Over the weekend I had the pleasure of viewing the Rembrandt in America exhibit at NCMA. The exhibition was full of facts about Rembrandt (and paintings, of course). What most struck me was the question of "connoseurship"--whether Rembrandt had actually put brush to canvas to create a given painting.
[caption id="" align="alignright" width="281" caption="Rembrandt Himself"][/caption]
Estimates of the actual number of Rembrandt paintings range from about 250 to over 700. The lower extreme is taken by the Rembrandt Research Project, begun in 1968. On the high end is NCMA's own director from the 1950's, William Valentiner.
There is some "true" number that may never be actually known. Sound familiar? This is exactly the situation that statisticians encounter every day.
An important first question to ask yourself when trying to make a judgment under uncertainty is, do you care more about Type I (false positive) or Type II (false negative) errors? In the case of Rembrandt, the answer depends on who you are. Museums and private collections, whose values depend on authenticity, likely care much more about avoiding Type II errors. Scholars with an eye toward accurate attribution likely care more about avoiding Type I errors.
I don't currently have any data on hand to arbitrate between the two sides, but I would love to see their estimates of uncertainty. NCMA did a great job bringing this issue to the public in an accessible way. Even in the art world, statistics is a rhetorical practice.