Timeline for Simple case of MNAR missing data
Current License: CC BY-SA 3.0
6 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Nov 25, 2012 at 21:45 | vote | accept | Matteo | ||
| Dec 15, 2011 at 18:21 | comment | added | jbowman | @dmk38 - In this case we can construct the exact censored-data likelihood function with ease, enabling MLE, so imputation buys us nothing. | |
| Dec 15, 2011 at 18:18 | comment | added | jbowman | @Matteo - I just set the censored data to NA out of some misplaced sense of purity of numbers in data. You could use 0 instead; the test in the code would have to be changed appropriately. In your case, the data is a special type of missing - "censored" - which enables us to ignore the NMAR qualification, as we know (or can deduce easily) the true probability distribution of the observed data. | |
| Dec 15, 2011 at 8:35 | comment | added | Matteo | Thanks for the answer. I'm not very familiar with R. Why did you set the censored data to NA, instead of 0 (which is what I observe)? Also, re: your comment on using E-M: my understanding was that E-M cannot be used when data are NMAR. Is this correct? | |
| Dec 15, 2011 at 0:34 | comment | added | dmk38 | why shouldn't he use multiple imputation? Zelig/Amelia in R? | |
| Dec 14, 2011 at 22:55 | history | answered | jbowman | CC BY-SA 3.0 |