Matt N
CF Legend
Thanks very much @HeartlineCoaster; much appreciated! Out of interest, how does the Mitch Hawker strategy work?
Out of interest, how does the Mitch Hawker strategy work?
Pair Wins | Pair Losses | Win rate | ||
1 | Nemesis | 36 | 0 | 1 |
2 | Icon | 30 | 1 | 0.968 |
Wild Mouse (BPB) | 18 | 3 | 0.857 | |
3 | Stealth | 25 | 5 | 0.833 |
4 | Wicker Man | 19 | 4 | 0.826 |
5 | Swarm | 26 | 6 | 0.813 |
6 | Inferno | 26 | 7 | 0.788 |
7 | Ultimate | 14 | 8 | 0.636 |
8 | Oblivion | 20 | 12 | 0.625 |
9 | Big One (BPB) | 12 | 9 | 0.571 |
10 | Thirteen | 10 | 9 | 0.526 |
Thanks very much for the in-depth explanation, and for doing the top 10 for me too! I must admit that is a very good methodology, but I must ask; is there any sort of computer algorithm or something that does that, or is it less complex than it looks, because it looks like some extremely complex maths would be required to find the results manually there, at least from an observer’s perspective?Ah yes, now I have your blessing I can elaborate.
To stick with the example coasters I used above, it asks how many times someone preferred Icon to Stealth and how many times someone preferred Stealth to Icon.
In this instance it was 6-5 in favour of Icon. That grants Icon a 'win'.
This comparison happens between every single combination of coasters and you get a number of wins and losses for each coaster.
Essentially it asks the following question, because not every ride appears on everyone's list and we don't have proof of why:
When this ride does appear, how well does it do against everything else?
Pair Wins Pair Losses Win rate 1 Nemesis 36 01 2 Icon 30 10.968 Wild Mouse (BPB) 18 30.857 3 Stealth 25 50.833 4 Wicker Man 19 40.826 5 Swarm 26 60.813 6 Inferno 26 70.788 7 Ultimate 14 80.636 8 Oblivion 20 120.625 9 Big One (BPB) 12 90.571 10 Thirteen 10 90.526
Unsurprisingly Nemesis won against every single pair comparison - in each case more people preferred it to any given ride than the other way around.
Icon only lost to Nemesis in direct comparison with a 4-9 defeat, but again more people rated it higher rather than lower against any other named coaster in the data.
It gets a bit woolly at the bottom because there's simply not enough data, but statistically it's the most sound way forward even if we're only asking people to list a top ten and not everything.
We can probably take for granted that most UK voters have ridden all of the significant UK coasters but if you plan to move on to somewhere like the USA, there will be a lot more regional difference and gaps in representation of ridership across the board, meaning the popularity side of things like the Golden Ticket Awards starts to rear its ugly head.
I also think this formula looks like a pretty good one, as it would probably remove some of the bias towards more ridden coasters over less ridden coasters!Good work on the list Matt, I think it's come out pretty much as expected as there aren't too many surprises there.
For the formula I probably would have gone with:
[Average position of coaster in lists] * ([Total No. of lists] / [No. of times the coaster appears in a list])
So Nemesis would probably appear in every list somewhere so that multiplier value would just be 1, but for a coaster that only appeared in 75% of the lists for example, the multiplier would be a number bigger than 1 which could push it down the list compared to others. So a coaster with a higher average ranking but less appearances in lists overall would potentially rank lower than one with a lower rating that appeared in more lists.
I suspect though that no changes to the formula would alter the list very much except for the last few positions
Subtle difference here but Nemesis doesn't get more points overall, just Icon doesn't gain any. It has less of an impact on this overall result than if you had directly contributed to the number of Nemesis riders, which then directly affected the original scores. By simply mentioning that you liked Icon more than Nemesis, Nemesis would have done slightly better in the original method.I'm a perfect example here, having placed Nemesis outside of the Top 10 list. In this case, the duel Icon-Nemesis did not take place, causing Nemesis to get more points than if I'd put it on #10.
The maths is far from complex, it just takes a bit of clever spreadsheet manipulation to get what you need out of it. I would imagine Mitch, Ellocoaster etc. use a specific program due to the way most of those ballots require the responses in a specific format which can then be harvested from, but with the number of entries we're dealing with here it can be done in other ways.is there any sort of computer algorithm or something that does that, or is it less complex than it looks, because it looks like some extremely complex maths would be required to find the results manually there, at least from an observer’s perspective?
That is unfortunate haha. The Big One has nostalgic value for me, being the first “big coaster” I went on in the mid 90s.^
Aside from slotting Swarm and Wickerman in there somewhere, you're probably not missing too many of the currently operating coasters
...and you may find that Air and The Big One are not as good as you remember!