Tinder Experiments — Answering Questions and Criticisms Part I: Mostly Questions
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. This is the first article in what I hope to be a short series. Sorry it took so long. Better late than never!
Whazzup Nerd, how did you analyze your data?
Stian Pedersen discovered something serious:
Cute, but a serious error in this analysis. How do you define payoff? Is it an awkward hookup, a few dates, a long term relationship? This matters to any analysis of “Tinder economics”.
…
In this analysis the payoff was getting an initial like. “Likes” are the currency in my model of the Tinder economy.
…
Nick Vance asked about methodology and understanding (and Alex Albright also had a similar comment):
I’m curious about the methodology you used. The way I read your ‘caveats’ section, it seems like you got 27 data points: each of which is the percentage of men that each woman ‘likes’. And then you extrapolate that [sic] data into your graph. Am I understanding that correctly? And how did you extrapolate that [sic] data into the graph you show?
…
You are correct. I got 27 data points from women that represented the distribution in the percentage of men individual women liked on Tinder. I then organized these data into rank order and calculated the corresponding rank order percent for women. I then graphed the male percent distribution versus the female ranked order percent. I labeled the female ranked order percent as “Female Attractiveness” although it is probably more accurate to label it something like “Female Pickiness.” So, in that first graph you mentioned, male attractiveness is defined by how many likes a man gets from women. In contrast, female attractiveness is defined by how many likes a woman gives compared to other women. I would assume this is highly correlated to other measures of attractiveness in general (but that is only an assumption). From these data I was then able to calculate the Lorenz curve by integrating the attractiveness data and doing a little algebra.
…
Garance Cordonnier asked about the differences between women and men:
Maybe you could conduct the same experiment with men’s likes in parallel, because you are only considering half of the data. It also implies women’s likes as the currency of your modelized economy which tends to represent the dating app world as a market of women (they being the sellers and men the clients somehow) which is kind of biased. I would like to know what result the opposite experiment would give.
…
Someone from the dating app Hinge saw my article about Tinder and did the same analysis for their user data. They were able to look at trends for both women and men. I talked at depth about it in an article on my medium account. The trends for men between Hinge user data and the Tinder data I collected were almost the same (surprisingly close actually). The trends for likes received by women on Hinge were much more uniform than likes received by men across different levels of attractiveness, although more attractive women still got significantly more likes than less attractive women. You can read my article on the Hinge data if you want more details.
Data from OkCupid also seem to indicate that there is also a clear trend between increasing attractiveness and increasing number of messages received for both men and women (unsurprisingly). In the case of OkCupid though, the relative size of this trend between women and men seems to be reversed. The ratio between the number of messages received by the most attractive women versus the least attractive women is higher than the ratio of messages received by the most attractive men versus the least attractive men according to the reported OkCupid user data.
Sup Bro, can I get some of those sweet data?
Ivano Malavolta asked to play:
Can you post also the raw data so that we can play with it as well? :)
Luis Badillo asked about availability:
Are you making the data available? I think some really cool data visualizations can come of this.
…
I wrote the initial articles 7 years ago and so I don’t think I have any of the raw data around still. It probably wouldn’t be super useful beyond what I already presented since it was such a limited data set. I was able to find a few of the plot files used to make graphs after digging through old files. If you have any specific visualizations you would like to see I might be able to create them by reverse engineering the analyzed data sets. If you are interested in more visuals and data analysis I would recommend my article analyzing a similar study done by the dating app Hinge.
Hey Daddy-O, speaking of data, how did you make those pretty graphs?
Claudia Silva Cabrera asked about software:
I just wanted to know if the graphs were created using Excel or another tool, such as GGPLOT or something in Python?
…
I created most of the graphs using the freeware graphing program SciDAVis. Some of the graphs were made with Excel when I was feeling lazy.
Yo Dawg, what’s up with that widget?
Tim Kenney asked for widget explication:
Your widget isn’t explained. And review of the article in the widget explanatory link just takes you back to the original article. How are we to calculate the percentage to enter in your widget? By the ratio of women we liked vs liked us? By the total number of women on Tinder? (Unknowable). Please explicate?
…
The correct input into the widget is the number of women that liked you divided by the number of women that you liked multiplied by 100. So, if you liked 200 profiles and got 6 matches you would input 3 into the equation:
attractiveness% = 16.8*ln(like%)+52.3 = 16.8*ln(3)+52.3 = 70.7%
This means that around 70% of men get less than 3 likes from every 100 profiles. It was only meant for brand new accounts in which you like every profile you see without discretion. Both the Tinder algorithm and the number of female bots has probably changed quite a bit in the last 7 years. Therefore, the true equation might be different now, but I am guessing the old one it is still close enough to let you know if you are below average, average, above average, or very high. An average person will get around 1 like in 100 and anything above 10 likes in 100 is very high.
…
Ludvig Westerdahl commented on my formulation:
The formula: y=16.8 * ln(x)+52.3 that you used is not correct in terms of percentage. When x = 100, y ~= 130 unless that’s what you were going for.
…
It actually is what I was going for. There are a lot of female accounts on Tinder that either are bot accounts or aren’t in active use. Therefore, one would expect that no man would obtain a 100% like percent. The widget takes into account data I collected in a previous study that shows that the most attractive men receive about 22% likes. In this case an attractiveness percent higher than 100% would be expected if x > 22.
…
Peter Kirk asked me to explain the impossible:
So, while this is interesting, I just tried it out — I got 19/100 matches. Your calculator puts me in the top -1.767%, which seems… impossible?
…
The widget is just a logarithmic fit of the data I collected and analyzed. The actual data doesn’t fit a logarithmic curve perfectly so there is some error in the equation. This is especially true near the high end of the curve. The error at the top is probably plus or minus 3 like percent. Needless to say, if you are getting 19 matches per 100 likes you are at the very high end of attractiveness of male Tinder users.
I could have avoided that problem if I had normalized the data (as described above) based on the projected fit instead of based on the pre-fit data. In this case the new formula would be
attractiveness% = 16.1*ln(like%)+50.3
For 19/100 matches that would equate to 97.7% attractiveness (not too shabby). In this new formula the average male Tinder user would fair slightly better than before and match with 1 in every 102 female users instead of 1 in every 115. It wouldn’t affect the trends or overall conclusions of the analysis though.
Dude! Why are you using misogynistic language? Not cool!
Tuzgai asked about my word choices:
Curious why you opted to call men ‘men’ and women ‘females’.
Sam Watson also commented on my word choices:
Ahh yes, ‘females’ is how women prefer to be described, yup.
…
This is a good conversation. After doing a little research, what makes the most sense to me grammatically is to strictly use women and men as nouns representing adult humans, and female and male as adjectives (since female and male are adjectives and woman and man aren’t). I have already started to do this in my latest posts (including this one). Or maybe I will just start using chicks and dudes, guys and dolls, and ladies and gentlemen — that seems like something a male would do.
I never meant for the content of my article or the language in it to be misogynistic, although I can understand how some people interpret it that way sometimes. I used both “female and male” and “women and men” interchangeably throughout the article. This does seem to be in line with the dictionary definitions and histories of these words, especially as used in scientific literature. To be fair to the question though, I did find cases where I used “men” (or guys) with “females” in the same sentence. There was no malevolent intent on my part, but in these cases I should have, at minimum, used the same structure for both sexes. As I stated in the previous paragraph, I have already changed the way I use these words going forward.
I found a good article from The Merriam-Webster Dictionary discussing the history of these words and their usage if anyone is interested in learning more:
https://www.merriam-webster.com/words-at-play/lady-woman-female-usage
Chill Man!
Tiffany Batsakis commented on my lack of geographical knowledge and worldly experience:
Totally off topic, but you should go to the Seychelles. Most beautiful place I’ve ever been. You’ll be too preoccupied to even think about Tinder!!
…
This sounds like a really good idea!
A reflective way to end…
Saurat Xiuhcoatl is asking the bigger questions :
I guess my question will be one of a reflective nature: what can those of us men who are ‘normal’, or just average (on a scale of 1 to 10, between 5, 6 and 7), do? I’m not looking for an absolute, epiphany-enabling answer, but I would at least like to hear a recommendation.
…
I am better at analyzing data than giving advice, but maybe use a dating app that better lets your amazing personality and good taste shine through, or you can join an afterwork social netball team.
Also, I hear visiting the Seychelles is nice.
