Individual plant size often determines the vital rates of growth, survival and reproduction. However, size can be measured in several ways (e.g. height, biomass, leaf length). There is no consensus on the best size metric for modelling vital rates in plants. Demographic datasets are expanding in geographic extent, leading to choices about how to represent size for the same species in multiple ecological contexts. If the choice of size variable varies among locations, inter-population comparative demography increases in complexity. Here, we present a framework to perform size metric selection in large-scale demographic studies. We highlight potential pitfalls and suggest methods applicable to diverse study organisms. We assessed the performance of five different size metrics for the perennial herb Plantago lanceolata, across 55 populations on three continents within its native and non-native ranges, using the spatially replicated demographic dataset PlantPopNet. We compared the performance of each candidate size metric for four vital rates (growth, survival, flowering probability and reproductive output) using generalized linear mixed models. We ranked the candidate size metrics based on their overall performance (highest generalized R2) and homogeneity of performance across populations (lowest total magnitude of, and variance in, population-level error). While all size variables performed well for modelling vital rates, the number of leaves (modelled as a discrete variable, without transformation) was selected as the best size metric, followed by leaf length. We show how to interrogate potential trade-offs between overall explanatory power and homogeneity of predictions across populations in any organism. Synthesis. Size is an important determinant of vital rates. Using a dataset of unprecedented spatial extent, we find (a) consistent size-based models of growth, survival and reproduction across native and non-native populations of this cosmopolitan plant species and (b) that several tested size metrics perform similarly well. This is encouraging for large-scale demographic studies and for comparative projects using different size metrics, as they may be robust to this methodological difference.