We show that meaning shifts in political dogwhistle expressions (DWEs) are explained by the expressions changing with regard to their “hidden” (in-group) and “public” (out-group) dimensions. We study the association between computational measures of Lexical Semantic Change (LSC) and the In-group/Out-group Ratio (IOR) of four Swedish DWEs. We use a combination of distributional modeling of DWEs in the online discussion forum Flashback and data collected from a lexical replacement survey of Swedish residents. We explore several vector-space meaning representation approaches and demonstrate that distributional methods can be used to identify semantic shifts relevant to dogwhistle development, particularly contextual representations from Swedish BERT, SBERT, and multilingual T5.