Assume that I have a set of dependent variables and at least two independent variables (categorical). Assume that I'm interested in running an ANOVA or MANOVA with a full factorial model but my dependent variables are completely missing data for some combinations of the independent variables. How should this situation be handled in general and how does SPSS handle this? For example, if a certain interaction effect with a lot of missing data for its different combinations was considered as insignificant, how should this be interpreted?
Dear Daniel,
as far as I know, SPSS delivers at least two options to choose from, how it should handle missing data. You can choose from pairwise or listwise exclusion of the data. Both methods have pros and cons and have to be chosen specifically for your analyses. For example, if you want to comapre results between the analyses, it seems plausible to choose a listwise exclusion, so that you have the same subjects for each analysis.
When you have completely missing data for some combination, then SPSS will not include this for the calculation, unless you specifiy how to handle the missing values. SPSS offers to replace missing values with the mean of the variable for example, which is simple, but has some disadvantages, e.g. reduced variance and therefore a biased standard error. There are far better methods to replace missing data. But have a look at the fantastic book from Tabachnick and Fidell - Using Multivariate Statistic on this topic, as well as the article I appended.
Another way to deal with missing data is to use hierarchical linear models/multi-level models instead of ANOVA/MANOVA. They are more robust in dealing with missing data.
Dear Daniel,
as far as I know, SPSS delivers at least two options to choose from, how it should handle missing data. You can choose from pairwise or listwise exclusion of the data. Both methods have pros and cons and have to be chosen specifically for your analyses. For example, if you want to comapre results between the analyses, it seems plausible to choose a listwise exclusion, so that you have the same subjects for each analysis.
When you have completely missing data for some combination, then SPSS will not include this for the calculation, unless you specifiy how to handle the missing values. SPSS offers to replace missing values with the mean of the variable for example, which is simple, but has some disadvantages, e.g. reduced variance and therefore a biased standard error. There are far better methods to replace missing data. But have a look at the fantastic book from Tabachnick and Fidell - Using Multivariate Statistic on this topic, as well as the article I appended.
Another way to deal with missing data is to use hierarchical linear models/multi-level models instead of ANOVA/MANOVA. They are more robust in dealing with missing data.
Hi Daniel
My personal preference is to run Expected Maximization (EM) procedure to deal with missing data in SPSS. EM procedure is sufficiently robust to deal with 'missing completely at random (MCAR)' data. Normally, we cap the missing value at 20%.
However, literature favours multiple imputation approach.
You can find procedures and interpretations for both methods in the publication below (page 15 onward). Cheers.
Saiyidi
Article Introduction to SPSS
Hi,
For your question, kindly see the following:
https://www.youtube.com/watch?v=m0zV_wFGA1I
If you assume the missing data are "at random", there is no problem to fitting any linear model (regression, ANOVA). Your software should allow for that. If missingness is related to patient characteristics, then you can still run the linear model, but you may have more difficulty defining the validity of your inferenc
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software
- SPSS