Timeline for Very large odds ratio (558) in logit regressions
Current License: CC BY-SA 3.0
18 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Jul 21, 2017 at 20:51 | vote | accept | retrography | ||
| Feb 6, 2015 at 23:34 | history | edited | Scortchi♦ |
edited tags
|
|
| Jan 15, 2015 at 12:44 | comment | added | retrography | That makes a lot of sense. I think the comparisons interest me the way they are. The indicators point very well in the direction I expected, and very strongly so. But I also expected the addition of control variables to absorb a part of the hypothetical unobserved heterogeneity that was potentially boosting my IV coefficients, but what happened was the reverse. I guess the lesson is that I have to think of other potential controls, from a theoretical POV. Needs more thinking... | |
| Jan 15, 2015 at 12:31 | history | edited | retrography | CC BY-SA 3.0 |
removed obsolete info
|
| Jan 15, 2015 at 12:24 | answer | added | Scortchi♦ | timeline score: 1 | |
| Jan 15, 2015 at 12:06 | comment | added | Maarten Buis | My first step would be to think about that cell in substantive terms: Does it make sense that it is almost empty, do I want to compare that cell with others, which comparisons interest me most. Sometimes that is enough to find a solution. Only after that, I would start to consider more "technical" solutions. | |
| Jan 15, 2015 at 10:05 | history | edited | retrography | CC BY-SA 3.0 |
results corrected after modifying model according to suggestions
|
| Jan 15, 2015 at 10:03 | comment | added | retrography |
Thanks for all the comments (esp. @MaartenBuis). I just had a look at my data: (1) The whole sample is being used. Actually, in one case more than the whole sample was being used! There was one "+" too much in my model that I had not noticed. I fixed that and updated the question. (2) There is no separation in the data, but there is something close to one: nrow(df[!df$iv&df$dv,])) is less than 100 in all my samples - but the rest is fine. Now, I know my data and this is not necessarily wrong. The question is: Is this interpretable or is the data simply inappropriate for logit?
|
|
| Jan 15, 2015 at 8:51 | comment | added | Maarten Buis | First step: look at the sample size used by your model and compare that with the sample size in your data. Second step: just stare at cross-tabulations: look for (nearly) empty cells, figure out why those cells became empty due to datacleaning (i.e. did you "overclean" your data). Third step, revisist every datacleaning step and every control variable till you find the answer. That is a lot of work, but that is normal: preparing your data is usually by far the most time consuming part of a statistical analysis. | |
| Jan 15, 2015 at 8:21 | comment | added | kjetil b halvorsen♦ | This looks more like a stats question, even if there is some mention of stata and R. The main question is stats, shoiuld be open. | |
| Jan 15, 2015 at 7:36 | comment | added | Glen_b | Do you perhaps have complete separation anywhere? | |
| Jan 15, 2015 at 7:21 | review | Close votes | |||
| Jan 15, 2015 at 10:54 | |||||
| Jan 15, 2015 at 4:05 | history | edited | retrography | CC BY-SA 3.0 |
minor correction to results
|
| Jan 15, 2015 at 3:46 | comment | added | retrography | Isn't listwise deletion about handling missing data? I don't have any missing data in my control variables. | |
| Jan 15, 2015 at 3:44 | comment | added | rolando2 | Since logistic regression requires listwise deletion (complete case analysis), chances are your inclusion of control variables has markedly reduced the sample size, in a way that biases the iv's coefficient upwards. | |
| Jan 15, 2015 at 3:27 | history | edited | retrography | CC BY-SA 3.0 |
enhanced readability of results
|
| Jan 15, 2015 at 3:22 | review | First posts | |||
| Jan 15, 2015 at 3:40 | |||||
| Jan 15, 2015 at 3:20 | history | asked | retrography | CC BY-SA 3.0 |