Re: st: Centering variables for interactions and using mi regress

[Richard Williams <richardwilliams.ndu@gmail.com>]

Interesting question, and I am not sure if there 
is any official wisdom on this.  One reason to 
center is so that 0 is a meaningful value.  For a 
variable like education, though, it might make 
sense to deduct 12 from each case, so that a 
score of 0 on the new variable corresponded to 
high school graduate. But in any event, if my 
life depended on it I think I would feel better 
doing the centering before imputation, rather 
than, say, have the centering be done differently with each imputation.

Also, I am not sure if this is a good idea, but I 
suppose you could run the analyses without 
centering, use the -mean- command to get the 
imputed mean for each variable, and then center using the imputed means.

As for the interactions, Paul Allison posted a 
while back that you should go ahead and compute 
interactions with the unimputed data and then let 
imputation fill in the missing values.  This may 
seem counter-intuitive, in that, say, if X1 has 
an imputed value of 3, and X2 has an imputed 
value of 7, the imputed value for the interaction 
X1X2 need not be 21! But Allison argued that you 
get biased results if you compute the interaction 
after X1 and X2 have been imputed.

I am just sort of guessing with all this though; 
listen to the experts if they come forward!

At 04:12 PM 1/15/2011, Amy S wrote:
> Hello everyone!  This may be a very basic 
> question for most of you, but I am running 
> regressions using multiple imputation (mi 
> regress) in stata 11.  Because my models have 
> interactions, I need to center the continuous 
> variables used in those interactions.  However, 
> I am unsure how to do this for the variable I am 
> imputing.  In the past, without multiple 
> imputation, I have just centered the variables 
> first, and then used the new centered variables 
> (and interaction terms with those center 
> variables) in the regression.  However, because 
> the missing data on the variable is imputed 
> based on the regression model itself, I donât 
> think I can do that.  Please correct me if Iâm 
> wrong, but  I doubt itâs legit to simply begin 
> by centering the variable and then use mi 
> regress to impute with the centered form of the 
> variables, is it?  Even if that is legit, how 
> would I create the interaction term before 
> putting it into the regression if all the data
>  has not been imputed yet and therefore there 
> are missing variables?  Maybe there is a 
> simple way to do all of the centering and 
> creating interaction terms and runnign the 
> regression model with imputation all in one step - I'm hoping so!  Â
> Â
> Thanks in advance!
> Â
> -Amy
>
>
>
>
>
> *
> *   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/

-------------------------------------------
Richard Williams, Notre Dame Dept of Sociology
OFFICE: (574)631-6668, (574)631-6463
HOME:   (574)289-5227
EMAIL:  Richard.A.Williams.5@ND.Edu
WWW:    http://www.nd.edu/~rwilliam


*
*   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/