Saturday, March 28, 2009

Confidence in Marriage

I've thought a fair amount lately about dating, marriage, and statistics. Here are the hypotheses I am working off of:

Null Hypothesis: The person you are dating is not who you should marry.
Alternative Hypothesis: The person you are dating is who you should marry.

I guess I've decided that since I do not believe there is one and only one soulmate out there for me, Type II errors shouldn't be too concerning. Or at least as concerning as Type I errors should be.

So, Type I errors are bad, but where do I set my level of confidence? 95%? 99%? I'm not sure. But should I set it too high, I'll never get married. The higher I set alpha, the more likely I am to commit Type II errors.

So what do you do? I don't know. The odd thing is that despite these conclusions, I often am not very concerned about Type I errors, while the tragic feeling of committing a Type II error scares me a lot more. Perhaps I trust my judgment more than I should? Maybe I should be more afraid of Type I errors. What do you think?


Josh said...

In a detection task for which there is sufficient training data, you can sweep through the range of possible alpha values to find a threshold that yields a combination of Type I and Type II errors that minimizes some cost function Cost(type1errors,type2errors).

Unfortunately, available training data for this problem is minimal so you're left to tune your threshold manually. Good luck.


Gabe said...

Maybe this is a problem for Hidden Markov Models.

cking said...

A very interesting intellectual twist you have put on the subject matter. Have you done any further research or come up with any 'significant' findings? Very intriguing.