2005-2009 list: http://www.metalstorm.net/users/list.php?list_id=5838
2010-2014 list: http://www.metalstorm.net/users/list.php?list_id=5828
2015-2019 list: http://www.metalstorm.net/users/list.php?list_id=6722

I sure do love this site and it's rating system, it makes finding great music so easy and, even more importantly, when an album is overlooked for a while but starts getting a buzz a few months later, boom, there it appears in the top 20.

All that being said, I still wish there was some way to get a weighted ranking that not only takes the rating of the album into account, but also the number of votes that were cast. For example, an album with a rating of 8.8 and 1000 votes would rank higher than an album with a rating of 8.9 and 20 votes. So, I scraped all the albums on the site with at least 1 vote into a database. Then, taking a leaf out of IMDb's book, I used Bayesian estimation (like the IMDb's Top 250) to calculate a weighted rating for each album based on its Metal Storm rating and the number of votes it had received.

The formula for calculating the Top 100 Albums gives a true Bayesian estimate:
Bayes rating (BR) = ( C × m + R × N ) ÷ ( m + N ) where:
C = average expected rating given...
m = minimum number of votes
R = average rating for the album = (Metal Storm rating)
N = number of votes for the album = (votes)

The way the Bayesian estimator works is that all albums in the database are given and additional m fake votes with an average rating of C, irrespective of how many true votes it already has. Then a new weighted rating is calculated...
If the album has no true votes ( N = 0 ), then BR = ( C × m + R × 0 ) ÷ ( m + 0) = C × m ÷ m = C (the average expected rating).
If the album has a number of true votes equal to the minimum ( N = m ), then BR = ( C × m + R × m ) ÷ ( m + m ) = ( C + R ) × m ÷ 2m = ( C + R ) ÷ 2 (halfway between the two ratings).
If the album has a very large number of true votes ( N >> m ), then BR = ( C × m + R × N ) ÷ ( m + N ) ≈ R × N ÷ N = R (the Metal Storm rating).
This means that the Bayes rating will never equal the Metal Storm rating but will approach it as the true votes grow arbitrarily large.

Now that the maths bit is out of the way, here was my methodology:
Only Studio and EP album types, from the years 2000 to 2004, with at least 1 vote, were included.
The average expected rating (C) was determined by averaging all the Metal Storm ratings from 2000-2004.
The minimum number of votes (m) was determined by calculating the 75th percentile of the votes, rounded to the nearest 10.
I also decided that if an album's Bayes rating was within 0.005 of #100, it would be included too; hence the +5.

C for 2000-2004 = 7.599
m = 30

Number of albums from each year in the list:
2000 - 15
2001 - 27
2002 - 14
2003 - 20
2004 - 29

Ratings are accurate as of 30/10/2023