Timeline for GLM missing data
Current License: CC BY-SA 4.0
17 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| S May 5, 2020 at 3:26 | history | suggested | Remy | CC BY-SA 4.0 |
correct typos and an updated post as examples
|
| May 4, 2020 at 18:18 | review | Suggested edits | |||
| S May 5, 2020 at 3:26 | |||||
| Jun 16, 2016 at 8:28 | comment | added | gregorp | I guess the problem was with how I was deleting the data. I deleted the first 10%, since I thought the data was ordered randomly. Tried another example where I deleted the data randomly and MI gives much better results than the dummy variable method in every parameter. Thank you for all your help! I also checked for convergence, seems that it converged. | |
| Jun 14, 2016 at 8:58 | comment | added | Robert Long | @gregorp it depends on how you delete the data. It would really be better if you asked a new question and provided a reproducible example. | |
| Jun 14, 2016 at 8:54 | comment | added | gregorp | Ok, I'll run the plot() command, when R Studio is finished doing another imputation (takes some time), thank you. One more question. If you make a model, then delete some data, impute the deleted data and make a model based on those imputations, the results of the both models should be very similar right? | |
| Jun 14, 2016 at 8:40 | comment | added | Robert Long |
@gregorp yes, you simply run plot() on the object returned by the mice() function. At this point I'd also recommend reading some of the online tutorials such as this one
|
|
| Jun 14, 2016 at 8:09 | comment | added | gregorp | I haven't checked for convergence as I haven't learned yet how to do it. Is there a function for it? I'll try to make a reproducible example and let you know. | |
| Jun 14, 2016 at 7:10 | comment | added | Robert Long | @gregorp 50 imputations and 50 iterations should be more than enough - but did you actually check convergence ? Please post a reproducible example with data (simulated if you like) and code in another question so that we can look at this, since this is now beyond the scope of the current question. | |
| Jun 14, 2016 at 5:28 | comment | added | gregorp | I'm sorry my comment was unclear. I wanted to see how MI is doing on an example so I made a glm1 with one variable without missing data. Then I deleted 10% of the data and did glm2 with MI and glm3 with a dummy variable for comparison. The coefficients of glm3 were closer to coefficients of glm1. But at that point I used 50 imputations and 5 iterations. Later I've left the PC running and tried with 50 imputations and the results of MI seem better, but still not in all coefficients. How many iterations are usually enough for a good model? | |
| Jun 13, 2016 at 13:41 | comment | added | Robert Long | @gregorp why did you delete 10% of the observations for a variable before doing multiple imputation ? If you are going to use MI you shouldn't be deleting anything. When you say the results are "more far off" than the original model, what was the original model ? In what way are the different ? The results can certainly be different with MI, because listwise deletion and making a dummy variable are biased methods. How many imputation and iterations did you use ? Did you check that convergence had been achieved ? | |
| Jun 13, 2016 at 10:54 | comment | added | gregorp | So I've tried the MICE package. I made a glm model on all data. Then I deleted the first 10% observations for a variable and did a multiple imputation. Unfortunately the results of the multiple imputation seem to be more far off from the orinigal model, than if I make a dummy variable for these 10%. What could be the cause of that? Am I using the imputation wrong, or can such a thing happen? | |
| Jun 13, 2016 at 5:31 | vote | accept | gregorp | ||
| Jun 10, 2016 at 8:02 | history | edited | Robert Long | CC BY-SA 3.0 |
Further detail
|
| Jun 10, 2016 at 7:40 | comment | added | Robert Long | @gregorp I have updated the answer to address your comment. | |
| Jun 10, 2016 at 7:40 | history | edited | Robert Long | CC BY-SA 3.0 |
Added further detail
|
| Jun 10, 2016 at 6:24 | comment | added | gregorp | What exactely do you mean by "can yield"? That it usually yields unbiased results, or that it can sometimes yield unbiased results, but more often it doesn't? Also, is the method I am currently using bad, or is it a reasonable way to taclke this problem, just not as accurate as multiple imputation? | |
| Jun 9, 2016 at 11:59 | history | answered | Robert Long | CC BY-SA 3.0 |