Dear all, please excuse this humble query by a statistically and otherwise insignificant researcher: I'm would like to compare slope and intercept parameters from linear regression analysis between four different groups. I thought about producing distributions for these 2 parameters by bootstrapping, and then comparing the distributions by KS. My problem is: one of the groups has much more data points at one end of scale than on the other, and due to the randomness of the resampling process this end is 'overrepresented'. Would it be o.k. to randomly exclude points at one end of the scale, just in order to equalise the density of points along the regression line, or would that be similar to data cooking? -------------------------------------------------- Paul H Artes, BSc(Hons.) MCOptom Research Optometrist, University Department of Ophthalmology Manchester Royal Eye Hospital Oxford Rd, Manchester M13 9WH ENGLAND Tel +44 (0161) 276 5615 Fax +44 (0161) 273 6354 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%