[BUGZILLA #17604] approx's new na.rm=FALSE does not play well with 'ties' argument. (R-devel post 3.6.1)

[MichaelChirico]

In R-devel (post 3.6.1) approx() and approxfun() have a new argument, na.rm=TRUE. If FALSE, input x,y pairs with NAs in only the y values are left around so that predictions based on xout values close to x become NA. The bug occurs when there are duplicate x values for which some y values are NA and some not: the ties function is only called if there is more than one non-NA y value of a given x value. I think it should be called if there is more than one y value, NA or not, for a given x.

E.g., shouldn't the following two example give the same results?

data.frame(approx(c(2,3,3,4), c(12,NA,13.2,14), method="constant",
xout=seq(2,4,by=1/2), ties=function(x){print(x);mean(x)}, na.rm=FALSE))

x    y

1 2.0 12.0
2 2.5 12.0
3 3.0 13.2
4 3.5 13.2
5 4.0 14.0

data.frame(approx(c(2,mean(c(3,3)),4), c(12,mean(c(NA,13.2)),14),
method="constant", xout=seq(2,4,by=1/2), ties=function(x){print(x);mean(x)},
na.rm=FALSE))

x  y

1 2.0 12
2 2.5 12
3 3.0 NA
4 3.5 NA
5 4.0 14

---

METADATA

• Bug author - Bill Dunlap
• Creation time - 2019-08-26 19:37:48 UTC
• Bugzilla link
• Status - UNCONFIRMED
• Alias - None
• Component - Analyses
• Version - R 3.5.0
• Hardware - Other Other
• Importance - P5 normal
• Assignee - R-core
• URL -

[github-actions]

I think what you want is

data.frame(approx(c(2,mean(c(3,3), na.rm = TRUE),4), 
c(12,mean(c(NA,13.2), na.rm = TRUE),14),
method="constant", xout=seq(2,4,by=1/2), 
ties=function(x){print(x);mean(x)},

              na.rm=FALSE))

x    y

1 2.0 12.0
2 2.5 12.0
3 3.0 13.2
4 3.5 13.2
5 4.0 14.0

because not having the na.rm=TRUE in mean(c(NA,13.2)) results in NA for the second element of y.

This can be closed.

---

METADATA

• Comment author - elin.waring
• Timestamp - 2020-06-18 14:21:32 UTC