Source apportionment models have been extensively applied to ambient pollution data to estimate source-specific pollution. With increasing availability of personal and low-cost air pollution monitors, pollution can be measured at multiple scales (e.g., personal and ambient). A remaining question is how to combine data from these scales to better estimate source-specific air pollution for personal exposure studies. Source chemical compositions, represented by profiles in source apportionment models, should be similar across multiple scales. However, variations can occur because of source characteristics, for example traffic chemical composition can vary by the mixture of gasoline vs. diesel vehicles. We propose a Bayesian source apportionment model that can incorporate multi-scale data and model differences in source profiles between scales. Instead of fixing source profiles to be the same between scales, we allow deviations between source profiles. We demonstrate our model and compare it to existing Bayesian source apportionment models using personal exposure and ambient pollution data from a commuter study in Washington, DC.