2K23: Brandon Marsh

[onejayhawk]

Before COVID he was a hot prospect. Since then he has pieced together a full MLB season, slashing .248/.303/.374 with 16 SB and 13 HR. Still, he plays excellent OF defense and put together a nice 2nd half following a trade to the Phillies.

His early ADP was about #270 but since Christmas (16 drafts) he has dropped past #300. Paul Sporer liked him at #270, see below. What do you think of him now?

https://fantasy.fangraphs.com/4-duds-who-could-be-studs-in-2023/

[harmon]

He looks like Jayson Werth but doesn't hit like him.

35% K rate the last 2 years is too scary for me.

[revo]

He looks like Jayson Werth but doesn't hit like him.

35% K rate the last 2 years is too scary for me.

LOL. Jayson Werthless?

[harmon]

He just went #342 in our 15 team draft and hold.

May be a little harsh for Marsh but lots of skeptics.

[onejayhawk]

Would you rather have Marsh or Oswaldo Cabrera? ADP is about the same.

[Ken]

Would you rather have Marsh or Oswaldo Cabrera? ADP is about the same.

Marsh goes ~3-4 rounds before Cabrera in NFBC drafts

[onejayhawk]

Marsh goes ~3-4 rounds before Cabrera in NFBC drafts

Not lately. Since 24 December, 7 drafts, Marsh is #284 and Cabrera is #300. That's just over a round and they are trending toward each other.

[heyelander]

Phils just traded away most of his ready competition in Veirling and Matos so he should get full-time ABs.

[Ken]

Not lately. Since 24 December, 7 drafts, Marsh is #284 and Cabrera is #300. That's just over a round and they are trending toward each other.

A 7 draft sample is not useful

[Big Tymer]

A 7 draft sample is not useful

I get what you’re saying but as someone who has done 2 Draft Champions now I disagree. You need to look at more recent drafts to ensure you capture signings and role changes and then also filter for the specific format to get a sense for when guys are going. I also then go into the trend charts for each player to further drill down.

100% it’s not enough of a sample to have high confidence but even with a few drafts you will see ADP be more sticky then it should be.

[Ken]

I get what you’re saying but as someone who has done 2 Draft Champions now I disagree. You need to look at more recent drafts to ensure you capture signings and role changes and then also filter for the specific format to get a sense for when guys are going. I also then go into the trend charts for each player to further drill down.

100% it’s not enough of a sample to have high confidence but even with a few drafts you will see ADP be more sticky then it should be.

If you are looking at general ranges sure, but not with the precision that was intended in the post above.

My statement was that the 2 players being compared are 3-4 rounds apart typically. The response, based on 7 drafts, was that no, they are just going 1 round apart. However, from that 7 draft sample if you remove just 1 of the drafts (the one on 12/25, for example), the results move by an entire round.

Randomly choosing a start date and applying 7 samples to counter a general statement is not useful in a statistical analysis. Too many outliers to use it to come up with a result with that type of precision.

[onejayhawk]

If you are looking at general ranges sure, but not with the precision that was intended in the post above.

My statement was that the 2 players being compared are 3-4 rounds apart typically. The response, based on 7 drafts, was that no, they are just going 1 round apart. However, from that 7 draft sample if you remove just 1 of the drafts (the one on 12/25, for example), the results move by an entire round.

Randomly choosing a start date and applying 7 samples to counter a general statement is not useful in a statistical analysis. Too many outliers to use it to come up with a result with that type of precision.

As a trained statistician I disagree that it is too small to be useful. Even one sample has value. For sparse data like this I like at least five, with the date chosen to service that figure. Generally, a shorter time is better.

The offseason began with the two considered multiple rounds apart. This was in spite of Marsh having what appears to be assured PT and Cabrera does not. If anything, Marsh's PT is more assured but his ADP has dropped around two rounds while Cabrera has eased up.

I wonder what Cabrera's role will be. Is he the regular SS in spite of his defense? Is he a place holder? If so, for how long? If he plays then he has value but will he play?

[Ken]

As a trained statistician I disagree that it is too small to be useful. Even one sample has value. For sparse data like this I like at least five, with the date chosen to service that figure. Generally, a shorter time is better.

Really odd response, this doesn't read like the response that a "trained statistician" would post if that were the case. Not saying you aren't, you certainly might be, but if that were the case I'd expect to see you reference data on distribution, sampling error, confidence levels, uncertainty bars, etc.

Consider me... dubious. Anxious to hear the explanation.

To your "even one sample has value" statement, that's odd. Obviously all data points are evenly weighted, but the problem creeps in with small samples because anomalies are more heavily weighted than they would be in a large sample. If we have 100 NFBC drafts where Brandon Marsh is picked 300th and then his cousin picks him 1st overall you are saying that 1 draft is useful? And when we look at the last 7 drafts we should say that he's much more likely to be picked earlier now by ADP? I... disagree.

The "trained statistician" flex is funny though, thanks for the laugh!

[onejayhawk]

Really odd response, this doesn't read like the response that a "trained statistician" would post if that were the case. Not saying you aren't, you certainly might be, but if that were the case I'd expect to see you reference data on distribution, sampling error, confidence levels, uncertainty bars, etc.

Consider me... dubious. Anxious to hear the explanation.

To your "even one sample has value" statement, that's odd. Obviously all data points are evenly weighted, but the problem creeps in with small samples because anomalies are more heavily weighted than they would be in a large sample. If we have 100 NFBC drafts where Brandon Marsh is picked 300th and then his cousin picks him 1st overall you are saying that 1 draft is useful? And when we look at the last 7 drafts we should say that he's much more likely to be picked earlier now by ADP? I... disagree.

The "trained statistician" flex is funny though, thanks for the laugh!

I have a degree in math with a statistics emphasis. Sue me.

All data point should not be equally weighted, which is the point. Recent data is more probative of current thinking.

Your example is of an outlier. There are things you can do about that. One is insisting on at least five data points. Another is check the range of the picks. You can tell when one draft skews a whole group.

[Ken]

I have a degree in math with a statistics emphasis. Sue me.

All data point should not be equally weighted, which is the point. Recent data is more probative of current thinking.

Your example is of an outlier. There are things you can do about that. One is insisting on at least five data points. Another is check the range of the picks. You can tell when one draft skews a whole group.

Where is the number 5 coming from? It's very arbitrary. There are ways to calculate it by looking at the data. If you haven't done that, that's fine, but suggesting that I'm wrong about 7 being way too small without supplying any data at all doesn't sound like something a math/stat guy would do. I was truly interested in hearing how you came up with it.

The rule of 30 samples was thrown out a long time ago, the size of sample you need is dependent on population distribution, static numbers like that aren't useful.