boot() does not calculate bias and std error

[LenaH]

Hi,

I'm afraid boot() doesn't want to become friends with me.

I tried to boot a very simple regression and I did it exactly as told by the Quick-R: Bootstrapping site. But it won't calculate the bias and the standard error...

My script looks as followed:

TPfunction <- function(formula, data, coefficients) {
  d <- data[coefficients, ]
  fitTP <- lm(formula, data = d)
  return(coef(fitTP))
}

boot_JSTP <- boot(data = data_JS2, statistic = TPfunction, R = 1000, formula = JS.All2 ~ TP.All2)


boot.ci(boot_JSTP,
        conf = .95,
        type = "bca",
        replace = T, index = 2)

And the output like that:

> boot_JSTP

ORDINARY NONPARAMETRIC BOOTSTRAP


Call:
boot(data = data_JS2, statistic = TPfunction, R = 1000, formula = JS.All2 ~ 
    TP.All2)


Bootstrap Statistics :
     original  bias    std. error
t1*  5.784360       0           0
t2* -0.174887       0           0

Does anybody know what I miss?

[nirgrahamuk]

That does look strange.
I think it might depend on what data_JS2 is.
What is your output when you do:

TPfunction <- function(formula, data, coefficients) {
  d <- data[coefficients, ]
  fitTP <- lm(formula, data = d)
  return(coef(fitTP))
}

data_JS2 <-data.frame(
  JS.All2 = c(rep(0,500),rep(12,500)),
  TP.All2 =c(rep(0,450),rep(6,550))
)

group_by_all(data_JS2)%>%summarise(n=n())


boot_JSTP <- boot(data = data_JS2, statistic = TPfunction, R = 1000, formula = JS.All2 ~ TP.All2)
boot_JSTP

[LenaH]

Hi,

thanks for the reply.

With your script my output looks like this:

> TPfunction <- function(formula, data, coefficients) {
+   d <- data[coefficients, ]
+   fitTP <- lm(formula, data = d)
+   return(coef(fitTP))
+ }
> data_JS2 <-data.frame(
+   JS.All2 = c(rep(0,500),rep(12,500)),
+   TP.All2 =c(rep(0,450),rep(6,550))
+ )
> group_by_all(data_JS2)%>%summarise(n=n())
Fehler in group_by_all(data_JS2) %>% summarise(n = n()) : 
  konnte Funktion "%>%" nicht finden
> boot_JSTP <- boot(data = data_JS2, statistic = TPfunction, R = 1000, formula = JS.All2 ~ TP.All2)
> boot_JSTP

ORDINARY NONPARAMETRIC BOOTSTRAP


Call:
boot(data = data_JS2, statistic = TPfunction, R = 1000, formula = JS.All2 ~ 
    TP.All2)


Bootstrap Statistics :
         original       bias     std. error
t1* -2.049203e-13 2.003240e-13 7.424014e-14
t2*  1.818182e+00 3.931745e-04 2.406952e-02

I used data_JS2 for another bootstrap. Its a data frame of 6 variables and 327 observations.

[nirgrahamuk]

ok. with my made up data_JS2, you get the same results as me.
In principle can you share data_JS2 ?
I think we only need the two variables you use.

library(dplyr)
dput(data_JS2 %>% select(JS.All2 ,  TP.All2))

and copy and paste the results

[LenaH]

That worked
I created a new data frame with only the two needed varibales and it works well.

Thank you so much!!!