Bayesian BBQ, by Tom Freeland

The reponse to our challenge has been a good one. And, while some folks have written in with some rather implausible explanations of Bayesian Networks, we were a little concerned when our friend, Tom Freeland, wrote in with an example problem that actually makes us think that he understands BNs. At least he made it interesting for me, with the food connection! His thoughts follow. And, for more thoughts by Tom, check out his blog here.

While Baysian networks may be inadequate to resolve the question “what is good barbecue,” this approach may bring some focus to what seems to many an intractable problem. Prior approaches have tended to propose answers in terms of relatively simplistic dichotomies—inside or outside meat? Sliced or chopped? whole hog or shoulder? Vinegar or ketchup?—with complicating arguments for geographic dominance. The literature is also plagued with nuisance variables, early examples being arguments about barbecue made from meat other than pork, or with a sauce that is actually based on yellow mustard, and more recent examples arising from the marketing of gas ovens that do not use wood or charcoal as suitable for cooking barbecue.

The approach proposed by this paper would call, first, for an identification of the appropriate pairings from the literature, along with the elimination of nuisance variables, and then the creation of appropriate interrelationships in a directed graph.

A secondary problem in framing the question is that some pairs that seem entirely independent (whole hog/shoulder and vinegar/catsup) seem to spontaneously arise in tandem. Examples include the barbecue traditions of Memphis, Tennessee (shoulder/ketchup), Lexington, Tennessee (whole hog/vinegar), Goldsboro/Ayden, North Carolina (whole hog/vinegar) and Greensboro/Lexington, North Carolina (shoulder/ketchup). Even here, geography is a complicating factor, with the existence of “blended” barbecue (shoulder served with vinegar sauce in Chapel Hill, North Carolina) on the border between traditions that follow the more usual combinations.

--Tom Freeland