|
|
 |
| Missing outcome and classification qu... |
 
|
| Message/Author |
|
|
| Shufang W posted on Thursday, August 02, 2012 - 12:43 pm
|
|
|
Hi,
I understand that entropy is a measure for classification quality and 0.8+ indicates a good classification.
I am fitting a GMM. There are about 45% missing values in the repeatedly measured outcomes. I suppose that the classification could not be very good,i.e. the entropy couldn't be very high given such a high proportion of missing data. I wonder whether you know the relationship between missing proportion and reasonable cutpoint of entropy. The highest entropy I can get is around .77. Do you think it's fairly high?
Thanks a lot. |
|
|
|
|
How entropy is affected depends on which part of the growth model you are most interested in. It is likely more affected at later time points when there is probably more missing data.
I think .77 is a quite high entropy. |
|
|
| Back to top |
|
|
