This research uses five consecutive annual waves of the Household Behavioral and Financial Survey conducted by Hiroshima University and administered online by Nikkei Research in February–March of 2020, 2021, 2022, 2023, and 2024. The baseline wave (February 20–25, 2020) collected 17,463 effective responses. Follow-up waves were fielded in the same late-winter to early-spring period each year. The 2021–2023 waves were conducted while COVID-19 was an ongoing pandemic; the 2024 wave occurred after the World Health Organization declared on May 5, 2023 that COVID-19 no longer constituted a Public Health Emergency of International Concern.
After matching respondents across all five waves, 2,817 individuals were available in the merged longitudinal dataset. The analytic sample was narrowed to 1,673 adults after excluding observations with missing information on variables required for the analysis (complete-case exclusion). The authors report that the reduction from the matched five-wave sample to the final analytic sample stemmed from requiring complete data on items such as employment status, financial literacy, household income, and household assets. Baseline characteristics differed between those included versus not included in the final analytic sample on several observed measures (for example, sex, age, education, financial literacy, and assets), and these differences are noted as relevant to interpretation.
Loneliness was assessed using the three-item UCLA Loneliness Scale (UCLA-3) developed by Hughes et al. Respondents answered how often they felt they lacked companionship, felt left out, and felt isolated, with response options “hardly ever or never,” “some of the time,” and “often.” The primary, inclusive coding classified respondents as lonely if they answered “some of the time” or “often” to any of the three items; respondents answering “hardly ever or never” to all three items were classified as not lonely. Cronbach’s alpha for the three items ranged from approximately 0.8444 to 0.8637 across waves, indicating good internal consistency.
The study also incorporated demographic, socioeconomic, and psychological variables. Gender, rural residence, university education, and presence of children were taken from the 2020 baseline. Age and current socioeconomic and psychological characteristics were measured in 2024. Variables indicating changes in living arrangements, employment, financial conditions, health, future anxiety, financial satisfaction, and depression were constructed using information from 2020 and 2024.
The analysis is descriptive. Annual prevalence of loneliness was estimated using the UCLA-3 inclusive threshold. Five-year individual-level sequences (patterns of lonely vs not lonely across the five waves) were tabulated to summarize stability and transition. Because the primary coding is inclusive and captures any moderate or greater frequency of loneliness, the authors performed a sensitivity analysis using a stricter definition that classified loneliness based only on “often” responses. The sensitivity check assessed whether the main descriptive patterns—particularly the timing of peak prevalence and the relative frequency of stable vs transitional sequences—held under the stricter threshold.
Among respondents retained in the five-wave panel and included in the analytic sample, the prevalence of loneliness rose to a peak in 2021. Prevalence declined in 2022 and then remained relatively stable from 2022 through 2024. The reported pattern highlights an early-pandemic increase followed by partial recovery and stabilization in the later years covered by the panel.
When participants’ responses were tracked across all five waves, the most common patterns were stable trajectories: respondents who were consistently classified as lonely across the five waves, and respondents who were consistently classified as not lonely across all five waves. A smaller fraction of respondents exhibited transient or switching patterns—reporting loneliness in some years but not others—during the pandemic period. The sensitivity analysis using the stricter loneliness definition preserved the overall finding that stable trajectories were more prevalent than transitional ones, although the ranking of specific five-year sequences shifted under the stricter threshold.
This study provides descriptive longitudinal evidence on how reported loneliness evolved among a cohort of online panel respondents in Japan during and after the COVID-19 period. Strengths include repeated measurement with the same instrument over five consecutive years and demonstrated internal consistency of the UCLA-3 items across waves.
Key limitations are related to sample composition and generalizability. The analytic findings reflect respondents retained in a longitudinal online panel and restricted to complete cases; the authors explicitly caution against interpreting the results as nationally representative prevalence estimates. Attrition and complete-case exclusion produced baseline differences between included and excluded respondents on several observed characteristics, which may bias descriptive patterns. The study is descriptive and does not attempt causal inference about pandemic-related drivers of loneliness.
In this five-wave online panel of adults in Japan, loneliness prevalence among retained respondents peaked in 2021, declined in 2022, and remained relatively stable from 2022 to 2024. Stable five-year trajectories—either persistently lonely or persistently not lonely—were more common than transitional sequences. A sensitivity analysis using a stricter loneliness threshold maintained the key patterns. These results document longitudinal patterns among the study cohort but should not be taken as nationally representative estimates.