OddThinking

I was using Google Search today to search for a person’s first name. (No, I wasn’t playing a Name Game, but I was trying to figure out the most popular spelling of a name.) I noticed a new feature had been added to Google Search. The search results may include images from Google Images.

I noticed this because when I searched for a woman’s name, the top result was a set of bikini-clad glamour shots. (For the record, I checked and found I had the default ‘moderate’ behaviour for ‘SafeSearch’, which is Google’s filtering against explicit sexual content.)

I noticed that I got pictures of women when I searched for women’s names, but no pictures when I searched for the male equivalent.

Here’s a forward peek at what I mean. The 80th most popular woman’s name is Erika. Coincidentally, the 80th most popular man’s name is Erik. At the time of writing, Google returns glamour shots for Erika but not for Erik.

Is Google objectifying women, by suggesting the most important thing about them is photos of their bodies?

I decided to do an experiment on the sexism of Google Search. I promised myself I would publish the result here whatever it was, in an attempt to avoid publication bias.

I went to the US Social Security Data, which I have found useful before. I extracted the top 100 male and female names for 25 year old Americans.

I wrote some software to search Google for each name and record if images were the top link.

I knew at the time I wrote it that it was violating the Google Terms of Service. I granted myself an exemption from the rules, because it was only a small batch of searches, it was a once-off run (not an ongoing tax on the servers) and because I was testing Google itself – Google shouldn’t be able to avoid scrutiny by pointing to small print. Unfortunately, the Google web-site didn’t recognise my self-declared exemption, and immediately detected my searches as foul play (most probably the User-Agent string gave me away). Rather than pushing my luck and being more stealthy, I reverted to a slower manual method, which meant a smaller sample size.

I decided to limit the sample size to 25 of each gender. In 50 names, only three returned images (one male, two female). This wasn’t enough to draw a conclusion.

Rather than publish a dead-boring blog article, I decided to plow on to get more data. (Danger: Publication bias slipped back in here!)

I pondered whether the popularity of the names meant there was so many pages to choose from with high Page Rank, that Google didn’t need to resort to Google Images. I moved to the bottom of the list and repeated the experiment – e.g. the 76th to 100th most popular male and female names.

This time, I got eight sets of images, one male and seven female.

So overall, 4% of men’s names and 16% of women’s names were associated with images. The men’s names (Jason and Todd) were not associated with glamour shots, while almost all of the women’s images were.

Of course, I am not really blaming Google here; it is merely reflecting the gender issues of the world-wide web. However, I thought it was an interesting result.

To get a more conclusive result, I would look at automating the process using the proper Google Search API and also use rarer names. I would also do a proper statistic analysis to work out the level of significance of this result. Obviously, more research funding is required.