Older adults across the globe exhibit increased prosocial behavior but also greater in-group preferences

Population aging is a global phenomenon with substantial implications across society1,2. Prosocial behaviors—actions that benefit others—promote mental and physical health across the lifespan3,4 and can save lives during the COVID-19 pandemic. We examined whether age predicts prosociality in a preregistered global study (46,576 people aged 18–99 across 67 countries) using two acutely relevant measures: distancing during COVID-19 and willingness to donate to hypothetical charities. Age positively predicted prosociality on both measures, with increased distancing and donations among older adults. However, older adults were more in-group focused than younger adults in choosing who to help, making larger donations to national over international charities and reporting increased in-group preferences. In-group preferences helped explain greater national over international donations. Results were robust to several control analyses and internal replication. Our findings have vital implications for predicting the social and economic impacts of aging populations, increasing compliance with public health measures and encouraging charitable donations.

for test-retest reliability over one month) and existed before the pandemic precede 71 the prosocial behaviours, which applied specifically to the recent pandemic context. We ran a 72 structural equation model for each of the three prosocial outcomes: distancing, national 73 donations, and international donations. Each had a direct path from age to the prosocial 74 measure, indirect paths via each trait factor, and the relevant control variable predicting 75 prosocial behaviour (perceived risk for distancing and subjective wealth for both types of 76 donations). Donation amounts were logit transformed and all variables were z-scored as in 77 the main models. Covariances between the trait factors were also included in the model. For

103
Results are the same when excluding the 10% of data available for preliminary analysis 104 105 Before our preregistration, we received a randomly selected 10% of the overall dataset and 106 ran preliminary analyses relevant to the preregistered hypotheses. These data were included 107 in the full dataset to increase power, particularly for country-level effects. Participants whose 108 data were in this 10% were divided evenly between the two subsamples. We also showed that 109 the key findings are the same when excluding these participants (see Table S2). Results in the main text are reported from analyses using participants' raw age as predictors.

115
We also tested whether key findings remained the same when using participants' age adjusted having a non-binary gender or did not report their gender, the life expectancy for the whole 120 population in the country was used. All results reported in the main text for raw age were the 121 same when using age adjusted for life expectancy (Table S3). Adjusted age predicted higher 122 distancing scores, larger overall donations, and more national bias in giving.

143
show fitted linear models, shaded areas show 95% confidence intervals.

The relationship between age and donations to national & international charities is 145
robust with or without controlling for subjective wealth 146 147 As described in the deviations from the preregistration, we included a fixed and a random

COVID-19 severity and country-level wealth predict total donations and national bias in 194
giving 195 196 In addition to the three-way interactions between age, charity location and COVID-19 deaths 197 predicting donations reported in the main text, the model also showed significant two-way 198 interactions and main effects (Table S6). Two-way interactions between COVID-19 severity 199 and charity location showed that higher death totals, both in the participants' country (β=0.06   Table S7).

Individual difference measures correlate with age and prosocial behaviours 219 220
To test the relevance of the individual different factors for age-related changes in prosocial 221 behaviour, we calculated correlations of the factor scores with age and the prosocial measures 222 (Table S8 and see Results). We next tested whether there were differences in the strength of 223 correlations between the prosocial measures for each factor. As this analysis was exploratory, 224 we only report differences significant at p<0.0001 Bonferroni-corrected. For positive traits, 225 negative traits, and interpersonal morality, these comparisons showed significant differences 226 in the absolute size of the correlations such that distancing > national donations > international 227 donations (Table S8) Table S9

271
Indirect paths (a & b) where the overall indirect effect (a*b) was not significant are in grey (for 272 example general morality significantly predicted prosocial behaviour but general morality is 273 not predicted by age so there is no indirect effect).