I once heard John Chambers (the inventor of the S language, and member of the R Core Group) say, "Show me a programming language no-one complains about, and I'll show you a language no-one uses". The R language has its fair share of complainants, to be sure -- and that's to be expected for a language with more than 2 million users.
- The Packages: "Packages for all manner of problems have been written, frequently exclusively for the R universe"
- RStudio: "makes interactive coding easy-peasy"
- The Language: "Yet despite the simplicity, and thanks to the fact that everything is vectorized, the language is expressive enough to accomplish an impressive number of things in one line."
- The Data Structures: "the data frame and the factor stand out in particular"
- Serializability: "Code, even closures, can be treated as first-class values—you can serialize it and send it around."
- CRAN: "I don’t know if there’s something about the way R package authoring/publishing works that makes distribution particularly robust and straightforward, or what. But shit just works."
- Embedded R and Rserve: "There’s some good interop in the form ofRPy2 for Python and REngine for Java"
Read Yang's full post for all the details. One thing I'd add to this list myself is R's formula notation for statistical modeling (see this StackOverflow answer comparing R and Python for regression for a good example). Our own take on what's to like about R can be found in our series, "Why R is Hot".
@yaaang's blog: What’s there to like about R?