The Curious Statistician

Gregory Berns and Sara Moore of the Economics Dept. at Emory University downloaded 120 relatively new and unknown songs , evenly spread over six genres, and had 32 teenagers listen to short clips while undergoing an MRI scan of the brain.  The teenagers also filled out a short assessment of how much they liked each song.  Over 3 years later, the researchers reviewed sales figures for the 120 songs.

The songs that did well were not the ones that the teenagers said they liked (in statistical terms, the likeability survey had no predictive value).  However, increased neural  activity in the orbitofrontal cortex and the ventral striatum did correlate with song sales.   So, although the teenagers' conscious perceptions predicted nothing, their unconscious brain activity did.

Good news for MRI vendors, and for neural researchers eager to make the leap to popular culture.

One important technical note, on the statistical side.   The logic in this experiment runs as follows:  Let's identify the best selling songs, then look at the brain images for all teenagers and all songs, and see if we can find any activity patterns that are especially prevalent for the best sellers.  Given the complexity of the brain, some interesting patterns are sure to emerge merely as a result of chance, the same way that (say) a hundred individuals flipping a coin ten times will produce some interesting patterns.

This is known as the "multiple comparisons" problem, or, more generally, the multiplicity problem.  The probability of finding something interesting just by chance increases with the number of variables (brain areas, in this case), the number of questions being asked, and the number of statistical tests being performed.  This study used a Monte Carlo approach (and a program called AlphaSim) to control for this tendency and set a significance threshold to minimize the chances of being fooled in this way.

The real proof of the pudding, though, will come when a record company starts its own lab and tries out this theory on a new set of teenagers.  I'm hoping it will not pan out.  The music industry has always had a strong element of producing music via formulas and focus groups.  Will singer-songwriters now have to have their own MRI labs?