Tuesday, February 17, 2009

Type II Errors

Sign of the nerdiness of my colleagues: he was describing a student with passionate, constantly changing opinions of him, "her attitude could be graphed like a sine wave."

Sign of the nerdiness of me: I took (okay, gave blood for) a prenatal sceening test this morning. The doctor reminded me that it was good for ruling stuff out (if negative, 95% chance baby is fine), but not good for confirming anything (if positive, 97% of time nothing is wrong). I had to try and make this into a statistical test for my senior seminar class, if Type I error rate is 5% (5% of negatives are false negatives) and Type II error rates are 97% (97% of positives are false positives), can we determine anything about probability of underlying conditions? I don't think so unless we know the actual positive/negative reporting rates. Anyone geekier than I.

Alloicious Nukular (aka E. Mervivan Phogg) heart is beating just fine, by the way.


Sparkling Squirrel said...

Okay, I did some further research and found out that it comes back positive about 5% of the time. All sorts of interesting stats can ensue. Just based on pure statistics, I have something of a 90% chance of falling in the category I want to (reject false, or true negative).

Irene said...

I'm glad to hear that the heartbeat is going strong!

Debbie said...

Maybe you could get a publication of of this blog.

Sparkling Squirrel said...

After all that I mixed up my type ones and type twos. I suppose part of the issue is what's the null-- is a positive result accepting the null or a negative. Anyway, the probabilities work out really strangely-- the probability of a false negative (.0475) is about that of a true positive (.045), so if there are abnormalities, there's about a 50/50 chance of catching them with this test.