The use of format.pval in R (and with \Sexpr in Sweave)

Question

I have read the help file for format.pval and was wondering if this is a right tool to output numbers with proper format as far as significant digits are concerned

In this post:

R / Sweave formatting numbers with \Sexpr{} in scientific notation

a solution was proposed and I wonder if the function format.pval is already here to do this.

And I guess that we can just use it in Sweave as \Sexpr{format.pval(a value/variable here)} together with options for the number of significant digits.

Thanks a lot...

`format.pval` uses `format` which is more general in purpose. So you should probably use `format`, `prettyNum` or `formatC` instead. — Andrie, Dec 09 '11 at 07:30
@ Andrie Thanks. Can you consider converting your comment in a a solution here? 1 vote up. — yCalleecharan, Dec 09 '11 at 07:36

score 3 · Answer 1 · answered Dec 09 '11 at 18:29

3

I have included that functionality in the knitr package, which makes knitr really smart now -- it automatically uses $a \times 10^b$ if your \Sexpr{} produces a number and it is either too big or too small (if you output HTML instead of TeX, it automatically uses a × 10<sup>b</sup>). You can probably start switching from Sweave to knitr now: http://yihui.github.com/knitr/

answered Dec 09 '11 at 18:29

Yihui Xie

28,913
23
193
419

Thanks. Is knitr stable? I understand that it is much more recent than Sweave, 1 vote up. – yCalleecharan Dec 28 '11 at 06:18
The first version will be released to CRAN very soon, and it can be considered as stable then. – Yihui Xie Jan 13 '12 at 01:25

AdamO · Answer 2 · 2017-06-14T21:09:53.860

I think format.pval should probably not be used. We (statisticians) generally don't use conventional scientific notation for $p$-values because the leading 0s are also a reflection of precision. Thus we use total digits. You will find this in most scholarly publications.

For instance, reporting three digits means "trace" values are reported with "<0.001" as in the greatest precision afforded is to the thousands-th place. If I specify digits=3 and eps=0.001, a p-value of 0.0040523 is reported as 0.00405 suggesting my precision is to the hundred-thousandsth place when 0.004 would have been desired.

I use the following little wrapper:

format.p <- function(p, precision=0.001) {
  digits <- -log(precision, base=10)
  p <- formatC(p, format='f', digits=digits)
  p[p == formatC(0, format='f', digits=digits)] <- paste0('<', precision)
  p
}

score 1 · Accepted Answer · answered Dec 09 '11 at 08:51

format.pval is specifically designed to format printed p.values, for example when viewing the output of lm.

The workhorse for format.pval is format which is more general in purpose. So you should probably use format or its cousins - prettyNum and formatC instead.

Footnote: for formatting dates (POSIXct or POSIXlt) you need the date formatting function strptime

score 0 · Answer 4 · answered Jan 11 '19 at 16:24

To be conservative, a p-value of 0.32312 should be rounded to 0.324, not 0.323. Here's my solution:

format.p = function(p, precision=0.001) {
  digits = -log(precision, base=10)
  p = formatC(ceiling(p/precision)*precision,format = 'f', digits=digits)
  p[p == formatC(0, format='f', digits=digits)] = paste0('<', precision)
  p
}

The use of format.pval in R (and with \Sexpr in Sweave)

4 Answers4