如何可视化线性混合效应模型预测

Question

I have a very simpel linear mixed-effects model:

Rodlangde means root length Mehlich means plant available phosphorus Lokalitet means locality

model<-lme(Rodlangde~Mehlich,random=~1|Lokalitet)

I would very much like to produce a plot where you can see the 10 (I have 10 localities) different linear graphs with the same slope but different intercept that the model is composed of. I have now tried to search for at solution on the Internet for 2 days, but those codes I can find are too complicated for me to understand, or I can't find out which packages I need before I can use the codes. Can anyone help me with a simple code to visualize the 10 different graphs in the same plot?

Kind regards

Data:
Lokalitet   pH  Mehlich Kol Rodlangde
Odrup   7.02    0   0.919642857 6.362373845
Odrup   6.87    0   0.875   5.372476457
Odrup   7.09    0   0.868421053 14.23942792
Odrup   6.64    0   0.939393939 4.640122704
Orhoje  6.81    12.13896843 0.83    2.842893319
Orhoje  7.44    7.062027912 0.741666667 4.399108137
Orhoje  7.23    6.915193254 0.917355372 3.597793514
Orhoje  6.73    3.930162033 0.909090909 5.28750758
Melby   5.2 28.20132199 0.669642857 2.541898484
Melby   5.35    14.97459413 0.519685039 2.790724941
Melby   5.04    8.352860756 0.596153846 2.927228501
Melby   5.02    10.51701575 0.596153846 1.538074359
Kallingedal 8.4 17.47092431 0.458646617 8.059178499
Kallingedal 8.33    21.74560339 0.703703704 10.50345245
Kallingedal 8.3 21.34370501 0.762295082 7.610537154
Kallingedal 8.37    25.06114498 0.770491803 11.88896483
Ravrigtigkalk   5.61    5.117349119 0.952380952 9.307948512
Ravrigtigkalk   5.92    3.400217532 0.85046729  12.80110763
Ravrigtigkalk   5.77    3.358878819 0.607476636 14.82758346
Ravrigtigkalk   5.82    2.854552095 0.9375  4.231563699
Karsemose   5.28    0   0.813084112 12.06213863
Karsemose   5.36    1.312479611 0.838095238 7.341806594
Karsemose   5.32    0   0.898148148 10.1038273
Karsemose   5.34    0   0.821782178 8.16704508
Lergraven   8.43    44.62536835 0.847457627 13.48193914
Lergraven   8.41    39.52348256 0.884297521 12.67270404
Lergraven   8.39    43.26503035 0.880597015 21.24738813
Lergraven   8.41    40.8293479  0.770491803 16.12249983
Hvidtjorn   7.98    41.68676311 0.923076923 19.46781449
Hvidtjorn   8.16    43.89098256 0.827868852 14.39349303
Hvidtjorn   8.19    37.35675233 0.942857143 34.98582813
Hvidtjorn   8.2 29.90406084 0.927927928 17.09668084
Ravsurt 5.17    5.061969924 0.821782178 5.956222014
Ravsurt 5.31    9.271879523 0.842975207 12.71456674
Ravsurt 5.47    9.796946179 0.692307692 4.772145446
Ravsurt 5.33    12.27335664 0.852173913 5.802874149
Eskebjerg   5.6 5.866279787 0.805309735 10.41055981
Eskebjerg   5.78    11.59095638 0.961538462 6.981631906
Eskebjerg   5.34    0.381918387 0.789473684 8.218044532
Eskebjerg   5.52    7.376130558 0.942857143 4.018040528

Answer 1

Fit model (it's always a good idea to use an explicit data argument -- among other things it's necessary if you're going to use predict with new data)

library(nlme)
model <- lme(Rodlangde~Mehlich,random=~1|Lokalitet,data=dd)

Create a prediction frame. length=51 is overkill when the model is linear (could just be length=2), but is useful for nonlinear or generalized linear models ...

pframe <- with(dd,
               expand.grid(Lokalitet=levels(Lokalitet),
                      Mehlich=seq(min(Mehlich),max(Mehlich),length=51)))

Predict at "level 1", ie level of localities (rather than level=0, the population level):

pframe$Rodlangde <- predict(model,newdata=pframe,level=1)

Plot with lattice::xyplot :

library("lattice")
xyplot(Rodlangde~Mehlich,group=Lokalitet,data=pframe,type="l")

or with ggplot2 :

library("ggplot2")
ggplot(dd,aes(Mehlich,Rodlangde,colour=Lokalitet))+
     geom_point()+
     geom_line(data=pframe)