Testing this model for MLB 2011

**Maverick22** · 01-26-11, 12:44 AM

You did not build the model using the 2009 or 2010 data did you?

**ljump12** · 01-26-11, 12:47 AM

Originally posted by Maverick22

You did not build the model using the 2009 or 2010 data did you?

This.

**OMGRandyJackson** · 01-26-11, 01:04 AM

What do you mean?

**josie88** · 01-26-11, 03:08 AM

Sharp Sports Betting by Stanford Wong chapter 7 - testing W/L records for significance

**lovesbaseball1** · 01-26-11, 08:05 AM

why did you scrap all the data before you made the model?

**Maverick22** · 01-26-11, 08:07 AM

how was he supposed to make the model with no data?

**Maverick22** · 01-26-11, 08:16 AM

The same data that you used to make the model. (Unless you simply pulled this model out of your ass/backside.) Should not be used to "test" it.

There should be two sets of data. One for construction of your model. One for testing.

For example, NBA. You might build your model using the 2000-2005 Seasons. And you might validate/test it using the 2006-2009 and 1995-1999 seasons.

The book mentioned above will explain why, but I just gave you the readers digest version.

**OMGRandyJackson** · 01-26-11, 01:14 PM

Ok I think I see what you mean. When we tested it on the 2009 season no data from 2009 or 2010 was used. When we tested on 2010, we did add the 2009 data back in. And then for 2011 we plan on adding 2010 data in.

Is this what you mean?

**blackbart** · 01-26-11, 04:38 PM

how may degrees of freedom?

**LLXC** · 01-26-11, 04:40 PM

Cross-validation

**OMGRandyJackson** · 01-26-11, 04:54 PM

We are new to this what does degrees of freedom mean and what does cross validation mean?

Thanks.

**Maverick22** · 01-26-11, 09:31 PM

How many years of MLB data do you have?

If you only have 2009 & 2010. "They" will tell you, that you need more data than this...

**OMGRandyJackson** · 01-26-11, 11:33 PM

We have used many more seasons then just 2009 and 2010 lol.

**OMGRandyJackson** · 01-31-11, 12:23 PM

bummpp

**DirkDiggs** · 01-31-11, 05:31 PM

Take 2009 and 2010 out of the model.

Use one set of data to build the model and use a different set to validate the model. So, for example use 2000-2008 to build the model. Then test what you have come up with on the 2009 and 2010 seasons to see if it holds true.

**OMGRandyJackson** · 02-01-11, 01:53 PM

That is what we did...we took the seasons we used before 09 and tested on 09, then we added 09 to test on 10, and then we are going to add 10, to use on 11.

Are you saying do not add 09 and 10 data for 2011? And just stick to the data from seasons before that?

**bztips** · 02-01-11, 04:21 PM

Originally posted by OMGRandyJackson

That is what we did...we took the seasons we used before 09 and tested on 09, then we added 09 to test on 10, and then we are going to add 10, to use on 11.

Are you saying do not add 09 and 10 data for 2011? And just stick to the data from seasons before that?

You could add in the 09 and 10 data, then test for whether those added years are statistically different from the prior years, using a Chow test (a form of an F test). If they are different, then that's an indication that something structural/fundamental has changed and your model based on pre-09 data may no longer be valid/relevant.

**OMGRandyJackson** · 02-02-11, 07:23 PM

Ok this is what we've done so far. We created a model using seasons 05-08, we tested this on 09, and it showed profit. We then added 09, and tested on 10 and it showed similar profit. So for 2011 we want to add 2010 to it and use it in 2011.

**batigol** · 02-03-11, 09:28 AM

good idea !

**dwaechte** · 02-03-11, 11:50 AM

Originally posted by batigol

good idea !

gpia