followers_count unreliable

coleott · 2013-07-29T21:21:21.000Z

The followers_count field returned in each User object (in both the users/lookup endpoint and from site streams) has been returning data that seems incorrect. I was wondering if anybody has insight into why it’s inaccurate and/or how to work around that.

I’ve been polling the users/lookup endpoint and comparing those results to the incoming follower events from site streams as well as the ads.twitter.com Analytics dashboards. Fairly regularly I find that the user.followers_count will increase by a greater number than there were follower events for a time period. I also was looking at one user who, according to the User objects, dropped from 820 followers down to 806 in one day, but (according to the ads.twitter dashboard) only had 6 unfollows for that period. It seems this user never reached 820 followers, so the user object isn’t stale, it’s just wrong. This is happening with basically all the users I’ve looked at so far. I’m wondering if Twitter’s using some sort of approximate count or if followers’ accounts being suspended for spamming would show a dip in followers without actually creating “unfollow” events (or both).

If anybody has insight into this or an idea of how to get more consistent data (e.g. some way to get the number of unfollows for a period, like the ads dashboard does), I’d love to hear it!

episod · 2013-07-30T15:10:30.000Z

Unfollows aren’t the only event that can result in a loss of followers. When accounts are suspended, deactivated, or otherwise cleaned up for abuse, you’ll find that the accounts are mostly silently removed from the follower inventory. The numbers are also not meant to be real-time accurate but instead an eventually consistent approximation – definitely behind layers of cache that might be different depending on which server you’re hitting.

coleott · 2013-07-30T20:46:00.000Z

Ok, that’s what I suspected. Thanks for verifying.

It seems like the most likely explanation for why I would be getting situations where followers_count increases by more than the number of follower events I get for a period, then, is that at the beginning of a period the followers_count is undercounting because it’s stale, but by the end of the period it’s caught up more?