Monday, October 28, 2013

Gay Couples? What Are The Odds?

In this Huffington Post piece, an antique photograph collector explains that he believes he has a special collection of photographs of gay couples from bygone days.  He even blogs about it, but I won't include the link because it doesn't seem to work.

Jeffery Gent is passionate about what he does.  The article quotes him saying "Unfortunately, so many of these photos were purposely destroyed by horrified family members..." He continues "For every photo that I may have mistakenly identified as gay, thousands more were burned or torn into pieces to keep a family secret..."

To those who hated college statistics, it may not be obvious that we can examine this one pretty easily.  To those of you who hated college statistics and passed the class, remember, you performed no worse than I did.  (D in undergrad stats.  You can only imagine what now Nobel Prize winning Lars Hansen, my dissertation adviser, said when he learned this.)

We'll attempt to apply Bayes Law to see what we might know about the probability that anyone in these images is gay.  Bayes Law is a great tool for drawing statistical inferences.

Where do we start?I'll formulate an answer so you can easily play with a Bayesian calculator on your own.

What is our hypothesis? The image shown portrays two gay men.

What is the data? The data is a picture of two men.

We want to know, conditional on the existence of a photograph, the probability that the picture depicts two gay men.

What is the unconditional probability an image of two men is of gay men?  This is somewhat arbitrary (and political, potentially.)  So, let's say 20%.  It doesn't matter much.  We could say that number is high (or low) for the total population of men.  We could say that number is low (why do two straight guys get photographed together?) We could say that number is high (Don't many brothers appear in photographs together? Grooms and best men?)

The key, really, is the the claim by Gent that most photographs of known gay couples were destroyed by family and friends out of a desire for secrecy.

Images are gone for two reasons: Passive loss (accidental disposal, fire, time) and active loss ("We're small minded, bigoted people who are ashamed, so we are destroying his memory.")

We need to establish the probability that a picture exists today conditional on straight men in it, and the probability that pictures exist today conditional on gay men in it.

Suppose 50% of all images are lost passively, but that a further 25% of gay men images are lost actively.  This is incredibly conservative based on Gent's statement.  We're saying only 50% of known gay images are destroyed actively.  He's implying the vast majority.

So, in our calculator, we have Pr(Image Exists | Gay) = 0.25 and Pr(Image Exists | Not Gay) = .5.

What's the result? Pr(Gay | Image Exists) is about 11%. 

If the vast majority of images depicting known gay men were destroyed (a la Gent) and we set Pr(Image Exists | Gay) = 0.01, then it is virtually impossible that any of the images depict gay men.

Even if most of the images taken historically were of gay men (90%) and the spiteful relatives destroyed 99% of them, it remains the case that only 15% of surviving images depict gay men.

Nice collection of old pictures, though.