[BUGZILLA #17897] all.equal(x,y) wrongly reports TRUE for factors containing different sorts of NAs

[github-actions]

x <- structure(3:1, .Label = c("a", "b", NA), class = "factor")
y <- structure(c(NA, 2L, 1L), .Label = c("a", "b", NA), class = "factor")
all.equal(x, y)
# [1] TRUE

But is.na() gives different results for x and y.
all.equal(is.na(x), is.na(y))
#[1] "1 element mismatch"
is.na(x)
#[1] FALSE FALSE FALSE
is.na(y)
#[1] TRUE FALSE FALSE

I am not sure if ordinary R functions, aside from structure, would ever produce 'y', but the vroom package does and its tests use all.equal() to conclude inappropriately that x and y are equivalent. See tidyverse/vroom#262 for details.

---

METADATA

• Bug author - Bill Dunlap
• Creation time - 2020-08-26 20:02:35 UTC
• Bugzilla link
• Status - UNCONFIRMED
• Alias - None
• Component - Accuracy
• Version - R 4.0.x
• Hardware - All All
• Importance - P5 normal
• Assignee - R-core
• URL -

[github-actions]

NA

---

METADATA

• Comment author - Benjamin Tyner
• Timestamp - 2020-09-12 20:47:41 UTC

[github-actions]

'y' can be produced by
factor(c("c", "b", "a"), levels = c("a", "b", NA), exclude = NULL)
factor(c("c", "b", "a"), levels = c("a", "b", "c", NA), exclude = "c")

---

METADATA

• Comment author - Suharto Anggono
• Timestamp - 2020-09-12 22:08:11 UTC

[github-actions]

NA

---

METADATA

• Comment author - elin.waring
• Timestamp - 2020-09-14 19:03:21 UTC

[github-actions]

It looks like the current behavior may be intentional, based on the log of revision 56668,
------------------------------------------------------------------------
r56668 | ripley | 2011-08-08 13:36:08 -0400 (Mon, 08 Aug 2011) | 1 line

all.equal.factor predates character NAs
------------------------------------------------------------------------

---

METADATA

• Comment author - Benjamin Tyner
• Timestamp - 2020-09-15 00:13:54 UTC

[github-actions]

NA

---

METADATA

• Comment author - elin.waring
• Timestamp - 2020-09-15 01:02:54 UTC

[github-actions]

Since the 2 factors act differently, e.g. in modelling functions, I think all.equal should report that they are not equivalent.

x1 <- structure(rep(3:1,3:1), .Label = c("a", "b", NA), class = "factor")
x2 <- structure(rep(c(NA, 2L, 1L),3:1), .Label = c("a", "b", NA), class =
"factor")
all.equal(x1,x2)

[1] TRUE

y <- 11:16
summary(lm(y~x1))

Call:
lm(formula = y∼ x1)

Residuals:
1 2 3 4 5 6
-1.000e+00 -2.498e-15 1.000e+00 -5.000e-01 5.000e-01 -4.441e-16

Coefficients:
Estimate Std. Error t value Pr(>|t|)
(Intercept) 16.0000 0.9129 17.527 0.000405 ***
x1b -1.5000 1.1180 -1.342 0.272228
x1NA -4.0000 1.0541 -3.795 0.032119 *
---
Signif. codes: 0 ‘’ 0.001 ‘’ 0.01 ‘’ 0.05 ‘.’ 0.1 ‘ ’ 1

Residual standard error: 0.9129 on 3 degrees of freedom
Multiple R-squared: 0.8571, Adjusted R-squared: 0.7619
F-statistic: 9 on 2 and 3 DF, p-value: 0.05399

summary(lm(y~x2))

Call:
lm(formula = y∼ x2)

Residuals:
4 5 6
-0.5 0.5 0.0

Coefficients:
Estimate Std. Error t value Pr(>|t|)
(Intercept) 16.0000 0.7071 22.627 0.0281 *
x2b -1.5000 0.8660 -1.732 0.3333
---
Signif. codes: 0 ‘’ 0.001 ‘’ 0.01 ‘’ 0.05 ‘.’ 0.1 ‘ ’ 1

Residual standard error: 0.7071 on 1 degrees of freedom
(3 observations deleted due to missingness)
Multiple R-squared: 0.75, Adjusted R-squared: 0.5
F-statistic: 3 on 1 and 1 DF, p-value: 0.3333

---

METADATA

• Comment author - Bill Dunlap
• Timestamp - 2020-09-15 01:21:13 UTC

[github-actions]

NA

---

METADATA

• Comment author - elin.waring
• Timestamp - 2020-09-15 12:29:40 UTC

[github-actions]

NA

---

METADATA

• Comment author - Martin Maechler
• Timestamp - 2020-09-16 22:11:44 UTC