Replace Missing Values - Expectation-Maximization - SPSS (part 1)
HTML-код
- Опубликовано: 15 окт 2011
- 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.
I love all your videos. You touch on very useful and relevant topics. You obviously know a lot and you have a great voice for presentations.
You really go round and round at first!
this video is so helpful!!! thank you so much!
What should I do if the MCAR test is significant?
Thank you very much for helping :)
thank you so much
Hello, Thank you very much for all of your helpful videos. For days I am trying to create drop-out variable for my longitudinal data to do drop-out analyses. Do you have any lecture about this topic?
Thanks!
@how2stats. Thanks for the video-what about Multiple Imputation video? (cannot't find it)...
Thank you for this great Video. I am using SPSS20 for analyzing my survey that has categorical and quantitative missing responses. I have found that multiple imputation (MI) works well with categorical variables but not with scale, even after log10 and square root transfer. For instance, I got negative number for the age (scale variable). I found the opposite with expectation-maximization algorithm. The EM works well with square root transferred variable but not with categorical variables. My question is can I use both methods (MI and EM), MI for quantitative and EM for categorical variables, for the same data in same publication. I would greatly appreciate if you kindly send some literature on using two or more methods to replace missing values.
Hi, thanks for this useful tutorials. Just want to ask if it's fine to exclude first manually those cases (i.e., demographic information) with missing values then proceed with EM? Will there be any issue for me to do this?
From memory, I think what I meant when I said that more than 2% missing values is problematic is that how you decide to deal with your missing values will make a difference. When you have a very small percentage of missing values (say, < .50), it does not matter what method you use, you'll very likely get the same results. I think I meant 2% to be the relatively arbitrary cut-off. After that, it does not matter, and you need to use a sophisticated method such as EM.
Hello, good videoes. I got a question I hope you can answer even tho this is an old video. I'm looking for a good reference saying that an EM imputation is a good method to impute missing values?
Thanks so much for this video, extremely helpful. I have a dataset of 531, analysing 18 variables (a questionnaire ade up of 18 items). There are only 7 missing values, and MCAR came out as sig at .002! Could I argue that MCAR can be ignored because there were so few missing values (.2 or .4 % on a few items)? thanks
You suggest adding the items that make up particular construct (a) into EM model. I'm presuming you would then run construct b and combine these 2 datasets? If this is the case, how would you use auxiliary variables to improve model? I'm trying to run an analysis with with multiple key variables. I could run 5 different EMs and combine into dataset, but I also have auxiliary variable, highly correlated with my key variables, but that wont be used in my analysis. What would you advise in this instance?
the info you need is at 3:30
I'm really struggling with what to do, since Little's MCAR test shows significance (.000). Most of my items have missing data, but no item has more than 3.8% missing. Given the small amount, could I get away with using EM without too much worry, or is Multiple Imputation preferred? You mentioned you would make a video on MI - are you still planning to do this? I can't find this topic on your channel.
at these steps missing values of quantitative variables are replaced but not the missing values of categorical variables.....? would it be a good idea to replace missing values with zero (0) ?
Do you have tutorial of logistic regression?
hi, I'm assuming from this that it is important to do EM imputation on the individual items prior to averaging into composite variables. Is this correct? (This would mean that I am imputing significantly more individual data points than if I created composites first and then dealt with missing data) Will I run into problems if I perform imputation for missing values on composite variables?
I'd Impute on the items; then create a composite variable based on the items which do not include any missing values.
The entire first video is just him going on and on and on and on and on....
May I ask why it gives you the same number for each missing value in each column??
Why don't most of you upload the link to the database so that someone can follow along?
Can anyone help me?
I use SPSS 24 and when I click on Analyze - missing values is not a choice- Do you know how to make it appear? Is it an add in?
Patti - you have to buy the "missing values" package from IBM to do this.
Good Video. However, I was using this method, however it imputed a negative value on a likert scale of 1-5. I don't trust this method anymore. Which means I have to run all of the past imputations through a different method.
watch this video from 3:29 , he talks a lot(
where is the second part ?
Usually, it comes up automatically. You might have to search in RUclips "expected maximization spss" part 2
before before before before before before
All these videos are, at best, 50% content and 50% waffle. Sorry, but that's the truth.
Have you tested that statistically? ; - )
;-D
Why you talk too much... oh my god.. go to the point for the sake of god!
Dude just get to the point
at these steps missing values of quantitive variables are replaced but not the missing values of categorical variables.....? would it be a good idea to replace missing values with zero (0) ?