[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]
"Maarten Buis" <M.Buis@fsw.vu.nl>
Mon, 2 Oct 2006 11:59:23 +0200
You are fitting a nonlinear curve, so the effect of x changes as x changes. See for instance slide 21 in my presentation. With -mfx- you are summarizing this with one number, so you are approximating this non-linear curve with a straight line. In other words you are estimating the effect of a unit change for an average individual (default), or an individual with the values of x you specified. If the unit of your x is pretty small this approximation is usually fine, but if it is large you get strange results like you have just shown. -mfx- won't standardize the explanatory variable for you, so if you haven't standardized the variable uba, than -mfx- will show you the effect of a unit change, where the unit is the original unit in your dataset. The solution is to choose a smaller unit, for instance by dividing uba by some number (10 or 100) to improve the approximation.
Maarten L. Buis
Department of Social Research Methodology
Vrije Universiteit Amsterdam
1081 HV Amsterdam
Buitenveldertselaan 3 (Metropolitan), room Z434
+31 20 5986715
From: email@example.com [mailto:firstname.lastname@example.org]On Behalf Of Cheng, Xiaoqiang
Sent: maandag 2 oktober 2006 11:44
Subject: st: RE: st: ´ð¸´: st: How to choose a proper model if the dependent variable is within bounds?
However, there is a problem now:
-mfx- predicts that:
variable | dy/dx Std. Err. z P>|z| [ 95% C.I. ] X
uba| -3.05144 1.55749 -1.96 0.050 6.10406 .001175 .004839
And the effect of uba on dependent variable exceeds unit if one std. dev. Of uba changes, certirus paribus?
* For searches and help try: