Why isn't my mixed model loop working? (RStudio, Crossover design) - r

Question

I can't figure out why my loop isn't working.
I have a database (36rows x 51columns, its name is "Seleccio") consisting of 3 factors (first 3 columns: Animal (12 animals), Diet (3 diets) and Period (3 periods)) and 48 variables (many clinical parameters) with 36 observations per column. It is a 3x3 crossover design so I want to implement a mixed model to include the Animal random effect and also Period and Diet fixed effects and the interaction between them.
A sample of the data (but with less rows and columns):
Animal Diet Period Var1 Var2 Var3 Var4 Var5 Var6
<chr> <chr> <chr> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
1 A A A 11 55 1.2 0.023 22 3
2 B A A 13 34 1.6 0.04 23 4
3 C B A 15 13 1.1 0.052 22 2
4 A B B 10 22 1.5 0.067 27 4
5 B C B 9 45 1.4 0.012 24 2
6 C C B 13 32 1.5 0.014 23 3
> dput(sample[1:9,])
structure(list(Animal = c("A", "B", "C", "A", "B", "C", NA, NA,
NA), Diet = c("A", "A", "B", "B", "C", "C", NA, NA, NA), Period = c("A",
"A", "A", "B", "B", "B", NA, NA, NA), Var1 = c(11, 13, 15, 10,
9, 13, NA, NA, NA), Var2 = c(55, 34, 13, 22, 45, 32, NA, NA,
NA), Var3 = c(1.2, 1.6, 1.1, 1.5, 1.4, 1.5, NA, NA, NA), Var4 = c(0.023,
0.04, 0.052, 0.067, 0.012, 0.014, NA, NA, NA), Var5 = c(22, 23,
22, 27, 24, 23, NA, NA, NA), Var6 = c(3, 4, 2, 4, 2, 3, NA, NA,
NA)), row.names = c(NA, -9L), class = c("tbl_df", "tbl", "data.frame"
))
I want to make descriptive analysis (normality testing and checking for outliers) of each variable sorted by Diet (which is the treatment) and also run a mixed model and make an ANOVA and a Tukey test for the fixed effects.
I can do the analysis one by one, but it takes a lot of time, I have tried several times to create a for loop to automate the analysis for all the variables but it didn't work (I'm pretty new to R).
What I got so far:
sink("output.txt") # to store the output into a file, as it would be to large to be shown in the console
vars <-as.data.frame(Seleccio[,c(4:51)])
fact <-Seleccio$Diet
dim(vars)
for (i in 1:length(vars)) {
variable <- vars[,i]
lme_cer <- lme(variable ~ Period*Diet, random = ~1 | Animal, data = Seleccio) # the model
cat("\n---------\n\n")
cat(colnames(Seleccio)[i]) # the name of each variable, so I don't get lost in the text file
cat("\n")
print(boxplot(vars[,i]~fact)$out) #checking for outliers
print(summary(lme_cer))
print(anova(lme_cer))
print(lsmeans(lme_cer, pairwise~Diet, adjust="tukey"))
}
sink()
This code runs but doesn't do the job, as it gives me wrong results for each variable because they are different from the results that I get when I analyse each variable one by one.
I would also like to add to the loop this normality testing sorted by Diet (Treatment) code. I wonder if it would be possible.
aggregate(formula = VARIABLENAME ~ Diet,
data = Seleccio,
FUN = function(x) {y <- shapiro.test(x); c(y$statistic, y$p.value)})
Thank you very much in advance to all of those who are willing to lend me a hand, any help will be greatly appreciated

Answer 1

I don't think i can run the model with only 6 observations, so i couldn't find why would your loop doesn't return the same as doing it one by one. Maybe the problem is with cat(colnames(Seleccio)[i]): you only want the Var names, and for i=1, 2 and 3, that code will return "Animal", "Diet" and "Period", thus messing up how you're comparing the results. Using cat(colnames(vars)[i]) might correct that. If you find a way to include more observations of Seleccio i might be able to help more.
I would suggest you to create a list to store the output:
vars <- as.data.frame(Seleccio[,c(4:51)])
fact <- Seleccio$Diet
dim(vars)
output = list() #Create empty list
for (i in 1:length(vars)) {
var = colnames(vars)[i]
output[[var]] = list() #Create one entry for each variable
variable <- vars[,i]
lme_cer <- lme(variable ~ Period*Diet, random = ~1 | Animal, data = Seleccio) # the model
#Fill that entry with each statistics:
output[[var]]$boxplot = boxplot(vars[,i]~fact)$out #checking for outliers
output[[var]]$summary = summary(lme_cer)
output[[var]]$anova = anova(lme_cer)
output[[var]]$lsmeans = lsmeans(lme_cer, pairwise~Diet, adjust="tukey")
output[[var]]$shapiro = aggregate(formula = variable ~ Diet, data = Seleccio,
FUN = function(x) {y <- shapiro.test(x); c(y$statistic, y$p.value)})
}
This way you have the results in you R enviroment, and have better visualisation options: do output$Var1 and get all the results for Var1, which should fit in the console; do for(i in output){print(i$summary)} to get all of the summaries; etc.