R. Handling dates and wide format from an imported Stata file - r

Question

I have been given a Stata data file (counts.dta) that contains daily counts for the years 1975 to 2006 stored in wide-format. The columns are labelled month (full name of the month as a character string), day (numeric with values 1-31), and then the years from 1975 to 2006 with labels '_1975', '_1976' ... '_2006'. I assume that the underline is a consequence of something in Stata. There are dummy counts of zero (0) inserted for the date 29 February when the year-column is not a leap year.
I want to do several things. First, convert to long form with a sensible representation for year. Second, change the tri-partite representation of the date to something more sensible.
My approach has been to change the character string month to a factor and then to get it into the correct order:
require("foreign")
counts <- read.dta(file='counts.dta')
counts[['month']] <- as.factor( counts[['month']] )
counts[['month']] <-
factor(counts[['month']], levels( counts[['month']] )[c(5,4,8,1,9,7,6,2,12,11,10,3)])
I then have
str( counts )
'data.frame': 366 obs. of 34 variables:
$ month: Factor w/ 12 levels "January","February",..: 1 1 1 1 1 1 1 1 1 1 ...
$ day : int 1 2 3 4 5 6 7 8 9 10 ...
$ _1975: int 515 649 745 599 445 667 725 749 646 740 ...
$ _1976: int 485 685 529 467 630 723 712 685 715 504 ...
$ _1977: int 505 437 489 588 634 734 682 537 453 673 ...
and so forth. Converting to long format
lcounts <- reshape(counts,
direction="long",
varying=list(names( counts )[3:34]),
v.names="n.counts",
idvar=c("month","day"),
timevar="Year",
times=1975:2006)
str( lcounts )
gives
'data.frame': 11712 obs. of 4 variables:
$ month : Factor w/ 12 levels "January","February",..: 1 1 1 1 1 1 1 1 1 1 ...
$ day : int 1 2 3 4 5 6 7 8 9 10 ...
$ Year : int 1975 1975 1975 1975 1975 1975 1975 1975 1975 1975 ...
$ n.counts: int 515 649 745 599 445 667 725 749 646 740 ...
plus some further lines relating to the original Stata file.
My questions are: (1) what is now a good way to convert to factor-month, numeric-year and numeric-day to a useful date format, so that I can determine, for example, the day of the week, the interval between two dates and so on? (2) Was there a better way to have tackled the problem from the start?

Answer 1

This should be pretty easy because all you have to do is paste together the rows of your data.frame and use as.Date to create a Date class vector.
Let's start with some data similar to yours:
dat <- data.frame(month = c(rep("January",31), rep("February",29)),
day = c(1:31, 1:29),
Year = 1975,
n.counts = 515)
Then the creation of the date variable is simple:
dat$Date <- as.Date(with(dat, paste(as.numeric(month), day, Year)), "%m %d %Y")
str(dat)
# 'data.frame': 60 obs. of 5 variables:
# $ month : Factor w/ 2 levels "February","January": 2 2 2 2 2 2 2 2 2 2 ...
# $ day : int 1 2 3 4 5 6 7 8 9 10 ...
# $ Year : num 1975 1975 1975 1975 1975 ...
# $ n.counts: num 515 515 515 515 515 515 515 515 515 515 ...
# $ Date : Date, format: "1975-02-01" "1975-02-02" "1975-02-03" "1975-02-04" # ...

Answer 2

The main focus in this thread is naturally what to do in R after data import, but here I bundle together various details on the Stata side of this.
It is longstanding advice that data of this kind are much more easily handled in Stata in a long shape and reshape long is a standard command to do that conversion for data arriving with each year's data in a separate variable (R users: please read "column" as a translation). So, if possible, you should ask a provider of such Stata files to do that before export.
What the OP calls labels such as _1975 are legal variable names in Stata, and as the OP guesses the underscore is needed because variable names in Stata may not start with numeric characters.
On the information given, it would have been possible to export the data without loss from Stata in file formats other than .dta, notably as the usual kinds of text files (.csv, etc.).
Stata's preferred way of holding daily dates is as integers with origin 0 = 1 January 1960 (so 26 March 2015 would be 20173), which presumably is trivially easy to convert to any date representation in R.
In short, the particular and indeed peculiar form of the data as presented to the OP is in no sense either required by any Stata syntax or even recommended as part of good Stata practice.