How Do You Test Out A New Look? Dating Apps!

People have described a dystopian future where machines control your dating life by presenting you with images of single women and convincing you that feeding the machines will lead to a life (or a night) with one of them.

While I have no desire to live in such a world, the existence of an environment where people make snap judgments about the quality of a future mate intrigues me because it is a source of data! And when my friends suggested that more women would be interested in me if I shaved my beard, I went to the data stream to once and for all determine…

Do women within 50 miles of my surrounding area prefer me with a beard or without a beard?

Before I get into the meat and potatoes of the study, I want to provide a caveat. The results of this study do not show whether women generally prefer beards or even whether women in my surrounding location (the Bay Area) prefer beards. The study only looks at comparative interests in my beard. Who knows, maybe other people look good clean-shaven.

With that caveat out of the way, let’s get into the study!


For this study to work, I needed to reduce any bias unrelated to the beard. So to start, I took five pictures of myself with my beard in different settings with different outfits. For the purposes of anonymity (and my own amusement), I have blurred out my face and eyes in the below pictures:

This slideshow requires JavaScript.

The above photographs represent the different types of images that I felt could increase my odds of a match (and therefore increase my data pool). They include a business casual image, an image with an animal, an active image, and a social image. My friend, depicted in the social image, helped with the image capture and photo selection.

Once we finished creating the first set of images, I proceeded to shave my beard completely. We then captured a second set of images with the same outfits and same positions, but without the beard:

This slideshow requires JavaScript.

The above images represent our best attempt to recreate the initial images without the beard. This process was way more difficult that we originally envisioned and some of the non-beard images have slight variations from the beard images. Overall, the similarities were great enough that we felt the study could proceed.

The next step was to create two semi-identical dating profiles. I chose Tinder for running this experiment due to its relatively quick swipe rate and relatively high population. The profiles were relatively barebone, including a short description, age, and occupation.

In order to run both accounts simultaneously, I installed an application called Parallel Space which creates a separate environment for duplicating applications.  The free version of Tinder limits each profile to 100 swipes every twelve hours – a perfect way to control my sample sizes.

For five days (Sunday night – Friday night), every 12 hours I would open one profile, swipe right 100 times consecutively, then open the other profile and do it again. Each evening, I would add each match to an excel spreadsheet with additional available data.


First let’s start with the overall results. What do the dating women of Tinder statistically prefer to see on my face: Beard or No Beard?

Beard v. Non Beard

The vast majority of the matches I received (64 out of 94) were on the beard profile. These results do not shock me in the least. I have never been able to pull off a clean-shaven look and I have a glorious red beard that is a shame to lose. Where things get fun is in the breakdowns:

Age Distribution of Beard MatchesAge Distribution of Non Beard Matches

The age distribution interests me because it appears to follow a normal distribution centered around 28 (with a single spike around 32 or 33). Given that I am 31, it tells me that women on Tinder tend to be, on average, younger, or that the distribution of women interested in me tends to be younger. Either way, the distributions between the two are roughly the same and therefore provide little data on preferences by age.

Beard Matches by DateNon Beard Matches by Date

I had not originally planned to create a date chart, but I kept the data as a means of tracking inputs. When I started feeling a drop-off in overall matches, I checked back on my dates to discover that a large portion of the matches came soon after I created the account (since I created the account Sunday night, Monday was my first day of matching).

The early spike leads me to believe that higher visibility is given to recently created accounts, but only for a very short period of time. This is likely to increase interest in new users. I have also been told that making changes to an account will increase the match rate, but I have yet to test this theory.

Beard Matches by Race

Non Beard Matches by Race

Finally, the largest disparity: race. Now as a caveat, I had to make a few generalizations when taking down the data for this category. Tinder does not have a race option and the only women who indicated race were the ones who were tired of being asked. Thus, there may be some minor errors in the race category where a person was listed as one race incorrectly.

That said, there was a huge drop-off in the match rates of white women between beard and no beard.  The beard profile generally received twice as many matches from white women as the non-beard profile. While small decreases appeared for non-white women, the results were not large enough to be statistically significant, especially given the likely errors in data capture.

There was a little bit of overlap between the two profiles where the same woman matched on both. Only a few of the women who matched both profiles actually realized that they were both the same person. The ones who did generally found the whole thing amusing once I explained my experiment.

I also had a problem with data degradation caused by women unmatching my profile before the data could be recorded. Whether this was because they saw both profiles and felt like there was something fishy or because they removed Tinder I may never know.


I look better in a beard. Sorry, that was a personal conclusion. The conclusion of the study is that the beard attracts more initial interest than the absence of a beard, especially for white women.

One caveat is that every Asian woman I messaged on Tinder told me that they preferred the clean-shaven look to the beard look. When told about the initial findings of the study, they stuck firm to their preference of no beard over beard.


This study was exhausting to complete, but also very rewarding. There were a few extensions of this experiment that I thought could be interesting for future studies:

  1. Perform the experiment on a different platform. Performing the experiment on Bumble could add the data point of messages received and possibly type of messages received. Though type of message is generally subjective, it would be useful to determine if the beard has an effect on long term match interest versus short term match interest.
  2. Perform the experiment with a different option. A possibility exists that while the clean-shaven look is not particularly strong, the stubble look could be a contender. This experiment only compared full beard to no beard. Future experiments could include different levels and types of facial hair.
  3. Try different profile descriptions. I have a theory that women tend to read the profile descriptions more than men do. That said, I have always been curious how much the profile description affects the match rate. It would be interesting to perform an A/B experiment using the same images, but having one garbage profile and one interesting profile.
  4. Poll the communityI mostly performed this experiment for my own fun, but a few of my friends thought it was amusing enough to publish so I am putting it out here. If anyone has an idea for a better experiment, I would love to hear about it!

18 thoughts on “How Do You Test Out A New Look? Dating Apps!

  1. According to the ever reliable internet grapevine, Tinder reduces the visibility of a profile if you constantly swipe right. The thinking behind this is Tinder think you are either a bot or a bad match (as you are not picking carefully). It may be worth adopting a swiping patter like ‘left-left-right’ to try and account for this.

    It may be the reason for your significant decline in matches after the initial period.


  2. It may been better to represent the matches in regard to the total amount of woman who saw your profile of each age group/ race. The charts now mostly show what your local tinder userbase is like.

    Liked by 1 person

    1. Or he could have contacted Tinder direct and asked them to set his profile up different for a research project he was doing…I am pretty sure they would be extremely intrigued and help you out.


    1. I believe you misread it. He said the similarities were great enough, which is an oddly written way of saying the similarities were close enough. Or written another way, the differences were small enough to proceed.

      If written the way you corrected him as, “Overall, the similarities were small enough that we felt the study could proceed.” completely changes the meaning of what the author wrote.

      What he wrote originally was correct, just oddly phrased in my opinion


  3. Could be interesting to have a ratio match/total swipe right instead of a absolut number for each graph that you make.
    You would have to collect a set of datas for each swipe right, e.g age, but you could plot a graph that say “for the girls age 28 with my beard I have 27% of match but 12% without”.


    1. Both actually. I created two dummy facebook accounts for creating the profiles but then had to verify them using two different phone numbers. I ended up creating a Google voice line just so I could verify the second account.

      Thanks for the comment!


  4. You should maybe not care that much about your ‘mean attractiveness’ – Your look is not up for a vote. If my girlfriend likes blue shirts on me, it doesn’t matter much what the majority is saying, does it ; )


  5. You said that the spike the day after indicated to you that some kind of preference is given to new accounts. This is not true. What you are seeing is an influx of attention on what is referred to as “singles day” (11/11). Most likely a lot of lonely women brought increased traffic to the app, skewing your results for that day.


  6. Hi The Scientist,

    Interesting study. I’ve meant to do some work with A/B testing in Tinder, however I’ve wanted to make sure I could control as many variables as possible.

    I created a python command line application to automatically like or dislike tinder users based on your historical preference. It’s called tindetheus, and there is a white paper which describes how it works.

    The model would be deterministic, so the outcome of the model for a given profile is always the same. My intention was the tindetheus would be very useful for A/B testing, since you would always like or dislike the same profile.

    Also you may want to do an A/B test on brand new profiles only, since Tinder has some form of ELO, or internal ranking scheme which prioritizes how often your profile shows up. Other commentators have already hinted at this.



Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s