent trustworthiness beliefs (B = .00, SE = .03, p = .964).