i have looked @ prediction lme4 on new levels r documentation quoted allow.new.levels=true.
i dont understand "if allow.new.level=true, prediction use unconditional (population-level) values data unobserved levels (or nas)" means.
in case have data.frame called exdata
subjects y x1 x2 1 0 1.6179339 0.9517194 1 0 1.4128789 1.0248514 1 0 0.9127448 1.8073684 1 0 1.5729219 2.1003925 1 0 1.6254359 1.2471660 2 0 1.6626074 8.5102559 2 0 1.3903638 6.0425018 2 0 1.1438239 2.5654422 2 0 1.1393088 2.9982242 2 0 1.1564141 2.8395960 3 0 1.1688192 13.9791461 3 0 0.9255715 18.8544778 3 0 1.2369097 4.2376671 3 0 1.3021943 9.6894289 3 0 1.2296961 12.4789910 4 0 1.0978131 2.0577688 4 0 1.1405409 1.4339044 4 0 1.0355546 1.9496732 4 0 1.1370849 1.7402332 4 0 1.1942591 1.3509880 5 0 0.4141535 2.1723957 5 0 0.9129311 0.8274350 5 0 0.9658796 1.2754419 5 0 0.8370701 2.1998756 5 0 0.5509546 2.3590774 6 0 1.2827411 1.5474088 6 0 1.1636606 0.7746669 6 0 1.1782936 1.1566909 6 0 1.1630238 1.7486415 6 0 1.1565711 0.6984409 7 0 0.8600331 0.1382253 7 0 0.8303510 0.1927431 7 0 0.8087967 0.6065926 7 0 0.7815187 0.9464185 7 0 0.7532042 0.9771646 8 0 1.1638190 1.3456340 8 0 0.5867126 1.4862727 8 0 0.6523964 0.5138441 8 0 0.9513971 2.3932337 8 0 0.9278743 2.3273670 9 1 1.0978606 1.2585635 9 1 1.0414897 1.2946008 9 0 0.6215353 0.2907148 9 0 1.0267046 1.0173432 9 0 1.1470992 0.7014759 10 0 0.9505266 0.4247866 10 0 0.8624758 0.2276577 10 0 0.8279061 0.2314898 10 0 0.7856832 0.3143003 10 0 0.7569739 0.7880622 with 10 subjects, y response (0-1) , 2 explanatory variables. want model logit glm random effect model, use
model <- glmer(y~x1+x2+ (1|subjects), family=binomial(link="logit"), data=exdata[exdata$subjects!=5,], nagq=1) i use first 4 subjects want use last 1 prediction. result of model variance of random effect on 82.55 , estimates
(intercept) -14.8377 x1 4.0366 x2 0.1056 the prediction new subject is
prediction <- predict(model, newdata=exdata[exdata$subjects==5,], type="response", allow.new.levels=true) giving
2.408299e-06 1.564704e-05 2.031392e-05 1.331586e-05 4.266690e-06 how these values predicted? hope ranef() subject=5 calculated, , coefficients model used calculate proability. can in way print ranef() of subject=5?. dont understand r documentation allow.new.levels cant understand how prediction made.
in specific example fixed part of model used (since population-level model model single random effect):
y1 <- as.matrix(cbind(1, exdata[exdata$subjects==5, c("x1", "x2")])) %*% fixef(model) c(exp(y1)/(1+exp(y1))) #[1] 2.408298e-06 1.564704e-05 2.031392e-05 1.331586e-05 4.266689e-06 
Comments
Post a Comment