Tinder Experiments — Answering Questions and Criticisms Part II: A response to a criticism that mostly turned into a long post about OKCupid data from 2009 and 2014
I wrote an article several years ago about how Tinder can be analyzed similar to an economy if you consider “likes” as currency. To my surprise the article went somewhat viral without me even realizing it until recently (several years later). Over the years many people have left questions and comments about my original article. I figured I should take the time to answer some of their questions since they took the time to read my article. Sorry it took so long. Better late than never! This is the second part of an at least two-part series (Part I can be found here).
Ashley Gelwix () asked:
I was asked to comment on an article by Carlyn Beccia () from the sexography blog entitled, ““80% of Women Choose Only the Top 20% of Attractive Men” is a Big Fat Incel Lie (and I have the data to prove it)”. In the article the author claims that the conclusions from my original articles on Tinder are a big fat incel lie and that she has the data to prove it. What do I think and, more importantly, what do the data say?
To be blunt, I think a lot of her article is verifiably incorrect and greatly mischaracterizes my article. Based on what Carlyn wrote I am not sure if she actually read my article. Her article does contain a few interesting points, and I will try to highlight those, but it includes more factual errors and I will address those first. I am not going to go through every point she makes, but instead I will highlight a few I find the most egregious.
After a fair amount of vitriol, Carlyn’s first criticism was focused on my data collection methods:
“He got his data from polling 27 male Tinder users.
No sources. Just 27 disgruntled men. But in a world where critical thinking skills flew away with the dodo birds, 27 angry male opinions snowballed into a viral meme. “
There are valid criticisms of my data collection method, but this is not one. First, I polled a random selection of 27 WOMEN not 27 men. Now, I may not know if these women were disgruntled or angry, but I do know they were not men. Secondly, I didn’t ask for opinions. I asked for quantitative data. I pointed out in my article how these data could possibly be biased, but biased quantitative data are not the same as opinions. If I had polled men instead of women and the data were unbiased the conclusions would have been just as valid as from polling women. Finally, I admit that 27 is small sample size for this type of analysis, but it isn’t so small as to be statistically insignificant. In fact, when the dating app Hinge recreated my study using their full data set of thousands of users their results were almost the same.
Her next criticism was based on an article on OkTrends — the official OkCupid blog. The blog is no longer available, but a copy can be found using the internet archive wayback machine:
Right off the bat there is a problem with this comparison. My article was written exclusively about Tinder. Although subsequent data from Hinge backed up my Tinder claims, one wouldn’t necessarily expect all dating apps to show the exact same trends. This is especially true of dating sites like OkCupid which focus on personality matching and also allow sending direct messages before both parties have matched. To be fair to her though, by this point in her article she was criticizing how other people characterized my article as much as she was directly criticizing my original article.
Still, she seems to misconstrue the findings of the OkTrends post. She showed and described two graphs found in the OkCupid article that relate attractiveness to messaging rates:
From these graphs she concluded the following:
“Women messaged more average-looking men and were less likely to message the most attractive men.
..
This shouldn’t shock anyone but the neckbeards. Women are far more likely to message those average-looking 2s and 3s because women don’t care as much about a man’s appearance.”
Unfortunately for Carlyn, this is not at all what those data say. To be fair, I think these graphs are a little misleading, but they were made how they were with a purpose in mind. They were configured to demonstrate how men and women differ in the way they rate attractiveness and how that is different from who they tend to message. Carlyn’s conclusion is incorrect because the two lines on the graph can’t be viewed independently. The attractiveness distribution affects the message distribution. The most attractive men get a lower percentage of total messages not because women prefer to message less attractive men, but because women don’t rate very many men as highly attractive to begin with. Christian Rudder shows what happens when you combine the two lines later in his blog post:
“Finally, I just want to combine the two charts to emphasize how much fuller the inboxes of good-looking people get. I have scaled this graph to show multiples of messages sent to the lowest-rated people. For instance, the most attractive guys get 11× the messages the lowest-rated do. The medium-rated get about 4×.”
From this graph we can see that the most attractive men and women both get far more messages on average than their average looking counterparts. This was basically the whole point of my original article. The most attractive female and male OkCupid users both get about 4 times the number of messages/week as the average women and men. The most attractive women get 25 times the number of messages/week as the least attractive female OkCupid users, whereas the most attractive men get about 11 times the number of messages/week as the least attractive male OkCupid users. It is true that men seem to be more superficial than women on OkCupid, but I never said that men weren’t superficial. It can be perfectly true that a high percentage of attention goes to the most attractive people for both men and women. This would at most indicate that dating sites might be just as horrible for women as they are for men. I think some of the data coming up next shows why this isn’t necessarily the case (as well as the data from Hinge that I discussed in a previous post). We will see that it is much better to be a woman on a dating site than a man.
Even multiplier graph is still somewhat deceiving. The attractiveness scale is not linearly proportional to attractiveness percentile and therefore not the same for men and women. Christian Rudder talked about this more in depth in his book “Dataclysm” (2014) in “Chapter 7. The Beauty Myth in Apotheosis.” In that chapter he shows similar data to those previously mentioned, but now normalized to attractiveness percentile instead of attractiveness rating. In effect, he linearizes the attractiveness scale. He also graphs messages in an absolute scale instead of a normalized scale. This shows the true differences in the messages men and women receive on OkCupid (or at least the messages they received in 2014).
So, what can we learn from these data? First, there is (unsurprisingly) a steady rise in messages received as a function of attractiveness percentile for both women and men. Second, there is also an abrupt rise for both men and women at the highest percentiles (above 80%), but the rise for women seems more dramatic. Third, we can estimate that there are approximately twice as many male OkCupid users as female OkCupid users by comparing the total number of messages in the men, women, and all categories. Finally, we can see that total number of messages received by men is significantly less than the total number of messages received by women. Women in the 30th attractiveness percentile receive more messages than even the most attractive men. This is a major component of why being a man on dating apps is so terrible in general (especially for men below the 80th percentile).
So, now that we have a nice data set for OkCupid, how does that compare to the Tinder and Hinge data? I digitized Rudder’s graph using and analyzed the data the same way I analyzed the original Tinder data I collected. First though, I calculated the messages/week normalized to the most attractive users for each gender as a function of attractiveness percentile. The normalized data show that the relative messages/week for men and women compared to their highest attractiveness profiles are fairly similar. The rise in messages/week for the most attractive women isn’t as dramatic as it first seemed.
Next, I graphed the Lorenz curve for the three data sets and calculated the corresponding Gini coefficients. For more information about this analysis I will refer you to my previous posts (here and here). The Gini coefficients are 0.398 for women, 0.380 for men, and 0.399 for all. The Gini coefficient for men on OkCupid is similar to the Gini coefficient for women on OkCupid and is much smaller than the Gini coefficient for men on Tinder (0.58) and Hinge (0.54). The Gini coefficient for female OkCupid users (0.40) is almost the same as the Gini coefficient for female Hinge users (0.38).
Lower Gini coefficients indicate a more equally distributed economy. These results show that the distribution of messages on OkCupid for men is much more equally distributed than the distribution of likes on Tinder for men. This could be for many reasons. OkCupid is built upon finding people with similar interests and ideas as you. Maybe this system is working. It is also possible (and likely I would guess) that the distribution in OkCupid messages/week is more equally distributed than Tinder likes because the men of lower attractiveness send more messages than attractive men do. The Tinder data assumed that attractive men and unattractive men would swipe right on the same set of women. If an average man on Tinder swiped right on twenty times as many women as an attractive man the average man might get a similar number of likes as the attractive man in return.
These data also demonstrate one of the biggest problems with the Lorenz curve and the Gini coefficient. They only show wealth distribution in an economy — they don’t account for the total wealth of the economy. So, in this case the Gini coefficient for men and women are almost the same although women receive 8.8 times as many messages as men on average. The Gini coefficient only tells you how big your slice of pie is compared to the other slices of pie; it doesn’t tell you how big the pie is to begin with.
In conclusion, I don’t see any data that persuasively counters the data I presented in my previous blog posts. Additionally, there are several assertions in Carlyn’s post that are either misinterpretations or just plain incorrect. Still, I am glad I examined her post deeper. There are indications in it that it might be possible to create a better dating app than Tinder. I hope someone tries. The rest of the post by Carlyn diverges into a discussion of how men are becoming obsolete. There are some interesting points in that section, but it doesn’t directly relate to my article, so I won’t delve into it here.
