Rich Ulrich <[EMAIL PROTECTED]> wrote in message news:<[EMAIL PROTECTED]>... > On 14 Sep 2003 14:57:43 -0700, [EMAIL PROTECTED] (Bastian) wrote: > > > does it make any sense to do a factor analysis when there are lots of > > missing values (i.e. 320 from 378 values)? > > It sounds as if you have a per-item response rate of less than 20%. > That's right.
> And it sounds as if you want to do a factor analysis on 58 scores, > split among some unstated number of variables and number of > individuals. I hope that is not right. But that is how I read it. > Technically, that could be done if there were complete cases, > but I imagine that you should be able to derive a better > data reduction from examining the available scatter-charts > or cross-tabulations. > The factor analysis comprises 58 variables and there are 11 variables with the problem of too less data (as already said about 320 of 378 values are missing). Nevertheless I don't want to omit these variables completely... > But you don't provide enough information, anyway. > Why are they missing? Is it missing-at-random? > If not, your best hope of making progress might be > an analysis of missing versus non-missing. > Unfortunately the values are missing because the value haven't been risen in some of the evaluations. > > After doing the principal > > component analysis I want to use the factors for a regression analysis > > and I assume it is not very useful to mix-up the variables with lots > > of missing values with my "correct" data without missing values. > > Should I omit the variables? Is there any "rule of thumb" or something > > else to pay attention? > > If a variable is not there, you cannot predict from its score. > Do you have hopes that it will be present, in the future? > > Can you predict from "Missing, Yes versus No?" > > Hope this helps. Thank you very much for your constructive help. Bastian . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
