How to rank youtube videos according to like/dislike percentage

#1
This seems like it would be a basic question but im fairly lost on how to do it.

How do you rank a list of youtube videos from the highest rank, to the lowest?

You could rank them according to their like/dislike % ratio but that can be extremely misleading at low numbers. You obviously don't want to put something with 95% like/dislike ratio, and only 20 votes to be above something with 94% dislike and 10000 votes.


But what is the most accurate way to do this? How do you accurately take into consideration sample size with this data? And how do you easily do this for thousands of videos so you can comprise a list easily?
 

obh

Active Member
#2
Hi,
I think that some website don't rank before getting a specific number of votes.

I can suggest to use the lower range of proportion confidence interval. You may decide to take a lower confidence level than the common 0.95. What do you think?

For example:
For example 0.95 with 20 votes:0.763869
0.94 with 2000 votes: 0.928724

I tried 0.95 but you can try with lower value ...
http://www.statskingdom.com/41_proportion_confidence_interval.html
 
Last edited:

Karabiner

TS Contributor
#3
You could use the [like - dislike] difference. Or whatever.
Without context (who asks this, and why? what for, and how
will the results be used?) and without an exact research question,
we can only guess how a useful proposition might look like.

With kind regards

Karabiner
 

obh

Active Member
#4
Hi Karabiner, Origomart,

I think that CI of proportion is better then like-dislike. Since (p=0.95,n=40) in my opinion is better then (p=0.8,n=1000)
Just choose the best confidence level for you.
I would choose several options, then I would decide what I think is the best rank for the options.
Then I would use the method that supports my rank.

If for example, I think that (p=0.95,n=40) is a similar rank as (p=0.9,n=1000) then I would choose CI 0.8

After creating the following table I would probably use CI 0.95 (confidence interval with a confidence level of 0.95)
So it probably makes more sense statistically.

1577591078314.png