Error in the application of the Boxcox function of the MASS package

Question

I’m getting the following error while using the function boxcox() package MASS:

Error in as.data.frame.default(data, optional = TRUE) : 
  cannot coerce class ‘"function"’ to a data.frame

I don’t understand how to solve it, follow my script:

df <- read.table("https://raw.githack.com/fsbmat/StackOverflow/master/teste.txt",header = TRUE)
names(df)[1:4] <- c("a","b","rep","y")
str(df)
df$a <- as.factor(df$a)
df$b <- as.factor(df$b)
# análise de variância 
m0 <- aov(y~a*b, data=df)
summary(m0)
# checagem das pressuposições
par(mfrow=c(2,2)); plot(m0); layout(1)
#
#------------------------------------------------------------------------------------------
# testes
shapiro.test(residuals(m0))
bartlett.test(residuals(m0)~interaction(df$b,df$a)) 
car::leveneTest(m0) 
car::leveneTest(m0, center="mean")
#------------------------------------------------------------------------------------------
# precisa-se de tranformação para normalidade e homocedasticidade
require(MASS)
boxcox(m0)

Answer 1

In addition to the model adjusted to the data, it is necessary to inform the function MASS::boxcox where this data is stored:

boxcox(m0, data = df)
Error in boxcox.default(m0, data = df) : 
  response variable must be positive

But see that R gives us an error message. This is because the Box-Cox transformation is not set to zero and negative numbers, but your response variable is assuming exactly less than zero:

summary(df$y)
    Min.  1st Qu.   Median     Mean  3rd Qu.     Max. 
-0.02981 14.55843 21.65648 18.96480 25.39923 28.93967
sum(df$y < 0)
[1] 1

In this way, I would ask myself if this value can exist in this context. After all, there are 259 observations and only one is negative. Since I know absolutely nothing about the origin of this data, I find it strange. But it may be that this observation makes sense yes and then another transformation should be applied in the data. I suggest researching the Yeo-Johnson transformation.