Replace Missing Values - Expectation-Maximization - SPSS (part 2)

Learn how to use the expectation-maximization (EM) technique in SPSS to estimate missing values . This is one of the best methods to impute missing values in SPSS.

[howstats]

This is one the those situations in statistics where you don't have much in the way of options. In my opinion, you would have to restart the study and re-examine your measure to understand why certain types of people are not responding to certain items. It's a very serious issue that a lot of people simply disregard, but any interpretation of the results will be compromised. That's not to say that plenty of people simply carry on after deleting the cases with incomplete data (unjustifiably).

[younesmaknassi]

Hi there! Thousand thanks for your tutorial. It has helped me a lot because I could not calculate a total score wit missing Now it's possible. Have a good day. Waiting for more videos...

[r]

Thanks so much...your video and instructions are clear and concise - great help!

[melcatsss]

Most helpful stats vids I've found, THANK YOU.

[angelameadows]

Am loving all of the videos from your series I've seen so far. I can't find one from you on multiple imputation - did you not make one? Thanks.

[abdulkarim.jamal.kanaan]

I have a question, why we should do the replacement process separately for each variable's items as you did in the video for (a and b)?
Thank you

[TheKillerDeer]

Great video! It would have been nice to have an explanation of the Little's test within this video (since it provides the output anyway) to avoid having to go watch another video.

[Mzcentric]

Do you have to analyze each subscale separately? I tried analyzing everything together and I got some very large numbers and some negative numbers... I am wondering if analyzing each scale separately would have made a difference

[howstats]

Yes, I do believe it is an add-on. I suppose an alternative is to simply use mean replacement. If you were good a multiple regression, you could build a regression equation to predict missing values.

[jprmaps]

Very useful and well explained. Thanks!

[natachaemerson]

Awesome video!! Thanks for being so clear

[tonyyoung]

This is a big help.  Thanks very much!

[MilliVanilli]

What if the MCAR with ALL your values and subscales was non-significant but now, doing all subscales separately as you suggest here, the MCAR under the EM Means table for one of the subscales comes up as significant?

[alexanderstevens]

For the imputed values, do you recommend rounding up or down rather than retaining the mean? Likert scales are discrete. Say we have a scale of 1 to 5, a value of 2.5 doesn't exist. So, for any value > .5, would it be ok to round up and round down any value < .5? So, if 2.5, that would be rounded to 3.

[heejinlee]

Thanks for the video. Could you elaborate why MI is called more sophisticated than EM?

[ohoodali]

hello, i did EM for the missing data, and the value spss estimated did not correspond with the variables' coded. what could be the problem?

[aldaxhabrahimi]

What if the EM method replaces the missing data with higher values outside of the value range that a variable should have? for example, a variable should have min value 1 and max value 5, but the EM method replaces some of the missing data with numbers bigger than 5.

[matejfiskus]

Hi,
I am interested in using EM technique on 2 scales: one has 21 items and another 12 items. All are asymmetrically distributed (Kolmogorov is <0, 01). Is it possible to use it on that data (N=158)?

[sputaccount]

Should missing DVs be imputed this way (along with other missing variables), and if so, should the cases with imputed DVs be used in an analysis?

[jesseludenyo]

Must it be done for each subscale separately?