r - Does the varIdent function, used in LME work fine? -
i glad if me solve problem. have data repeated measurements design, tested reaction of birds (time.dep
) before , after infection (exper
). have fl
(fuel loads, % of lean body mass), fat score , group (experimental vs control) explanatory variables. decided use lme
, because distribution of residuals doesn’t deviate normality. there problem homogeneity of residuals. variances of groups “before” , “after” , between fat levels differ (fligner-killeen test, p=0.038
, p=0.01
respectively).
ring group fat time.dep fl exper 1 xz13125 e 4 0.36 16.295 before 2 xz13125 e 3 0.32 12.547 after 3 xz13126 e 3 0.28 7.721 before 4 xz13127 c 3 0.32 9.157 before 5 xz13127 c 3 0.40 -1.902 after 6 xz13129 c 4 0.40 10.382 before
after have selected random part of model, random-intercept (~1|ring
), have applied weight parameter both “fat” , “exper” (varcomb(varident(form=~1|fat), varident(form=~1|exper)
). plot of standardized residuals vs. fitted looks better, still violation of homogeneity these variables (same values in fligner test). do wrong?
a common trap in lme
default give raw residuals, i.e. not adjusted of heteroscedasticity (weights
) or correlation (correlation
) sub-models may have been used. ?residuals.lme
:
type: optional character string specifying type of residuals used. if ‘"response"’, default, “raw” residuals (observed - fitted) used; else, if ‘"pearson"’, standardized residuals (raw residuals divided corresponding standard errors) used; else, if ‘"normalized"’, normalized residuals (standardized residuals pre-multiplied inverse square-root factor of estimated error correlation matrix) used. partial matching of arguments used, first character needs provided.
thus if want residuals corrected heteroscedasticity (as included in model) need type="pearson"
; if want them corrected correlation, need type="normalized"
.
Comments
Post a Comment