Timeline for Does taking the logs of the dependent and/or independent variable affect the model errors and thus the validity of inference?
Current License: CC BY-SA 3.0
9 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Nov 26, 2017 at 22:12 | comment | added | DeltaIV | thanks for the link, but I'm already familiar with OLS, which assumes additive errors. In the log-transform case, it looks like we assume multiplicative errors: see Chris Haug & Bill's comments to my original post. | |
| Nov 26, 2017 at 18:19 | comment | added | Hugh Perkins | There is a description of how to obtain confidence intervals for OLS on wikipedia, en.wikipedia.org/wiki/Simple_linear_regression section 'Model-based properties'. Is this what you mean? | |
| Nov 25, 2017 at 18:37 | comment | added | DeltaIV | in my experiments reducing smoothing increased ringing, if with ringing you mean the oscillations: I cannot edit the question now, but I will next week. But still, my issue is with the confidence intervals, not with the regression curve: your answer doesn't cover them. | |
| Nov 25, 2017 at 13:58 | comment | added | Hugh Perkins |
added some thoughts on effect of changing smoothing on ringing; and a possible use of logit to transform the graph into something closer to a straight line
|
|
| Nov 25, 2017 at 13:57 | history | edited | Hugh Perkins | CC BY-SA 3.0 |
added 682 characters in body
|
| Nov 25, 2017 at 13:47 | history | edited | Hugh Perkins | CC BY-SA 3.0 |
added 682 characters in body
|
| Nov 24, 2017 at 18:58 | comment | added | DeltaIV | your answer is not correct, or at least incomplete. As I proved in my experiment, the effect of the log-transformation on the confidence intervals is not due to the increased smoothing - if I don't log-transform but I apply even more smoothing, I get confidence intervals that are much larger. See my edited question. | |
| Nov 23, 2017 at 7:57 | comment | added | DeltaIV |
If what you say is correct, I should be able to obtain the same result simply by increasing the smoothing parameter of the GAM by hand, instead than letting gam choose it by cross-validation. Interesting: I will test that. I appreciate the insight (+1) but it doesn't answer the original question of the effect of log-transformations on the confidence intervals.
|
|
| Nov 22, 2017 at 23:25 | history | answered | Hugh Perkins | CC BY-SA 3.0 |