Modelling personality, plasticity and predictability in

9 Behavioural assessments of shelter dogs (Canis lupus familiaris) typically comprise 10 standardised test batteries conducted at one time point but test batteries have shown 11 inconsistent predictive validity. Longitudinal behavioural assessments offer an 12 alternative. We modelled longitudinal observational data on shelter dog behaviour using 13 the framework of behavioural reaction norms, partitioning variance into personality (i.e. 14 inter-individual differences in behaviour), plasticity (i.e. individual differences in 15 behavioural change) and predictability (i.e. individual differences in residual intra16 individual variation). We analysed data on 3,263 dogs' interactions (N = 19,281) with 17 unfamiliar people during their first month after arrival at the shelter. Accounting for 18 personality, plasticity (linear and quadratic trends) and predictability improved the 19 predictive accuracy of the analyses compared to models quantifying personality and/or 20 plasticity only. While dogs were, on average, highly sociable with unfamiliar people and 21 sociability increased over days since arrival, group averages were unrepresentative of all 22 dogs and predictions made at the individual level entailed considerable uncertainty. 23 Effects of demographic variables (e.g. age) on personality, plasticity and predictability 24 were observed. Behavioural repeatability was higher one week after arrival compared to 25 arrival day. Our results highlight the value of longitudinal assessments on shelter dogs 26 and identify measures that could improve the predictive validity of behavioural 27 assessments in shelters. 28