Development and psychometric validation of the Kanazawa University loneliness scale for children and adolescents Authors: Yasuo Murayama, Daiki Soma, Masafumi Kameya, Makiko Nishiura, Ai Uemiya, Sanae Tanaka, Minehisa Ueda, Masatsugu Tsujii, Mitsuru Kikuchi Published: 2026 ABSTRACT AbstractLoneliness is a key risk factor for youth mental health; however, existing scales often fail to capture its multidimensional and relational nature. This study developed the [blinded] University Loneliness Scale (KULoS) for students in Grades 1–9 and evaluated its psychometric properties using cross-sectional and longitudinal survey data collected in Ishikawa, Japan (Time 1: n = 856; Time 2: n = 709). Exploratory factor analyses conducted separately for grades 1–4 and grades 5–9 supported a two-factor structure primarily reflecting item wording (direct vs. indirect items). The four-item direct subscale (KULoS-D) demonstrated good internal consistency (ω = 0.801–0.862), whereas the indirect subscale (KULoS-I) showed lower reliability (ω = 0.662–0.666). Confirmatory factor analyses at both time points indicated a good fit for the two-factor model. KULoS-D showed strong associations with single-item loneliness and depressive symptoms (rs ≈ 0.50) and moderate correlations with aggressive and prosocial behavior in the expected directions, whereas KULoS-I demonstrated weaker convergent and criterion-related validity. Cross-lagged panel analyses revealed that KULoS-D at Time 1 significantly predicted depressive symptoms at Time 2 after controlling for baseline depressive symptoms, whereas KULoS-I did not predict depressive symptoms. Overall, these findings support the KULoS-D as a brief and psychometrically robust measure of loneliness in children and early adolescents, with potential utility for large-scale research and early screening.Keywords: Loneliness, Emotional loneliness, Social loneliness, Factor analysis, Depressive symptoms, Adolescence Introduction Loneliness is a distressing emotional experience arising from unmet social needs owing to insufficient or unsatisfying relationships [45]. Although loneliness affects individuals across their lifespan, it is particularly prevalent during adolescence, a critical period of social reorientation and identity formation [46]. Recent surveys have highlighted the growing concern; for example, 38.7% of Australian adolescents reported moderate-to-severe loneliness during the COVID-19 pandemic [14], while over a third of Chinese adolescents reported similar feelings during the same period [57]. Even before the pandemic, approximately 30% of Japanese children and adolescents reported experiencing loneliness [56]. Furthermore, more recent evidence suggests a continued increase in loneliness among youth aged 15–16 years [55]. These findings underscore that loneliness among young people is not a transient response to pandemic-related disruptions, but a persistent and widespread phenomenon that demands careful attention. The implications of loneliness during childhood and adolescence are profound. Accumulating evidence has linked loneliness to a wide range of psychosocial difficulties. For instance, recent review studies targeting children and adolescent have reported that loneliness is moderately to strongly positively correlated with depressive symptoms ([25]: r =.33–0.66; [33]: r =.12–0.81). Furthermore, longitudinal investigations have demonstrated a bidirectional relationship between loneliness and depressive symptoms. A meta-analysis including child and adolescent samples confirmed the existence of a vicious cycle in which loneliness increases later depressive symptoms, and in turn, depressive symptoms intensify feelings of loneliness [8]. In addition to depressive symptoms, loneliness has been associated with various other psychosocial difficulties. For example, among university students, loneliness has been moderately positively linked to aggressive behavior, a core externalizing symptom [53]. This relationship has also been supported by experimental studies [6]. Moreover, loneliness is negatively associated with prosocial behavior [58], suggesting broader implications for social functioning and adjustment. Collectively, these findings underscore the importance of assessing and addressing loneliness during childhood and adolescence in order to support healthy psychosocial development. To better understand loneliness in young populations, it is essential to consider its multidimensionality. A widely accepted distinction exists between emotional loneliness, defined as the absence of close and intimate relationships, and social loneliness, which refers to a lack of a broader social network [34]. Furthermore, loneliness is embedded in specific relational contexts; children and adolescents may experience loneliness differently depending on whether it occurs in relation to parents, peers, or other adults [10]. Relationship-specific scales, such as the Loneliness and Aloneness Scale for Children and Adolescents (LACA; [35]) and the Relationship Provisions Loneliness Questionnaire (RPLQ; [22]), have attempted to address these distinctions. However, existing tools often fail to simultaneously assess both loneliness subtypes (emotional vs. social) and relational domains (e.g., peers vs. adults), limiting their conceptual coverage and practical utility. Another concern is the psychometric properties of widely used instruments. Although the Children’s Loneliness Scale (CLS; [2]) has been extensively applied in research, it was originally developed for younger children and does not clearly specify the conceptual definition of loneliness or the relational context in which loneliness is experienced. Although the LACA was designed for adolescents and includes both parent- and peer-related domains, it does not explicitly distinguish between emotional and social loneliness, which may lead to an underrepresentation of the emotional aspects of loneliness [34]. In addition, previous research has suggested that the factor structures of loneliness measures may be influenced not only by substantive conceptual distinctions (e.g., emotional vs. social loneliness) but also by the effects of item wording. For example, a reanalysis of the CLS, which includes items intended to reflect both emotional and social loneliness, supported a factor structure primarily differentiated by directly versus indirectly worded items rather than by the intended loneliness dimensions [27]. Similar wording-driven factor structures have been reported in other loneliness measures for adolescents [24, 31]. Consequently, an ongoing debate remains regarding whether direct or indirect wording provides a more valid approach to assessing loneliness in youth populations [34]. The present study addresses these limitations by developing the Kanazawa University Loneliness Scale (KULoS), a multidimensional instrument for children and adolescents in grades 1–9. The KULoS includes both directly and indirectly worded items and was designed to integrate the emotional and social dimensions of loneliness while capturing relationship-specific experiences with peers and adults. To establish criterion-related validity, we examined the associations between KULoS scores and both a single-item measure of loneliness and psychosocial difficulties that have been linked to loneliness in previous research—namely, depressive symptoms, aggressive behavior, and prosocial behavior ([6, 25, 33, 53, 58]). In addition, predictive validity was evaluated by testing longitudinal associations between loneliness, as measured by the KULoS, and subsequent depressive symptoms, based on prior evidence suggesting a bidirectional relationship between these constructs (e.g.,, [8]. By providing a developmentally appropriate and relationally nuanced tool, this study aims to advance the assessment of loneliness and contribute to the more precise identification and support of children and adolescents at risk of mental health problems. Methods Participants This study used data from the Mental Health Survey for Elementary and Junior High School Students, conducted as part of a support initiative in response to the Noto Peninsula Earthquake in January 2024. As of 2025, this educational support program has included all public elementary and junior high schools in Wajima City (nine elementary schools and three junior high schools), as well as one elementary school in Komatsu City and two in Nanao City, all located in Ishikawa Prefecture. The program was jointly implemented by the Ishikawa Kokoro no Care Center and the Research Center for Child Mental Development at Kanazawa University. The present study used data collected at two time points: October 2024 (Time 1) and February 2025 (Time 2). At Time 1 (T1), a total of 856 students from Grades 1 through 9 participated (435 boys and 421 girls). The sex ratio (1.03:1) closely reflected the national distribution of Japanese elementary and junior high school students in 2023 (1.05:1; [52]). The distribution across grades was as follows: Grade 1 (n = 82), Grade 2 (n = 78), Grade 3 (n = 105), Grade 4 (n = 92), Grade 5 (n = 106), Grade 6 (n = 105), Grade 7 (n = 96), Grade 8 (n = 84), and Grade 9 (n = 108). In the Japanese education system, students generally do not repeat grades, so school grade corresponds closely with age—from approximately 7 years old in Grade 1–15 years old in Grade 9. Of those who participated in the T1 survey, 709 students (369 boys and 340 girls; retention rate = 82.8%) also took part in the Time 2 (T2) survey. Procedure The survey was administered to all the students using tablet devices provided by the Japanese government. The students completed the questionnaires in their classrooms under the supervision of their homeroom teachers. Completion took approximately 10 min. Participation in the study was voluntary. Students provided informed assent through their homeroom teachers. An information sheet explaining the purpose and procedures of the study was distributed to parents by each school. The document informed parents that, if they wished to decline their child’s participation, they could notify the school administration (e.g., the principal) or the homeroom teacher by the specified deadline, and that failure to submit a refusal by the deadline would be regarded as consent. Parents were also informed that they could decline participation at any time without disadvantage. This procedure is an accepted method for minimal-risk, anonymous school surveys in Japan. In addition, before allowing participation in the survey, homeroom teachers confirmed—through school-wide parent meetings and related communication channels—that no refusal had been submitted by each child’s parent or guardian. This process ensured that parental intentions regarding participation were appropriately verified within the school’s standard procedures. All study procedures were reviewed and approved by the Chukyo University Research Ethics Committee (Approval Number: 2024-048). Measures Development of the KULoS The KULoS was developed to assess loneliness in students from Grades 1–9. Four psychologists (a clinical developmental psychologist, a cognitive psychologist, and two clinical psychologists) created the initial item pool based on the existing scales and original items designed to be comprehensible to younger children. The items aimed to capture both emotional and social loneliness across peer and adult relationships. Previous studies have noted that item wording, especially the distinction between direct and indirect items, can affect the psychometric properties of loneliness measures [11, 44]. While some argue that indirect items reduce response bias, others have found them to compromise validity and reliability [11, 13]. This study adopted a balanced approach by including both types and determining item retention based on the psychometric results. The final draft of the KULoS consisted of 10 items: seven direct and three indirect (see Table 1). Among the direct items, four assessed emotional loneliness, and three assessed social loneliness. Among the indirect items, two assessed emotional loneliness and one assessed social loneliness. Some items were framed around peer relationships, others around relationships with adults, and others used inclusive phrasing. Table 1The items in the initial draft KULoS and the final result of the factor analysis (Maximum likelihood method with promax rotation)NoTypesItemsGrades 1–4Grades 5–9 M SD FL M SD FLF1F2F1F21ELI feel sad because I don't have friends to play or talk with1.4990.827 0.666 0.0251.3850.728 0.791 −0.0412*ELI feel like I can rely on my friends3.1090.973−0.014 0.555 3.1781.0540.027 0.672 3ELI feel like my family and teachers don't understand me1.7000.923––1.7760.970––4SLI feel like the other kids and adults ignore me1.4620.724 0.536 0.0781.3610.657 0.584 0.1015ELI feel like my family and teachers won't help me when I'm in trouble1.5520.865––1.6850.996––6*SLI feel like my family and teachers care about my feelings3.3450.975−0.084 0.722 3.2870.986−0.084 0.615 7ELI feel sad because I don't have friends I can talk to about everything1.3840.735 0.833 −0.0841.3210.678 0.846 0.0238SLI feel sad when I want to be with someone but no one is there1.5070.767 0.791 0.0031.4030.717 0.874 −0.0569*ELI feel like I have friends I can talk to about happy and sad things3.3030.9790.156 0.607 3.4390.9450.082 0.593 10SLI feel like the people around me don't care about me1.7481.046––1.8741.113––Items marked with an asterisk (*) are reverse-scored. SL = social loneliness; EL = emotional loneliness; FL = factor loading. Items 3, 5, and 10 were excluded because they did not meet the predefined criteria; therefore, their factor loadings are not reported in the final exploratory factor analysis. Bold values indicate the primary factor for each item, defined as the factor on which the item exhibits the highest loading. The items in the initial draft KULoS and the final result of the factor analysis (Maximum likelihood method with promax rotation) Items marked with an asterisk (*) are reverse-scored. SL = social loneliness; EL = emotional loneliness; FL = factor loading. Items 3, 5, and 10 were excluded because they did not meet the predefined criteria; therefore, their factor loadings are not reported in the final exploratory factor analysis. Bold values indicate the primary factor for each item, defined as the factor on which the item exhibits the highest loading. Participants responded to the prompt: “Recently, how often have you experienced the following?” Based on prior research [13], items were rated on a 4-point Likert scale (1 = not at all, 4 = always), with higher scores indicating greater loneliness. External criteria for validity testing In order to evaluate the criterion-related validity of the KULoS, four external psychological constructs were assessed: loneliness, depressive symptoms, aggressive behavior, and prosocial behavior. Notably, depressive symptoms were assessed at both T1 and T2. Previous studies, including those involving children and adolescents, have sometimes assessed loneliness using single-item measures (e.g., [3, 29). These items typically ask directly about loneliness, such as “I feel lonely at school” or “I felt lonely.” Previous research has shown that loneliness assessed via single-item measures tends to correlate strongly with loneliness assessed using multi-item scales [38], with reported correlation coefficients ranging from 0.4 to 0.6. In the present study, we used the item “I feel very lonely” from the Japanese version of the Birleson Depression Self-Rating Scale for Children (DSRS-C; [5, 40]) to assess students’ subjective feelings of loneliness. Responses were rated on a three-point scale (0 = never, 1 = sometimes, 2 = most of the time), with higher scores indicating greater perceived loneliness. Depressive symptoms were measured using the short version of the Japanese DSRS-C [5, 41], a nine-item self-report scale using the same three-point response format. Higher total scores reflect greater severity of depressive symptoms. Internal consistency of the scale was acceptable at both time points (T1: McDonald’s ω = 0.793 T2: McDonald’s ω = 0.792). Aggressive behavior was assessed using the short version of the Buss–Perry Aggression Questionnaire for Children [49, 54], consisting of eight items. This version has demonstrated a high correlation (r =.91) with the original version [54] and has been validated for use with Japanese middle school students [20]. The internal consistency was high (McDonald’s ω = 0.888). Prosocial behavior was measured using the Prosocial Behavior subscale of the Japanese version of the Strengths and Difficulties Questionnaire [18, 21]. This subscale includes five items rated on a three-point scale, with higher scores reflecting stronger prosocial tendencies. This measure was administered only to students in grade 3 and above, and its internal consistency was acceptable (McDonald’s ω = 0.739). Data analysis Factor analysis To examine the factor structure of the KULoS, we used the T1 data. Prior to analysis, indirectly worded items were reverse-scored so that higher scores consistently indicated greater loneliness. Item suitability was evaluated using the Kaiser–Meyer–Olkin (KMO) measure of sampling adequacy. Following Howard’s [23] recommendation, items with KMO values below 0.60 were excluded from subsequent analyses. To determine the optimal number of factors, Velicer’s Minimum Average Partial (MAP) test and parallel analysis were conducted. These procedures are considered more accurate than the conventional Kaiser–Guttman rule, which tends to overestimate the number of factors [23, 59]. To select an appropriate extraction method, the item distributions were assessed using the Kolmogorov–Smirnov test. Maximum likelihood estimation was used when the data met the assumption of normality; otherwise, the least-squares method was applied. To obtain a stable and interpretable factor solution, items were retained according to the robust criteria proposed by Howard [23]: a primary factor loading of at least 0.40, a secondary loading below 0.30, and a difference of at least 0.20 between primary and secondary loadings. Because the present study targeted children and adolescents in the developmental stages, the factor structure of the KULoS may differ across grade levels. Therefore, we conducted factor analyses separately for each grade group. In the Japanese educational system, students are institutionally divided into elementary school (grades 1–6) and junior high school (grades 7–9). Elementary school grades are often further categorized into lower (Grades 1–2), middle (Grades 3–4), and upper (Grades 5–6) levels. Considering these developmental distinctions and the recommendation that a sample size of at least 300 is desirable for factor analysis [39], the participants were divided into two groups (Grades 1–4; n = 357 and Grades 5–9; n = 499), and factor analyses were conducted separately for each group. To confirm the factor structure identified in the exploratory factor analysis, confirmatory factor analyses (CFA) were conducted using T1 and T2 data. CFA was first performed separately for each grade group (Grades 1–4 and Grades 5–9). When the same factor structure was supported across the two grade groups, an additional CFA was conducted using the full sample to confirm the overall model fit. Model fit was evaluated using multiple goodness-of-fit indices, including the RMSEA, CFI, TLI, and SRMR. Following the criteria proposed by Hu and Bentler [26], model fit was considered satisfactory when CFI and TLI were ≥ 0.95, RMSEA was ≤ 0.06, and SRMR was ≤ 0.08. Examining validity of the KULoS Criterion-related validity was examined using both zero-order and partial correlations between the KULoS scores and the external variables described above. Partial correlations controlled for grade (i.e., age) and sex, given that both factors are known to influence the experience of loneliness during childhood and adolescence [4]. The interpretation of effect sizes was based on the guidelines proposed by Funder and Ozer [17]. Although socioeconomic status (SES) has been found to correlate negatively with loneliness in previous research [37], SES data were not collected in this study because of ethical considerations. In the immediate aftermath of the Noto Peninsula Earthquake, it was deemed inappropriate to ask families about their financial situation, given the widespread economic disruption and distress in the affected regions. Predictive validity was examined using cross-lagged panel models to assess the longitudinal associations between the KULoS subscales (i.e., all factors identified in the factor analyses) and depressive symptoms. The magnitude of the cross-lagged effects was interpreted based on the guidelines proposed by Orth et al. [42] for standardized coefficients (β): small = 0.03, medium = 0.07, and large = 0.12. Multiple statistical software packages were used for the analyses. The number of factors and the results of the exploratory factor analyses were determined using the HAD [50]. CFAs were conducted using JASP (JASP Team, [28]). All other analyses were performed using the Predictive Analytic Software (PASW) Statistics (SPSS Version 27.0) and IBM SPSS Amos (version 26.0; [1]). Results Factor analysis Before conducting the factor analysis, we evaluated the suitability, data distribution, and factor structure of the original KULoS. The KMO values ranged from 0.772 to 0.906, comfortably exceeding the 0.60 criterion and confirming that all ten items were suitable for the factor analysis. Across both grade groups, parallel analysis supported a two-factor solution (Grades 1–4: one-factor = 2.855, two-factor = 0.783, three-factor = 0.484; Grades 5–9: one-factor = 2.930, two-factor = 0.870, three-factor = 0.526). In contrast, the MAP test favored a one-factor solution in both groups (Grades 1–4: one-factor = 0.0355, two-factor = 0.0447, three-factor = 0.0527; Grades 5–9: one-factor = 0.0404, two-factor = 0.0490, three-factor = 0.0514). Given evidence that the MAP procedure may underestimate the number of factors [43] and that wording effects (direct vs. indirect items) can meaningfully shape the factor structure in self-report measures [11], we retained a two-factor solution for subsequent analyses. Prior to conducting the exploratory factor analysis, we examined the normality of the item score distributions for the KULoS. The Kolmogorov–Smirnov tests indicated that all items significantly deviated from a normal distribution (D = 0.240–0.419, ps < 0.001). Accordingly, we used the least-squares method for factor extraction. Exploratory factor analyses using the least squares method with Promax rotation were conducted separately for each grade group. In both groups, the analyses yielded a clear, wording-based separation. Factor 1, designated KULoS-Direct (KULoS-D), comprised seven directly worded items, whereas Factor 2, designated KULoS-Indirect (KULoS-I), comprised three indirectly worded items. Across repeated analyses, three direct items (Items 3, 5, and 10) did not meet the loading criterion of 0.40 and were removed. The final analyses confirmed that all retained items satisfied the pre-specified thresholds (primary loading ≥ 0.40, secondary loading < 0.30, and loading difference ≥ 0.20). The factor loadings for the retained items are presented in Table 1. Regarding internal consistency, McDonald’s omega was 0.801 and 0.862 for the final KULoS-D for Grades 1–4 and Grades 5–9, respectively, indicating good reliability. McDonald’s omega for the KULoS-I was 0.666 and 0.662 for Grades 1–4 and Grades 5–9, respectively, which fell below the acceptable threshold [36]. The correlations between the two factors were r =.385 and 0.350 (ps < 0.001) for Grades 1–4 and Grades 5–9, respectively, indicating a moderate positive association. It is important to note that, although the parallel analysis supported a two-factor solution, the MAP test favored a one-factor solution in both grade groups. Thus, a unidimensional structure was also statistically defensible. When an exploratory factor analysis was conducted specifying a one-factor solution, the retained items corresponded to those comprising the KULoS-D in the two-factor solution. Therefore, the one-factor model yielded the same substantive item composition as the direct-item factor identified in the two-factor solution. Confirmatory factor analysis To evaluate the model fit of the factor structure identified in the exploratory factor analysis, CFAs were conducted separately for each grade group using data from T1 and T2. Given that the KULoS items were measured on a 4-point ordinal scale and that prior analyses indicated non-normal item distributions, CFA was conducted using a diagonally ordinary least squares estimator, which is appropriate for ordinal data and robust to violations of multivariate normality. At T1, the model demonstrated a satisfactory fit in both groups (Grades 1–4: CFI = 1.000, TLI = 1.000, RMSEA = 0.011, SRMR = 0.041; Grades 5–9: CFI = 1.000, TLI = 1.000, RMSEA = 0.007, SRMR = 0.037). Similar results were obtained at T2 (Grades 1–4: CFI = 1.000, TLI = 1.000, RMSEA < 0.001, SRMR = 0.041; Grades 5–9: CFI = 1.000, TLI = 1.000, RMSEA < 0.001, SRMR = 0.033). Because the same two-factor structure was supported across grade groups and time points, an additional CFA was conducted using the full sample at T1 and T2 to confirm the overall model fit. At T1, the model demonstrated a satisfactory fit (CFI = 1.000, TLI = 0.999, RMSEA = 0.017, SRMR = 0.031). Similar results were obtained at T2 (CFI = 0.999, TLI = 0.999, RMSEA = 0.024, SRMR = 0.030). Concurrent validity Table 2 presents the means and standard deviations of the four external criteria variables, as well as their zero-order and partial correlations with each KULoS subscale. The KULoS-D showed strong positive correlations with the DSRS-C single-item loneliness measure at both T1 (r =.520, p <.001) and T2 (r =.531, p <.001), and with depressive symptoms at T1 (r =.503, p <.001) and T2 (r =.519, p <.001)1. Moderate associations were found with aggressive behavior (r =.246, p <.001) and a negative association with prosocial behavior (r = −.179, p <.001). Partial correlations controlling for grade and sex yielded similar results: loneliness (T1 r =.518; T2 r =.530), depressive symptoms (r =.504), aggressive behavior (r =.256), and prosocial behavior (r = −.188), all ps < 0.001. Table 2Zero-order and partial correlations between KULoS scores and external criterion variablesExternal criterion variableMSDTime r prKULoS-DKULoS-IKULoS-DKULoS-IDSRS-C loneliness item0.2660.522T10.5200.2570.5180.2610.2130.459T20.5310.2520.5300.253Depressive symptoms3.8533.186T10.5030.3940.5040.4053.5733.018T20.5190.3300.5160.335Aggressive behavior14.3965.238T10.2460.1280.2560.112Prosocial behavior7.0892.137T1− 0.179− 0.238− 0.188− 0.226r = zero-order correlation coefficients; pr = partial correlation coefficients controlling for grade and sex. All p-values were less than 0.001 Zero-order and partial correlations between KULoS scores and external criterion variables r = zero-order correlation coefficients; pr = partial correlation coefficients controlling for grade and sex. All p-values were less than 0.001 In contrast, the KULoS-I exhibited moderate positive correlations with the DSRS-C loneliness item at T1 (r =.257, p <.001) and T2 (r =.252, p <.001), as well as with depressive symptoms at T1 (r =.394, p <.001) and T2 (r =.330, p <.001). Correlations with aggressive behavior (r =.128, p <.001) and prosocial behavior (r = −.238, p <.001) were small. Partial correlations controlling for grade and sex remained largely consistent: loneliness (T1 r =.261; T2 r =.253), depressive symptoms (T1 r =.405; T2 r =.335), aggressive behavior (r =.112), and prosocial behavior (r = −.226), all ps < 0.001. Predictive validity To evaluate the predictive validity of the KULoS, we first examined whether there were any baseline differences between students who dropped out of the study (n = 147) and those who remained in T2 (n = 709). Independent-samples t-tests were conducted on four psychosocial indicators assessed at T1: single-item loneliness, depressive symptoms, aggressive behavior, and prosocial behavior. Because the assumption of homogeneity of variances was violated for the single-item loneliness measure, Welch’s t-test was used for comparison. The results indicated no significant group differences for any variable (single-item loneliness: t = 2.459, n.s.; depressive symptoms: t = 1.270, n.s.; aggressive behavior: t = 0.879, n.s.; prosocial behavior: t = 0.701, n.s.). In addition, a chi-square test showed no significant difference in sex distribution between students who participated at T2 and those who dropped out (χ² = 1.476, n.s.). Next, prior to the cross-lagged analyses, we conducted multiple regression analyses to assess potential multicollinearity. In these analyses, the dependent variables were T2 KULoS-D, KULoS-I, and depressive symptoms, and the independent variables were the corresponding T1 variables and demographic covariates (grade and sex). All variance inflation factors were below the recommended threshold of 5 [51], indicating no evidence of problematic multicollinearity (VIF range: 1.006–1.518). Figure 1 presents the results of the cross-lagged panel model used to examine the predictive validity of the KULoS. After controlling for grade and sex, KULoS-D scores at T1 significantly predicted depressive symptoms at T2 (β = 0.113, p =.001), whereas KULoS-I scores at T1 did not significantly predict depressive symptoms at T2 (β = 0.029, n.s.). Conversely, depressive symptoms at T1 significantly predicted both KULoS-D (β = 0.121, p <.01) and KULoS-I (β = 0.169, p <.001) scores at T2. The overall model fit was acceptable to good (CFI = 0.989, TLI = 0.942, IFI = 0.989, RMSEA = 0.056). A cross-lagged model examining the predictive validity of the KULoS. Note : * p <.01, ** p <.001. All values represent standardized coefficients. Grade and sex were included as control variables in the model but are omitted from the figure for clarity. Correlations between error terms are also excluded from the figure. Discussion This study aimed to develop the KULoS, a novel measure of loneliness for children and early adolescents that was conceptually informed by the distinction between emotional and social loneliness, while also considering interpersonal relationships with peers and adults, as well as both direct and indirect item wording. Exploratory factor analyses conducted separately in two stratified grade groups revealed a two-factor structure driven by item wording, namely, direct and indirect factors. The direct-item subscale (KULoS-D) demonstrated strong internal consistency and robust concurrent validity through significant associations with established external criteria, whereas the indirect-item subscale (KULoS-I) showed weaker psychometric properties. Factor structure of the KULoS Factor analyses were conducted separately for the two grade groups in accordance with recommended procedures for determining the appropriate number of factors. The results consistently supported a two-factor solution, with the factors distinguished by item wording. Direct items explicitly assessed loneliness using negatively phrased statements, whereas indirect items assessed loneliness using positively framed statements that required reverse scoring. Furthermore, CFAs conducted in each grade group and in the full sample supported the factor structure identified in the exploratory factor analyses. Taken together, these findings suggest stability of this wording-based distinction across grade groups and time points. Importantly, although the scale development was theoretically informed by the distinction between emotional and social loneliness, the present factor-analytic findings did not support a latent differentiation between these dimensions in this sample. Rather, the observed structure reflected differences in item wording (direct vs. indirect), rather than theoretically distinct subtypes of loneliness. These findings indicate that the second factor identified in the two-factor solution reflects wording-related method variance, rather than a substantively distinct psychological dimension of loneliness. The finding that the factor structure was primarily based on item wording is consistent with previous research. Studies on self-reported measures of loneliness have shown that wording effects can influence the factor structure of loneliness scales (e.g., [11]). Similar patterns have been observed in other psychological assessments (e.g., [16]. Collectively, these findings suggest that the structure of self-report scales may sometimes be influenced by linguistic format in addition to the substantive constructs they are intended to measure. The absence of a clear latent distinction between emotional and social loneliness in the present study is also consistent with findings from widely used loneliness measures. For example, the UCLA Loneliness Scale has typically been conceptualized and validated as a unidimensional measure [47, 48], despite including items that reflect both emotional and social aspects of loneliness. Similarly, the Children’s Loneliness Scale (CLS; [2]) has been reported to exhibit a single-factor structure, with items reflecting both emotional and social experiences loading onto the same factor. These findings suggest that, although emotional and social loneliness are theoretically distinguishable, they are not always empirically separable at the latent level, particularly in child and early adolescent populations, such as those examined in the present study. One possible explanation for this lack of differentiation, particularly among children and early adolescents, may involve developmental characteristics of this stage. During this period, broader social connections may be closely intertwined with proximal relationships, such as those with family and friends [9]. From this perspective, emotional and social aspects of loneliness may not yet be clearly differentiated at the latent level. This possibility is consistent with prior empirical findings indicating that items intended to capture emotional and social loneliness often load onto a single factor in child and adolescent samples [2, 7]. Validity of the KULoS-D The KULoS-D demonstrated strong validity across multiple indices. First, it was highly correlated with the loneliness-related item in the DSRS-C (i.e., I feel very lonely), with a correlation coefficient exceeding 0.50 even after adjusting for sex and grade level. This level of association is consistent with previous research [19], which reported similar correlations between single- and multi-item measures of loneliness. These findings affirm the scale’s capacity to capture core experiences of loneliness. The KULoS-D also demonstrated adequate criterion-related validity, as evidenced by its significant associations with depressive symptoms. These associations remained robust after controlling for demographic variables and are consistent with meta-analytic evidence indicating a strong link between loneliness and depression among children and adolescents [12]. Previous studies have reported moderate to strong correlations between these constructs (e.g., r =.33–0.81; [25, 33]), and the coefficients observed in this study fell within this range. Importantly, these associations were largely unchanged even when depressive symptoms were recalculated after excluding the loneliness-related item from the DSRS-C, suggesting that the observed correlations were not solely driven by item overlap. Furthermore, the KULoS-D was positively associated with aggressive behavior and negatively associated with prosocial behavior, consistent with previous research across diverse populations [30, 53]. Consistent with these findings, the KULoS-D demonstrated predictive validity over a four-month interval. In the cross-lagged model, loneliness significantly predicted subsequent depressive symptoms after controlling for prior levels of depression, and vice versa. The magnitude of the cross-lagged effect (β = 0.113) approached the threshold for a large effect according to the guidelines proposed by Orth et al. [42]. These results indicate that KULoS-D scores provide incremental predictive information regarding later depressive symptoms. However, the findings should not be interpreted as evidence of a strong or deterministic causal relationship, but rather as supporting a statistically reliable longitudinal association between loneliness and depressive symptoms. In addition, although the KULoS-D consists of only four items, its content adequately reflects the conceptual definition of loneliness. Regarding scale length, while some established loneliness measures include approximately 20 items, brief measures with three to five items are also commonly used. Therefore, the brevity of the KULoS-D does not raise substantial concerns regarding its validity. In terms of item content, two items were originally written to reflect social aspects of loneliness and two to reflect emotional aspects. However, as noted above, these aspects were not empirically differentiated at the latent level. Accordingly, the KULoS-D should be interpreted as capturing a general loneliness construct rather than distinct subtypes. Moreover, regarding relational domains, two items focus on peer relationships, one targets relationships with adults, and the remaining item is applicable to both peer and adult contexts. Taken together, the KULoS-D appears to provide a balanced and developmentally appropriate assessment of loneliness in children and early adolescents while minimizing the respondent burden. Overall, these findings provide robust evidence for the convergent, criterion-related, predictive, and content validity of the KULoS-D, supporting its utility as a self-report measure of youth loneliness, grounded in established psychological frameworks. Validity of the KULoS-I In contrast, the KULoS-I exhibited relatively weaker associations with external variables that are typically associated with loneliness. Although moderate correlations were observed between the KULoS-I and depressive symptoms, these associations were substantially lower than those found for the KULoS-D. For example, previous research has reported correlation coefficients exceeding 0.50 between single-item and multi-item measures of loneliness [19]; however, in the present study, the correlation between KULoS-I scores and the single-item loneliness measure fell below 0.30. Moreover, this correlation coefficient was lower than that observed between KULoS-I and depressive symptoms. These findings raise concerns regarding the convergent validity of the KULoS-I. Consistent with these concerns, the results of the predictive validity analyses also suggested limitations of the KULoS-I. As described above, while the KULoS-D exhibited bidirectional causal associations with depressive symptoms over a four-month interval, aligning with a previous meta-analytic finding indicating a reciprocal relationship between loneliness and depression [8], the KULoS-I failed to significantly predict subsequent depressive symptoms. This discrepancy further calls into question the validity of the KULoS-I as a reliable measure of loneliness. The subscale’s associations with aggressive and prosocial behaviors were also weaker than expected. Although the direction of these associations was consistent with theoretical expectations, the effect sizes were notably smaller than those reported in previous research [30, 32, 53]. These discrepancies suggest that the KULoS-I may be limited in its ability to capture the full scope of loneliness-related psychological functioning at least in its current form. Indirect worded items, which often rely on positively framed statements, may be less effective in eliciting accurate self-reflection about loneliness, particularly among children with limited metacognitive insight or in cultural contexts where emotional disclosure is constrained. Final composition of the KULoS Based on the present findings, we recommend that loneliness be assessed using the four directly worded items (KULoS-D) as the scoring model. The total loneliness score should therefore be calculated exclusively from these four items. The KULoS-D consistently outperformed the KULoS-I across all psychometric indices, and its properties aligned with previous recommendations favoring the use of direct items in loneliness measures [15]. Furthermore, the internal consistency of the KULoS-D exceeded 0.80, whereas that of the KULoS-I fell below the commonly accepted threshold of 0.70 [36]. These results support the use of the KULoS-D as the primary indicator of loneliness in children and adolescents. Importantly, because the KULoS-I demonstrated limited convergent and discriminant validity, it is not recommended for use as a standalone measure of loneliness. In the factor-analytic results, the one-factor solution yielded the same substantive item composition as the direct-item factor identified in the two-factor solution. Thus, from a substantive perspective, both the one-factor and two-factor models converge on the same loneliness construct, represented by the KULoS-D. In this sense, the choice between one- and two-factor representations does not alter the substantive assessment of loneliness in the present study. The two-factor solution was retained to transparently represent the wording-based distinction between direct and indirect items. Modeling this distinction allows method-related variance associated with item phrasing to be explicitly identified rather than absorbed into a single undifferentiated factor. From a measurement perspective, this approach enhances interpretability and helps prevent potential response bias from contaminating the substantive loneliness score. However, direct-item scales have limitations. As noted in the literature, directly asking about sensitive experiences such as loneliness may introduce response biases, including social desirability or discomfort, particularly in younger populations or collectivist cultural settings. Although indirect items may theoretically mitigate such biases, the present findings suggest that their limited reliability and validity pose challenges for their use in substantive assessment. Accordingly, the KULoS should retain its two-subscale, seven-item structure to preserve transparency regarding wording effects and to allow the indirect items to serve methodological and exploratory purposes. However, only the four direct items (KULoS-D) should be used to derive the total loneliness score. This framework provides a psychometrically sound, conceptually coherent, and practically clear basis for assessing loneliness in childhood and adolescence. Limitations and future directions This study has several limitations. First, the scale was developed and validated using a sample of Japanese elementary and junior high school students, with participants exclusively drawn from Ishikawa Prefecture. Therefore, the generalizability of the present findings to students in other regions of Japan is unclear. In addition, the applicability of the KULoS to other cultural and linguistic contexts has not yet been examined, and future studies should conduct cross-cultural validation to establish broader generalizability. Second, the current version of the KULoS was not evaluated with older adolescents, such as high school students. Given that loneliness often intensifies during late adolescence, future studies should investigate the scale’s utility in this age group. Third, all external validation measures were based on self-report, which may be subject to memory biases or social desirability effects. Future research should consider incorporating teacher or parent ratings, behavioral observations, or physiological markers of emotional distress. Third, the criterion-related validity in the present study was evaluated using a single-item loneliness indicator derived from the DSRS-C. Although the use of single-item loneliness measures has been supported in prior research ([29, 3), some researchers have raised concerns about the limitations of assessing loneliness using a single item (e.g., [38]). Future research should replicate the present validity findings using independent multi-item loneliness scales and examine whether these associations remain consistent. Furthermore, data were collected within the context of post-disaster recovery, which may have influenced participants’ psychological states and response tendencies. Emotional numbing, increased social cohesion, or altered interpersonal dynamics following a collective traumatic event could have affected how children perceived and reported experiences of loneliness, depression, and related behaviors. These contextual factors should be carefully considered when interpreting the current findings and when designing future studies in non-disaster settings. Conversely, the post-disaster context may have served to accentuate psychological vulnerability, particularly among children and adolescents who were already at greater risk. This heightened psychosocial strain could have led to increased variability in mental health indicators, such as loneliness and depressive symptoms, thereby enhancing the sensitivity of the data to detect individual differences. From this standpoint, the timing of data collection may have been advantageous for the development and validation of screening tools, as it provided a context in which the full spectrum of emotional and behavioral responses could be more readily observed and assessed. Despite these limitations, the KULoS represents a theoretically grounded, psychometrically robust instrument for assessing loneliness in children and adolescents. Although the scale was conceptually informed by multiple emotional and relational aspects of loneliness, the present findings indicate that it functions primarily as a unidimensional measure of general loneliness in children and early adolescents. The scale also holds promise for practical application in educational and clinical contexts, particularly in disaster-affected settings, where rapid and accurate assessment is critical. With continued refinement and cross-validation, the KULoS can contribute meaningfully to both research and intervention efforts aimed at alleviating loneliness and supporting youth mental health. Conclusion The present study developed and validated the KULoS as a developmentally appropriate measure of loneliness in children and early adolescents. The direct-item subscale (KULoS-D) provides a concise yet psychometrically sound index of loneliness, offering a substantially lower assessment burden than many existing youth loneliness scales. Despite consisting of only four items, the KULoS-D was designed to reflect diverse emotional and relational aspects of loneliness. However, the present findings indicate that these aspects were not empirically differentiated at the latent level; therefore, the KULoS-D should be interpreted as assessing a general loneliness construct rather than distinct subtypes. This brevity supports its feasibility for large-scale research and school-based applications. From a research perspective, the KULoS-D offers a brief and conceptually grounded tool for large-scale surveys and longitudinal studies. From an educational and clinical perspective, it may serve as a practical screening indicator for identifying students at risk of elevated loneliness and related mental health issues. Notably, KULoS-D scores significantly predicted depressive symptoms four months later, with a moderate-to-large effect size, supporting its potential value for early monitoring and support. Footnote 1. Because the DSRS-C total score includes an item assessing loneliness, we recalculated the depressive symptom scores after excluding this item and then computed correlations with the KULoS subscales. For the KULoS-D, the correlation coefficients were 0.461 at T1 and 0.483 at T2; for the KULoS-I, the coefficients were 0.394 at T1 and 0.320 at T2. Although the correlations for the KULoS-D were slightly attenuated (by approximately 0.04), the overall pattern of associations was largely consistent with the coefficients obtained using the original DSRS-C total score. Acknowledgements This study was partially supported by grants from the Moonshot Research and Development Program (Grant Number: JPMJMS2297 and JPMJMS229C-14a). We would like to thank Editage (www.editage.com) for English language editing and journal submission support. The authors have authorized the submission of this manuscript through Editage. Author contributions Yasuo Murayama wrote the manuscript and conducted all work in this study. Daiki Soma and Masafumi Kameya conducted the survey via the internet. Ai Uemiya, Sanae Tanaka, and Minehisa Ueda contributed to drafting the items for the new scale developed in this study. Makiko Nishiura conducted the data analysis. Masatsugu Tsujii provided advice for the research project. Mitsuru Kikuchi managed the research project and provided funding. Funding This study was supported by grants from the Moonshot Research and Development Program: grant number JPMJMS2297 and JPMJMS229C-14a. Data availability The datasets utilized and/or analyzed during this study are available from the corresponding author upon reasonable request. Declarations Ethics approval This study was conducted in accordance with the Declaration of Helsinki and was approved by the Research Ethics Committee of Chukyo University (Approval No. 2024-048). Consent to participate Participation in the study was voluntary. Students provided informed assent through their homeroom teachers. An information sheet explaining the purpose and procedures of the study was distributed to parents by each school. The document informed parents that, if they wished to decline their child’s participation, they could notify the school administration (e.g., the principal) or the homeroom teacher by the specified deadline, and that failure to submit a refusal by the deadline would be regarded as consent. Parents were also informed that they could decline participation at any time without disadvantage. This procedure is an accepted method for minimal-risk, anonymous school surveys in Japan. In addition, before allowing participation in the survey, homeroom teachers confirmed—through school-wide parent meetings and related communication channels—that no refusal had been submitted by each child’s parent or guardian. Consent to publish Not applicable. Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. References Associated Data Data Availability Statement The datasets utilized and/or analyzed during this study are available from the corresponding author upon reasonable request.