Timeline for Mixed models: wrong random intercepts and interpretation issues
Current License: CC BY-SA 3.0
9 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Apr 22, 2015 at 9:07 | vote | accept | statastic | ||
| Apr 22, 2015 at 9:07 | vote | accept | statastic | ||
| Apr 22, 2015 at 9:07 | |||||
| Apr 22, 2015 at 9:07 | comment | added | statastic | Hi Karl, I didn't noticed this. Then indeed, this is exactly what I want! It's the first time I use mixed models, so this is all pretty new to me. Thanks again for the great answer! | |
| Apr 21, 2015 at 17:36 | comment | added | Karl Ove Hufthammer |
@statastic If I have understood you correctly, that’s exactly what the model does. For example, if you add two observations for a new race (race = 6), and set the observed speeds to, e.g., 100, the estimated mean speed of the race will depend on which two individuals participated in the race. If it is the two theoretically worst individuals (individual 1 and 2), the estimated race speed (based on the data set from the simulation) will be 105.3. If it is the two theoretically best individuals (individual 14 and 15), it will be 97.4. For the average individuals (7 and 8), it will be 100.6.
|
|
| Apr 21, 2015 at 2:32 | comment | added | statastic | Maybe a better approuch would be to firstcenter every racers speed according to their own average speed across all the races they did? | |
| Apr 21, 2015 at 2:30 | comment | added | statastic | Well, you have fast individuals and slow individuals. In difficult races all individuals will be slower relative to their own mean. But some races will attract more fast racers (eg championships) in the average speed in the race will be faster because of the faster individuals and not because it's a harder racer. I want to use the fact I have racers that did mulptiple races to compare races. So you have first a racers effect and then I would like to explain the variation in the residuals by race effect. | |
| Apr 20, 2015 at 19:40 | comment | added | Karl Ove Hufthammer | @statastic In the first model, there is a random effect (intercept) for each individual, just like you require. I’m not sure I understand what you mean by having ‘only individuals that did only one of the races’ and why this should imply that the race effect is 0. If you have few observations for an individual, that individual’s predicted intercept (i.e., conditional mean) will be heavily shrunken towards the global mean of the individuals. | |
| Apr 20, 2015 at 19:29 | comment | added | statastic | Hi, thanks for the answer! I'm still a bit puzzeled about the meaning of a random intercept in this case. Now it seems (your first model) that each race has it's own fixed intercept and that there is an additive effect for each individdual. I would like to go to a situation where I have an intercept for each individual and then an additive effect for each race. This also means if I would have in my dataset only individuals that did only one of the races, my race effect would be 0 since all information is allready in the individual intercept. How do this relate to you model? | |
| Apr 19, 2015 at 18:41 | history | answered | Karl Ove Hufthammer | CC BY-SA 3.0 |