what is your experience with statistics? You could use aregImpute or MICE in R. I think STATA and SAS have similar packages available. Altneratively, SPSS has built-in options to deal with missing data.

In general, multiple imputation is recommended to preserve the uncertainty related to missingness and allow data to be missing at random (MAR assumption). Recently, Stef Van Buuren has derived some work on MNAR structures, but Im not sure to what extent these ideas are implemented in software packages (and how robust they are to violations of posed assumptions).

ou can use 'R'- an excellent open source software. Several packages are available for missing data handling. Follow the links: http://gking.harvard.edu/amelia/

http://cran.us.r-project.org/web/packages/mitools/index.html

Stata (only the most recent version 12) has a built-in comprehensive and easy to use module for multiple imputation, including multivariate imputation using chained equations.

Extensive overview at:

http://www.stata.com/bookstore/stata12/pdf/mi_intro_substantive.pdf

try with unistat software or R-software tool

MicrOsiris incorporates IVEWARE from the Institute for Social Research for various imputation schemes as well as other built-in options. It's free at microsiris.com.

IVEware is very good free and works especially well with SAS and is much more robust, because of its approach, than SAS PROC MI. I am also impressed with the missing data module available for SPSS, but if you are buying the retail business version it is an exponsive add-on, whereas it is included in most academic version. Bob

BTW, MicrOsiris accepts data from SPSS, SAS, STATA, and Excel

Thank you Neal for the added information. I am a pretty big fan of IVEware, so it's nice to know that you have incorporated it in Osiris and also that you have these data import capabilities. I think that makes it a very formidible free solution that goes well beyond missing data. My compliments. I have bookmarked the linked, and I am sure I will download and install at some point.

I think SAS statistical software is the best one.

I am not sure if Alireza was answering directly about handling missing data. SAS creates multiply imputed data sets using PROC MI. Part of the imputation is done using EM (expected maximum), a good technique, but it can crash, mostly commonly in SAS with a matrix singularity message. It can take a lot of tweaking (by removing variables from the imputation) to get it to run. Once you have the imputed data, all of which will be in one SAS data set, you have a lot of analytic tools available in PROC MI ANALYZE. Alternatively, one version of IVEware (as mentioned above) can be accessed from within SAS. It does the imputation differently, using chained equations, and is much more robust. It creates a SAS data set that you can analyzed using either IVEware commands from within SAS or using PROC MI ANALYZE commands on the data set IVEware created. Additonally, IVEware has its own standalone interface so that you don't need SAS, although having looked over the MicOsiris link from Neal Van Eck, it seems like it might make more sense to go that route to run IVEware as there would be more data handling tools available than in the IVEware standalone, although I haven't actually tried MicOsiris yet. The IVEware standalone interface does work, but it's nothing pretty.

Robert, the MicrOsiris link to IVEWARE is through a dialog box, and the output is collected in one place.

I've used the imputation tools in both SAS and STATA. I think STATA does a much better job with less coding and data manipulation - or tweaking.

Hi Constance, I wonder what you mean by doing a better job. The issues I find with imputation programs is that eventually you can crash the imputation. In my experience SAS isn't terribly robust (Matrix is singular message). Some of it may be inherent in using EM and some ot it may be in how SAS uses EM. I haven't done head to head comparisons of SAS to STATA on the same problems, but I know you can push STATA to far also. What I do know is that IVEware, which doesn't use EM, goes much, much further than SAS in most circumstances. Bob

This was not a terribly large sample, so there wasn't a problem with pushing either too far. This was a head to head comparison and it appeared that STATA was more accurate. I agree that SAS is not very robust at all in this case. STATA, in my opinion, was more user friendly with less opportunity for error. However, that was just my opinion. Unfortunately, I think this is one of the weaker tools in SAS.

I'm certainly no fan of SAS in general, and don't run it myself. IVEware can run inside of SAS and has its own commands but SAS-like syntax. Even though I like the general idea of EM, at least the way SAS implements it, it seems pretty easy to crash. It wouldn't be the large sample that would crash EM, it would be the number of variables, the collinearity, and other things. The complete missing value suite for SPSS looks very good, but it costs a lot of money. Thanks for the feedback. Bob

I'm not expert in handling missing data. But I would do the following thing: First look at books about Missing data (for example in amazon). Then look if they provide information on software to handle with missing data. I found this: "Flexible Imputation of Missing Data" of Stef van Buuren. It delivers theory and R-code. It does makes sense to understand the various type of missing data (theory) and to have the code at hand.

IBM SPSS or even STATA are good widely used softwares for missing data

In my opinion, Amelia II-package for R is one of the best tools to impute missing values.

http://www.jstatsoft.org/v45/i07/paper

It is difficult to answer if we don't know what is the amount of missing entries ? Is it in a single vector ? or in a space-time field ? There is a huge number of software you can use to fill in missing entries (including free ones as R, octave, scilab etc.). To get a better help, you need to give details about your missing entries (nature, amount and so on).