Batter's Box Interactive Magazine - Predictions are Bunk

Like many things in life, this started off as one idea, and ended up somewhere else entirely.

My original plan was to do what I did last year: come up with the best case scenario and worst case scenario for the Jays' hitters and pitchers, and then use those numbers to estimate how many wins the Jays would pick up. Using this method, last year I confidently predicted that the Jays would score between 546 and 891 runs, and win between 45 and 107 games. Out on a limb, I was.

So I thought I'd try it again. First, the good numbers - here's the best-case scenario for each Jay batter for 2006.

AB R H 2B 3B HR RBI AVG Overbay 579 91 174 46 2 24 106 .301 Hill 574 102 178 41 7 16 79 .310 Adams 570 102 170 34 7 14 74 .298 Glaus 563 120 160 37 1 47 117 .284 Cat 419 56 126 29 5 8 59 .301 Sparky 398 55 107 21 6 8 58 .269 Wells 678 118 215 49 5 33 117 .317 Rios 514 92 155 31 8 22 89 .302 Molina 410 54 121 15 0 17 74 .295 Shea 544 87 163 38 2 23 90 .300 Hinske 211 37 59 15 1 10 37 .280 Zaun 202 32 54 10 0 6 31 .267

Here's how I came up with these figures:

Hill and Adams take significant steps forward.
Rios takes a huge step forward. (This is probably the least realistic of my optimistic estimates.)
Overbay repeats his best season, but with a few doubles turning into home runs.
Hinske is extremely effective as a spot player, as he has the platoon advantage.
For Glaus, Wells, Hillenbrand and Molina, I basically repeated their best seasons. For Zaun, I repeated his best season, but pro-rated.
For Cat and Sparky, I just put in last year's numbers. Neither of them are likely to surprise in any way.

This team of Uber-Jays winds up scoring 946 runs, which is more than any real-life team scored last year in all of baseball, and is 55 runs more than the Dream Jays of 2005 scored.

Now for the bad news - here are my extremely pessimistic projections:

AB R H 2B 3B HR RBI AVG Overbay 579 61 144 27 2 15 61 .249 Hill 574 58 139 19 4 10 46 .242 Adams 546 59 130 25 4 7 49 .238 Glaus 569 79 142 24 1 30 83 .250 Cat 419 56 126 29 5 8 59 .301 Sparky 398 55 107 21 6 8 58 .269 Wells 644 72 169 27 2 24 79 .262 Rios 481 66 126 19 6 10 48 .262 Molina 428 34 105 18 0 5 40 .245 Shea 515 60 144 32 1 17 66 .280 Hinske 211 16 47 9 1 5 21 .223 Zaun 185 18 41 7 1 3 19 .222

Here, everybody either takes a step back or reproduces his career-worst season. (Except, again, for Sparky and Cat, who repeat their 2005 performance.) This team of underachievers scores 634 runs, which is a significantly lower total than that compiled by any real-life team in 2005. (But, once again, this is significantly higher than the low-end total for last year. Yay, J.P.!)

This is the point at which I threw up my hands in despair: why bother making serious predictions about the upcoming season when almost anything could happen? Of course, there's no way that absolutely everything will go right, or go wrong. But we could come closer to either outcome than you might think. Recall the summer of 2003, when the Jays went into a collective hot streak and were scoring about a jillion runs a game. Remember when Delgado and Wells were 1-2 in the league in RBI?

And, for the worst case scenario, you need look no further than 2004. I rest my case.

So the theme for this article is now: predicting the future in baseball is a mug's game. There are too many variables. Luck is too much of a factor. I say let's just sit back and watch the games. Popcorn, anyone?

Posted by Dave Till on Tuesday, March 14 2006 @ 03:25 PM EST.

Predictions are Bunk | 6 comments | Create New Account

The following comments are owned by whomever posted them. This site is not responsible for what they say.

jjdynomite - Tuesday, March 14 2006 @ 04:13 PM EST (#142509) #

If Rios gets over 20 homers and almost 90 ribbies from the 7th/8th spot that would make me one very happy Jays fan.

In a recent article JP mentioned that Zaun "will play even more than he thinks he's going to play", so the ABs you have him slotted for seem a bit low.

Zaun and/or Molina could well DH for Hillenbrand when Shea switches to third to give Glaus a day off, or they could DH in general when Hillenbrand gets a day off. Though Hinske has been impressive this spring so far, for what it's worth.

(Note: This was my first official Bauxite "Jay homer fan" post. More to come).

CaramonLS - Tuesday, March 14 2006 @ 05:34 PM EST (#142514) #

Till, Can you please add OBP.

eeleye - Tuesday, March 14 2006 @ 05:52 PM EST (#142517) #

I think it'll be something more like this, somewhere in between (I've just changed the HR, RBI and AVG columns):

HR RBI AVG
Overbay 24 106 .301
Hill 11 82 .290
Adams 13 74 .280
Glaus 40 117 .270
Cat 8 59 .301
Sparky 8 58 .269
Wells 33 117 .307
Rios 18 80 .269
Molina 14 74 .295
Shea 23 94 .288
Hinske 14 49 .277
Zaun 6 31 .267

Glevin - Tuesday, March 14 2006 @ 06:50 PM EST (#142524) #

"Glaus 569 79 142 24 1 30 83 .250"

Not that it matters in real life (because Glaus walks so much) but he's hit between .250 and .258 every year for five years so .250 is actually pretty realistic. The only predicition that almost never fails is this: Most players (esp. hitters) will be about where you'd expect them to be. Some players will be better than you expect. Some will be worse. Some will get hurt. The key is the number of players who fall into each catagory. (And of course, certain players have higher predictability and injury likelyhoods).

Andrew K - Tuesday, March 14 2006 @ 06:54 PM EST (#142525) #

To add some statistics to this argument, let's suppose that a hitter has a "true" batting average of .280; that is, he has an independent 28% chance of getting a hit at each at-bat (I'm ignoring walks and so on, for the purposes of making the point). Let's say that he gets 500 at-bats in a season.

His "observed" batting average for that season is random, depending on these 500 lots of 28% chances. Some easy calculations show that he is 10% likely to attain a batting average of at least .306, and 10% likely to attain a batting average of at least as bad as 0.254.

We see that luck plays a huge part in his performance, even in as many as 500 at-bats. Although we often caution about "small sample sizes", even a whole season is a pretty small sample size! This is still true, to a lesser extent, if we consider the entire team's batting average, or runs created, or whatever. And it's magnified by the luck of some hits being more valuable than others (hits clustered together are good for scoring runs, isolated hits not so much).

What's the conclusion? Luck plays a big part. When the BPro people (I think it is them) do their 1 million simulated season experiments, what is striking is how widely the outcomes vary.

Better cross your fingers then!

Michael - Wednesday, March 15 2006 @ 05:43 PM EST (#142603) #

Yeah, and on a player-by-player level I think those "best" case/"worst" case are clearly not a true best/worst case in that they aren't extreme enough. To pick one number what do you think the chances are that Glaus hits less than 30 HR. I'd say relatively high. For one he could get injured which is clearly a worse worst case than the line above. For another 30 HR is a lot. BP's PECOTA does percentile breakdowns for players. And, in addition to the very valid points about the range of observed outputs even given a known true value, the ability can change a lot. PECOTA for Glaus has as his 50% percentile hitting 29 HR in 502 PA. They see his 25% percentile score being a .251 AVG (with 25 HR in 483 PA).

They have as his weighted mean 33 HR with a .269 AVG, but it isn't a worse case situation if he ends up with much fewer. BP says 10% of the time Glaus doesn't hits 16 or fewer HR and has an average below .220. A worst case is even worse than that.

But to estimate team scoring what you really want to do is take each players true range of talent (which is generally bigger than what you have above) and then run each player's line separately (I.e., pick a different random value between true worst and true best with whatever that true distribution is) and then sum these values as one run of the team. Then do that 100s or 1000s of times. The team as a whole that way will have a much smaller range of likely values than what you present here even if each player individually has a much larger range.

If you do that my gut says you probably come up with a prediction for the Jays that would be somewhere around 86 or 87 wins with a standard deviation of 5 or 6 wins.

That would give the Jays a 5 or 10% chance to win 95+ wins.

Predictions are Bunk | 6 comments | Create New Account

The following comments are owned by whomever posted them. This site is not responsible for what they say.