| 0 |
Subject: Unexplained Data
Posted by: APerfect10
- Leader [70412023] Mon, Apr 07, 2003, 13:29
Sampling from 500 teams of which 439 are valid
From Saturday's Roster Freeze to Sunday's Roster Freeze
244 total trades of which:
25 total (10.25%) Hank Blocck buys: Blalock gains 140k
78 total (31.96%) Griffey Sells
From Sunday's Roster Freeze to Monday's Roster Freeze
154 total trades of which:
35 total (22.73%) Hank Blalock buys: Blalock gains 120k
22 total (14.28%) Griffey Sells
Someone please explain... |
| 1 | SillySpheres at Work
ID: 582492810
Mon, Apr 07, 2003, 13:30
|
too small of a sample?
|
| 2 | APerfect10
Leader
ID: 70412023
Mon, Apr 07, 2003, 13:33
|
And post #1 is not a valid response :)-
A sample size of 500 leads to (atleast) 95% accuracy. To lazy to do the math. Even if every error possible was made with Blalock, it doesnt explain the data.
|
| 3 | tduncan
ID: 423452813
Mon, Apr 07, 2003, 13:41
|
Yersterday's price changes weren't normal.
|
| 4 | Matt G
ID: 43047610
Mon, Apr 07, 2003, 13:45
|
Thge only thing I can think of is more residual buys from the 3 prior days, if he was bought in mass, it might run over percentage wise...
|
| 5 | APerfect10
Leader
ID: 70412023
Mon, Apr 07, 2003, 13:55
|
I'm not sure what you mean by 3 prior days. Hitters prices are only affected by the prior days trading. Percentages doesnt explain it either, look above...
While I'm at it, I'd like to complain about the TSN servers. I'm tired of it taking 10 minutes to make 2 trades on a cable modem! We're paying for a service, provide us with the means of using it properly.
|
| 6 | The Left Wings
ID: 6142019
Mon, Apr 07, 2003, 13:57
|
On the other hand, I have no problems accessing my basic teams.
|
| 7 | johnnyrotten
ID: 20344619
Mon, Apr 07, 2003, 13:58
|
isn't the formula all about the ratio of buys to sells??
|
| 8 | johnnyrotten
ID: 20344619
Mon, Apr 07, 2003, 14:00
|
AP10 - and by the way - thanks loads for posting the roster sample stuff. great info!!
|
| 9 | Jags
ID: 212312612
Mon, Apr 07, 2003, 14:09
|
AP10..always enjoy ur informative posts..and I'm still trying to completely understand TSN's price changes(my first year)
Is it possible that the price change also factors in the present value of a player?
i.e. A $4M player and a $1M player both have same % ownership and are traded by exactly the same # of teams..does the more expensive player maybe show less of an increase..due to his already higher value?
And therefore Blalocks smaller increase(from higher trade %) was due to the fact he was $140,000 higher than the day before?
|
| 10 | Guru
ID: 330592710
Mon, Apr 07, 2003, 14:11
|
My best guess - TSN added back the logarithmic (square root) dampening factor that was in place last year. Perhaps they had not intended to exclude it this year, and yesterday's price change pointed out the inadvertant change. Or perhaps they got so many complaints about Griffey's drop yesterday that they chickened out.
|
| 11 | Guru
ID: 330592710
Mon, Apr 07, 2003, 14:14
|
BTW, I forget exactly how the square root process was applied last year. Perhaps Richard or someone with a better recollection could assess whether that could explain this degree of daily disparity.
|
| 12 | APerfect10
Leader
ID: 70412023
Mon, Apr 07, 2003, 14:21
|
isn't the formula all about the ratio of buys to sells??
That ratio is always 1 ;)
Guru, your logic is the same as mine. The point of my post was to prove that TSN changed the formula, which IMHO is unacceptable. You dont change the rules after the game starts.
|
| 13 | Guru
ID: 330592710
Mon, Apr 07, 2003, 14:23
|
I hope my guess is wrong, too. Changing from one year to the next is OK - even desirable, if there are features from one year that need to be adjusted.
But changing the formula during the course of a season - especially without any notice - is disturbing. As I recall, Small World did that same thing about 3 years ago, and took a lot of heat for it. We should be able to rely on day-to-day consistency.
|
| 14 | Farn
Donor
ID: 7822711
Mon, Apr 07, 2003, 14:24
|
I wish I was smart enough to run a spreadsheet.... :(
Anyway, I tend to agree with Guru. My guess is TSN changed the formula between noon yesterday and noon today due to people complaining. Or they just plain forgot to change it for the start of the season and didn't realize it until today.
Like I said yesterday though, I think its wrong they would change their procedures mid game, at least without telling us. What if I made a move today thinking that the same pricing used yesterday was being used today? I agree they shouldn't tell us what their formula is but they should be held to one formula and not different ones for different days.
|
| 15 | Matt G
ID: 43047610
Mon, Apr 07, 2003, 14:29
|
My point AP10 was looking at Friday to Saturday and Thursday to Friday...
If at Friday's freeze and Saturdays Freeze his buys were not equivalent to his Gain then its possible that it pushed over from the days before....I thought it was 5 day average so for Sunday we see Wednesday -> Sunday's buys/Sells and on Monday we see Thursday -> Monday. Meaning if Sunday's gain is more then Mondays, we can see that buying from the one period is less that the other.
I only see 2 days of buying samples above meaning 2 Freeze to freeze periods.
|
| 16 | Farn
Donor
ID: 7822711
Mon, Apr 07, 2003, 14:31
|
Matt G,
Hitter prices only factor in the previous 24 hours. Pitcher prices work on the 5 day period.
|
| 17 | Farn
Donor
ID: 7822711
Mon, Apr 07, 2003, 14:32
|
I wonder if Erik, Bernie, or Larry will ever respond to this. Anybody who emails them often want to see if they can get a fair response about this?
|
| 18 | APerfect10
Leader
ID: 70412023
Mon, Apr 07, 2003, 14:32
|
Actually every price change thus far seems logarithmic, minus Sunday.
ie)
Fri-Sat
Sanders (14.75% buys) -> +$130K
*This number seems accurate compared to todays pricing
Sat-Sun
Sanders (7.78% buys) -> +$200K
*This number is inaccurate
So it appears TSN tinkered with something on Sunday and only Sunday.
|
| 19 | Guru
ID: 330592710
Mon, Apr 07, 2003, 14:49
|
AP10 - I assume you saw Richard's analysis in [22] of this thread? He seemed to think that linear pricing was in place before Sunday.
We probably need more data to know for sure. But why are you and he coming to different conclusions about the changes for days prior to Sunday?
|
| 20 | johnnyrotten
ID: 20344619
Mon, Apr 07, 2003, 14:54
|
I'll bet when you do your next roster sampling, it will either make perfect sense, or will jighlight that they messed with the numbers Sunday. Griffey was owned by 32% on 4/3, and Blaylock 16%. I'll bet Griffey had built to around 36% by 4/5. He has gone down 690k since. wonder what those %s are now...
|
| 21 | APerfect10
Leader
ID: 70412023
Mon, Apr 07, 2003, 15:15
|
Guru, thanks for pointing that thread out. I did not read Richard's post. However, I am having a major discrepency with his numbers versus mine.
His net trade %'s are identical for some, and radically different for others.
|
| 22 | johnnyrotten
ID: 20344619
Mon, Apr 07, 2003, 15:22
|
i saw a post from zio, which said that griffey went from 34% on 4/5 to 11% yesterday
|
| 23 | Richard
ID: 10112315
Mon, Apr 07, 2003, 15:29
|
Interesting stuff, AP10. Unfortunately, it will probably be Tuesday night before I can work my data again. I'm only looking at 300 teams (randomly selected at the start of the season) so I'l have a bit more scatter/uncertainty in my sampling.
I'd like to write up what I do in a fair amount of detail and share it with you but that might take a day or two. I've got a fly fishing club board meeting tonight to distract me from TSN baseball pricing follies. I'll be back at the data Tuesday night unless taxes get in the way.
Richard
|
| 24 | Matt G
ID: 43047610
Mon, Apr 07, 2003, 15:32
|
That explains why im confused....stupid hitters, sorry AP10 I now see, and am displeased
|
| 25 | Zio
ID: 13255303
Mon, Apr 07, 2003, 15:45
|
AP10 and others who are doing sampling:
How are you "randomly" selecting which rosters you sample?
Are you sampling the same rosters every day or do you re-select who the sample population will be each day?
Are you including imcomplete rosters in the totals? Why or why not?
I'm not well versed enough in stats to analyze what the impact of these decisions might be, but it seems possible these could be sources of error.
However the sample size is more than adequate. Here is random sample calculator that tells us that a sample of 638 of 16692 rosters gives us a 99% confidence level. But that is if the sample is truly random of course.
For my samples so far, I have sampled team ids from 1 to 2000, which yields 1868 rosters, and I have included imcomplete rosters in my total because their trades count like the others.
I'm tempted to sample 100% of the rosters for a week or two so we can have solid numbers to work with, but I suppose that's somewhat excessive.
|
| 26 | CanEHdian Pride
Donor
ID: 48936413
Mon, Apr 07, 2003, 16:00
|
The numbers do seem odd but there is something missing from this thread.
Logic.
What would be the logically reason for TSN to change the pricing algorithm for one day? Was there anything wrong with it? Any major problems? Why would they be testing new algorithms if there was no major problems with the current one.
This leads me to believe that they have been using the same pricing algorithm since day one. Though it may have been changed from last year I find it hard to believe that they thought it would be a good idea to change it for one day.
Which leads to the next problem. Why did they change it back to last years model?
Perhaps it was due to the flood of emails I'm sure they received with complaints about the new algorithm. People arguing that it should have been stated that it had been changed and things of the sort. In a service business, if you get enough complaints, you act.
I believe that this is what transpired in the last couple of days and I don't fault them for changing it because everyone was up in arms about the price drop, eventhough it seemed inline with some of the hockey changes.
To answer the question of whether the news of a new algorithm should have been made known before hand, i'm not sure why it would have to. The question is, would it have changed anyone's strategy simply knowing that price changes were slightly different? I can't see how it would until data started to come into play about exactly how the algorithm had changed. Since it seems to be merely a change to the formula and not one to the entire mechanics of changes (as was the 5 day model, which was brought to our attention) I don't see how advanced notice would have prepared us anymore than none at all.
|
| 27 | Bandos
ID: 422571916
Mon, Apr 07, 2003, 17:03
|
While the numbers I have seen are slightly disturbing, Pride makes a point. I haven't been surprised by anything. Price changes seem to make sense based upon 4 years of playing and getting a sense of patterns.
That being said, I find it interesting that no one from TSN has even tried to address this either here oir on the unintentional comedy scale 100 TSN boards.
|
| 28 | RecycledSpinalFluid
ID: 42121814
Mon, Apr 07, 2003, 17:19
|
Well, I feel good with my 624 team sample I have run a couple of times so far. I wonder how many teams are being duplicated in all this sampling...
Random is as random does. I just used excel to create 1000 random numbers in a range. Ran it. Culled out the invalid numbers and came up with 624 good ones (including non-active teams).
|
| 29 | APerfect10
Leader
ID: 70412023
Mon, Apr 07, 2003, 17:29
|
Zio:
How are you "randomly" selecting which rosters you sample?
Originally with a random number generator.
Are you sampling the same rosters every day or do you re-select who the sample population will be each day?
Same rosters every day. Over time, I will throw out my old data and compile a new sample set.
Are you including imcomplete rosters in the totals? Why or why not?
I do not include incomplete rosters. We've had long discussions in the past, whether or not to include incomplete rosters. Either way, it lower the percentages, but everything is relative so it shouldnt alter the significant outcome.
Richard, I would love to share information with you. I believe I still have your email address, if you've lost mine, I'll shoot you an email. Let me know when.
Here is a small set of data, which appears to prove my theory. The left column is the percent of trades, the right column is the absolute value of the price change. The left set of data is for Thursday, Friday, Saturday, & Mondays price change, while the right set of data is for Sunday. While the data shows the sampling error, its accurate enough to determine which days are similar and that Sunday's price change doesnt fit.
|
| 30 | APerfect10
Leader
ID: 70412023
Mon, Apr 07, 2003, 17:29
|
| TR/Fri/Sat/Mon | | Sun | | % | abs(PC) | | % | abs(PC) | | 0.000 | 0 | | 0.000 | 10 | | 0.004 | 10 | | 0.008 | 20 | | 0.006 | 10 | | 0.016 | 50 | | 0.006 | 10 | | 0.020 | 30 | | 0.006 | 10 | | 0.029 | 100 | | 0.006 | 0 | | 0.061 | 90 | | 0.012 | 0 | | 0.078 | 200 | | 0.012 | 10 | | 0.102 | 140 | | 0.016 | 30 | | 0.320 | 660 | | 0.018 | 10 | | | | | 0.018 | 20 | | | | | 0.018 | 50 | | | | | 0.021 | 30 | | | | | 0.026 | 50 | | | | | 0.030 | 30 | | | | | 0.030 | 50 | | | | | 0.032 | 40 | | | | | 0.036 | 30 | | | | | 0.037 | 60 | | | | | 0.037 | 90 | | | | | 0.048 | 90 | | | | | 0.049 | 80 | | | | | 0.058 | 90 | | | | | 0.062 | 50 | | | | | 0.065 | 60 | | | | | 0.067 | 40 | | | | | 0.127 | 70 | | | | | 0.136 | 60 | | | | | 0.143 | 130 | | | | | 0.148 | 130 | | | | | 0.170 | 100 | | | | | 0.170 | 140 | | | | | 0.170 | 140 | | | | | 0.194 | 130 | | | | | 0.227 | 120 | | | | | 0.303 | 130 | | | |
|
| 31 | Richard
Leader
ID: 5410352713
Mon, Apr 07, 2003, 20:30
|
AP10 - drop me an e-mail (I've added my e-mail to this message) since I can't find yours. I've got my ownership data for the hitters ready for to share with you, if you can handle an Excel 2000 spreadsheet.
Richard
|
| 32 | Rogue Nine
ID: 58326714
Mon, Apr 07, 2003, 20:55
|
One thing I don't see discussed too much, is this: Does anyone think that the drop was too much?
My opinion on it: Here we have a player who just may have been the highest owned player in the game. (At worst top 5.) And then he gets so injured that everyone who has him and is still playing (since we're day 7 that should be ALOT of owners) sells him. Zio's sampling said he went from 35% to 11%. Why shouldn't someone who's sold THAT much not drop by a ton? A 200K drop wouldn't be nearly enough to demonstrate just how much ownership he lost. I can understand wanting to cap gains like that, since we don't want money making to be too easy, but I see no reason to cap losers. If your too dumb to save a trade when you own the most highly owned player in the game... you get what you deserve.
Also, contrary to what some have said here, losing 660k isn't completely ruining their season. Its a blow, but thats it. In hockey we had teams with only 60 million still in the top 100 going into the last week. JKaye competed with a much smaller roster value then everyone else two years ago. So they're mistake here means they can't be among the top roster value dudes at season's end. Well, thats what you get for making a blunder this big. Tough luck.
----------------------------------------------
I don't think this phenomenon is all that common either. How often does the highest owned player in the game get injured for 6-10 weeks? In hockey, that 600K Iginla drop was for the mid-season basic game. For those who don't know hockey, he was the best player in the game last year, but struggled mightily in the first half this year. His price for mid-season was obviously very low, and he started his old scoring ways before the break. Obviously he was very heavily owned. When Calgary's schedule turned to crap (not to mention he hadn't been doing much points-wise) he dropped like a rock. Similar circumstances, similar results. I have no problem with it, and I don't think it should be changed.
|
| 33 | Zio
ID: 13255303
Tue, Apr 08, 2003, 05:06
|
Richard and AP10,
I also would be interested in sharing data and thoughts on this puzzle.
I keep my data in MS SQL-Server, but it is easily exportable to Excel, CSV, or a different format if you like.
My roster sampling program is written in C#.NET. I don't know if either of you have experience programming, but I could send you the source code, or if you prefer, just the program to run on your computer. I tweaked it last night and I was able to download 15,946 rosters (I'm missing 800 or so for some reason) in about 8 hours. If you have a faster connection than I do it could take considerably less time.
When you are comparing ownership tags from one day to another, are you looking at changes at the individual team level? For example, if 200 people bought Roy Oswalt and 10 people sold him for a net of 190 buys, are you taking the 10 sales into consideration? If you look only at the net it looks like 190 trades were used, but in reality 210 were used. This occurred to me last night so I haven't had a chance to analyze the impact (maybe this is fundamental and it just took me a long time to realize it). If the calculation is based on NetTrades/TotalTrades then the denominator would be skewed down, but it would be skewed for every player and shouldn't be a factor. But I'm not sure yet, I'll have to take a closer look tonight.
BTW - I'll be around this week, but next week I'm going to the US for two weeks and I won't be able to check my work email.
|
| 34 | Richard
Leader
ID: 53824316
Tue, Apr 08, 2003, 10:09
|
Thanks for the offer of your code, Zio but I'll decline. I'm currently using an Excel spreadsheet I developed to look at the frozen rosters of 310 teams. I generated my list of 310 teams at the start of the season by randomly selecting team_ids between 1 and 15,000. I drop the daily results into an Access database to keep a historic record. A similar process to yours but using different tools.
For my quick analysis, I use the net change in ownership approach. In your example of 200 folks buying Oswalt and 10 selling him, I would capture that as 190 buys for Oswalt. I realize that this isn't the most accurate way to handle this because, as you state, it should really be captured as 200 buys and 10 sells. However, my shortcut still yields fairly useful data.
The movement of players on my small sample usually matches the movement of players from the entire universe of TSN Managers. I'm not using my sample to calculate/verify absolute dollar price changes. Rather I just track overall ownership levels so I can spot players that are either overbought and susceptible to big selloffs or players who are underbought and still might have some upward price potential.
Richard
|
| 35 | Ref
Donor
ID: 100261311
Tue, Apr 08, 2003, 11:33
|
I agree with Rogue 9 in that if the player was dropped by that much, he SHOULD go down that much. It was supposed to be a $1 Mil. cap. THen they went to a new pricing formula a year or two ago where it rarely went over 200K. I don't agree with that. If everyone drops a player on one day, his losses should represent that. Same thing if someone buys them on a certain day (or 5 days).
|
| Rate this thread: | If you wish, you may rate this thread on scale of 1-5. Ratings should indicate how valuable or interesting you believe this thread would be to other users of this forum. A '5' means that this thread is a 'must read'. A '1' means that this is a complete waste of time. If you have previously rated this thread, rating it again will delete your previous rating. If you do not want to rate this thread, but want to see how others have rated it, then click the button without entering a rating, or else click here. |
|
|
|
Post a reply to this message: (But first, how about checking out this sponsor?)
|