Rounding a time

[Bob Green]

Hello,

I have data in the following format. I want to examine when the most activity occurs by half hours. This code extracts the hour - gen double myhour = hh(testvar). Would I need to transform the data into a string and rename time units as 1pm to 1.30 pm etc, or is there a more efficient way to do this.

Any assistance is appreciated.

Bob

* Example generated by -dataex-. To install: ssc install dataex
clear
input double testvar
     -1893406560000
     -1.8933828e+12
     -1893382140000
-1893379380000.0002
     -1.8933705e+12
     -1893384660000
-1893405239999.9998
     -1893401880000
     -1893401160000
     -1893401640000
     -1893393780000
end
format %tchh:MM testvar

[Nick Cox]

Thanks for the clear example.

(It seems that I have to request such an example in most of the threads I have recently posted in.)

My preference is to round down and tag each period by when it begins. In this case there are 30 * 60 * 1000 ms in a half hour.

. l

     +---------+
     | testvar |
     |---------|
  1. |    1:44 |
  2. |    8:20 |
  3. |    8:31 |
  4. |    9:16 |
  5. |   11:45 |
     |---------|
  6. |    7:49 |
  7. |    2:06 |
  8. |    3:02 |
  9. |    3:14 |
 10. |    3:06 |
     |---------|
 11. |    5:17 |
     +---------+

. gen double testvar2 = 30 * 60000 * floor(testvar / (30 * 60 * 1000))

. format %tchh:MM testvar2

. l

     +--------------------+
     | testvar   testvar2 |
     |--------------------|
  1. |    1:44       1:30 |
  2. |    8:20       8:00 |
  3. |    8:31       8:30 |
  4. |    9:16       9:00 |
  5. |   11:45      11:30 |
     |--------------------|
  6. |    7:49       7:30 |
  7. |    2:06       2:00 |
  8. |    3:02       3:00 |
  9. |    3:14       3:00 |
 10. |    3:06       3:00 |
     |--------------------|
 11. |    5:17       5:00 |
     +--------------------+

I see precisely no reason to use a string here. A numeric variable ensures proper ordering, among other things.

See also http://www.stata-journal.com/sjpdf.h...iclenum=dm0002

The same problem arose in different form earlier today: http://www.statalist.org/forums/foru...-in-panel-data

[Bob Green]

Nick,
Many thanks for the prompt reply. I had looked at round, ceil and floor but can now see why they didn't work the way I'd used them.
Bob

[Nick Cox]

Despite its wonderful name, round() is a function I usually avoid, except for rounding to the nearest integer, because otherwise I can never remember exactly what it does, especially at boundary cases, where very often there is a strong idea of what should happen. (Experiment! is a fair answer to that, however.)

floor() and ceil() are in contrast much easier to memorise and extend. For example, ceil(10 * runiform()) is an easy way to get a uniform sample from the integers 1 to 10. (There are other ways to do that too.)