st: RE: RE: RE: mean, mode or median for missing values

["Verkuilen, Jay" <JVerkuilen@gc.cuny.edu>]

Nick Cox wrote:

>There is another take, orthogonal to several good points.  You could impute in all sorts of different ways and then compare the
results with each other and with those from the incomplete data. The ideal outcome is clearly that you might reach the same conclusion, so none of this matters. I can't tell you what to conclude if you get very different results except that the problem of missing data inhibits firm conclusions. <

That's true, missing data does inhibit firm conclusions. In any given dataset there may be missing data with different missing mechanism. Some are MCAR, others MAR and others NMAR (possibly of different flavors). As most general purpose missing data procedures such as EM or MI depend on MAR, what you're often trying to do is to put in enough information to hopefully approximate MAR. The big thing I've learned from doing several missing data analyses is that it's possible to be too inclusive. For instance that don't belong in the dataset shouldn't be kept for the imputation either. Furthermore, multicollinearity created by loading up your dataset with lots of auxiliary variables inhibits convergence substantially. 

Of course, models for censored data or mixed models ARE models for NMAR data, but they illustrate well the issue that you really need to know quite a bit about what's going on to model that situation properly. 



>For "likert" read "Likert", passim. 

Indeed: http://en.wikipedia.org/wiki/Rensis_Likert



*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/