Bill Connelly SP+ predictions for this week...

[A Clockwork Orange]

Has SU losing 29-24. 58% probability.

 

[TheCusian]

Has SU losing 29-24. 58% probability.

He is good - but the numbers still use last years data combined with this years results until week 4. So it's not really factoring in the new and improved offense yet.

And not accounting for the Shrader we've seen so far is a huge difference.

 

[721Comstock]

He is good - but the numbers still use last years data combined with this years results until week 4. So it's not really factoring in the new and improved offense yet.

And not accounting for the Shrader we've seen so far is a huge difference.

I think I can speak for OiG when I say - we will see Bill Connelly in HELL.

Cooking in last year's data for this team at this point, is absurd.
Our O is night and day different, due to Anae, Beck, the WR coach, and Shrader's vast improvement.
Plus the D is even better, which is also amazing, given how good they already were.

Based on last year's data, Llvll rolls us, and UCann't is probably just a 2 score win.


It IS his fault. He SHOULD have known.

 

[LeMoyneCuse]

Cooking in last year's data for this team at this point, is absurd.

It's not, though. Syracuse has played a maybe bad team in Week 1, which isn't the best indicator of how a team will be, and a cupcake. That's not enough sample size in a sport with small sample sizes.

Yeah, Syracuse appears to be an exception to an approach that needs rules and cannot be altered for exceptions.

 

[A Clockwork Orange]

He is good - but the numbers still use last years data combined with this years results until week 4. So it's not really factoring in the new and improved offense yet.

And not accounting for the Shrader we've seen so far is a huge difference.

Agreed. He had Louisville winning a close one in Week 1, with a 59% chance of being correct. He predicted we'd cover the spread against Louisville but not hit the over. I think as it goes throughout the year the predictions become a bit more on the nose, but it obviously isn't there yet.

 

[HRE Otto IV]

It's not, though. Syracuse has played a maybe bad team in Week 1, which isn't the best indicator of how a team will be, and a cupcake. That's not enough sample size in a sport with small sample sizes.

Yeah, Syracuse appears to be an exception to an approach that needs rules and cannot be altered for exceptions.

It is going to be off the first few weeks no matter what. So shouldn't it be off by this years small sample size vs using stale data?

 

[LeMoyneCuse]

It is going to be off the first few weeks no matter what. So shouldn't it be off by this years small sample size vs using stale data?

Doing so would completely disjoint the model from pre-season predictions by removing all continuity, which makes it even more useless out of season than even Bill C. would acknowledge that it already is.

 

[NineOneSeven]

It could happen. Not worth getting upset over.

 

[NJneer12]

It's a predictive model with a lot of pre season parameters.

I'll take this 100% in week 3.

 

[HRE Otto IV]

Doing so would completely disjoint the model from pre-season predictions by removing all continuity, which makes it even more useless out of season than even Bill C. would acknowledge that it already is.

If you are going to use a model that is supposed to take away bias and subjectiveness, you should not use data that is based off of bias and subjectiveness. It is ok to have Bama not in the Top 10 based on their performance right now. And if the model is garbage until week 4, don't publish the data until week 4.

 

[LeMoyneCuse]

If you are going to use a model that is supposed to take away bias and subjectiveness, you should not use data that is based off of bias and subjectiveness. It is ok to have Bama not in the Top 10 based on their performance right now. And if the model is garbage until week 4, don't publish the data until week 4.

I'm sure he'd love your feedback.

 

[Orangezoo]

It's not, though. Syracuse has played a maybe bad team in Week 1, which isn't the best indicator of how a team will be, and a cupcake. That's not enough sample size in a sport with small sample sizes.

Yeah, Syracuse appears to be an exception to an approach that needs rules and cannot be altered for exceptions.

Tough to say Louisville is maybe a bad team at this point unless UCF is complete garbage which I saw little sign that they are.

 

[TheCusian]

I think I can speak for OiG when I say - we will see Bill Connelly in HELL.

Cooking in last year's data for this team at this point, is absurd.
Our O is night and day different, due to Anae, Beck, the WR coach, and Shrader's vast improvement.
Plus the D is even better, which is also amazing, given how good they already were.

Based on last year's data, Llvll rolls us, and UCann't is probably just a 2 score win.


It IS his fault. He SHOULD have known.

I agree - but predicting what each team will be with no data in a current year is impossible. Most teams don’t have massive jumps or declines year to year - it’s easier to bank on a team being similar that doing a 180.

It will move eventually towards a more true showing of what this team is.

 

[LeMoyneCuse]

Tough to say Louisville is maybe a bad team at this point unless UCF is complete garbage which I saw little sign that they are.

That's really not the point of what I'm saying.

 

[Millhouse]

He is good - but the numbers still use last years data combined with this years results until week 4. So it's not really factoring in the new and improved offense yet.

And not accounting for the Shrader we've seen so far is a huge difference.

Same dumb nonsense every year from him, he has to know he has nothing to predict off of but he wants clicks all year instead of after it's over

 

[TheCusian]

If you are going to use a model that is supposed to take away bias and subjectiveness, you should not use data that is based off of bias and subjectiveness. It is ok to have Bama not in the Top 10 based on their performance right now. And if the model is garbage until week 4, don't publish the data until week 4.

The data from last year is unbiased. It is what it is based on 12-14 games for each team.

Any statistical morel needs data points. CFB is very limited as far as data points. If you limit what you have (1 or 2 weeks worth) you get an even more useless set of numbers.

The data says based on Bama made it to the finals of the CFP last year and were very good drive to drive, until last week. Throw in recruiting data into the mix and bam - it’s likely that they are still top ten.

 

[LeMoyneCuse]

Same dumb nonsense every year from him, he has to know he has nothing to predict off of but he wants clicks all year instead of after it's over

Ah, yes. Someone tries to look at football differently, and because you don't like it, it's click bait.

 

[Orangezoo]

That's really not the point of what I'm saying.

Small sample size - i got your point. Just was nitpicking the example within that. Small sample size alone is pretty much everyone's reality by week 3.

 

[Millhouse]

Ah, yes. Someone tries to look at football differently, and because you don't like it, it's click bait.

It's a good model when there is data to feed into it

when you feed a good model with worthless assumptions instead of data, you get worthless output

 

[LeMoyneCuse]

It's a good model when there is data to feed into it

when you feed a good model with worthless assumptions instead of data, you get worthless output

You'd rather just have this?

 

[HRE Otto IV]

The data from last year is unbiased. It is what it is based on 12-14 games for each team.

Any statistical morel needs data points. CFB is very limited as far as data points. If you limit what you have (1 or 2 weeks worth) you get an even more useless set of numbers.

The data says based on Bama made it to the finals of the CFP last year and were very good drive to drive, until last week. Throw in recruiting data into the mix and bam - it’s likely that they are still top ten.

So recruiting ranking are unbiased? Weighting it so the most recent class of FR has a bigger impact than SRs isn't biased? Putting little into transfers isn't biased? Using the last 4 years of team data is relevant how? Most teams do not have those same players, nor systems, nor coaches (which is a big deal). Sure a team like Bama has little variance but for teams 20-40 that is a big factor.

He isn't using stats to make these. That is my problem. Is it likely that after 12 games Bama is top 10? Sure, but why have them there now if your model is supposed to not have human bias? It is no better than the Media Poll.

 

[HRE Otto IV]

You'd rather just have this?

The point is he is lowering his model by introducing subjective data. Which makes his model no better than a human.

 

[Dinkyspond]

It's not, though. Syracuse has played a maybe bad team in Week 1, which isn't the best indicator of how a team will be, and a cupcake. That's not enough sample size in a sport with small sample sizes.

Yeah, Syracuse appears to be an exception to an approach that needs rules and cannot be altered for exceptions.

I don’t think we classify Lville as a bad team. Week by week, we'll learn more. They play at home vs. FSU this week, another good indicator where they're at.

 

[TheCusian]

So recruiting ranking are unbiased? Weighting it so the most recent class of FR has a bigger impact than SRs isn't biased? Putting little into transfers isn't biased? Using the last 4 years of team data is relevant how? Most teams do not have those same players, nor systems, nor coaches (which is a big deal). Sure a team like Bama has little variance but for teams 20-40 that is a big factor.

He isn't using stats to make these. That is my problem. Is it likely that after 12 games Bama is top 10? Sure, but why have them there now if your model is supposed to not have human bias? It is no better than the Media Poll.

Hey, man - do what you want with them. I find them a helpful starting point because I understand all of that stuff gets factored in and a suddenly better team like Syracuse this year still makes sense even after that factoring.

current numbers + historical data > media's myopic understanding of the sport on any given week

And the "bias" you see with 'Bama is just the historical data, ie. they've been really good for a long time. That's not bias. That's what the numbers say year over year.

If this stuff isn't for you, call up Nate Mink and you guys can swap anecdotes

 

[TheCusian]

The point is he is lowering his model by introducing subjective data. Which makes his model no better than a human.

yeah, he admits that including recruiting subjective recruiting analysis is not his preferred method of understanding a teams talent level - but until there is a better way, the correlation between stars and performance will have some value