Date and time zone conversion – ambiguous DST transition handling in as.POSIXct

Answer first

With an explicit tz = "America/New_York", common base R builds parse "2021-11-07 01:30" to the earlier fall-back occurrence: 01:30 EDT = 2021-11-07 05:30 UTC. The later occurrence would be 01:30 EST = 2021-11-07 06:30 UTC.

There is no documented base R option or argument to choose the later instance. as.POSIXct has no fold/ambiguous parameter. Omitting tz uses the system time zone and makes results non-portable.

Do not assume future R releases will standardise this. Base R does not guarantee earlier/later resolution for ambiguous wall-clock times; it depends on the R build and the underlying C time functions and timezone database.

Confirmed facts

• POSIXct stores seconds since the Unix epoch. tz controls how a character string is parsed and how the instant is displayed.
• Three-letter abbreviations like "PST" are not portable Olson zones. Using them can be silently ignored and yield UTC-like values.
• Base R provides no configurable switch for ambiguous DST times.

Likely explanation

The parsing is implementation behavior, not an R contract. On many builds the broken-down time 01:30 on 2021-11-07 is interpreted with the DST offset still in effect, i.e. EDT UTC-4, giving the earlier 05:30 UTC instant. This is why the default appears to be the earlier occurrence, but it is not documented and can vary by build and OS timezone data.

Portable steps for this case

1. Always supply an explicit tz. Never rely on the system zone.
2. Parse the ambiguous wall time with the location name and verify the instant.

x <- as.POSIXct("2021-11-07 01:30", tz = "America/New_York")
format(x, "%Y-%m-%d %H:%M:%S %Z %z")
as.numeric(x)

3. To force the later EST instance, parse with the fixed offset for that instance and then display in America/New_York. POSIX Etc/GMT+5 means UTC-5.

y <- as.POSIXct("2021-11-07 01:30", tz = "Etc/GMT+5")
format(y, tz = "America/New_York", "%Y-%m-%d %H:%M:%S %Z %z")
as.numeric(y)

When x is the earlier occurrence, as.numeric(y) - as.numeric(x) should be 3600 seconds.

Do not infer the offset from %Z alone; abbreviations repeat. Compare %z numeric offsets and the numeric epoch.

To make a recommendation specific to your environment, what R version and OS/timezone database is the target machine running? The resolved instance can differ by build, so the target-machine result is authoritative.