Timeline for glmer.nb does not converge when one variable is included in the model
Current License: CC BY-SA 4.0
9 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Mar 8, 2020 at 3:37 | comment | added | PeterPanPan | oh yeah you are right. thank you about that. thank you | |
| Mar 6, 2020 at 14:23 | comment | added | josliber | @PeterPanPan there's probably no need to scale the 1/0 dummy variables introduced from a categorical variable -- they are already on a reasonable scale. But scaling your continuous variables would be reasonable. | |
| Mar 6, 2020 at 4:34 | vote | accept | PeterPanPan | ||
| Mar 6, 2020 at 3:15 | comment | added | PeterPanPan | oh I got confused more. It generates a new problem ? Before I use scale() to variables, all variables ,you know , are discrete data, therefore I am capable of using poisson or negative binomial if fitdistr checks ok. But after scale() applying all of variables, they will be continuous data .,and then the norm or lnorm distribution fits the data. But fitdist() displays countable data with negative binomail is pretty good , as above graphic, wheras the norm and lnorm with scale(y) are not so good. So is that correct that I only scale(x) but others variables keep countable data? Thank u. | |
| Mar 6, 2020 at 0:48 | comment | added | josliber | @PeterPanPan folks often will, e.g. by dividing each by their standard deviation. | |
| Mar 6, 2020 at 0:34 | comment | added | PeterPanPan | I totally agree with you. But I'm still confused if I need to rescale other variables when I proceed x/10000 like y/10000 , z.fac/10000 . Do they need to proceed at the same time ? | |
| Mar 5, 2020 at 16:08 | history | edited | josliber | CC BY-SA 4.0 |
added 468 characters in body
|
| Mar 5, 2020 at 15:51 | history | edited | josliber | CC BY-SA 4.0 |
added 19 characters in body
|
| Mar 5, 2020 at 15:28 | history | answered | josliber | CC BY-SA 4.0 |