This paper presents an experimental test of a common assumption in the phonological literature on Lexical Conservatism, namely that the effects of the phonologically-advantageous stem allomorph that licenses a markedness-reducing phonological alternation in a derived form yield outcomes where the derived form specifically resembles the licensing allomorph. We report the results of a wug-test with adult native speakers of English that uses trisyllabic stimuli, and find that the base-specificity assumption is supported: derived forms licensed by stem allomorphs with penult stress exhibit greater propensity for penult stress than those with no licensing forms and those with final-stressed forms, and derived forms licensed by final-stress stem allomorphs exhibit more final stress than those with no licensing forms and with penult-stressed forms. We fit formal grammatical models to the experimental data implementing three different models of Lexical Conservatism from the literature, and use model comparison techniques to quantify how much support each theory receives from the data. We find that of the three theories investigated, only the Voting Bases model proposed in Breiss (2024) is flexible enough to capture the salient characteristics of the experimental data.