I know lots of folks are talking about Monthly Active Users when it comes to health of the Fediverse.
We use that to compare social medias and even ourselves, a social network, to each other.
I argue we should be focused on user engagement. I know LinkedIn has “impressions”, but idk what that means.
So I wonder if there’s a good way to generate this. Someone posting is the highest, commenting, subscribing, liking, disliking, and follow on down. I guess that would be a statistical model? But with diminishing returns. One SUPER ACTIVE ANNOYING poster does not a network make, but “media” it does.
I don’t have a clue how this would work statistically. But I theorize, that while we’re smaller MAU, our user engagement is significantly higher when population size is accounted for.
Is there any data anyone knows of to back this up or disapprove it? I’m pro small.social though, so maybe I’m wrong. Any data scientists in the Fediverse? :-D
I was actually working on something to measure this today. I may have something for you tomorrow. Otherwise, it’ll probably be next week.
AWESOME! Please don’t hesitate to ping me! I’d love to share it when you’re ready. <3 :)
Exciting!
I guess with statistics, you’d always better ask a very specific question. I mean, these are just numbers, I guess? And if you’re fixing an old Linux computer, there is no point in lots of people commenting on meme posts. You want the one person who’s done this before to be part of the network, read your post and then reply… Or if you want to discuss politics, all the people re-posting the news articles on geopolitics don’t really count, you’ve already read the newspaper, now you’d like nuanced opinions in the comments. I’m a bit unsure whether a single abstract number means anything.
For the health of the overall network, I think MAU isn’t even all that bad. There’s probably a strong connection between “health” of a place, and how many people think it’s worth subscribing and then coming back on a regular basis.
MAU afaik is the easier number to find closer to actual engagement.
But I think engagement could be measured by (comments + votes + downvotes + boosts if available) / (number of posts).
However, size of a given instance, specially for smaller ones, may cause odd fluctuations, so standard deviation or a threshold of small/medium/large instances may be needed when ranking.
Also an instance that had a massive initial engagement metric, or an abysmal one, would have its overall score affected for quite a while, affecting the interpretation of others about its metrics. So to mitigate, maybe taking the numbers by time period may be better, e.g. one ranking for the past week, one for the month, one for the semester, one for the year, and one for all time.
If my quickly written down SQL query is right, those are the numbers for the last month from my instance’s perspective (my subscribed communities):
num_comments | upvotes_on_posts | downvotes_on_posts | num_posts --------------+------------------+--------------------+----------- 188597 | 1646685 | 46461 | 13928So without the boosts, it’d be a total score of 135.
“Impressions” is the amount of time a piece of content was viewed. This is a key metric in advertising-based situations because you want to know the ratio of clicks on the thing vs how many impressions it has to judge how effective a bait it is.
" the amount of time" so is this by the minute? That’s really gross imo and makes me want to stay off there even more. Thank you Fedica! :D
