Again about do(.) in dplyr

[Stuart Luppescu]

I had a problem like this a couple of months ago. Dennis Murphy helped by suggesting using do(.). It worked then, but when I try to apply it to another problem it's failing.

I wrote a function to calculate reliability:

calc.rel <- function(x, se) {

    var.x <- var(x, na.rm=TRUE)

    mse <- sum(se^2, na.rm=TRUE)/sum(!is.na(se))

    rel <- (var.x - mse)/var.x

    data.frame(rel)

}

I have a data frame that looks like this:

 head(dm.both)

   meas   se agency     type

1  4.99 1.10    CPS Combined

2 -0.11 0.62    CPS Combined

3  2.29 0.74    CPS Combined

4  3.42 0.78    CPS Combined

5 -1.19 0.59    CPS Combined

6 -0.84 0.59    CPS Combined

I would like to calculate basic statistics and reliability for every combination of agency and type:

    dm.comp <- dm.both %>%

        group_by(type, agency) %>%

            summarize(mean=mean(meas),

                      stddev=sd(meas),

                      rel=do(calc.rel(meas, se))

                      )

But I'm getting this error:

Error in summarize(., mean = mean(meas), stddev = sd(meas), rel = do(calc.rel(meas,  :   argument "by" is missing, with no default

It works fine without the do(calc.rel()) line. 

Any ideas?

--

Stuart Luppescu -- pixbuf .at. gmail.com
I made up a new word: plagiarism.

  --jpuller

[holger brandl]

Stuart,

Just remove do within summarize! The grouping is taken care of by summarize

However, I think there's another mistake which is that calc.rel should be invoked with the summarized mean and stdev:

    summarize(

         mean=mean(meas),

         stddev=sd(meas),

         rel=calc.rel(mean, stddev)

   )

Cheers,

Holger