Predicting population collapse in the face of unprecedented anthropogenic pressures is a key challenge in conservation. Abundance-based early warning signals have been suggested as a possible solution to this problem; however, they are known to be susceptible to the spatial and temporal subsampling ubiquitous to abundance estimates of wild population. Recent work has shown that composite early warning methods that take into account changes in fitness-related phenotypic traits - such as body size - alongside traditional abundance-based signals are better able to predict collapse, as trait dynamic estimates are less susceptible to sampling protocols. However, these previously developed composite early warning methods weighted the relative contribution of abundance and trait dynamics evenly. Here we present an extension to this work where the relative importance of different data types can be weighted in line with the quality of available data. Using data from a small-scale experimental system we demonstrate that weighted indicators can improve the accuracy of composite early warning signals by >60%. Our work shows that non-uniform weighting can increase the likelihood of correctly detecting a true positive early warning signal in wild populations, with direct relevance for conservation management.