Most comments talk about efficiency of the code relative to RAM and speed, but I think there is another issue that is usually overlooked.

Assuming the number of records and variables is modest enough that R, SAS,SPSS and MATLAB can all handle the problem size, the differences in program speed are rarely important enough to be substantive.

But for some problem types, the human time dimension differs greatly among program environments. For example if a large number of combinations of relatively routine regression/anova/glm analysis is needed, and formatted output is required, SAS may get from start to finish in dramatically less time than R or MATLAB. Less lines of code in SAS and much more formatted output ready to incorporate directly into reports translates into more efficient project completion in these situations.

This efficiency on a few projects can pay for the SAS license in short order. Less time debugging means more spare time.

Conversely, when a project requires a large degree of more specialized methods, perhaps methods testing and development or simulation, then R or MATLAB are generally the most efficient environments to get a problem up and running. The choice between the two is a narrow split depending more on which environment one likes to work in. MATLAB has an easier programming language, but R has far more built in statistics methods, so it depends on balancing those needs.

In my experience, I can optimize my work by maintaining licenses for SAS (Base and STAT), and also having MATLAB and R available for more specialized work. The trick is forecasting the most efficient environment for a given project and getting the job done.

I regularly use all three programs.

Regarding the computational side of things:

I would agree with others that SAS is preferred, and at time the only choice, when large databases with millions of records and hundreds of variables are at play. Much larger problems can be handled at much greater speed than either MATLAB or R.

I disagree. R is a powerful and free environment, but it is also considered to be a relatively difficult programming environment to work in. It will always call to 'open-source' and highly technical people. However, there are also lots of labs without the technical resources to use something like R, that can pay the premium for a package like SPSS, Matlab or others, that will allow a much less experienced user to do similar things with just a couple of clicks. R will always be a powerful environment for programming and automation, but I don't see it drawing in those types of users anytime soon.

My point here is that increasing number of universities are learning R as a main computing tools. This trend is a sign for decreasing in numbers of customers for non free programs. In my field I see that the number of R users are increasing every day, at least researchers probably going to entirely switch to R

I agree. Our university also uses R in its Biostatistics classes (and department). But I do see a lot of resistance among certain non-computational types. I was just saying that I think there will always be a market for the more user-friendly programs, even if more people move to R.

Ditto. For undergraduate teaching purposes, using R is fine when you are teaching Statistics / Mathematics (or even Computer Science) majors. However, for the majority of students, using command line software is alien; it is already difficult enough to encourage Biology / Business / Psychology students to see the importance of understanding statistical methods without scaring them by using command line based programs. As such, there will remain a need for user friendly programs [not just some of the GUIs for R such as Rcmdr] that cover the majority of techiques covered at undergraduate level.

Furthermore, many companies relying on the output of statistial models to make important decisions will want a warrantee (someone else to blame) of the sort that does not come with open source programs.

I like R too. Programming is quite easy in R.

R is a tricky language and very difficult to use it properly. One of the most important thing in R and any open source project is that you can look inside the code. Closed code software are like black box you put something and get something but you cannot check how it was transformed. Only when you can verify the implementation of some algorithm you can be sure that it work in proper way. Especially in science it is important to know how it's work.

The most important advantage in R is that you can implement your own algorithm or improve someone's code. You can also publish your code in a simple and fast manner.

In my experience, R is not very good for large databases (e.g. 6M by 500K data matrix) that I had to work with. While I was trying to figure out tricks on how to make things work with a reasonable speed and w/o overusing memory, I (surprisingly) found out that SAS has no problem with this and is optimized in terms of memory use and speed. Further, once I had to report confidence intervals for a certain statistic, and would have to program it myself to get them (there was no function to do so), while several types of the needed confidence interval were readily available in SAS and (interestingly) MatLab.

While I do like R and use it a lot, I think it needs significant improvements before it can be a stand alone package.

I did not test either SAS or MatLab, however I know that MatLab is faster and more powerful programming language which is especially suitable for mathematical problem solving. But I think it is more suitable for programmers then researchers (e.g. ecologists). SAS is known as a suitable tool for very large data sets however is not as powerful programming language as R and its licence is very expensive. For very large data R is not very suitable as it runs in RAM and needs many times much RAM than is a size of input however you can see the link below for solving this problem:

http://www.revolutionanalytics.com/products/enterprise-big-data.php

Neither of them (SAS and MatLab) gained so big popularity as R because it is really very useful in everyday and multyway use.

Now a days R keep on improving in handling large data as well. R has some packages like bigmemory and ff for big data. And i am sure in future more packages will be available for handling large data :)

R is a great tool for Open Science (http://en.wikipedia.org/wiki/Open_science), so it is an unvaluebale tool when a reserach should be reproducible.

In my opinion, R is the 'lingua franca' for statistical computing with a really strong community behind it. And yes, R is difficoult beacuse it is, first at all, a programming language. But, on top of the R's core you could install graphic user interface like Rstudio.

Sometimes R's slowness depends on the use of loops with the 'for' command (scalar code). Avoiding them and using vectorization over the data improves speed dramatically.

Most comments talk about efficiency of the code relative to RAM and speed, but I think there is another issue that is usually overlooked.

Assuming the number of records and variables is modest enough that R, SAS,SPSS and MATLAB can all handle the problem size, the differences in program speed are rarely important enough to be substantive.

But for some problem types, the human time dimension differs greatly among program environments. For example if a large number of combinations of relatively routine regression/anova/glm analysis is needed, and formatted output is required, SAS may get from start to finish in dramatically less time than R or MATLAB. Less lines of code in SAS and much more formatted output ready to incorporate directly into reports translates into more efficient project completion in these situations.

This efficiency on a few projects can pay for the SAS license in short order. Less time debugging means more spare time.

Conversely, when a project requires a large degree of more specialized methods, perhaps methods testing and development or simulation, then R or MATLAB are generally the most efficient environments to get a problem up and running. The choice between the two is a narrow split depending more on which environment one likes to work in. MATLAB has an easier programming language, but R has far more built in statistics methods, so it depends on balancing those needs.

In my experience, I can optimize my work by maintaining licenses for SAS (Base and STAT), and also having MATLAB and R available for more specialized work. The trick is forecasting the most efficient environment for a given project and getting the job done.

I regularly use all three programs.

Regarding the computational side of things:

I would agree with others that SAS is preferred, and at time the only choice, when large databases with millions of records and hundreds of variables are at play. Much larger problems can be handled at much greater speed than either MATLAB or R.

Good review! I would like to work with SAS. Unfortunately it is not available for me and for many others also I guess. In such a case there is only choice between octavia and R and the later is undoubtedly the best choice.

Great answer from John Kern.

Like multilingualism is inevitable in programming and computer science, in data analysis we should consider the coexistence of different software as a 'toolbox' from which we choose the best tool for the analyses we want to perform.

Realistically we should balance between (a) using the tools we are better trained to and (b) the 'best' tool but with a steep learning curve.

I am inclined to use the Free and Open Source tools, but I consider beneficial for the software ecosystem the existence of excellent non-free software alternatives and I think that non-free software will not die but they will 'evolve or die' exactly like Open Source Software.

Fabio, Well stated in many fewer words!