Bookmark and Share

Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: st: Multiple Imputation

From   daniel klein <>
Subject   Re: st: Multiple Imputation
Date   Mon, 7 Nov 2011 15:41:50 +0100


did you consult the help file, the pdf-manual, and/or other sources
_before_ writing to Statalist as suggesteted in the FAQ
( It
seems, at least some of your questions can easily be solved using this

You quote the error message, which is generally a good idea. However,
all you probably need to solve your problem is already on your screen.
Stata tells you to use -cd- to change your directory. So you should
either type

. cd <whatever directory you have write permission>

directly, or, for more informartion

. help cd

Note that none of us can tell which directories you may write to and which not.

Concerning your first question, I am not quite sure why you want to do
univariate regression (I assume you mean univariate imputation using
-regress-) when you have more than one variable with missing values.
Univariate imputation is only appropriate if you have missing values
in only one of your variables, as can be seen from the help file and
pdf documentation. So you will have to use a multivariate method to
impute your missing values.

Questions about the propper number of imputations to be added are
briefly discussed in the pdf manual. This number depends, besides
other things, on how large your proporition of missing values is. If
you do not want to go into greater detail here, the number should be
something between 5 and 20, where 5 was a rule-of-thumbs for some
time, manly because computers were slow and it would have been
computationally to burdensome to do more. I think I am already
starting to quote the pdf, so just start there. You will also find
additional literature there.

Concerinign -rseed()- please see the entry

. help seed

or follow the link to -[R] set seed- in the help file for -mi impute-.

Personally, I do not believe the -seed- to be that important, if your
are not running simulations, but others (knowing more about potential
problems with computers generating "random" numbers) might well
disagree here. I usually -set seed- only so I (and others) can
replicate my results, and for thta purpose you may simply set your
seed 42 (or any number).

Note that there is an ado -setrngseed- by Antoine Terracol and Bill
Gould, available from SSC that allows you to set "ture" random


I am having a problem with -mi-. I am using Stata version 11.2.

[...] Firstly, after registering variables with missings as imputed
and registering other variables as normal, what variables do i include
into the univariate regression? All variables in the data set? Or just
the variables i need for a specific model? Because including all the
imputed variables as well as the normal variables into the regression
brings up a mesage that says : "Option add( ) is required when no
imputation exist." How do i specify what number to have in add( ) and
rseed(  )?

This leads me to use only one of the registered imputated variables
and all the registered normal variables in the univariate regression.
I get this message afterwards: "Could not find a filename for
temporary flongsep file. I tried _mitmpfile1.dta through
_mitmpfile200.dta. Peharps you do not have write permission in the
current directory . This may occur, for example, if you started Stata
by clicking directly on the Stata executable on a netweork drive. You
should make sure you have written permission for the current
directory or use cd to change to a directory that has write
permission. Use pwd to determine you current directory." I dont
understand how to change to a directory that has write pwermission.
*   For searches and help try:

© Copyright 1996–2018 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   Site index