Time series plot gets offset by 2 hours if scale_x_datetime is used

Question

Problem:

I'm trying to plot a time series with ggplot but the data gets offset by two hours for some reason.

Data:

> test <- structure(list(interval = structure(c(1465423500, 1465423800, 
1465424100, 1465424400, 1465424700, 1465425000, 1465425300, 1465425600, 
1465425900, 1465426200, 1465426500, 1465426800, 1465427100), class = c("POSIXct", 
"POSIXt"), tzone = ""), mean = c(0.339622641509434, 0.132075471698113, 
0.150943396226415, 0.0754716981132075, 2.09433962264151, 0.528301886792453, 
0.867924528301887, 0, 1.47169811320755, 0.30188679245283, 0.132075471698113, 
0.320754716981132, 0.679245283018868)), .Names = c("interval", 
"mean"), class = c("tbl_df", "data.frame"), row.names = c(NA, 
-13L))

> test
    Source: local data frame [13 x 2]     

interval      mean     
(time)     (dbl)     
1  2016-06-09 00:05:00 0.3396226 # First value: 5 minutes past midnight    
2  2016-06-09 00:10:00 0.1320755     
3  2016-06-09 00:15:00 0.1509434     
4  2016-06-09 00:20:00 0.0754717     
5  2016-06-09 00:25:00 2.0943396     
6  2016-06-09 00:30:00 0.5283019     
7  2016-06-09 00:35:00 0.8679245     
8  2016-06-09 00:40:00 0.0000000     
9  2016-06-09 00:45:00 1.4716981     
10 2016-06-09 00:50:00 0.3018868     
11 2016-06-09 00:55:00 0.1320755     
12 2016-06-09 01:00:00 0.3207547     
13 2016-06-09 01:05:00 0.6792453

Example

This works fine:

g <- ggplot(interval.steps, aes(interval, mean))
g + geom_line()

But this doesn't:

g <- ggplot(interval.steps, aes(interval, mean))
g + geom_line() +
    scale_x_datetime(date_labels = '%H:%M') # offsets times by -2 hours

Question

What am I doing wrong? Thanks in advance.

Answer 1

It looks like scale_x_datetime is changing the timezone of interval from your local timezone to UTC. The function below should resolve the problem.

# Source: http://stackoverflow.com/a/11002253/496488
# Put in your local timezone. I've inserted mine just for illustration.
date_format_tz <- function(format = "%H:%M", tz = "America/Los_Angeles") {
  function(x) format(x, format, tz=tz)
}

g <- ggplot(interval.steps, aes(interval, mean))
g + geom_line() +
  scale_x_datetime(labels = date_format_tz())

Answer 2

Time zone-independent implementation

Eipi10's answer above is a good workaround. However, I wanted to avoid hardcoding a time zone setting into my program in order to make it reproducible in any locale. The way to achieve this is very simple, just leave out the tz parameter:

# Generator function to create 'hh:mm' labels for the x axis
# without explicit 'tz' specification  
date_format <- function(format = "%H:%M") {
    
    function(x) format(x, format)
}

Advantage

The advantage of this method is that it works regardless of the time zone parameter of the original variable and the current locale.

For example if your time values were read in with something like this:as.POSIXct(interval, format = '%H:%M', tz = 'Pacific/Honolulu'), the graph will still be plotted with the correct X axis labels, even if you're in, say, Zimbabwe.