andersonwinkler/PALM

Missing data

Closed this issue · 2 comments

Hello,

I was trying to handle missing data (inputs) with PALM and I'm facing an error. To my understanding, missing data are handled using the -imiss missing.csv, where missing.csv is a table of same size as Y, with 0 if the data is not missing and either -1, 1 or 2 if the data is missing. In my case, I set these missing instances to 1, although I don't know what does it change to set them to -1 or 2.

I run the following command line in octave:
palm -i Data.csv -d DesignMatrix.csv -t contrast.csv

It works fine. When including the -imiss option:
palm -i Data.csv -d DesignMatrix.csv -t contrast.csv -imiss missing.csv

It works fine if missing.csv is actually only 0s. However, if I set some elements to 1 (or -1 or 2), I have many warnings and an error. Here is the log:

-imiss missing.csv
warning: With missing data MAR/MNAR, the option "-zstat" is mandatory.
         Adding it automatically.
warning: called from
    palm_takeargs at line 1342 column 9
    palm_core at line 33 column 11
    palm at line 81 column 1
    GenerateFakeData at line 47 column 1
warning: With missing data, the option "-cmcx" is mandatory.
         Adding it automatically.
warning: called from
    palm_takeargs at line 1348 column 9
    palm_core at line 33 column 11
    palm at line 81 column 1
    GenerateFakeData at line 47 column 1
warning: With missing data, the partitioning must use the "Guttman".
         method. Adding automatically the options
         "-pmethodp Guttman" and "-pmethodr Guttman".
warning: called from
    palm_takeargs at line 1360 column 9
    palm_core at line 33 column 11
    palm at line 81 column 1
    GenerateFakeData at line 47 column 1
Reading input 1/1: Data.csv
Reading design matrix and contrasts.
Elapsed time parsing inputs: ~ 1.4873 seconds.
Preparing designs for missing data [Design: 1/1, Contrast 1/1]
error: 'idx' undefined near line 192 column 12
error: called from
    palm_misspart at line 192 column 4
    palm_core at line 362 column 42
    palm at line 81 column 1
    GenerateFakeData at line 47 column 1

Do you have an idea where it may come from ?

Besides, is it possible to investigate a correlation (-pearson option) in case of missing inputs ?

Thank you !

Hi Anderson,

I'm curious if the missing data functionality has progressed enough to be useful. Thanks for your help!