Individual’s daily behaviour and intergenerational mixing in different social contexts of Kenya

We investigated contact patterns in diverse social contexts in Kenya and the daily behaviours that may play a pivotal role in infection transmission to the most vulnerable leveraging novel data from a 2-day survey on social contacts and time use (TU) from a sample of 1407 individuals (for a total of 2705 person days) from rural, urban formal, and informal settings. We used TU data to build six profiles of daily behaviour based on the main reported activities, i.e., Homestayers (71.1% of person days), Workers (9.3%), Schoolers (7.8%), or locations at increasing distance from home, i.e., Walkers (6.6%), Commuters (4.6%), Travelers (0.6%). In the rural setting, we observed higher daily contact numbers (11.56, SD 0.23) and percentages of intergenerational mixing with older adults (7.5% of contacts reported by those younger than 60 years vs. less than 4% in the urban settings). Overall, intergenerational mixing with older adults was higher for Walkers (7.3% of their reported contacts), Commuters (8.7%), and Homestayers (5.1%) than for Workers (1.5%) or Schoolers (3.6%). These results could be instrumental in defining effective interventions that acknowledge the heterogeneity in social contexts and daily routines, either in Kenya or other demographically and culturally similar sub-Saharan African settings.

Individual social interactions are critical drivers of transmission processes underlying the spread of infectious diseases such as measles, influenza, and the novel coronavirus infection (SARS-CoV-2).
The availability of age-specific data on social contact patterns is fundamental to properly estimate the agespecific transmission parameters of an infectious disease transmitted directly from person to person, and to mimic the potential age heterogeneity characterising different infection episodes 1,2 . Furthermore, these data could be used to estimate critical quantities such as age-specific matrices of secondary cases, i.e., the next generation matrices, which allow to identify the transmission routes bringing the disease into the most vulnerable population segments, to estimate reproduction numbers 3 , and to evaluate the impact of health policies aimed at mitigating the infection spread 4,5 .
The need for information on social contact patterns becomes crucial when considering resource-poor settings, where the quality and availability of health care requires special efforts to define feasible and sustainable strategies to prevent transmission and reduce the disease burden. Evidence from social contact surveys has shown that, in low-and middle-income countries (LMICs), contact patterns are generally characterised by (i) a marked age assortativeness (i.e., individuals tend to interact more often with peers of similar age) concentrated among school-age children and young adults, and (ii) higher rates of intergenerational mixing compared to high-income settings. Common features from previous studies are that the largest daily number of contacts is experienced by school-aged children between 6 and 18 years old and working-age adults, while intergenerational contacts are mainly reported between household members [6][7][8][9][10][11][12][13][14][15][16][17][18][19] .  Fig. S2). Specifically, people living in the rural setting were on average younger, lived in larger HHs, and had a lower socio-economic status, both in terms of SES index and current education level.
Social contact patterns by setting. A total of 23,532 contacts were reported over the two survey days.
The overall mean number of contacts per person per day was 8.92 (median 8, IQR [5][6][7][8][9][10][11][12] and it was higher in the rural setting (mean 11.56) than in the two urban ones. After adjusting for the other participants' characteristics, we found that the average number of contacts was 34% and 24% lower in the slum and the mixed urban area, respectively (Table 1). Across all setting, most of contacts were non-physical, of long duration (more than 1 h) and with people outside the family, although with percentages much higher in the urban settings than in the rural one. In the rural site, we found that the percentages of physical contacts as well as contacts with relative (likely belonging to the extended family) were around 30% of the total ( Supplementary Fig. S3). www.nature.com/scientificreports/ For each of the three settings, we built contact matrices by 5-year age groups defining the intensity of interactions between the age groups ( Fig. 1). In each setting, we found predominantly age-assortative contact patterns among younger individuals, with a larger number of contacts occurring among school-age children and adults up to 30 years. To quantify the importance of contacts occurring between individuals of the same age group in the overall contact matrix, we calculated the assortativeness index Q 39 . The closer the value of Q to zero, the lower the assortativeness and the higher the mixing between different age groups (Fig. 1). Overall, we found little difference between the three settings, as Q was equal to 0.061 for the rural setting, 0.056 for the urban slum setting, and 0.069 for the urban mixed/non-slum setting. However, a comparison with the estimates of Q for other LMICs countries (e.g., 0.051 in Zimbabwe 12 , and 0.075 in Vietnam 8 and Peru 7 , own calculations), and for a group of high-income countries in Europe (ranging from 0.14 in the United Kingdom 2 , own calculation, to 0.21 in Belgium 40 and 0.22 in Russia 41 ) clearly showed the higher degree of mixing between different age groups (and the lower assortativeness) in Kenya and in other LMICs compared to high-income countries. Although contacts with children (age group 0-14) were reported by participants of all ages, their intensity was higher among child participants. Higher contacts with children across all age groups were especially reported in the rural setting (Fig. 2C), possibly because of the higher rate of intergenerational residence and the larger HH size ( Supplementary Fig. S4). A strong mixing between adults older than 20 years and younger than 60 years was also found across all settings.
Overall, the contact distribution was right-skewed, especially in the rural setting, where individuals in the last decile of the distribution reported between 18 and 42 contacts per day, while only between 12 and 21 in the slum ( Fig. 2A). Regardless of the setting, most contacts were reported at home (from 59.7% in the mixed area to 67.9% in the slum) and in the general community (between 20.1% in the slum and 30% in the rural area) (Fig. 2B). Indeed, contacts at home and in the general community together represented 92.5% of all contacts in the rural setting, a share that was significantly larger than in both urban settings (87.9% in the slum and 86.6% in the mixed area), even after adjusting for participants' socio-demographic characteristics in a statistical model (sex, age group, education level, HH size, living arrangements, SES status, TU profile, and contact day type, see Supplementary Fig. S5).
Finally, across the settings, intergenerational mixing, defined as the mixing between different generationschildren (aged 0-14 years), teens and adults (aged 15-59 years), and older adults (aged 60 years or more)-was predominant for older participants. Indeed, 90.3% of contacts reported by older adults were intergenerational, of which 23.3% (95% CI 20.7-26.2%) were with children. Lower rates of intergenerational mixing were reported by children (37% of their contacts, of which 1.7%, 95% CI 1.5-2.0%, with older adults) and by teens and adults (26.9% of their contacts, of which 3.2%, 95% CI 2.9-3.5%, with older adults). Moreover, likely because of the age distribution of the different settings, we found that the percentage of intergenerational contact reported with both older adults and children was higher in the rural setting, while the percentage of intergenerational contact reported with teens and adults was higher in the urban settings (Fig. 2C). These results were robust with respect to whether the day of the reported contacts was a weekday or a weekend day, as we did not find any relevant difference in the percentages of contacts between the two types of day ( Supplementary Fig. S6).

Differences in participants' daily routines.
To understand the behavioural determinants of the number of social contacts, we identified six profiles based on individuals' daily routine behaviour, which were subsequently used to investigate differences in contact patterns between different groups of individuals. To this  www.nature.com/scientificreports/ purpose, we leveraged on the combination of a principal component analysis (PCA)-which we used to reduce the data dimensionality while retaining most of the information (transferred into newly created independent variables called "principal components")-and a hierarchical cluster analysis (HCA)-used to create groups of individuals with similar characteristics based on the principal components 43 . In our analysis, we first applied PCA to the time use (TU) data-a set of indicators for the time spent in different locations/activities within specific time slots during the day-and retrieved 11 principal components that summarized 79% of the total information. Next, we applied HCA to the principal components to group individuals into meaningful profiles that can be described based on the most frequently visited location during the day, in terms of time slots, by individuals included therein (Fig. 3).
Most individuals (92.3%) reported the same TU profile over the two survey days, especially in the rural site (99%). The largest group by far (Homestayers) accounted for 71.1% of p.d. and identified people staying at home or within the village the whole day (respectively 90.8% and 7% of their daily time). The remaining TU profiles were characterised by a specific location, other than the home. The second largest group (Workers: 9.3% of total p.d.) identified those spending over half of their day at work (56%); this profile was more prevalent in the urban settings, especially the slum, and only for the group of adults. The third profile group (Schoolers: 7.8% of p.d.) gathered those spending a large share of the day at school (48.6%), which was uniformly present in all settings, and mostly populated by children.
The last three groups encompassed individuals spending a relatively larger share of their time in the general community, and were gradually more prevalent in the urban settings and among teens and adults as the distance from home increased. The fourth group (Walkers: 6.6% of total p.d.) included those spending, outside the home, 37.5% of their time in locations, likely at walking distance, within the study site (i.e., within the rural site of   Social contact patterns by TU profile. The TU profiles described above were used to assess differences in contact patterns between people characterised by different daily routines. No matter the TU profile, most of contacts were non-physical, of long duration (especially those of Homestayers, Schoolers, and Travelers), and with people outside the family (especially those of Workers and Schoolers, but also of Walkers and Commuters) ( Supplementary Fig. S3).
With regards to age mixing, we found that the degree of age-assortativeness varied across TU profiles (Fig. 4). This was lower for Workers ( Q = 0.0030), Travelers ( Q = 0.023), and Commuters ( Q = 0.046), for which age mixing, apart from contacts between parents and their children, was scattered among teens and adults of different age. The Q index increased to 0.057 for Homestayers and 0.066 for Walkers, reflecting a moderate level of ageassortative mixing (therefore, suggesting marked intergenerational mixing) among children and adults up to the age of 30 years. Finally, the highest level of age-assortativeness (and thus the lowest level of intergenerational mixing) was found among Schoolers ( Q = 0.10), where the intensity of contact among children reached the peak.
The contact distribution under each TU profile was generally right-skewed, especially for Schoolers, who also reported a higher median number of contacts (Fig. 5A). Workers reported 43.3% of their contacts at work, Schoolers 48.5% of contacts at school, while Walkers, Commuters and Travelers reported between 55% and 64.4% of contacts in the general community, showing a positive relationship between the time spent and the reported number of contacts in each location (Fig. 5B). Contacts at home usually represented the largest share, except for Commuters and Travelers, who spent a larger share of time in the general community at increasing distance from home (Figs. 3A, 5B) and reporting a higher number of contacts as occurring in this location.
TU profiles also varied in terms of intergenerational mixing. Contacts with children were mainly reported by Schoolers (55.4%), followed by Homestayers (43.2%) and Walkers (34%) (Fig. 5C). Within these three profiles, contact with children was mainly reported by child participants; however, we also observed a high share of intergenerational contact reported by older adults with a Homestayer profile (26%). Conversely, contacts of Workers with children remained below 16%. Contacts with older adults represented a small fraction for all TU profiles, although we found significant differences between Workers and Schoolers (less than 2% of contacts), Homestayers (almost 3% of contacts), and Walkers and Commuters (almost 5% of contacts). Such contacts with older adults were mainly reported by participants aged 60 years or more; however, we also observed a substantial level of intergenerational contact reported by children with a Walker profile (3.2% of their reported contacts) or a Commuter profile (3.5%) and by teens and adults with the same profiles (4% of contacts for Walkers and 5.1% for Commuters), suggesting that interactions with the older adults occurred mostly with those individuals spending a high proportion of their time in the general community (Fig. 5C).
After adjusting for participants' socio-demographic characteristics in a statistical model, we estimated that Workers and Schoolers reported overall 38% and 23% more contacts than Homestayers, respectively (Table 1). We did not find any significant difference in contact numbers between Workers and Homestayers at home and in the general community ( Supplementary Fig. S7). Conversely, for Schoolers we estimated a lower relative number of contacts than Homestayers both at home and in the general community, supporting the finding that most of their contacts occurred at school (Supplementary Fig. S7). Finally, Walkers and Commuters (for whom we estimated 10% and 21% more contacts than Homestayers, respectively) had less contacts at home and much more in the general community.

Discussion
Although most social contact studies in LMICs 7, 8,11 have been predominantly conducted in rural settings, while those conducted in high-income countries 2,17,18,41,44,45 have been largely carried out in urban settings, only a few have so far presented a direct comparison of social mixing patterns between such different levels of urbanization within the same country or region 6,10,12,13 . Moreover, only one study sought to investigate whether differences in social behaviour induced by daily routines entailed differences in the average numbers of contacts 12 . Our contact study in Kenya contributes to filling this gap by (i) providing a robust comparison of social mixing patterns across diverse demographic settings along the rural-urban gradient, and (ii) identifying which individuals' profiles, based on their age mixing patterns and daily time use behaviour, can play a pivotal role in bringing different age groups in contact. The latter aspect is of great importance to identify those individuals who present a higher risk of bringing an infection to the more vulnerable segments of the population. Even though this study was not the first one conducted in Kenya 10,46,47 , it was the largest and most diverse one in terms of surveyed settings, describing three geographical areas that significantly differed in terms of socioeconomic status (the rural setting being the poorer in relative terms, the urban mixed/non-slum area being the wealthier), age and household size distribution (a younger population and a higher frequency of large extended households in the rural setting, and a higher frequency of small nuclear households in the slum), and TU profiles (a higher percentage of people spending time at work in the urban settings, especially in the slum, as well as a higher percentage of people spending time in the general community in the rural setting).
The number of contacts reported by our study participants were higher in the rural setting than in the urban ones, even after adjusting for the HH size (higher in rural site) and the living arrangements (with higher intergenerational cohabitation in the rural site), as a result of the higher contact numbers reported at home and in the general community. Even though these findings differed from what was found for Zimbabwe 12 , they were consistent with what was observed by previous studies in the same area in Kenya 10 , in Zambia and South Africa 6 , as well as with the evidence from the synthetic matrices for low-income countries constructed using setting-specific survey data (e.g., on household, school, classroom, and workplace composition, and on contact patterns) 48 . Despite a lower degree of age assortativeness in our three settings than what observed in high-income countries, we found both a higher intensity of contacts among children and a higher share of intergenerational contact with older adults in the rural setting than in the urban ones. As we found no evidence of a difference for these contact patterns regardless of being reported on weekdays versus weekend, we believe they were not entirely motivated by school attendance or the working routine during the weekdays, but rather a consequence Average number of contacts www.nature.com/scientificreports/ of the HH structures and living arrangements more common in the rural site. Such mixing behaviours should be taken into account when considering the transmission of infections with strongly age-specific severity, such as in the case of RSV or SARS-CoV-2 26,27,31,33 . Compared with previous studies conducted in Kenya, we found that our survey participants reported a lower number of contacts: a study carried out between 2011 and 2012 in Kilifi county reported 18.8 physical contacts per day in a rural setting and 16.5 in a semi urban area 10 , while a small study conducted in a slum in the capital Nairobi in May 2020 found an average of 18 overall contacts per day 47 . Such discrepancies with our study might be due to differences in the sample selection, especially in the urban site. First, since we lacked a sampling list for our urban site, conversely to what happened with the study in the Nairobi 47 slums, we employed a modified version of the WHO Expanded Program on Immunization (EPI) cluster sampling technique 49 (see Supplementary  Information), which may have caused some bias in the household selection. Second, the inaccessibility to some homes, especially the guarded houses in the non-slum area, may have been another possible source of bias in the household selection. Finally, the requirement that interviewers worked between 7 a.m. and 6 p.m., due to both security measures and working time restrictions, may have prevented the inclusion of working adults who might be outside the home when interviews were conducted. On the other hand, to improve data quality, especially that collected from illiterate respondents, instead of using self-reporting paper diaries 10 or phone interviews 47 , we used face-to-face interviews conducted by trained interviewers, during which the data provided by respondents were thoroughly revised and entered into an electronic questionnaire.
The combined collection of time-use and contact data within the same study allowed us to investigate how people differed in terms of daily behaviour and how their behaviour was associated with a distinct age mixing  Number of contacts   2  5  2  6  1  =  N  2  5  2  6  1  =  N  2  5  2  6  1  =  N  0  5  4  2  =  N  4  5  2  2  =  N  4  5  2  2  =  N  4  5  2  2  =  N  4  5  2  2  =  N  2  5  2  6  1  =  N  0  4  8  1  =  N  0  4  8  1  =  N  0  5  4  2  =  N  0  5  4  2  =  N  0  5  4  2  =  N  3  8  =  N 3  8  =  N  3  3  2  1  =  N  3  3  2  1  =  N  3  3  2  1  =  N  0  4  www.nature.com/scientificreports/ and contact numbers. At least 70% of participants spent most of their day at home or in the village, without any time at work or at school, and reported the lowest number of contacts (8.45 on average), 74% of which on the premises of their home. Intergenerational mixing patterns were found to be less relevant for Workers and Schoolers and more substantial for Homestayers, Walkers and Commuters. On the one hand, the level of intergenerational mixing increased for people spending more time farther from home, suggesting that people who spend most of their time and report most of their contacts in the general community might be relevant for the transmission of COVID-19 as they are characterised by a larger number of interactions between teens and adults and with older adults. These behaviours, as well as those of Workers and Schoolers who concentrate their contacts among people of the same generation, may be those mostly affected by social distancing and isolation, which are strategies designed to reduce the number of contacts with the most vulnerable age groups. On the other hand, the rather large mixing reported by older adults with children at home may be more difficult to control, as people who spend most of their time at home are less likely affected by public restrictions targeting individual movement, social activities, and school closures 50 .
Our results highlight that the higher number of reported contacts and the higher level of intergenerational mixing, especially between children and older adults in the rural site, as opposed to the urban slum, and mainly reported by individuals who spend most of their day at home or in the general community, may have important consequences for COVID-19 disease transmission to older adults. Preliminary results on the spread of the COVID-19 pandemic in Kenya showed that the first wave (April-August 2020) was particularly concentrated within those individuals of lower SES living in the urban areas, such as individuals working in the informal sector, likely less able to restrict movement, while the successive wave (October-December 2020) also spread among the higher SES in the urban areas, as well as in the rural areas 51 . However, at the moment of writing (summer 2021), the little available information on the age distribution of the COVID-19 cases, mostly due to the poor case reporting, the demand-driven testing, and a focus on younger people 52 , does not allow us to properly assess the epidemiological impact of the infection on the different generations, especially the older adults. Nonetheless, we believe that, considering the heterogeneity in contact numbers and social mixing entailed by differences in time use behaviour, rather than simply looking at the variations in contact patterns per location, may achieve a better identification of groups of individuals at higher risk of transmission, define effective measures to prevent the spread of the diseases, and design appropriate targets of future vaccination efforts. Study design. The study design was cross-sectional, targeting recruitment at the household level, and was carried out through a series of face-to-face interviews with trained fieldworkers. Individuals of all ages were grouped into seven age classes reflective of key social or behavioural groups: < 1 year (infants), 1-5 years (preschoolers), 6-14 years (primary school), 15-18 years (secondary school), 19-34 years (younger adults), 35-59 years (middle-aged adults), and > 60 years (older adults).

Methods
Estimates from a previous contact study conducted in Kenya were used to parameterize the sample size calculation 10 . To account for the larger heterogeneity in the social structure of the urban area in terms of household composition and social interactions, 1000 participants were targeted in the urban area and 500 from the rural area. Although the sample size was allocated to the seven age groups proportional to the population of the corresponding age strata, children were oversampled due to the critical role they were expected to play in infectious disease transmission and hence the need to reduce the level of uncertainty in the estimated contact rate for this age group. Recruitment was staggered over a six-month period (June 2015 to December 2015). Inclusion criteria included (i) giving informed consent, either directly or by parents if participants were minor of age, and (ii) planning to remain in the site for at least two weeks. Information sheets and consent forms were provided to all participants.
Each survey respondent was assigned two randomly picked consecutive days and requested to keep track of social contacts and visited locations during these 2 days by keeping a mental note, for those with fewer contacts, or using a tracking sheet on which they could write a few details such as name initials, presumed age, time and location where they met the contacts, colour of clothes or any other details that would to help them recall. During the interviews with the fieldworkers, background information as well as data on social contacts and time use were collected, revised, and entered into an electronic questionnaire.
Background information on the participants regarded their age and sex, their education career and occupational status, their HH size and characteristics (such as owned assets, use to derive a SES measure), and their school and work environments (e.g., distance from home, school class and workplace size).
Similarly to previous studies 2,12 , contacts were defined as an interaction between two individuals, either physical (involving skin-to-skin contact), or non-physical (involving a two-way conversation with three or more words in the physical presence of another person, but no skin-to-skin contact). For each contacted individual, www.nature.com/scientificreports/ participants provided information on the sex and the age of the contacted person, the type of contact, the location(s) where the contact occurred (home, school, work, general community), and the relationship with the contacted person (e.g., sibling, parent, non-relative). To distinguish short-lived contacts with long-duration contacts, and to account for the possibility that an individual may be encountered several times during the day, we collected the information on the total duration of the contact with such an individual. Finally, participants provided information on their time-use by recording all the visited locations (the same used for contacts) during their day. For this purpose, each survey day was divided into nine time slots reflecting the position of the sun, spanning from a minimum of two hours (dawn, early morning, late morning, midday/lunch time, early afternoon, late afternoon, evening, and dusk) to a maximum of eight hours (late night). Respondents reported whether, for each time slot, they had visited or not each of seven given locations (home, school, work, and the general community within the village/estate, within the study site, within the county, or outside the county), which resulted in a matrix of 63 dummy variables for each combination of time slot and location.
The study was approved by the Kenya Medical Research Institute-Scientific Ethics Review Unit (KEMRI-SERU) and the University of Warwick Biomedical and Scientific Research Ethics Committee (BSREC). The parent study, DECIDE, had received ethical approval by the ERC Ethics Committee and the Ethics Committee of the Italian National Institute of Health. Individual consent was obtained from each of the study participants. All research was performed in accordance with relevant guidelines and regulations.

Statistical analysis.
We analysed all collected data by adjusting for sampling weights (the inverse of the probability that an observation is included because of the sampling design), which were calculated for each site separately to compute site-specific estimates. These weights were based on both the KHDSS and the Mombasa County population age distribution, taken as reference populations.
We created two new variables for inclusion in the statistical models: a continuous SES index and a TU profile variable. The SES index was computed by applying principal component analysis to data on household characteristics and owned assets, also collected in the study, and taking the first principal component as the index 55 .
To construct the TU profiles, we applied principal component analysis to the collected TU data-a set of 54 dummy variables (nine time slots for six locations, as the location "general community within the village/estate" was excluded and kept as baseline) to summarize all the information contained in the variables with a smaller number of dimensions, constructed to be independent of each other, while retaining as much information as possible. We used the obtained components as input for a hierarchical cluster analysis to create groups of persondays characterised by similar time use behaviour profiles 43 .
Social contact matrices by study site and TU profile were constructed to show the average number of contacts between 5-year age classes in the respective populations (plus a single age class from 75 years or more). Each matrix element, m (rec.) ij , contained the mean number of contacts per day in age class j as reported by participants in age class i . To take into account the reciprocal nature of contacts 1 , the elements of the matrix were adjusted for reciprocity at the population level (see the Supplementary Information for further details on the construction of the contact matrices).
We measured the age-assortativeness with the Q index 41,56,57 , which is calculated as Q = [Tr(P) − 1]/(n − 1) , where P is the matrix whose elements represent the fraction of total contacts of age group i with age group j , P ij = M ij / j M ij , M ij is the matrix with the number of total contacts between age groups, and Tr(•) is the trace of the matrix, i.e., the sum of its diagonal elements. The Q index takes values one as the assortativeness becomes maximal (all contacts are on the main diagonal, i.e., with individuals of the same age), while, for increasing homogenous (or proportionate) mixing, it tends to zero 39 .
Generalised estimating equations (GEE) were chosen to evaluate the effect of the different participants' characteristics on the daily number of contacts 58 . The negative binomial distribution was preferred over the Poisson to allow for a higher variance in the number of contacts reported by participants than what expected under the former distribution (overdispersion) 2,12,59 . We tested, however, for the possibility of reducing the distribution to a Poisson by checking whether the 95% confidence interval (CI) of the estimated overdispersion parameter contained zero. Contact data were clustered within participants over the survey days under the assumption that responses from the same individual were correlated, while responses from different participants were assumed to be independent.
We fitted the GEE model to all contacts and then separately for each of the three settings. Except for the setting, which was included only in the overall model, the variables included in all models were the sex, the age group (0-14 years, 15-59 years, and 60 years or more), the HH size (approximately grouped in five quantile groups, i.e., 1-3, 4, 5, 6-7, and 7 or more individuals), the living arrangements (one-generation, two-generation, or three-generations HHs), the SES index, the current education level, the TU profile, and whether the contact day was a weekday or a weekend day. We interpreted the exponentiated model estimates from the negative binomial GEE as incidence rate ratios (IRR), giving the relative number of contacts per day with respect to the reference category of each variable, and we accompanied them with 95% CI.
Data cleaning and wrangling, matrix construction, and result visualization were carried out in R (using, in particular, the package "FactoMineR" for the PCA and the HCA 60 ); GEE models were estimated in Stata, using the procedure "xtgee".

Data availability
The datasets and the source codes necessary for the reproducibility of our results are available at the following repository: https:// osf. io/ ebwrh/? view_ only= 19d34 32658 3e47a 19510 668c6 e1c17 e5.