Several things I think should be changed in R to make it less nasty for new users:

1. summary.data.frame on numeric columns should produce standard deviation by default (turnable off with an option); it should offer several other common statistics as optional possibilities. If making sd default is going to break too many things, it should be an option for the function.

2. wilcox.test should be renamed wilcoxon.test. wilcox.test should still work as an abbreviated form of the name (?wilcox.test should show the help for wilcoxon.test). [Similarly - but less of a 'new user' issue - shapiro.test should be shapirowilk.test or shapiro.wilk.test or similar, kruskal.test should be kruskal.wallis.test etc.]

3. Basic table functions are a bit all over the place. There should be a new, somewhat more helpful function that does most basic stuff by calling them as needed that formats tables nicely.

@Glen_b ifelse should work sensibly. Help messages ought not be attempts to prove how smart the author of the message is. There's TONS of stuff wrong with R.

ifelse is somewhat above the level I'm going for in my rant, but maybe I missed something; what's ifelse doing nonsensibly?

the usual element by element construct if condition <do this> else <do that> becomes the vectorized function ifelse(condition, <do this>, <do that>) ... which makes sense to me.

All the ways (without loading or writing functions) to add sd to a summary output are very clunky but it's a bog standard thing to want.

I agree about help being hard on very new users (as an experienced user, I find help is usually very ... well, helpful, but it's not what new users need at all). I think help as it stands could be called manual and help could then be more new-user-helpful.

ifelse (at least if I remember right) does weird things with vectors and length of expressions. It's caused me lots of odd errors. I agree with you about table() the whole thing is weird.

I think the best short description of R is that it is "expert friendly".

Ah, yes, I think I know the problem you're talking about -- it happens with some other functions as well.

I use ifelse a lot and have never run into it, but it might just be the things I am doing with it.

But this probably isn't the right spot to get into a long discussion of R's merits/flaws.

Oh, gung gives a list of what he thinks it's for when telling people to post stats questions on the main site and always includes "gripes" in the list -- this is one of those.

Talking about it may help clarify things enough to come up with a concrete proposal.

Not volunteering, but perhaps a team of people should just start all over again with R and get everything right this time. That would be S -- no, been there -- perhaps Q would work. More seriously, @Glen_b's (1) better summary.frame defaults (2) sensible use of names like Wilcoxon etc. could be handled by yet more functions.

(3) on table functions is reminiscent of discussions about my favourite software where there is widespread approval for leaner basic table functionality until quite what it should cover is discussed, at which point people can't agree on elephant, giraffe or rhinoceros as the ideal horse.

yes most of my issues could be dealt with by adding a few functions

to the basic distribution of R

I'd really like read.table to give a warning when double-precision isn't enough to losslessly represent a number from a text file.

While we are daydreaming, where do I submit my request to finally agree on a convention on using dots or not? I have been using R for 15 years now and still have to look up the rownames() function vs. the row.names parameter for read.table(). And why does read.table() have a dot, and cumsum() not?

Dots all the way, @StephanKolassa! Down with underscores!

I've seen so many questions about Kaggle's Titanic dataset that it should probably have its own tag.

It will fade with time, @Kodiologist. I don't think the tag would have lasting value. Nor do I think a more generic [kaggle] tag would be sufficiently coherent to help organize the site's information in a useful way. You could bring it upon meta.CV, if you wanted.