CS Rarity List [WIP] - status update on page 75

[Solloby]

^ Not when you start weighting them to compensate for the lack of adopt limits. For my calculations I assumed 400 dogs were adopted per litter outcome. So I assumed 800 White July dogs were adopted compared to 1600 Sun dogs, because there are 2 and 4 outcomes per litter respectively. To compare the August dogs to July dogs, we would need to weight the August dogs differently, but I am unsure what number to use.

[Solloby]

Using that same methodology, and aside from Dark Heart Rat (went OMGSR) slotting in all August pets directly alongside and under the July pets (Nons went OMGSR briefly so we know they go above JulyAugs and Augusts, GWJ is fairly common for a July though so I put it on the same tier as the rarest Augs) we get this:

Rarity Tier List Draft

1 - Sunjewel
2 - July PPS - Moonswirl - Dogtag
3 - Coontail - Heavy Speckle Rat - Medium Speckle Rat - Light Speckle Rat - Dark Heart Rat - UR Butterbun
4 - Sunback - Noncoon - Nonjewel - Nonswirl - Nontag - Light Pawprint Rat
5 - GWJ - Spotted Tribal - Purple Toxic - BEG
6 - Pink Balloon - Nonballoon - Green Toxic - Blue Toxic
7 - Gearback Tribal - Blue Balloon - Yellow Toxic - Line Tribal - Red Toxic - CEG
8 - Light Heart Rat

Notes:
> Dark Heart Rat sat at the top of the Aug list, I left it amongst the July pets it went OMGSR with.
> Unsure if it is correct to put GWJ on the same level as BEG/Purple Tox/Spotted Tribal. We know they go below the Nons that went OMGSR briefly. We don't know if they are more common than the GWJ but since BEG was available at a similar time I thought it ok to put them together. This was the only non data-driven decision in the tier list thus far - does anyone have opinions on whether they should go together or not?
> Nonballoon has same rarity as Pink but not Blue. Unsure if this was intentional or an original release error that was corrected in future rereleases to put the Pink & Blue together? We should do a rerelease count of balloon dogs to check (might look at the existing rerelease count data we have after lunch)

[nickjr]

IIRC the galaxy dogs were adopted a ton, weren't they? Wonder how much the original adoptions influence rarity today. Not much, I guess...

The White July litter and the Galaxy litter were both put up in early July, but wasn't the White July litter taken down at the start of August? The galaxies were adoptable to late August (I adopted one myself on August 27) + their design probably made them super popular (because star dogs). If only we didn't have to think about rereleases...

[Solloby]

That's the reason I weighted the outcomes, to accommodate for dogs like the galaxies being heavily adopted while rats and dogs like white julies less so c:

As for the impact of the rereleases, not entirely sure. Happy to revise the rarity list draft with rerelease data added, but until we get the full dataset for all rereleases, the best we can do is work with the original adoption data combined with info from the rarity changes archive.

[nickjr]

Ah right. Too many things to keep track of -head goes boom-

[lil rascal]

So did the count show the non-coon is worth the same as the other nons even though it never went OMGSR? It makes sense, the others turned OMGSR just before the re-release so with the re-release if the non coon was even one pet less then the OMGSR threshold it wouldn't have turned but was just curious if that's what the data shows.

[Solloby]

Yes, it showed the sunback and bobs together... Lol autocorrect, NONS.

Note for self when I get home - can use the Dark Heart Rat's numbers to calculate diff between July and August! It will give us an approximate modifier value. Unlikely to be the same for diff between August and September though. Since all Augusts were always VR and no Septs omgsr, cannot use rarity change data to compare them. Would likely be less than July/August gap though.

[Lya]

One thing I did try doing is using the data combined with rarity change archives data (that thing being out of date is very problematic; searching the rarity changes thread helps but not everyone used text when posting images of rarity changes).

This. So. Much.

I actually asked the Mod who's in charge of it if they were still working on getting in up to date like... a year ago? And never got an answer.
I'd have gladly taken it over, brought it up to date and enhanced it with a spreadsheet archive (on throwaway account) plus an index of the VRs+ at any given time (would take some time to set up, but whatever ).


And thanks for the ranges.
I'll set away time for counting when I need to take breaks from other stuff (5k pets take 5-10 mins -- no, I don't have a RL ).

[Solloby]

I asked about it being updated a few years ago and it never was, so I don't think there would be an issue in someone taking it over? It's been years since the last update so it's clearly abandoned? Trying to find the missed changes would be a nightmare but it really needs to be done >.<

[Erythro]

So I don't think y'all are going to like what I have to say but from a stats PhD to a student to yall, here we go.

We never got a code to get the rest of the data so I could take to my professor but I went to talk to him anyway. When I finally got around to seeing him and explaining the data set and what we are trying to do(create rarity based tiers). He tried to come up with a way to compute what we are trying to compute. He said he'd do some research and get back to me. A month later and he sent me an email to come see him cause he had an answer.

He said he way Solloby was using chi-squared was okay and that it is possible to use it if you really want to but a better way to do it is treating each total number of pets as a mean and using a difference of means test. However this wouldn't be realistic in the real world and if someone were to look at it from a real world stand point, we'd be laughed at.

Then it goes onto talking about an F distribution which is essentially a ratio of chi-squared distributions but the work we would have to put into for that is unrealistic and I don't know enough about R/SAS to be able to use a program to calculate the data for us. I do know enough about the programs to do the difference of means test so don't let that idea escape our mind.

Then here comes the bad news, in a real world situation, we would never be able to do this. He asked me if I knew of an example of a limited series product that had value (he knows I'm super into fashion) and a YSL skateboard deck came to mind. There's only 100 of them sold and each one has immense value(think thousands). Now think of the aftermarket, the value only goes up as the demand increases and so people are willing to pay more. We know this is the case on an extreme level because of the situation we are dealing with that we call Greed Demand. Hence, why we are making this list.

His advice is completely count each and every pet we are looking at the rarity of (listed here) and set boundaries on what we want the tiers to be. IE; below:

for tier 1: Sunjewel - July PPS - Moonswirl - Dogtag [within 15 adopted of each other]
for tier 2: Coontail - Heavy Speckle Rat - Medium Speckle Rat - Light Speckle Rat - Dark Heart Rat - UR Butterbun [within 25 adopted of each other]
and so on until we have listed the pets we want to list
I totally understand if that's not something you all feel comfortable with. We could treat the number of pets released as the means and do difference of means testing, or continue with chi-squared testing or go with the suggestion my professor made. Regardless of what we choose, we need to start counting the pets again. Is anyone able to find someone who knows Python or knows Python themselves to do a program like the one created before? If not, we need to start delegating groups of pets for people to count.

TL;DR: We really only have 4 options; chi-square (what Solloby is already doing), difference of means testing, F distribution testing (really long and difficult), or setting bounds on tiers.