11-12-2014 05:00 AM
I would like which is the criterion to plot of marginal residuas Vs predictor lineal in the GLIMMIX proc.
I have done a GLMM with the seasons and months of the year as categorical variables (I put month nested into seasons).
I got the following plot (attached), can anyone explain the next questions?
1) Why are not the categorical residuals distributed homogeneously along the X axes? (some "months" are closer than others).
2) Which is the criterion to order the residuals? by name, by value...
_I get in all my plots of marginal residuas Vs predictor lineal a downward trend, and I don't know whether it is an artefact or it is real.
thank you all in advance
11-12-2014 08:13 AM
I would be elated to get plots like these for most data. Very little evidence of violations of assumptions.
Regarding the questions:
1. Without knowing the exact model being fit, and especially how your data is constructed, I can't come up with a good reason for the uneven spacing. I suspect something in the coding of month and season.
2. That looks like a trend that could be eliminated by differencing, but I am not sure.
11-12-2014 11:17 AM
I am studying the tend of a parameter (VarX) along the differents seasons.
I have collected the data from multiple studies and know I am re-analyzing all data together. To take this into accoutn I have put ID_study as random factor.
As I have repeated measures from a same population, so I have nested Month with Season in the model; in addittion to put Population as Random factor.
As I have data from multiples species I have nested de Family-Genus-Species in other random factor.
proc glimmix data=data method=laplace plots=(ResidualPanel(conditional marginal));
class Season Species Genus Family Poblacion ID_study Month;
model VarX = Season Month (Season)
random intercept / subject = Poblacion (ID_study);
random intercept / subject= Species (Genus*Family);
lsmeans Season/cl diff PLOT=diff (NOabs center) adjust=tukey;
lsmeans Season/cl ilink;
When I put only the seasons (without the Months) I have four columns in the marginal residuals, and they are also plotted in a downward trend.
So I don't know whether SAS plots it in an default way from higher residual values to lower residual values (so the downward trend is an artefact) or whether it is the real order of the months.
I have put the Months like: Jan, Feb, Mar, Apr, May, Jun, Jul, Aug, Sep, Oct, Nov, Dec. At the beginning I though that maybe it were plotted alphabetically, but the proximity of the columns do not correspond with it.
11-13-2014 07:31 AM
Try recoding the months as 1 to 12 and see if it makes any difference. I don't believe that the plot is from higher to lower residual values, but it wouldn't be the first time I was wrong on something regarding the ODS graphics.