Watch a Data Analyst Nails Meta Case Study Interview Question: Romantic Reduction

[@garyboy7135]

In addition to the weakness from the interviewer as I responded as a reply to the other comment (not exploring the root cause), I also think interviewer should improve on the experiment set up. Given this impact is from couples, it will be a nice opportunity to understand if candidates know if a user-based assignment is enough or a cluster assignment is needed

[@SamStudious]

This is very helpful - thanks so much to both video participants

[@razvanepe]

I likes this. Any thoughts on confounding variables? Let's say for example that the age of the user is the one that actually determines both the number of posts and relationship status. This will change how you should approach the solutions

[@garyboy7135]

Yeah. I found the interviewee is weak on the initial question of figuring out why coupled users are less likely to post. They jumped immediately to see if the difference is statistically significant and solutions without understanding the possible root cause. Then all the solution would be in vain if they are not solving the root cause.

[@justinc5043]

While I agree this is the proper route to take to decompose the source (i.e. who/when/what/where/which/how) of the perceived post degradation, a "typical" response from a "typical" interviewer may be: 'everything' that pretty much invalidates this often necessary step to understand what the problem is BEFORE root cause identification.

[@Arupchakraborty2004]

Thanks for posting I see the opportunity for performing A/B testing here That will change the route of the answers

[@iqjayfeng]

We hope this helps! I'm curious to see how you might approach the problem. Try it here: www.interviewquery.com/questions/romantic-reduction?

[@connie473]

Isn't the fact that users self-identify as single, being in a relationship, etc., a problem? Also, how many users don't even fill out their relationship status? I'm going to assume that's a lot given, at least, what I see for my connections. Are we assuming relationship status is predicted somehow?

[@elsggonz]

these sounds like metrics from 2010, todays metrics are way different: 1. history views 2. how long you watch a history 3. how fast you see a history after it is posted 4. how many times you repeated the history. and so on, there are new metrics for social media, and this are just 4 examples.