PrimalKyogre
AG Circuit Champion
That's a lot of words to saySo one day bored I figured I would use basic statistics to get a feel of how much each player would go for in the draft. So I spent a few hours compiling data from each of the 11 mocks to find out
Can mock drafts be used to predict a player's price in the real auction?Methods.
I took each player's price in each mock that they were drafted in and compiled the average and 95% confidence interval, which gave me an appropriate estimate of how much a player should go for in the draft if the managers for real played like the managers in the mock. This is also why I didn't participate in these mocks, as I did not want to alter the data since I was doing this. I use the number of drafts a player was chosen in as the sample size to base the projected price interval for that player.
Results.
Confidence intervals are at the 95% confidence interval. Red represents an inaccurate projection and green represents an accurate projection. The projected intervals are not in multiples of 500 like the draft due to the nature of the calculations. N/A represents the interval of a player only drafted in 1 mock since no interval can be calculated.
Interpretation and Conclusions.
From the data above, out of 56 players drafted (managers don't count and Leru, 64 Squares, and MAMP were not involved in any mocks), 26 (46%) were correctly predicted, 24 (43%) were drafted for less than predicted, and 6 (11%) were drafted for more than predicted. So it does not seem that mock drafts are an accurate indicator of a player's real price. However, mock drafts can still be used as a good gauge for managers to base their draft plans on as 89% of players went for equal or less than their expected price!
There are inherent issues with this method. All 11 mock drafts were included in the data so the players available to draft do not reflect the pool at the end (i.e. Dockiva withdrawing, pichus joining late).
Overall, some very interesting data, AG is unique in that there is a mock draft nearly every day so there was a lot of data to use for this. Hope you all found this as interesting as I did.
Also go crobats! View attachment 352711
id be interested in seeing data for the people who may have been drafted in mocks (perhaps for more than base) but subsequently were not drafted in the real draft...So one day bored I figured I would use basic statistics to get a feel of how much each player would go for in the draft. So I spent a few hours compiling data from each of the 11 mocks to find out
Can mock drafts be used to predict a player's price in the real auction?Methods.
I took each player's price in each mock that they were drafted in and compiled the average and 95% confidence interval, which gave me an appropriate estimate of how much a player should go for in the draft if the managers for real played like the managers in the mock. This is also why I didn't participate in these mocks, as I did not want to alter the data since I was doing this. I use the number of drafts a player was chosen in as the sample size to base the projected price interval for that player.
Results.
Confidence intervals are at the 95% confidence interval. Red represents an inaccurate projection and green represents an accurate projection. The projected intervals are not in multiples of 500 like the draft due to the nature of the calculations. N/A represents the interval of a player only drafted in 1 mock since no interval can be calculated.
Interpretation and Conclusions.
From the data above, out of 56 players drafted (managers don't count and Leru & 64 Squares) were not involved in any mocks), 26 (46%) were correctly predicted, 24 (43%) were drafted for less than predicted, and 6 (11%) were drafted for more than predicted. So it does not seem that mock drafts are an accurate indicator of a player's real price. However, mock drafts can still be used as a good gauge for managers to base their draft plans on as 89% of players went for equal or less than their expected price!
There are inherent issues with this method. All 11 mock drafts were included in the data so the players available to draft do not reflect the pool at the end (i.e. Dockiva withdrawing, pichus joining late).
Overall, some very interesting data, AG is unique in that there is a mock draft nearly every day so there was a lot of data to use for this. Hope you all found this as interesting as I did.
Also go crobats! View attachment 352711
id be interested in seeing data for the people who may have been drafted in mocks (perhaps for more than base) but subsequently were not drafted in the real draft...
Since you have it formatted already, what’s probably most significant is whether, at the 95% level, we can say there’s statistical evidence for whether the number of people above and below their (say, median) mock draft price are different. By the 43% too low, that’s certainly true. This way you’re testing on a larger number of data points and having a safer assumption for a normal distributionSo one day bored I figured I would use basic statistics to get a feel of how much each player would go for in the draft. So I spent a few hours compiling data from each of the 11 mocks to find out
Can mock drafts be used to predict a player's price in the real auction?Methods.
I took each player's price in each mock that they were drafted in and compiled the average and 95% confidence interval, which gave me an appropriate estimate of how much a player should go for in the draft if the managers for real played like the managers in the mock. This is also why I didn't participate in these mocks, as I did not want to alter the data since I was doing this. I use the number of drafts a player was chosen in as the sample size to base the projected price interval for that player.
Results.
Confidence intervals are at the 95% confidence interval. Red represents an inaccurate projection and green represents an accurate projection. The projected intervals are not in multiples of 500 like the draft due to the nature of the calculations. N/A represents the interval of a player only drafted in 1 mock since no interval can be calculated.
Interpretation and Conclusions.
From the data above, out of 56 players drafted (managers don't count and Leru & 64 Squares) were not involved in any mocks), 26 (46%) were correctly predicted, 24 (43%) were drafted for less than predicted, and 6 (11%) were drafted for more than predicted. So it does not seem that mock drafts are an accurate indicator of a player's real price. However, mock drafts can still be used as a good gauge for managers to base their draft plans on as 89% of players went for equal or less than their expected price!
There are inherent issues with this method. All 11 mock drafts were included in the data so the players available to draft do not reflect the pool at the end (i.e. Dockiva withdrawing, pichus joining late).
Overall, some very interesting data, AG is unique in that there is a mock draft nearly every day so there was a lot of data to use for this. Hope you all found this as interesting as I did.
Also go crobats! View attachment 352711
oh no hypothesis testing lol I was just using it in the way I do for research which is just assume that if the value falls within 95% CI it isn't statistically significant. I'll send the raw data your way tho.Since you have it formatted already, what’s probably most significant is whether, at the 95% level, we can say there’s statistical evidence for whether the number of people above and below their (say, median) mock draft price are different. By the 43% too low, that’s certainly true. This way you’re testing on a larger number of data points and having a safer assumption for a normal distribution
n being 11 at most means most of the bounds of the CIs are really, really conservative (and then some with a lower bound less than 3k). Idk if you used Excel but switching to a t over z distribution for the critical score might be a better estimate, even though neither are super useful when bid prices are not continuous (multiples of 500, so discrete).
If you can pass a file with the raw data I could probably throw it into excel or R.