15 Comments
User's avatar
Zachary Gomperts-Mitchelson's avatar

So, I went and did some maths as well, what I found was, amongst the teams he's reffed the most, what's REALLY interesting is the distribution, because it's SO WEIRD that he doesn't really seem to give Liverpool or City red cards. oh no I've made a table:

https://docs.google.com/spreadsheets/d/e/2PACX-1vSel8zFa1OPYTNqxodZbpxfAdkFupIwfgl4fPlwHgn5h6YDiNP7LadzHbi_AbRq0GJpuR9U_CUnXCRK/pubhtml

It's the distribution of 7 in 54, 0 in 50 and 1 in 56 that has me reaching for my tinfoil hat. It pushes the SSZS beyond what ranomness would explain. Despite the pretty normal numbers for the other teams he's officiated regularly.

Expand full comment
Zachary Gomperts-Mitchelson's avatar

I've changed my mind.

If you add in other teams he officiates regularly, the probability of observing something like this falls comfortably within an expected range of randomness. Despite notable deviations for three teams, it's not unusual enough to render the overall distribution statistically significant.

The sum of squared deviations from expected values falls well within the threshold for random variation. This means that while the numbers for Arsenal, Liverpool, and City might seem unusual in isolation, when you factor in all the teams, the overall pattern doesn't indicate anything out of the ordinary. It's just statistical noise.

Which is boring, obviously.

Expand full comment
Scott Willis's avatar

Yeah, I come to roughly the same unsatisfying conclusion. It looks suspicious but is just on the line of this can still happen randomly and our pattern seeking minds might be making too much of it.

Expand full comment
Josh Hoffman's avatar

What assumption are you making about the true underling distribution of red cards? That impacts the expected standard deviation, no?

Expand full comment
Scott Willis's avatar

I used the teams overall rate for all games, for games without Oliver, and Oliver’s red card rate. All are pretty close regardless but trying to estimate that is the key assumption and is hard to do.

Expand full comment
mathieu dauner's avatar

I love this approach. Curious what patterns emerge if we broaden the sample to current and upcoming opponents of Man City that Oliver has officiated where cards distorted match outcomes. I hope the club is also putting some data science resource to scrutinise Oliver’s highly unusual refereeing behaviour.

Expand full comment
Matt's avatar

Now what about him giving City 0…

Expand full comment
Scott Willis's avatar

Yeah that seems unlikely. I might pull that at some point too

Expand full comment
Jude Mariadas's avatar

Where did you get the data from for the last Arsenal and Manchester City data with red cards for matches and without Michael Oliver?

Expand full comment
Orcadian's avatar

Do you have the data for readiness to red card a team's opponent? Thinking of the Kovacic non-card. Do he have form for penalising City's opponents but letting our opponents get away with fewer cards than normal?

Expand full comment
Scott Willis's avatar

I don’t. It would be a lot of manual work to try and do that. It would actually be a great use of AI/Machine learning to estimate the call from previous calls but that’s beyond me

Expand full comment
Josh Hoffman's avatar

What if you layered on a Bayesian-type analysis that captures the "unusualness" of the decisions? We've got 4 calls (Martinelli's 2 yellows, trosssard's delay of game, this call, the kovacic non call) that come to mind that seem...unusual or atypical. What's the probability that all of those are just random noise given the expected likelihood of such calls?

Expand full comment
Rory's avatar

I thought he had given us 8 red cards?

Expand full comment
Scott Willis's avatar

It does look like 8 from 55. Damn out of date data.

Expand full comment
Scott Willis's avatar

The data I saw on referees seems to include this match and the red card and totaled to 7

Expand full comment