This study draws on longitudinal data from six waves (2017–2022) of the Swiss Household Panel-SHP [18, 19]. This nationally representative survey interviewed all household members older than 14 years of age from a random sample of private households in Switzerland since 1999. Questionnaires were administered mainly by computer-assisted telephone interviewing (CATI) (96–98%), with the rest of the participants using computer-assisted web interviewing (CAWI).

The analytical sample includes a subsample of SHP participants who were at least 65 years old in 2019 and had at least one valid measure of SWB between 2017 and 2022 (n individuals = 3086; n observations = 13,780). Table 1 provides participants’ descriptive information.

Among the entire sample, 49.6% had no missing information across the dependent variables, meaning they had completed information across all six waves. Additionally, 12.6% had only one measure of dependent variables across the time points (out of six possible). The missing data pattern was monotone, with the number of participants having missing outcomes increasing by 6.0–7.5% each observation year (2017–2022) (see Table 2).

Two components of SWB—positive affect and life satisfaction (PALS) and negative affect (NA)—were derived by summing up the relevant items ranging from 0 to 10. A higher score indicated greater satisfaction in each life domain and more frequent positive and negative affect. Table 3 details these measures. The psychometric properties of both measures were extensively examined [20], showing high internal consistency and scalar measurement invariance across time, age, sexes, and survey modes.

Positive affect and life satisfaction (PALS) comprised six items: (1) life satisfaction, (2) satisfaction with health, (3) satisfaction with personal, social, and family relationships, (4) satisfaction with leisure time activities [21, 22]; and two positive affect items [23] (5) the level of energy and optimism and (6) the level of joy. The internal consistency was strong, ranging between 0.75 and 0.76 across study years. The PALS score could range from 0 to 60.

Negative affect (NA) [23] comprised four items that assessed the level of four negative emotions—anger, sadness, worry/anxiety, and depression. The internal consistency was strong, ranging between 0.75 and 0.77 across study years. The NA could range from 0 to 40.

Stress [24] was measured using a single item that assessed an individual’s stress level, ranging from 0–“never” to 5–“very often.”

Sociodemographic indicators were included as potential effect modifiers of the age-SWB association to examine differences in the population-average trajectories across subgroups of the population: gender (1 = male; 2 = female), age groups (aged 65–74 years; those aged 75 years and above), and equivalized household income in quartiles.

The longitudinal population-average trajectories in SWB in 2017–2022 were described using piecewise growth curve analysis. This approach accounts for the multilevel structure of longitudinal data, where occasion-specific measurements are nested within individuals, and incomplete or unbalanced data, taking advantage of maximum likelihood estimation [25]. Our model included both fixed effects and random effects. Fixed effects represent the population’s average effect of time while controlling for age (and age2). Random effects include information about variance around the starting point of the trend (an intercept) and the trend itself (a slope).

Time is conceptualized as four separate periods (slopes), representing an overall trend between 2017 and 2022. The four slopes were (1) pre-pandemic—2017–2019; (2) into-pandemic—2019–2020; (3) pandemic—2020–2021; and (4) out-of-pandemic—2021–2022. This allowed us to directly compare the change in different periods. For instance, we could test whether a potential decline in SWB before the pandemic equaled a potential improvement in SWB post-pandemic using a Wald test, which formulates the null hypothesis that these two slopes equal zero (i.e., slope2017–2019 − slope2021–2022 = 0).

We also compared the trajectories of SWB between men and women, age groups (65–75 vs > 75-year-old), and quartiles of the equivalised household income by allowing the slopes to vary by these groups. We did this by including interaction terms between the age groups and slopes (e.g., gender*slope2017–2019, gender*slope2019–2020, gender*slope2020-2021, gender*slope2021–2022), and testing for differences with the Wald test (at p < 0.05). We further investigated the differences by running pairwise comparisons of estimated marginal means. All models were also controlled for the survey mode (CATI, CAWI).