TWO BAYESIAN STATISTICS SPOOF SONGS, AS PERFORMED BY GARETH ROBERTS AND JEFFREY ROSENTHAL AT THE "VALENCIA 7" CONFERENCE IN TENERIFE, JUNE 2002. ========================================================================= "CONFUSING PRIORS" (to the tune of "Confusing Wonder") In these days of confusing priors I have opinions but I don't know why Based on hunches, and vague intuition I'll believe them 'til I die I'll believe them 'til I die I once was so objective I once let the data speak But I've found, that tight prior bounds Ensure the results I seek In these days ... I once used a uniform prior So simple, easy to defend But transformed, it gets deformed Conclusions start to bend In these days ... ========================================================================= "AN MCMC SAGA" (to the tune of "Jailhouse Rock") Had some data ready to inspect I modeled the relation as a random effect The number of parameters just grew and grew I had to get some help from you-know-who Run run, Markov chain run Programming you was fun, but I'll be happy when you're done! I coded up a simple MCMC To do all the difficult work for me The sampler went funny and refused to mix 'Caused me a problem that I couldn't fix! Run run ... It seemed that my posterior required more A super-duper sampler it couldn't ignore With langevin, and tempering, a hybrid chain I had to tweak it again and again Run run ... I knew that my algorithm was no joke When my computer started spewing smoke My plan wasn't working so I had to sub I drowned my MC sorrows at the local pub! Run run ... (repeat twice) =========================================================================