使用glm预测0到1之间的连续变量family = binomial（link ='logit'）会产生错误

Question

I'm trying to use glm to estimate a logistic regression on a continuous variable between 0 and 1 using the following code, but am getting the attached error:

> glm(y ~ x, data=test_data, family=binomial(link = 'logit'))
Error in eval(family$initialize) : y values must be 0 <= y <= 1

However, when I do a summary on test_data, the df has y values that are entirely between 0 and 1...

> summary(test_data)
       y                  x         
 Min.   :0.000000   Min.   :0.0000  
 1st Qu.:0.001510   1st Qu.:0.0000  
 Median :0.003664   Median :1.0000  
 Mean   :0.025847   Mean   :0.5386  
 3rd Qu.:0.009054   3rd Qu.:1.0000  
 Max.   :1.000000   Max.   :1.0000

Can anyone help me understand what the issue here is? If I check the type of the variables, they are both numeric:

> class(test_data$y)
[1] "numeric"
> class(test_data$x)
[1] "numeric"

Answer 1

Suggest you try:

which(as.numeric(test_data$x) < 0 | as.numeric(test_data$x) > 1)
which(as.numeric(test_data$y) < 0 | as.numeric(test_data$y) > 1)

Answer 2

I found the issue here - after drilling down into the data, there are a small number of rows with very small, negative values of y (likely due to rounding errors), eg,:

> test_data[276,]
# A tibble: 1 x 2
          y     x
      <dbl> <dbl>
1 -1.47e-17     0

However, these out-of-range values do not show up in summary.