[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

st: RE: Imputing values for an Index

From   "Nick Cox" <>
To   <>
Subject   st: RE: Imputing values for an Index
Date   Wed, 17 Sep 2008 14:08:50 +0100

This raises questions on several levels. 

I am queasy at your tactic of finding a reference that supports what you intend to do. 

I don't think the alternative of saying that you can't do this if you have missing values is that outrageous, but I don't know your context. 

All methods of dealing with missing data other than going out and measuring again -- when that is possible -- are deeply unsatisfactory, but some are more deeply unsatisfactory than others. 

There definitely are different takes but 

1. I would impute upstream, and then do -pca-. If you do -pca- and then impute, your interpretation problems seem even greater. 

2. I would always compare results for complete observations and for complete plus imputed to see what differences there are. You gain from knowing how bad the problem is, 

3. -impute- is among the lousiest ways to do this. -ice- (use -findit-) is generally considered much better, it seems. I've never use it....


John Ataguba

I am implementing the -pca- command to generate an index of socioeconomic status.  Stata will not provide values for any observation that is missing in any variable when you use the -predict, score- after the -pca- command.

I decided to use the -impute- command to fill in the missing values of the index.  Noting particularly that the missing values are not much and could be assumed to be random.

My problem is that I need a valid reference to backup the process or to see if there is a more neater way to get around this.

*   For searches and help try:

© Copyright 1996–2017 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index