Date: Sun, 4 Feb 2007 09:02:33 +0330
Reply-To: Siavoush Dastmalchi <firstname.lastname@example.org>
Sender: "SPSSX(r) Discussion" <SPSSX-L@LISTSERV.UGA.EDU>
From: Siavoush Dastmalchi <email@example.com>
Subject: suitable statistical tests
Content-Type: text/plain; charset="iso-8859-1"
I will appreciate it if you could kindly suggest me a suitable statistical test for the following problem. I have used two tests, but need to know if they are okay.
Suppose I am testing three different methods to predict an structural property for integral membrane proteins called alpha angle (which determines the rotational orientation of the alpha helical transmembrane segment of membrane). I have selected several protein with total of 125 alpha helices. I have measured the alpha angles (125 angles) and then predicted them using three different methods. Then I have performed two different test to compare the performances of three prediction methods:
1. The measured alpha angles were compared with the predicted values and the mean differences are calculated for three methods. Then the mean differences were compared with each other using two-tailed paired t-test.
2. the second way of comparing the prediction methods was as following:
A criterion was set to judge on the performance of a method. For example if a predicted alpha angle was in the range of +/- 20 degree from the measured experimental value, it is considered a correct prediction. Then the number (% or ration) of correct predictions for three different prediction methods were determined. For example method A in 50% of 125 cases predicted the alpha angle with absolute deviation less than 20 degree, method B in 40% of cases and method C in 65% of cases. Then the performances of prediction methods were compared using the statistical test for comparing difference in proportions (comparing P1 and P2).