Home > Statistics Every Writer Should Know > The Stats Board > Discusssion
logistic regression
Message posted by j. boulton (via 142.106.180.134) on November 28, 2001 at 11:39 AM (ET)
My dependent variable is presence or absence with several independent variables. My problem is that values over 1000 (about 1/3 of each variable was over 1000) could not be verified, therefore anything over 1000 was recoded as 1001. do i have to transform the data for the regression analysis?
READERS RESPOND:
(In chronological order. Most recent at the bottom.)
Re: logistic regression
Message posted by Mark Waters (via 128.243.220.46) on November 28, 2001 at 12:06 PM (ET)
I would imagine - but I am not sure, wait for more responses - that no trabsformation will fix the fact that you'll have about 1/3 of all your points registering 1001. Well, I know that will be the case. I think the only way of fixing it is to try and get a regression from all the points below 1000 and, I'm afraid, junk the rest.
Re: logistic regression
Message posted by j.boulton (via 142.106.180.134) on November 29, 2001 at 11:24 AM (ET)
thanks for the brutal suggetion but the sample size would be too small if i were to junk values over 1000...
Re: logistic regression
Message posted by Darius (via 200.23.217.10) on November 29, 2001 at 12:42 PM (ET)
Mark sounds ok, but other way could be suppouse that is a normal distribution, 'dump' the truncated data and make an histogram, and replicate the bellow the mean (left side of the histogram) to the other side of the histogram (the zone of the truncated data).
Re: logistic regression
Message posted by j.boulton (via 142.106.180.134) on November 30, 2001 at 1:18 PM (ET)
i really can afford to dump the 1000+ data and the mirror tactic(Darius) is too risky, i shouldn't have to make up data when it's there---would a weighted regression be too messy?
Re: logistic regression
Message posted by j.boulton (via 142.106.180.134) on November 30, 2001 at 1:21 PM (ET)
i meant : i can't afford to dump the 1000+ data
Your $5 contribution helps cover part the $500 annual cost of keeping this site online.