Timeline for Is My Data MNAR (Missing not at Random)?
Current License: CC BY-SA 3.0
12 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Mar 16, 2017 at 0:31 | comment | added | EJ16 | I read about imputation, but it also confused me since I'm dealing with asylum numbers. I'm not sure how to fill in the blanks without affecting the asylum rates. Say, I just put zeros for all the years that were missing, the asylum rates will be missing as well, since there were no recognition. | |
| Mar 16, 2017 at 0:23 | comment | added | rolando2 | A common strategy is to impute values for missing data and then compare regression (?) results under such imputation to results solely dependent on original data ("complete case analysis"). If they hardly differ, it adds justification for the latter analysis. | |
| Mar 16, 2017 at 0:19 | comment | added | EJ16 | Indeed. Just scratching my head as to what to do with pending numbers alongside the rate calculations. | |
| Mar 16, 2017 at 0:17 | comment | added | rolando2 | I think this answer adds useful and thoughtful considerations, but there might be much more involved. E.g., "missing due to pending decision" might reflect special circumstances that make such cases substantially different from the rest in the way the independent variables relate to the dependent. | |
| Mar 16, 2017 at 0:11 | comment | added | EJ16 | I should have been more clear, so thank you! I do include a lot of controls, and pooled as much as I could in order to lag. I wanted to use a GEE model, but I know it's not biased if it's MCAR only. That's what really got my head scratching. | |
| Mar 16, 2017 at 0:08 | comment | added | salient | Ah, sorry misunderstood - thought it was disaggregate data at the individual level. I guess you will have to try and really use as much of the panel data setup as possible (e. g. lags of vars) with appropriate controls to convincingly disentangle the effects. Maybe utilize elections etc to identify such effects? | |
| Mar 16, 2017 at 0:02 | comment | added | EJ16 | Very simply, it looks at what are some domestic explanations that affect asylum-granting. So, for example, do far-Right parties in power in the host state really affect asylum-granting? If the host state has a different majority religion compared to those applying for asylum, does that affect asylum-granting? etc. The asylum-rates are the DV. | |
| Mar 15, 2017 at 23:57 | comment | added | salient | If we take a step back, what is your research question? | |
| Mar 15, 2017 at 23:56 | comment | added | EJ16 | Some of the numbers really look random. For example, some individuals from country X will only go to country Y once in a twelve year period. Or not go in year T1 but then later in T3. Some of it is obvious, such as conflict. Others are maybe not fleeing conflict but for other reasons. This has given me a headache for some time now, so I decided to throw a life line out here on stackexchange. | |
| Mar 15, 2017 at 23:50 | comment | added | salient | Yes exactly. But then I guess you would need a "time since applying for asylum"-metric. I see. Just trying to be creative :) Maybe there are more relevant things guiding the process, such as size of previous asylum groups from that country etc. | |
| Mar 15, 2017 at 23:46 | comment | added | EJ16 | I didn't think about a survival model. You mean like a Cox proportional? Also I thought about controlling for distance, but there are lot of migrants that immigrate to places that are not border states, such as in W. Europe. | |
| Mar 15, 2017 at 23:43 | history | answered | salient | CC BY-SA 3.0 |