I still don't understand.

[Quincy A. Wagstaff]

Maybe I'll understand after Saturday, but why are expectations so low for us?

SP+ ratings:

• Kentucky: 13.0
• Ole Miss: 7.8
• Missouri: 5.0
• Mississippi State: 4.2
• Arkansas: -3.2
• Vanderbilt: -11.6

FPI:

• Kentucky: 8.1
• Ole Miss: 6.4
• Missouri: 2.3
• Arkansas: 1.3
• Mississippi State: -0.7
• Vanderbilt: -6.8

FPI projected wins:

• Kentucky: 4.6
• Ole Miss: 3.9
• Missouri: 3.3
• Mississippi State: 2.5
• Arkansas: 2.2
• Vanderbilt: 1.4

On paper, we have the same if not better talent than the five other teams listed above.

Location shouldn't matter as much this year, which helps @ Kentucky and @ Ole Miss. But to the extent that there is a home field advantage, we have done pretty well at home against four of our five home game opponents (Missouri has never played in Starkville.):

• vs. Auburn: 3-1 dating back to 2012
• vs. Arkansas: 3-1 dating back to 2012
• vs. Texas A&M: 3-0 dating back to 2014
• vs. Vanderbilt: 7-0 dating back to 1973

Historically, Mike Leach has done well with less talent than his opponents. 247's Team Talent Composite goes back to 2015. In four of those five seasons, Washington State was dead last in the Pac-12 in talent (on paper). In 2018 they were 11th out of 12 (and went 7-2). So that means in 45 conference games from 2015-2019 Mike Leach had better talent in one game. He went 29-16. It will be interesting to see how well he does when he has as good if not better talent in half of his games. The experts say he'll win two or three of ten.

Who is buying 2-8 or 3-7?

 

[Drebin]

FPI is straight up garbage, as we learned in Jomo's first year. I think they had us favored in every game but one.

 

[Quincy A. Wagstaff]

FPI is straight up garbage, as we learned in Jomo's first year. I think they had us favored in every game but one.

"All models are wrong. Some are useful."

Of course you can pick games and teams in which FPI didn't accurately predict results, but you can do that with any model. That doesn't negate the model. Take an entire season of college football games and FPI is very accurate.

Also, you are misusing the model if you look at number of games favored vs. wins. Sometimes your win probability is 97.8% (South Alabama 2016). Sometimes it's 63.9% (Florida twentyeighteen*). Just looking at number of games favored vs. wins would treat those as equal upsets.
Another example: Imagine a team plays ten games and FPI projects its win probability in each game to be 55%. This does not project a record of 10-0. It projects a record of 5.5-4.5. So despite being "favored" in each game according to FPI, a 5-5 or 6-4 record would align with FPI's projection.

My point is that I think these metrics are underestimating Mississippi State in 2020, not that the models aren't useful.

*-won't let me type 8 + parenthesis

 

[Russ Wheeler]

Models work best over the long haul, so it's definitely a futile effort to try and pick one game to disprove the model.

That said, they are more effective in pro ball than college, where there is more structure. Players are much less mature too, so never know what you're going to get with 18-22 year olds. And the models work pretty good in conference only stats, though, so should be relevant this year.

 

[johnson86-1]

FPI is straight up garbage, as we learned in Jomo's first year. I think they had us favored in every game but one.

That seems pretty accurate as long as FPI doesn't take into account coaching. We should have been favored in every game but one based on talent and past performance. I'm guessing FPI in 2018 reflected our performance under Dan and overhyped us, and now FPI in 2020 reflects our performance under Jomo and underrates us.

 

[Go Budaw]

Who is buying 2-8 or 3-7?

I personally predicted 4-6, but truthfully I only feel good about the Vanderbilt, Arkansas, and Missouri games. The rest are all toss-ups at best, with 3 of them (Bama / UGA / LSU) probably all being guaranteed losses. I am optimistically hoping we can steal 1 or 2 out of 4 between OM, UK, Auburn, and A&M, but I fully expect us to possibly be underdogs in all of them. So, 3-7 wouldn’t shock me. 2-8 probably would.

 

[oxfordrebel22]

"All models are wrong. Some are useful."

Of course you can pick games and teams in which FPI didn't accurately predict results, but you can do that with any model. That doesn't negate the model. Take an entire season of college football games and FPI is very accurate.

Also, you are misusing the model if you look at number of games favored vs. wins. Sometimes your win probability is 97.8% (South Alabama 2016). Sometimes it's 63.9% (Florida twentyeighteen*). Just looking at number of games favored vs. wins would treat those as equal upsets.
Another example: Imagine a team plays ten games and FPI projects its win probability in each game to be 55%. This does not project a record of 10-0. It projects a record of 5.5-4.5. So despite being "favored" in each game according to FPI, a 5-5 or 6-4 record would align with FPI's projection.

My point is that I think these metrics are underestimating Mississippi State in 2020, not that the models aren't useful.

*-won't let me type 8 + parenthesis

Does this model not extrapolate those single game projections across the season? If it really does what you’re saying, and if you’re a 55% favorite in 4 games, and it projects you to go 4-0 instead of 2.2-1.8 in those games, then I agree with you, it’s absolute GARBAGE.

I just assumed, based on the fact that it’s got decimal places, that it spreads those averages out? I could be wrong.

But, to simplify the math for my feeble Ole Miss brain, if you were expected to win at a 70% clip in 5 games and at a 30% clip in 5 games, your projected win total should be 5.0 (or 5-5 record).
.70 x 5 = 3.5
.30 x 5 = 1.5

3.5 + 1.5 = 5

Obviously many of these systems and projections are flawed, but surely it’s not THAT flawed.

 

[oxfordrebel22]

But to answer the original posters questions, my two cents on it are this - most of these systems are based on previous year record and returning production. They aren’t as complex as they pretend to be. So, there are naturally flaws in them. If you want a true gauge on and independent outsiders view of your team, look to the Vegas season win totals. They combine analytics with expert opinions, and then allow the market to make minor corrections. I have no idea what those totals are for OM or MSU, but I’m sure it’s a good indicator of where we are. But, Vegas is wrong a lot, too. But they’re right more than they’re wrong.

But I’ve always felt like if you’re willing and able to take your bias out of it, that we could win money as passionate OM and MSU fans by betting on only those two teams. You’ve just got to be willing to bet against MSU (or OM in my case) and for OM (MSU in my case), at times. Because, while Vegas studies us, they don’t know the rosters and situational things like we do. I’ve won a lot of money on and against both teams over the years. I’ve lost a lot on others...

But, you’ve got to do it over the long haul. For example, this weekend I’ll bet Florida at -13.5 if it will ever drop to there, and MSU at +17. I need each game to move a half point, currently. I’ll bet MSU much higher than Florida, based on confidence level. And, the way I look at it, is if OM covers (or somehow, someway wins), I’ll be happy, and not worried about money. And if LSU blows State out, I won’t be too worried about it. We both know these two in-state programs pretty dang well.

 

[Quincy A. Wagstaff]

Does this model not extrapolate those single game projections across the season? If it really does what you’re saying, and if you’re a 55% favorite in 4 games, and it projects you to go 4-0 instead of 2.2-1.8 in those games, then I agree with you, it’s absolute GARBAGE.

I just assumed, based on the fact that it’s got decimal places, that it spreads those averages out? I could be wrong.

But, to simplify the math for my feeble Ole Miss brain, if you were expected to win at a 70% clip in 5 games and at a 30% clip in 5 games, your projected win total should be 5.0 (or 5-5 record).
.70 x 5 = 3.5
.30 x 5 = 1.5

3.5 + 1.5 = 5

Obviously many of these systems and projections are flawed, but surely it’s not THAT flawed.

You are right. In the first example (55% favorite in four games), the model projects a record of 2.2-1.8, not 4-0. In the second example (five games at 70%, five games at 30%), the model projects 5.0-5.0.

To prove that this is how ESPN uses FPI:
@ LSU-5.6%
Arkansas-49.5%
@ Kentucky-21.4%
Texas A&M-16.4%
@ Alabama-2.6%
Vanderbilt-72.1% (only game in which we are "favored")
Auburn-7.5%
@ Georgia-4.5%
@ Ole Miss-26.1%
Missouri-46.7%

Add these percentages. The result is 2.524.

It projects a record of 2.5-7.5, not 1-9.

 

[patdog]

Agree. IF we could bet with our minds and not our hearts, we could all make money betting on MSU and UM. Few "experts" know those teams as well as we do.

 

[oxfordrebel22]

You are right. In the first example (55% favorite in four games), the model projects a record of 2.2-1.8, not 4-0. In the second example (five games at 70%, five games at 30%), the model projects 5.0-5.0.

To prove that this is how ESPN uses FPI:
@ LSU-5.6%
Arkansas-49.5%
@ Kentucky-21.4%
Texas A&M-16.4%
@ Alabama-2.6%
Vanderbilt-72.1% (only game in which we are "favored")
Auburn-7.5%
@ Georgia-4.5%
@ Ole Miss-26.1%
Missouri-46.7%

Add these percentages. The result is 2.524.

It projects a record of 2.5-7.5, not 1-9.

Awesome, I appreciate you putting that together. Now, the expected results can obviously be flawed, because everyone is guessing to some degree.. but at least I know it’s not completely stupid beyond belief.

 

[oxfordrebel22]

Agree. IF we could bet with our minds and not our hearts, we could all make money betting on MSU and UM. Few "experts" know those teams as well as we do.

It’s easier said than done. I’ve found if I just look at the lines and go with my immediate gut reaction, I can win more than I lose, on those two teams, only.

Obviously, if I let my heart get involved I can sit around and convince myself that Florida and Grantham will just have no answer for Kiffin’s offensive genius, and our patchwork DL will hold their offense full of blue chip players at bay. I’ll also convince myself that Myles Brennan will pick State’s secondary apart like Drew Brees against Yazoo City HS and Mike Leach’s offensive system will be wishing it was back in the PAC-12 by the 3rd quarter. But just because my dumbass, biased brain convinced me of something, doesn’t make it true. No matter how much I may wish it was.

Trust your gut.

 

[kired]

I have very low expectations for this season.

• New coaching staff, no spring practice.
• Completely new offensive system that does not fit our personnel. Find a non-MSU fan that can name 2 of our WRs. Kylin Hill does not count.
• Our strength should be our run game. That's a perfect scenario for Leach**
• Our most productive returning WR apparently isn't good enough to be a starter, and that doesn't surprise me considering the passes I've seen him drop. (sorry, huge fan of Mitchell... but no denying he's had some big drops)
• Our defensive talent - or lack thereof
• QB1 is a grad transfer, QB2 is a true freshman
• Our offense has looked terribly unprepared and confused for 2 years. Most of those guys obviously never grasped what Moorhead was teaching. I have to wonder if the upperclassmen have less knowledge of the game in general due to the previous coaching staff. Same goes for the work ethic, S&C program, etc. If that slipped over the last 2 years like we all believed, you can't just flip a switch and correct it.

 

[drumrcraig]

Find a non-MSU fan that can name 2 of our WRs. Kylin Hill does not count.

Can you name 2 WRs that Leach has coached outside of Crabtree and Wes Welker? He has made an entire career of putting up big numbers with no name WRs. I can't name a single WR that played at WSU last year, but that doesn't mean the WRs sucked. I can't name one from when Gardner played there either.

 

[archdog]

How the hell are we not favored over Arkansas? they haven't won a SEC game in like 3 years.

 

[Quincy A. Wagstaff]

How the hell are we not favored over Arkansas? they haven't won a SEC game in like 3 years.

According to FPI Arkansas would be a two-point favorite on a neutral field. So having us as a slight underdog in Starkville means they are valuing home field advantage at less than two points, which is low compared to non-COVID years.

 

[drumrcraig]

According to FPI Arkansas would be a two-point favorite on a neutral field. So having us as a slight underdog in Starkville means they are valuing home field advantage at less than two points, which is low compared to non-COVID years.

Which even the fact that they would be favored in a neutral site makes no sense. This is a team that hasn't beaten a single P5 opponent (much less SEC) since coming from behind to beat Ole Miss by one on October 28, 2017. I wouldn't think they would be favored in a single game this year. They would have to show me they can beat a P5 opponent before I would bet on them. They even have 4 G5 losses in the last two years. Funny how coaching hires make people want to think that Ole Miss and Arkansas are going to be dramatically improved, but a new hire with an established coach is going to make us regress to our worse season in 14 years. Our current roster was barely alive the last time we only won 2-3 games.

 

[patdog]

Funny how coaching hires make people want to think that Ole Miss and Arkansas are going to be dramatically improved, but a new hire with an established coach is going to make us regress to our worse season in 14 years.

Yeah. That's really pissing me off too. I'm not expecting miracles from Leach and I'm overall pretty meh about the hire, but I'm damn sure expecting more from him than Mississippi and Arkansas will get from their new coaches.

 

[Quincy A. Wagstaff]

UPDATE: After watching the SEC games today, I still think 2-8, 3-7 is underestimating us.

 

[oxfordrebel22]

UPDATE: After watching the SEC games today, I still think 2-8, 3-7 is underestimating us.

Homer alert!!***

 

[Drebin]

As I said the other day, the FPI is straight-up garbage.

 

[kired]

Certainly addressed my concerns - I’ll gladly admit to being wrong

 

[Quincy A. Wagstaff]

As I said the other day, the FPI is straight-up garbage.

And as I said the other day, you can't judge a model like FPI on one game.

Which model do you prefer?

 

[mount lefroy]

Models work best over the long haul, so it's definitely a futile effort to try and pick one game to disprove the model.

That said, they are more effective in pro ball than college, where there is more structure. Players are much less mature too, so never know what you're going to get with 18-22 year olds. And the models work pretty good in conference only stats, though, so should be relevant this year.

Statistical analysis is a complicated thing. It indicates a good amount of important things to prepare for, but it does not predict. It also reveals things that were not noticeable at the time.

Something else about them: you have to understand the rules first before attempting to analyze numbers. The gamblers dilemma is a great example. The difference in correlation and causation is another. An output (say the final score) becomes less and less relevant to a previous input over time. People say they understand these things, but get caught in them all the same. Alabama's dominance over us last year was not the result of any year previous Saban's arrival or the arrival of the kids on each team, but people will quote without thinking that bama has beaten us for decades.