At Bocconi, I’m teaching some classes in marketing research. A core component is to give students an idea of experimental design. It’s one of my favorite lectures, as it invites to open discussions and learning-by-doing. Last semester, I asked my students to start two new and near-identical Tinder profiles, come up with an interesting hypothesis of what constitutes a successful Tinder profile, and then start swiping right and counting which profile actually got more matches.
We had some interesting and some weird results.
For instance, one group found that showing yourself at the gym vs. not doubles your matches. If you’re a girl, that is. Bad boys score better than good guys. You get more matches if you’re indicating interest in something short-term vs. something long-term. Sorry, Sean Rad. If you’re hot, people will ignore that your description shows you as a homophobic racist, one group found. You could be a total jerk. Or you could openly state that you have a prosthetic leg: if you don’t show it in your pictures, it won’t affect your matches. (Take that, people who think that the bio actually matters – if it matters, you’re already on the defense.) But maybe Tinder is not all about sex: Two groups independently thought that a profile with only one full-body picture would receive more matches than one with only a facial picture, but the opposite was the case. And another group found that if your profile is less provocative, you might get less questions, but they tend to be pleasant and interested rather than dirty and hardcore.
One could rightfully point out that Tinder is not the best background for this kind of exercise. It requires students to connect their Facebook accounts, and it exposes them potentially to some very strange messages. While I would not want to ignore that the majority of students has a Tinder account anyways, and is likely exposed to much worse stuff by their own choice, there is truth in that. Tinder, like other social media, are attractive to use in research and teaching, but ethically new ground. I was actually lucky to have just one group of students in one of the three classes in which I did this experiment that refused to participate. I accepted that, obviously, and decided that I would need to find a similar, but ethically not or at least much less questionable arena.
I chose Instagram. Almost every student already uses Instagram. Students can create a new secondary Instagram account. There is no dating component, and there are not going to be creepy messages. Pictures can be as controversial or not as the students want them to be. And in fact, all groups made very uncontroversial choices. The disadvantage is that by starting with a fresh account, most Instagram posts were likely not to receive much feedback. Also, while as an outcome of the Tinder experiment one of my students received the invitation to become the bass player of a metal-rock band in Modena (this one, maybe?), such outcomes are unlikely with Instagram. Still it takes the art of experimental design and puts it right in the context of social media that they are using day-in and day-out. After all, the point is to show the relevance of experimental design with a very simple and ad-hoc exercise.
And with that, let’s jump into it in a bit more detail. I have to skip one submission of which I received only one screenshot.
Lesson 1: More hashtags, more likes. Unless weird things happen.
This one should be obvious, as Instagram is built around hashtags. When people just browse through Instagram, they will see photos that match the hashtags they have used, and they will see photos that people from their contact list liked. Note that in this experiment, there is another slight variation, i.e. one photo indicates some place in Thailand as the location, the other one Maui in Hawaii.
But really? Here is another group that tested the same. And, nope, there is no difference in how unsuccessful either post was. For what to me seems an attractive motive with a good collection of hash-tags. But nothing. Not a single like. I can’t explain it.
Lesson 2: No black&white, use color filters.
Two photos from Lake Como. Well, one photo from Lake Como, but one time with a rather heavy pink saturation, and one time in black&white. None of them uses a hashtag. There are two take-aways here. First, my students followed my advice to make a trip to the lake (and so should you), and second, colors work better. But I’m not sure about this one. See, I had asked students to use similar and user names, but one of their user names – mktgresearch1234 – kind of indicates a mismatch between the profile name and the post. This user-name quite much sounds like spam: that might have been an issue here.
A second group tested the very same hypothesis. They also took a black&white photo of Lago di Como and compared it with a colored version. The only difference is that their colored version does not make use of any filter. It wins against the black&white version with 25:0 likes. And, in fact, across both groups, the outcome is 30:0. Not a single like for a black&white version. In defense of black&white: Maybe results were different if we were looking at black&white shots of something other than landscapes. Let it be people or specific objects. Someone could certainly make this a thesis project.
Lesson 3: Wear make-up? Don’t wear make-up? Doesn’t matter: You need visibility.
These screenshots were rather small and a bit difficult to read. Let me explain. On the left, there is a photo with a woman wearing light to no make-up. On the right, the woman is wearing much make-up. Or maybe it’s the other way around; the group didn’t tell me. To me, this woman is wearing tons of make-up in both photos.
But there are some other things going on. First, there are no hash-tags, only a funny tagline (“Do you ever feel like a plastic bag?”). The tagline I find so entertaining, it kept me laughing for a full minute. And then, for the right photo, we see the profile description as well. It says “Send Dm lol ;)” and then gives a link to… let’s generously call it a Russian dating website. I’m not sure that was present in both profiles. I suspect that by itself could have implied some differences had there been more visibility. But we already figured: no hash-tags, no visibility.
Lesson 4: Don’t get in the way of beautiful landscapes if you don’t show your face
This group thought that a photo with a person would get more likes than a photo of the very same landscape without a person. To their surprise, they found the opposite. Everything else was their same. Even the user names of both accounts, except for one tiny detail. Both these photos were among the three most successful ones, so it’s worth to have a look once more at the actual hash-tags. Those are many. One-by-one, they are pretty big and general, yet together they come across a bit witty.
In the context of this contribution, I think it’s important to distinguish between selfies and portraits. Also, we do not see the face of em.ilymilano. I’m not sure the effect would hold if we played around a bit with how exactly the girl appears in the picture. Please note also that the landscape-only picture has a dynamic element to it that is not visible in the landscape-plus-girl condition: the motor boat.
Wrap-up
Students got away from this with mediocre responses. In a next iteration, I certainly need to instruct them not to forget hash-tags. With a very notable exception, they matter. There were some interesting hypotheses, and I would have liked to have more groups to see more diversity.
It leaves me with a very simple question. All these posts were made from fresh accounts. We did not actually test what constitutes a successful Instagram post of a random person. Instead, we tested what contributes to the success of people who just start their Instagra profiles. And that, Instagram freshness (or social media freshness in general), might be a rather interesting avenue of research. How should you get started, and can you recover from certain early mistakes?
Interestingly, I noticed that academic studies on how things go viral online focus mostly on Twitter and Facebook. Few exceptions. Data scientists, however, are playing around with this already – inside and outside of universities. Some offer their algorithms as tools to the masses to predict how many likes they will get on their next post. Some sell it.
Onabike.cc 