Bivariate associations of problem gambling with participation in particular game types often decrease when adjusting for demographics or consumption behavior (e.g., number of game types played). A summary of 14 peer-reviewed studies showed inconsistencies as well as conceptual and methodological challenges. The aim of this study was to expand previous research by a combination of (1) sophisticated feature-engineering, which disaggregates gambling intensity into facets within and beyond a game type of interest, and (2) the application of mediation models. Data comprised last year gamblers of three merged cross sectional Icelandic gambling surveys of 2007, 2011, and 2017 (N = 4422). For each of 15 game types (12month time frame), a parallel multiple mediation model was applied to disaggregate bivariate associations of last year game type participation and problem gambling (Problem Gambling Severity Index) by six mediating mechanisms: (1) demographic problem gambling propensity, (2) number of game types played, (3) gambling frequency within the type, (4) maximum gambling frequency across all types beyond, (5) usual spending within the type, (6) maximum usual spending across all types beyond. Games showed two distinct profiles via which mediator they mostly impacted problem gambling: Electronic gaming machines offline, scratch cards offline, live betting online, and poker offline as well as online impacted problem gambling mostly via gambling frequency within, whereas all other types mostly impacted via the number of game types played. The applied mediation models answer the question by which mechanism game types impact problem gambling in a more exhaustive way than previous research.