One of the lessons the pandemic has reinforced is to be an active, questioning reader of data presented as fact in articles and news. Not in the crazy, conspiracy-theorist “do your own research” in which you hunt down articles that only support your point of view no matter how questionable the source, but in true academic inquiry, true curiosity. I was reading a post by Rand Fishkin the other day which cited a few stats that posts on Instagram which used “link in bio” substantially underperformed.
I wondered whether that was true or not. When I dug into the cited sources, I found them… well, less rigorous than I’d like in terms of drawing a conclusion from data. This was an interesting question, one I want the answer to.
Does putting “link in bio” or its variants cause less engagement, lower performance on Instagram?
Did I know? Nope.
Could I know? Yes.
The challenge with whether we could know the answer to something is fivefold, something I borrow all the time from cooking:
- Outcome: do I know what I’m trying to achieve?
- Recipe: do I have a process for achieving the desired result?
- Ingredients: do I have the raw materials I need to make the thing?
- Tools: do I have the tools necessary to process the ingredients?
- Talent: do I have the necessary skills to create the outcome?
Let’s break down what we’d need to understand the answer to our question about whether “link in bio” matters or not.
- Outcome: clear. We want to know if putting “link in bio” dampens Instagram engagement and performance.
- Recipe: from a data science perspective, this is also clear. There are many different ways to accomplish this, but the recipe I’m going with is something called propensity modeling. We classify Instagram posts by whether they use the target phrase or not as a treatment, and then use propensity modeling to see what kind of lift that treatment gets, of similarly matched content.
- Ingredients: this is the challenge for a lot of social media inquiries. A lot of data is hard to come by. I sourced my data from Facebook’s Crowdtangle software.
- Tools: clear. I’ll be using R Studio and the R programming language.
- Talent: I have the necessary skills for this particular outcome.
Now, let’s dig in. To make this work, we need to do some feature engineering on our dataset, tagging posts that use “link in bio” so that we can classify them appropriately. We also need to remove accounts that have missing or broken data. Because of the size of the dataset – nearly a million records in its raw form – I’ll need to use a random sampling of it instead, about a third of the records.
Once we’ve done that, we build our model and match similar records to take out as many confounding factors as possible. That’s the beauty of propensity modeling – by matching similar records, we can reduce the impact of variables and relationships we don’t know to look for or can’t see, looking for the statistical difference JUST along the “treatment”, which is the use of “link in bio”.
Propensity modeling, in other words, is more or less a retroactive A/B test when you can’t run a large-scale A/B test.
What are the results of our assessment?
What we see is… no, there’s pretty much no difference or not in terms of performance and the use of “link in bio”. We see there’s a slight difference in account size – accounts that use “link in bio” have a mean number of followers that’s slightly higher than those that don’t.
However, there’s no real difference in terms of the mean number of likes, and a 1% difference in the mean of comments on posts that use “link in bio”.
The conclusion we draw? Using “link in bio” has no impact on engagement in Instagram.
We can test this further by using automated machine learning. With IBM Watson Studio’s AutoAI, we feed the ENTIRE dataset (instead of just a sample) into Watson and ask us to tell it which variables have the strongest correlation to the outcome – engagements – that we care about:
What we see is fairly straightforward: the number of views a post earns delivers engagements, as does the size of the audience. Whether or not a post has a “link in bio” statement is utterly irrelevant, in terms of variable importance.
The key takeaway here isn’t whether or not “link in bio” matters to Instagram engagement or not, though that is a handy little tidbit. No, the key takeaway that I’d like you to remember is to read and question the validity of these little data tidbits we read every day. For things that genuinely pique your curiosity, ask whether you do know the answer, and if not, could you know the answer.
Again, that doesn’t mean hunting down things that agree with your point of view – that’s called incuriosity (as well as confirmation bias) and it’s literally the opposite of the takeaway. No, the point is to be MORE curious, to wonder what the answer really is, and to conduct valid, methodologically-sound research to get to the answer. If you have the capability to truly do your own scientific and statistical research, do so and publish it so that the greater community can benefit.
And feel free to tell people they can learn more with the link in your bio on Instagram.
Methodology and Disclosure Statement
I extracted 970,000 Instagram posts from Instagram public content, sampled from curated lists by Trust Insights in Facebook’s Crowdtangle platform. The sample population includes 7,856 Instagram brand accounts and 10,341 Instagram personal accounts across a variety of industries. The timeframe of the study data is July 1, 2021 – September 12, 2021. The date of the study itself is September 13, 2021. I declare no competing or conflicting interests, and I was neither given nor gave any compensation for this research, other than applicable service fees to vendors for the operation of their software itself.
You might also enjoy:
- How to Set Your Public Speaking Fee
- How To Start Your Public Speaking Career
- It's Okay to Not Be Okay Right Now
- Is Social Listening Useful?
- What Content Marketing Analytics Really Measures
Want to read more like this from Christopher Penn? Get updates here:
Get your copy of AI For Marketers