As part of the larger BaSAlt project on PA promotion and counselling in nursing homes (‘Verhältnisorientierte Bewegungsförderung und individuelle Bewegungsberatung im Setting Altenwohnheim ‘ – ein biopsychosoziales Analyse– und Beratungsprojekt, funded by the German Federal Ministry of Health 2019–2023, grant no. ZMVI1-2519FSB114), our present study occurred in eight different nursing homes in Southwestern Germany. Institutions differed by environmental context (periphery and urban); management (non-profit institutions), and resident population composition [17]. Living places varied from 33 to 52 and nursing homes contained one (ground level) to three living areas. More women than men lived in all homes; two homes included protected areas for residents with dementia—the reason the number of cognitively impaired residents was higher compared with other homes. Table 1 shows detailed information about nursing home sites.

We conducted observations guided by the system for observing play and recreation in communities—SOPARC—direct observation method [23] to collect data on PA patterns in everyday lives of residents and Thiel et al.'s [19, 20] systematic direct observation tool developed for an observational study on social dynamics of physical (in)activity. Table 2 shows our observation instrument categories.

We collected temporal and spatial-related information (category 1), such as date, weekday, or time, to classify our large amount of data and identify activity hotspots throughout the day or week [24, 25]. To gain knowledge about the indoor infrastructure (category 2), we investigated barriers and facilitators for promoting PA (walking aids or PA-provoking objects, such as balls). We developed items concerning person-related information (category 3) following McKenzie and colleagues [24, 25]. Observers classified observed persons into personal categories (resident, caregiver, significant other, such as relatives), assigned gender (men, women), and activity categories adapted for nursing home settings (passive, sitting, standing, seated rolling, walking +). We adapted resident gait patterns for the sample [26] into five categories (overlapping, foot to foot, crotch length one foot, crotch length ≥ two feet). To collect group-related information (category 4), in line with Thiel et al. [20], we defined all people who entered the observed area as the sample (total number of people observed). To gather further information, we documented verbal and non-verbal (e.g., feeding) interpersonal interactions [24, 25] as well as guided low threshold activities as an important and integral part of daily life in nursing homes. Guided low threshold activities included all PA proposals spontaneously integrated into everyday life by staff. Observers also recorded field notes [25, 27] between screenings and collected weekly activity nursing home schedules to enrich quantitative data with more detailed descriptions of observed PA patterns.

The first observation period ran January to March 2020, the second February to March 2021. We observed various living areas of participating homes to obtain realistic impressions of everyday life. We chose the same observation period both years to avoid seasonal effects. We observed community areas—where meals are served, small activities held, visitors received, or people simply lingered or talked—with a minimum size of 40 m2 within nursing homes since a large proportion of residents spent their time there during the day. We did not observe residents in outdoor areas since few residents went outside due to winter season-related cold weather with snowfalls and rain. To capture resident daily PA fluctuations, observation intervals ranged from 10 am to 6 pm (weekdays) and 9 am to 5 pm (weekends). To ensure inter-rater reliability, observers were introduced to all items used in the screening instrument and in a group session, the observers were shown pictures of nursing home residents. Gender, gait patterns and activity categories were discussed together to generate a common understanding. Each observer was accompanied by a developer of the instrument on the first day of data collection and the examples from the group session were available at all times. Nine trained observers collected data with the observation instrument in predefined observation areas in 15 min intervals. In a pilot phase, developers tested the instrument to identify and address problem areas as well as set observation intervals. Intervals were defined by 15 min since nursing homes tend to be low-activity settings [5,6,7]. Every 15 min, observers overlooked areas and documented all PA-related information with the screening instrument (Table 2). Between screenings, observers wrote fieldnotes about special incidents.

Table 3 provides information about screening days, number of screenings, observation hours, and person observation units in 2020 and 2021. Since the same persons were monitored several times a day, an enormous number of person observation units resulted. In 2020, on-site observations stopped after 34 days from the Covid-19 outbreak and regulations in Germany limiting access to nursing homes for external people.

For statistical analyses, we considered the following factors, derived from the existing literature [5,6,7,8,9,10, 12,13,14]:

(1) Day of the week (weekday/weekend)

(2) Food intake (mealtime/no mealtime)

(3) Men residents (present/not present)

(4) Women residents (present/not present)

(5) Staff or significant others (present/not present)

(6) Daytime (morning [am]/afternoon [pm])

(7) Activities (guided low threshold activity/unstructured being)

To compare PA, regardless of the total number of people in the observed area, residents were classified into five activity categories. For data evaluation, each activity category was assigned a MET value (metabolic equivalent unit) according to existing literature (Table 4) [28,29,30,31,32]. Overall, we selected rather low MET values (0.95–2.6) for data analysis since residents tended to perform all activities very slowly and with low energy consumption [29].

We defined passive (0.95 MET) as lying down or sleeping [28]. We rated sitting (1.0 MET) as resting energy expenditure during quiet sitting [29]. Sitting rolling (1.5 MET) described moving around independently in a wheelchair using legs but not arms. Elsewhere, the activity was rated as predominantly sedentary, rarely physically active and thus the limit for sedentary behaviour [30, 32]. We defined standing (2.0 MET) as standing independently with (e.g., staff, cane) or without help [30], and we rated walking + (2.6 MET) as normal walking (on level surface) [31].

We performed statistical analyses supported by IBM SPSS Statistics 25. For data analysis, due to the high fluctuation of residents, we first calculated t-tests for independent groups to investigate the impact of possible influencing factors on resident PA before and during the Covid-19 pandemic. Additionally, we applied a Bonferroni correction to counteract the problem of erroneously rejecting a null hypothesis from calculating multiple comparisons. To check the practical relevance of differences, we calculated effect sizes of differences using Cohen’s d.

Second, to analyse interaction effects between predictors and differentiate the most pronounced contrast groups concerning PA, we carried out a classification tree analysis (CTA) to identify contrasting groups of PA and test influencing factors for possible interaction effects [33,34,35]. We used the Exhaustive CHAID algorithm ('Exhaustive Chi-squared Automatic Interaction Detector') for its possibility of a categorial merging for each predictor variable until only two categories remain for each predictor [35]. As a specific ‘stopping rule’ for the analysis, the significance level for splitting nodes and merging categories was set at p = 0.05. The depth was set at three and the minimum number of cases in parent nodes was set at 100 and 50 for child nodes. We calculated reliability measures using the risk estimate of misclassification (variance within the nodes). The quality of a tree model was calculated via the explained variance of the tree (variance between the nodes).

Third, we transcribed handwritten qualitative field notes and systematically scanned them for relevant aspects (MAXQDA, 2018) to interpret tree analysis results. We used qualitative data to contextualise and enrich quantitative data.