Hey all, remember these?
Back in the 2024 Favourite Coasters Ranking, we did what we always do — collected everyone's top 10 lists and tallied them up to see what the CoasterForce community loves most. Simple, honest, and it's worked well for years. But a few of you pointed out (fairly!) that a straight appearance count doesn't tell the whole story. A coaster that shows up on ten ballots at #1 and a coaster that shows up on ten ballots at #9 get counted the same way. And that got us thinking…
In coaster ranking terms, the approach idea was to do the same head-to-head setup for roller coasters. Steel Vengeance goes against Taron and beats it, gains points. Steel Vengeance goes against Eejanaika and loses, giving Eejanaika points. Simple enough right?
That’s what we thought too…
… well, it isn’t as simple an open and shut case for ELO in roller coaster rankings.
Initial passes and test runs we were doing started to end in very quirky, weird results. Like… Velocicoaster finishing 20th overall weird results. What gives?
With a community poll like ours, ballots trickle in over months, and not everyone lists the same coasters. We are also riding and rankings roller coasters at random intervals, under different conditions, and with different rider perspectives. That means some rides only ever get compared against a handful of opponents. If a coaster appeared on just one or two ballots and happened to sweep everything on those lists, ELO rewarded it heavily — even though it never got tested against the coasters most of you are actually arguing about (cough Iron Gwazi vs. Velocicoaster cough). Without that time sequence, order-dependence started to show up a lot; some roller coasters would get a runaway lead if they lucked out on their first few matches, or would tank if going against true behemoth coasters.
The clearest example: Eejanaika landed at #1 overall, ahead of Steel Vengeance, on the strength of 18 games from two ballots where it beat everything in its path. Compare that to Steel Vengeance's 126 games across sixteen ballots. Both numbers are “real” in the sense that the math is correct — but they're not remotely equivalent as evidence of community favour. Katun, AlpenFury, Storm Runner, and a handful of others near the top of the ELO board have the same issue: 9 games, a clean sweep, an inflated rating. Check out the below table - it's a very different Top 10 than what you would expect with the given scoring system!
It's not that ELO is “wrong,” it's that ELO needs a densely connected web of comparisons to work as intended, and a hobbyist poll like ours — where everyone's list is a little different — just doesn't give it that.
To correct for this, we thought we would run multiple passes – like thousands of passes – which is a mitigation strategy @coasterbot has also used for their calculations. Alas, our relatively small sample size and ballot count still meant inflated ratings would set in, and vary wildly. The “who beats whom” calculus had backfired.
Bother.
And so, without further adieu… because there has been a lot of adieu this year!... I give you the Top 2025(ish) Favourite Roller Coasters of CoasterForce:
Under Borda, the top of the board reads a lot more like what most of you would expect from browsing this thread: Steel Vengeance at #1, followed by Iron Gwazi, VelociCoaster, Ride to Happiness, and Zadra. No single-ballot sweepers sneaking into the top 1 (boo ELO!), and giving room for new coaster entries to still punch in.
And for those curious to come along for the ride, I've created a special edition Google Sheet that details all the above approaches: ELO, Borda, and our "traditional" ranking (including year-over-year ranking comparison): https://docs.google.com/spreadsheet...ouid=103548817647712600270&rtpof=true&sd=true
As always — keep the top 10 lists coming, and the more of you who submit, the tighter both rankings get. Cheers!
Back in the 2024 Favourite Coasters Ranking, we did what we always do — collected everyone's top 10 lists and tallied them up to see what the CoasterForce community loves most. Simple, honest, and it's worked well for years. But a few of you pointed out (fairly!) that a straight appearance count doesn't tell the whole story. A coaster that shows up on ten ballots at #1 and a coaster that shows up on ten ballots at #9 get counted the same way. And that got us thinking…
What If We Changed How We Rank and Sort? Hello ELO!
The idea started simple enough – what if we jumped to an ELO-style model, similar to how @coasterbot runs their rankings? ELO has been around for years (beyond roller coaster rankings), and is a favorite go-to for head-to-head matches like golf, chess, or esports. The core idea is simple: every competitor starts at the same rating. When two go head-to-head, the winner gains points and the loser loses points. How many points depends on who was expected to win – beating a highly-rated opponent earns you a lot; beating a weak one earns you very little. Over time, ratings settle into a number that reflects overall strength based on who beat whom.In coaster ranking terms, the approach idea was to do the same head-to-head setup for roller coasters. Steel Vengeance goes against Taron and beats it, gains points. Steel Vengeance goes against Eejanaika and loses, giving Eejanaika points. Simple enough right?
That’s what we thought too…
… well, it isn’t as simple an open and shut case for ELO in roller coaster rankings.
Initial passes and test runs we were doing started to end in very quirky, weird results. Like… Velocicoaster finishing 20th overall weird results. What gives?Where ELO Started Going Wrong
Well, ELO hinges on one critical assumption; that head-to-head matches are played within same time window, and happen chronologically. That is, one match is played at a time, and the points flow accordingly.With a community poll like ours, ballots trickle in over months, and not everyone lists the same coasters. We are also riding and rankings roller coasters at random intervals, under different conditions, and with different rider perspectives. That means some rides only ever get compared against a handful of opponents. If a coaster appeared on just one or two ballots and happened to sweep everything on those lists, ELO rewarded it heavily — even though it never got tested against the coasters most of you are actually arguing about (cough Iron Gwazi vs. Velocicoaster cough). Without that time sequence, order-dependence started to show up a lot; some roller coasters would get a runaway lead if they lucked out on their first few matches, or would tank if going against true behemoth coasters.
The clearest example: Eejanaika landed at #1 overall, ahead of Steel Vengeance, on the strength of 18 games from two ballots where it beat everything in its path. Compare that to Steel Vengeance's 126 games across sixteen ballots. Both numbers are “real” in the sense that the math is correct — but they're not remotely equivalent as evidence of community favour. Katun, AlpenFury, Storm Runner, and a handful of others near the top of the ELO board have the same issue: 9 games, a clean sweep, an inflated rating. Check out the below table - it's a very different Top 10 than what you would expect with the given scoring system!
It's not that ELO is “wrong,” it's that ELO needs a densely connected web of comparisons to work as intended, and a hobbyist poll like ours — where everyone's list is a little different — just doesn't give it that.
To correct for this, we thought we would run multiple passes – like thousands of passes – which is a mitigation strategy @coasterbot has also used for their calculations. Alas, our relatively small sample size and ballot count still meant inflated ratings would set in, and vary wildly. The “who beats whom” calculus had backfired.
Bother.
Enter the Borda Count
So the poll just kind of sat for… ever lol. Life got busy and we moved on. But one day at work, I was having to work on similar data regression, and discovered a new ranking system I was unfamiliar with – Borda! While Borda is similar to ELO in comparing ranking opinions, it considered the broader ranked options of ballots, making it a popular pick in elections or survey responses (what I was using for at work!). The scoring methodology is pretty simple, with a points-per-position scoring. So out of 10 positions (aka Roller Coaster Top 10’s), 10 points are awarded to #1, 9 points awarded to #2, etc. The core idea is that higher-ranked options receive more points, versus the dog-eat-dog approach of ELO, where the winner takes all the points. This would help account for the weird, runaway victories that were happening with earlier ELO testing. Borda is also designed for ballot-style voting – that’s us! – and calculated from one “fixed” position of collecting the ballots, rather than dynamically updating over time like ELO. That would help with the reality that coasters are ranked at random time intervals. Best of all, it's less mathematically wonky. Every ballot contributes independently, so a coaster's score reflects broad community support rather than the luck of who else happened to share its ballot.And so, without further adieu… because there has been a lot of adieu this year!... I give you the Top 2025(ish) Favourite Roller Coasters of CoasterForce:
Under Borda, the top of the board reads a lot more like what most of you would expect from browsing this thread: Steel Vengeance at #1, followed by Iron Gwazi, VelociCoaster, Ride to Happiness, and Zadra. No single-ballot sweepers sneaking into the top 1 (boo ELO!), and giving room for new coaster entries to still punch in.
Where we've landed
Going forward we're treating Borda Count as the primary ranking for this poll — it's stable, it's transparent (anyone can check the math by hand), and it holds up well even with the amount of ballot variance we get in a community poll like this. Full comparison table (ELO rank, Borda rank, and the gap between them) is attached below if anyone wants to dig into the numbers themselves.And for those curious to come along for the ride, I've created a special edition Google Sheet that details all the above approaches: ELO, Borda, and our "traditional" ranking (including year-over-year ranking comparison): https://docs.google.com/spreadsheet...ouid=103548817647712600270&rtpof=true&sd=true
As always — keep the top 10 lists coming, and the more of you who submit, the tighter both rankings get. Cheers!
