0

I look at the size of fish (same species) in ten different sites and different ages. For each individual I know size at age 1,2,3,4 etc. -> I have one value for one individual at certain age (no multiple measurments for a specific age!). [From the data I know at least one population grows larger in all age groups -> could be a population effect indeepended of other fixfactors].

> 'data.frame': 688 obs. of  24 variables: 
>$ FishID       : int  99191 99192 99193 99203 99206 99207 99208 99213 99215 99217 ...  
>$ Sex: Factor w/ 3 levels "F","M","U": 1 2 1 2 2 1 2 2 2 2 ...  
>$ Elevation: int  829 829 829 829 829 829 829 829 829 829 ...  
>$ Mean_Sed: num  14992 14992 14992 14992 14992 ...  
>$ Age : int  1 1 1 1 1 1 1 1 1 1 ...  
>$ Length: num  113 111 117 106 111 ...  
>$ Location     : Factor w/ 10 levels " 1 Tavanasa",..: 1 1 1 1 1 1 1 1 1
>$ Catchment_km2: num  598 598 598 598 598 ...      
>$ Slope : num  1.08 1.08 1.08 1.08 1.08 ...      
>$ No_Trout: int  36 36 36 36 36 36 36 36 36 36 ...      
>$ Qt: Factor w/ 197 Levels "1005109","1011605",..: 122 122 122 122 122 122 122 122 122 122 ...

To simplify I'm am comparing within age groups.

Of course fish do not only differ in site but also in independend population. I tried this nested Approach: Size~(1|Site/FishID)+.... but get an error: number of levels of each grouping factor must be < number of observations.

I think, the error is explained by the fact that I dont have multiple fish size readings for one Individual FishID. Obviously in my case this apporach/nested design (School-Teacher-Scores) can not be used.

=> How do I best tackle this Problem? Is there another way to tell lmer that fish within the location are probably more similar in size because they belong to the same popluation.

Many thanks in advance

FischAnLand
  • 1
  • 1
  • 1
    "That means for each individual I know size at age 1,2,3,4 " This is a textbook example of multiple measurements. You are contradicting yourself. – Roland Jun 01 '17 at 10:43
  • (1|Location/FishID) works if using the whole data set, yes! But I try to look at this for individual age groups, for example all fish at age 1. then in a seperate model for age groupe 2, etc. – FischAnLand Jun 01 '17 at 11:52
  • Why are you fitting separate models? Don't do that. Include age in your model. – Roland Jun 01 '17 at 11:56
  • if I use whole data set, but still want to check if size depends to age group (small fish grow more then older), I would add: (1|Age/FishID) ----> Length~Treatment1+Location+Age+(1|Location/FishID)+(1|Age/FishID)? – FischAnLand Jun 01 '17 at 12:26
  • No, age is not a grouping variable. You could include it as a random slope `(Age|FishID)` and test if that is significant. It's very unclear what hypothesis you want to test. Also, Location can't be a fixed effect as well as a grouping variable for a random effect. A sensible model might be `Length ~ Treatment1 * Age + (Age | Location / FishID) + (Treatment1 * Age | Location)` or a nested more parsimonious model. I suggest to find a local statistician you can talk to. – Roland Jun 01 '17 at 12:42
  • Thanks for your inputs. I guess your rigth about finding a statistican. Thanks for clearification any way! – FischAnLand Jun 01 '17 at 14:09

0 Answers0