Attach/detach in R behaving very strangely - r

Question

I want to subset a dataframe by applying two conditions to it. When I attach the dataframe, apply the first condition, detach the dataframe, attach it again, apply the second condition, and detach again, I get the expected result, a dataframe with 9 observations.
Of course, you wouldn't normally detach/attach before applying the second condition. So I attach, apply the two conditions after one another, and then detach. But the result is different now: It's a dataframe with 24 observations. All but 5 of these observations consist exclusively of NA-values.
I know there's lots of advice against using attach, and I understand the point that it's dangerous, because it's easy to loose track of an attach statement still being active. My point here is a different one; I see a behaviour in attach that I just can't understand. I'm using R Studio 0.99.465 with 64-bit-R 3.2.1.
So here's the code, first the version that is clumsy, but produces the correct result (df with 9 observations, all non-NA):
df <- expand.grid(early_vvl=c(0,1), inter_churn=c(0,1), inter_new_contract=c(0,1), late_vvl=c(0,1), late_no_reaction=c(0,1))
attach(df)
df <- df[(1-early_vvl) >= inter_churn + inter_new_contract + late_vvl,]
detach(df)
attach(df)
df <- df[early_vvl <= late_no_reaction,]
detach(df)
Now the one that produces the dataframe with 24 observations, most of which consist only of NA values:
df <- expand.grid(early_vvl=c(0,1), inter_churn=c(0,1), inter_new_contract=c(0,1), late_vvl=c(0,1), late_no_reaction=c(0,1))
attach(df)
df <- df[(1-early_vvl) >= inter_churn + inter_new_contract + late_vvl,]
df <- df[early_vvl <= late_no_reaction,]
detach(df)
I'm puzzled. Does anybody understand why the second version produces a different result?

Answer 1

Have a look at what happens here:
attach(df)
df <- df[(1-early_vvl) >= inter_churn + inter_new_contract + late_vvl,]
length(early_vvl <= late_no_reaction)
## [1] 32
df <- df[early_vvl <= late_no_reaction,]
detach(df)
So your logical vector early_vvl <= late_no_reaction still uses the original df, the one that you attached. When you subset the data.frame the second time, the logical is longer than the data.frame has rows and it behaves like this:
df <- data.frame(x=1:5, y = letters[1:5])
df[rep(c(TRUE, FALSE), 5), ]
## x y
## 1 1 a
## 3 3 c
## 5 5 e
## NA NA <NA>
## NA.1 NA <NA>
You could just use & to avoid the problem:
df <- expand.grid(early_vvl=c(0,1), inter_churn=c(0,1), inter_new_contract=c(0,1), late_vvl=c(0,1), late_no_reaction=c(0,1))
attach(df)
df <- df[(1-early_vvl) >= inter_churn + inter_new_contract + late_vvl & early_vvl <= late_no_reaction,]
detach(df)