# Probability factors?

1. Feb 27, 2014

### 15123

Hello

I am having a hard time understanding which factors should be taken into account and which ones should not be taken into account when calculating probability.

I have the following situation:

Last night I played FIFA12 on xbox live against an unknown opponent. He challenged me to a rematch after the match was over.

I would like to calculate the chance I would have had of winning if I would have said "yes" to the rematch.
However, upon trying to calculate the chance (using Bayes' theory for example), I quickly gathered countless factors such as:

• previous personal win/loss ratio
• the fact he first scored a goal in the first 10 minutes of the first half
• the fact I scored two goals the last 20 minutes of the second half
• his predictability
• my predictability
• my fatigue at that hour; mentally as well as phyiscally
• the knowledge the opponent has about my strategy after having played a match with me
• how tired my opponent is
• the opponent's fatigue; mentally as well as physically
• if he is a night person or not

I keep thinking of factors such as these and I am not sure which ones would be valid factors for probability calculation. I am getting lost.

Does anyone have an idea on this?

Thank you

2. Feb 27, 2014

### MarneMath

In the real world, we often have an extreme long list of variables that may effect an outcome. There are many methods on how to construct probability models and thus make a prediction on an outcome. The only way you are going to make a good model is if you test it repeatedly against the data you aggregate.

Once you make a model, how you constructed it, will often dictate how you can test if a variable contributed a significant amount of information or not. For example, if you choose to make a multivariate linear regression, you could use partial F test or additional sum of squares.

As a side note, I would definitely not recommend you make a Bayesian model using your current knowledge of Bayesian statistics. Bayesian models need a prior probability and finding such distribution for a parameter is will work for your model is non-trivial.

If you want to try something, you could go for a simple model and see how often you tend to win a rematch regardless of an opponent.