# MLB FORUM

• ballslaughter

I just wanted to get some opinions on how you all see using last years stats versus this years. As part of my mlb research, next to a team name I’ll write what place in the league they are in runs scored, home runs, etc… It would seem like 2019 would be the instinctive route to take, but being only so far in to the season, you have only so large a sample size. So when I deferred to last years, that had more than double the sample size, there were examples of a sizeable disparity. Last year Cleveland was 3rd in runs & 6th in HRs; this year, 24th & 25th.
I realize there can/will be lineup differences…as a quick fix, my thoughts were to take the # of gp last year & this year to find a ratio, & then assign each team my own ranking based on that. Like, if a team played 162 games last year & 54 this year, thats 3 to 1. Then, say the disparity between last years home run placement versus this years is 12; as in last year they were 7th, this year they’re 19th. I’d rank them 11th.
Doing it that way places all the emphasis on quantity & none on recency though, so, perhaps some math to move it slightly more towards the current years stats…
And also, for anyone of the using-previous-years-stats mindset, at what point into a season would you guess you could just focus on current stats—halfway? Three quarters?
Im also curious as to how to rank teams defensively. Some sites will list teams simply in runs scored, and others have things like fielding percentage, errors, & so on.

• d735123

@DFSx42 said...

so you just did the math analyzing it all, including standard deviation all in your head and never once wrote anything down, never used excel or R to do any of it where you could easily export the results to a shareable google sheet… how stupid do you think we are that you did that for 80 slates all in your head and would now need to write it down for the first time??? Seriously, have you even thought through the implications of what you’ve said?

I lean more towards your system to be honest, the second i saw his bvp i cringed a little, but while I think data detective reaches a little too far into the realm of noise, he’s a legit quantitative analyst. When I see his work, I see the work of an intelligent and trained professional. I wish I had the formalized data skills he possesses. For example, he would never just pull out of his ass that he magically computed something in his mind that you yourself admitted would take hours to write down. How is it again it would take you hours to write down yet you can just magically compute it all in your head?

While I can’t abide by data detective using things like bvp, he has my respect because he’d never attempt some outright nonsense like you are right now.

You’re taking something reasonable and making it absurd. When he accepts your challenge you could have just said you changed your mind, there’s no need to go heads up for rollz.
Instead you just state no need, you magically computed it all in your head and have thus proven his system to be worse. Are you 12?

You overextended yourself in a dick measuring contest and didn’t expect him to agree and pull out the ruler, this happens to the best of us, just back out with dignity next time.

Whoa dude, chill. I never said I calculated means and SD in my mind. I never said I didn’t write anything down. Stop putting words in my mouth. I wrote the players names and numbers down on a piece of paper. This took an hour. Then I entered just the numbers into Excel to calculate mean and SD. If you’re accusing me of faking the numbers, prove it. It’s all easily obtainable public information.

DataDetective, if you think a nightly h2h will provide a more valid test of my hypothesis (I don’t), then let’s do it. It seems this is the only way I can prove my manhood to DFSx42.

• DFSx42

Thanks, I appreciate the kind words! Just to clarify, when I mentioned “BvP” I meant it in more of a general “batter vs pitcher matchup” context as opposed to actual BvP stats. In order words, I place a minuscule weight on the past history of batter X vs pitcher Y because I weight everything by # of plate appearances, and these tend to be tiny. I meant that I look at how batter X has fared against pitchers similar to pitcher Y (e.g. same handedness) and how pitcher Y has fared against hitters similar to batter X. These sample sizes tend to be large enough to be statistically significant.

In the interest of improving my model I’d love to better understand what types of noise you refer to. Besides variance, which impacts us all, my biggest issue is that historical data (and therefore my app) aren’t able to identify fundamental changes in a player’s game. E.g. when a pitcher changes his pitch repertoire or when a hitter adjusts his swing. I do overweight recent performance, but in some cases this might not be enough. A good example is Steven Matz, who had an awful 2017 season that in hindsight appears to be an anomaly when taken in the context of his larger body of work…since I look back 3 seasons, that one bad season still impacts my projections.

Ahh weighing bvp by PA seems reasonable, but then again the amount of difference it would make makes it seem like wasted effort? If you weigh it heavily you’re looking at noise, if you weigh it just a smidge then what’s the point? There could obviously be something to it but the main issue is you’ll never have a decent sample and if you do have a decent sample, it’s problematic because that means a lot of it comes from a long time ago when both the p and the b probably had very different skill sets.

I don’t have the time to read your stuff regularly (it wouldn’t help me much either because I have my own system and do cash games primarily) and the way they bury the blogs it’s only when I come across it by chance.

I really admire how you go back and track results and don’t come up with excuses. I love how you never put in the asterisks like “but he hit the warning track twice” type stuff you normally see.

I’m more curious than anything with your righty, lefty, switch splits. I’ve literally never even considered that angle and while I haven’t researched it to state definitively one way or another, I can’t imagine any scenario where that’s relevant and yet the sabremetric community as a whole ignores it. I understand that not all switch hitters created equal but I think you’re going down a very specific and granular level that won’t improve much and probably hurt it more than anything. I don’t think that’s necessarily right or wrong, but I lean towards it being noise and then when I see you comment on bvp it just reinforces my own unfounded hypothesis you stretch the bounds a bit looking for an edge when it probably just results in tracking more noise.

I find your methodology of points given up per pa also very interesting but I’m very curious if you isolate the points. For example, are steals part of that cohort? Should they be? I honestly don’t know. I like the idea in concept, but I again think it may lead to counting some things that probably shouldn’t be counted and vice versa. Honestly though, I wouldn’t have the slightest clue how to go about it, right now just visualizing how that algorithm would work in my mind is a bit of a trip. But I’m not going to lie, I am definitely considering some variation of that pathway if I ever transition to MME – which I think is the only long term viable future for DFS. I know someone who’s breaks down a slate by measuring how the hitters fare vs handedness and predominant pitch type and I if I were to guess, I’d prefer hedging more in that direction if I were to choose to break down a slate in the manner you go about it.

I also struggle mightily with the same problems you face. I don’t adapt fast enough to changes in skill sets and am very slow to move onto the guys who are making the leap to the next step. I still can’t roster Martin Perez in good conscience and now that his price reflects his production this year, I don’t think I’ll end up playing him all year. I just can’t ignore that for a long while he was probably the worst pitcher in all of baseball and doesn’t come with any sort of pedigree. He’s also had flashes of goodness before, while then the peripherals didn’t back it up (and now they do) I’m still not a believer and would rather forever miss out on the Martin Perez and Mike Minors each year than get caught up in some guys off to hot starts. On the flip side, I continually get dragged down in the muck. I will continue rostering someone hitting below the mendoze line because they used to be good two years ago and just like you, rarely play Matz even though the more recent data suggests I should.

• DFSx42

@d735123 said...

No need, I just ran the experiment. I went through your daily blog postings going back to May 1 and tabulated the FD points for your top 3 highest projections. Then I did the same for the best 3 batting order spots for the team with the highest implied run total. Your modelβs top 3 guysβ mean points were 14.7. Just going with batting order+run total resulted in a mean of 16.5. And itβs far, far worse for you if you were adjust for salary because your top picks are always top dollar guys, whereas the simple methodβs guys are not always. Standard deviation was about the same (your was slightly higher, 13.7 vs. 12.7).

Followed by

“Iβm not wasting two hours typing names and scores into a spreadsheet.”

Followed by

“I never said I didnβt write anything down. Stop putting words in my mouth.”

Followed by

“Then I entered just the numbers into Excel to calculate mean and SD.”

wait… so first “you’re not wasting time entering information on a spreadsheet” but then a few posts later mention you already have done that. So which is it?

I’m perfectly chill btw. It’s just a real dick move for you to say “you’re wrong but my methods are top secret.” If what you computed is indeed correct, then only an idiot wouldn’t share it. Your backtracking, contradicting stories and insistence on not sharing but that “we should invest the several hours ourselves” notion tells a story that you just made it all up.

You made the declaration, the onus is on you to back it up. If I tell you gravity doesn’t exist and you ask me how and I respond with “If youβre accusing me of faking the numbers, prove it. Itβs all easily obtainable public information” then I wonder how many people would think I wasn’t caught in a lie.

None of your actions make any sense

• d735123

@DFSx42 said...

Followed by

“Iβm not wasting two hours typing names and scores into a spreadsheet.”

Followed by

“I never said I didnβt write anything down. Stop putting words in my mouth.”

Followed by

“Then I entered just the numbers into Excel to calculate mean and SD.”

wait… so first “you’re not wasting time entering information on a spreadsheet” but then a few posts later mention you already have done that. So which is it?

I’m perfectly chill btw. It’s just a real dick move for you to say “you’re wrong but my methods are top secret.” If what you computed is indeed correct, then only an idiot wouldn’t share it. Your backtracking, contradicting stories and insistence on not sharing but that “we should invest the several hours ourselves” notion tells a story that you just made it all up.

You made the declaration, the onus is on you to back it up. If I tell you gravity doesn’t exist and you ask me how and I respond with “If youβre accusing me of faking the numbers, prove it. Itβs all easily obtainable public information” then I wonder how many people would think I wasn’t caught in a lie.

None of your actions make any sense

Wow, this is really out there. I said I entered just the numbers in Excel. This was to quickly calculate the means and SDs. I had just two columns of numbers with no names attached. Typing out 6 players’ names for each night would take a lot longer. I did not do that. Again, you’re accusing me of things. I don’t know why.

• ### Blogger of the Month

@DFSx42…

Thanks for the thoughtful response…you make some excellent points. I won’t quote your response for the sake of brevity but will try to address your points.

I completely agree with you about traditional hitter X vs pitcher Y BvP stats. While I like the premise (it stands to reason that a player can have another player’s number), the sample size is inherently too small to be meaningful. I would have left it out altogether but it ended up being about 10 lines of code (out of ~6000 total) to include them so I did. Since it only includes the past 3 seasons and is weighted by PA, it either makes no difference or is just a tiebreaker of sorts. No harm, no real value.

I still debate whether it makes sense to isolate switch hitters in my platoon splits, etc and might very well decide to undo that and simply treat a switch hitter as a R or L depending on the opposing pitcher handedness. My current thinking is that it makes sense to treat them as a separate category because switch hitters often remain in games longer (i.e. they don’t get benched) when the bullpen takes over, which boosts FPTS per game. Of course this isn’t reflected in my FPTS/PA tables that I include in my blog because they’re normalized. Anyway, the only real downside is that the sample sizes for switch hitter splits are smaller, but I figure that by including the PA numbers in the tables my readers can make their own decisions about whether it’s a meaningful sample size for a given pitcher.

My FPTS/PA stats do indeed include everything, including stolen bases. I agree that in some cases this might be misleading but one thing I’ve learned in my professional experience working with data is that we should resist the temptation to make premature judgements about what the data tells us (hence my silly #TrustTheData tag). There are often hidden correlations that aren’t apparent at first glance. In this case, if a certain pitcher gives up way more stolen bases than other pitchers, it might be correlated to his catcher(s), his pitch speed, etc. But I certainly see your point.

• ### Blogger of the Month

@d735123 said...

DataDetective, if you think a nightly h2h will provide a more valid test of my hypothesis (I donβt), then letβs do it. It seems this is the only way I can prove my manhood to DFSx42.

Sure, let’s do it…win or lose, I think it’ll be a fun exercise. Please PM me your FanDuel username and I’ll set it up. Is \$1 per contest OK just to have a little skin in the game?

• d735123

Sure, let’s do it…win or lose, I think it’ll be a fun exercise. Please PM me your FanDuel username and I’ll set it up. Is \$1 per contest OK just to have a little skin in the game?

OK, sounds good. My FD name is same as the one here. Thanks for setting up.

I’ll use only run total+batting order and attempt to fit in the best team possible under salary constraints. The hypothesis I wanted to test was that model-based projections incorporating various batter stats provide no useful additional information beyond that implied by run total and batting order. This is a very roundabout way of getting at that hypothesis. But perhaps we’ll learn something.

• ### Blogger of the Month

@d735123 said...

OK, sounds good. My FD name is same as the one here. Thanks for setting up.

Ok, I set up a FD league for our little challenge. Their smallest league is 3 people so it might run as a free contest, but I figured the league would help us track results. Good luck! May the best strategy win π

• DFSx42

I would have left it out altogether but it ended up being about 10 lines of code (out of ~6000 total)

this is something i’m very jealous of, i work in tech but on the product side and don’t want to involve colleagues in a sports gambling project (i assume you feel the same) so most of what I do it via duct tape and trial and error

very jealous, i’m actually really tempted to just spend my next vacation as a staycation and just sit down at my home office and pound away at the keyboard until I have something that functions with such grace and beauty. Most of my time is spent trying to figure out why it’s not working, unable to fix it, wasting hours trying to learn how to diagnose it, unable to do anything and then mentally exhausted end up just copy pasting and doing a lot of stuff manually

it reminds me of the time i first started doing my golf model with a pad and pen while on vacation with no laptop – those were simpler times

• ### Blogger of the Month

@DFSx42 said...

Most of my time is spent trying to figure out why itβs not working, unable to fix it, wasting hours trying to learn how to diagnose it, unable to do anything and then mentally exhausted end up just copy pasting and doing a lot of stuff manually

Haha, welcome to software development. ;-) Seriously, though, it gets easier with more experience…you’ll spend more time with the fun stuff (solving problems, designing solutions, etc) and less on the frustrating debugging. If you’re open to learning a new programming language, I highly recommend Scala (which is like Java++ in a lot of ways) for apps that manipulate data. It’s an elegant language and what I used to write my MLB projections app.

Anyway, I want to thank you for getting me thinking again about the way I handle platoon splits. You’ve convinced me to tweak the way my app handles switch hitters. Starting with my next blog, I will no longer only consider how the opposing pitcher has fared against switch hitters but will instead use a weighted average (by PA) of the opposing pitcher’s FPTS/PA against switch hitters and either right- or left-handed hitters (depending on the pitcher’s handedness). This should address the issue where a small sample size against switch hitters skews the results. I’ll do some back-testing to make sure this improves the correlation between my projections and actual results, but I’m confident that this will be an improvement. Thanks for the feedback!

• DFSx42

Never once looked at scala and will definitely check it out. Experimented with bits of python, R & the standard excel vba and use a frankenstein’s monster combination of the 3. It’s not pretty, I need to spend way too much time hands on with it but at the end of the day it does some pretty neat stuff.

Thanks for the tip to look at scala and wish you the best with your own journey with your own model.

• ### Blogger of the Month

@d735123 said...

OK, sounds good. My FD name is same as the one here. Thanks for setting up.

I’ll use only run total+batting order and attempt to fit in the best team possible under salary constraints. The hypothesis I wanted to test was that model-based projections incorporating various batter stats provide no useful additional information beyond that implied by run total and batting order. This is a very roundabout way of getting at that hypothesis. But perhaps we’ll learn something.

Soooo…after about a month I’m winning 23-11. Ready to give up and concede that data and research matters or do you want to continue? π

• d735123

Soooo…after about a month I’m winning 23-11. Ready to give up and concede that data and research matters or do you want to continue? π

Nice run, but your average score has been 138.7 while mine has been 131.1 (with standard deviations of 37 and 41, respectively). A t-test says this difference is very much in the realm of chance (p = .38). I’m happy to continue if you are. Though if you’re bored with it all, I understand.

• ### Blogger of the Month

@d735123 said...

Nice run, but your average score has been 138.7 while mine has been 131.1 (with standard deviations of 37 and 41, respectively). A t-test says this difference is very much in the realm of chance (p = .38). I’m happy to continue if you are. Though if you’re bored with it all, I understand.

Sure, let’s continue…I plan to play MLB until at least the end of July. BTW, I do find it curious that you drew your initial conclusion about my projections from a small 30-slate sample (of only 3 hitters) but now you’re brushing off a similar 34-slate sample (of complete lineups). ;-)

• DFSx42

• d735123

Sure, let’s continue…I plan to play MLB until at least the end of July. BTW, I do find it curious that you drew your initial conclusion about my projections from a small 30-slate sample (of only 3 hitters) but now you’re brushing off a similar 34-slate sample (of complete lineups). ;-)

I wouldn’t say brushing off. I’m indeed paying attention to these results.

• ### Blogger of the Month

@d735123 said...

I wouldn’t say brushing off. I’m indeed paying attention to these results.

Just messing with you… let’s see how it looks in another month

• d735123

Just messing with you… let’s see how it looks in another month

OK, it’s another month. Here are the totals I have:

Out of 60 contests, you won 32, for a winning percentage of 53%. A binomial test tells us the outcome of our matches is not different from flipping a coin (p=.7).

Your lineups averaged 136 points and mine averaged 139 (with SDs of 37 and 40, respectively). A t-test tells us there’s no difference here either (p=.59).

I remember you said you were thinking of finishing with baseball at the end of July. Do you want to call it a wrap?