Burak,

A 80000x500 matrix is not large for R, especially considering available computers these days. I have worked with matrices of 1.3Mx250 in R with no problem (in a Linux environment, mind you).

If you don't find what you want in R, develop it yourself, it is not as difficult as you may think. Besides imagine the amount of people that would benefit from your effort!

We, modern researchers, have received a lot from other people, it is time for us to repay the debt! That is the philosophy of open source.

Regards,

Augusto

Take a look here, and decide for yourself.

http://www.burns-stat.com/interview-with-a-forced-convert-from-matlab-to-r/

Cheers,

Luiz

Definitely learning and/or using R would only do you good in addition to what the matlab statistical toolbox. I often use both softwares depending on the specify question I intend to answer.

R is a language. Definitely worth learning. Matlab and R are both very useful but R is free and has a hugh library of very useful packages.

Thanks for the answers, and especially for the link Luiz. I guess I'll give R a try :)

Dear Burak,

I am great fan of MATLAB since 22 years now and wrote two textbooks about it, so I'm biased :-) I am using, however, also R, Scilab, Python, Mathematica, Octave from time to time as I am trying to teach stats independently from a specific software. In particular teaching in African countries requires free tools such as R.

One thing, however, the Stats Toolbox is full of parameter-free tools such as the Mann-Whitney test, also known as Wilcoxon rank-sum test, introduced by Henry B. Mann and Donald R. Whitney (1947). The function is called RANKSUM. Others are the Ansari-Bradley test, the KS Test and also bivariate parameter-free measures of correlation such as Spearman and Kendall's correlation coefficients.

Based on a survey here at ResearchGate, I am currently updating the older of my two books which will now include new sections about it. By the way, following the philosophy of my books, I provide a MATLAB code independent from the functions included in the toolbox (also for the Wilcoxon test) before I introduce the functions RANKSUM.

Cheers, Martin

First, I'd just like to say "as suggested by Wilcox" is about the best one can get when it comes to suggestions regarding statistics. I couldn't agree more with his evaluation of SPSS or of R (for those who are neither teach with nor were taught with his introductory statistics textbooks that touch on this, his evaluation of the former was negative and the latter positive).

Second, the comparison is a bit like apples and oranges- both are fruit, but they're not as readily related as one might suppose given that both are mathematical software packages. R is thoroughly and completely a statistical software package and scripting language. For those who are (or were) used to writing code in text editors for C, C++, or various other scripting languages, R is enormously beneficial both due to it's inherent power and the vast online support community, freely downloadable scripts, tutorials, etc. However, MATLAB is not only a scripting language but an environment with a magnificent GUI. One can write scripts, but need not rely only on running them. The inline "command window" allows one to program and test dynamically in ways that are not, to my knowledge, possible with any current R GUI. Also, as much as the linear/matrix algebra is the foundation of multivariate statistics (which would seem to be a reason to require those who use multivariate statistics to take AT LEAST a survey course in linear algebra independently of any multivariate statistics course), it is not straightforwardly extendable either to much of probability theory (which involves calculus) and needlessly complicates some of the statistical methods one might use. This is not hard to overcome, of course, but I have found that a good many multivariate statistical methods I have really wanted to control over require more custom programming in MATLAB than R. A large reason for this has to do with why MATLAB is often not included among lists of statistical software packages: it is more oriented towards engineering and mathematical modelling than to (fairly) straightforward statistical methods. Then there is the price tag- if your lab doesn't pay for a license than it is not exactly in the price range of your average researcher (unless I'm getting cheated). Don't get me wrong- I love it and I use it, but that's partly because I don't have to pay for it and partly because I also use R.

So I guess my recommendation is about as bland and useless as it gets: use both if you can, and R if you can't obtain MATLAB.

I think for ecological studies R is better however for modeling Matlab is better

R is a very powerful statistical packages that worth learning and it is free. However, Matlab is sometimes more useful for modeling.

Judging by the amount of packages, books and documentation available from the R repository (CRAN) I would say that R is becoming the language for people interested in statistical analysis and modelling of data. Since I started learning R more than 6 years ago I have never gone back to matlab. For me the main characteristic of R is that it is open source software, I have adapted many R scripts to my own needs with no problem. Today I have a large collection of R scripts which I use for all my data analysis and modelling (mostly related to modelling natural disasters using extreme value analysis). Another advantage is that I can share my scripts with colleagues who are working on similar problems and also request scripts from them. That is the beauty of it, we are a community of users!

Do not hesitate for a minute: start learning R today!

Again, thanks for all the input! As the answers accumulate, I guess the picture emerges more clearly. For statistical analysis R is better! However, there seems to be another issue. Being a chemical engineer, I am using Matlab for modeling atomic fluctuations, and statistical analysis is ONE of the intermediate (but very important) steps I take during modeling. For instance I take Fourier Transforms (FFT) of large trajectory matrices (e.g. 80000x500), then do hypothesis tests, then perform SVD, followed by other analyses. I do not think I will be able to do all of these analyses and handle huge matrices (especially multidimensional ones) on R. Nevertheless, I may do the hypothesis test step on R, and then transfer the results to MATLAB and continue.

Those were very valuable responses, again thank you all!

O by the way, Martin you are right in mentioning that Matlab's statistical toolbox has a couple of rank-based statistics. However, those advocated by Wilcox are based on trimmed estimators and their confidence intervals. As far as I can see, they are not built in the statistical toolbox. That's why I have been in search for a different computer package.

Burak,

A 80000x500 matrix is not large for R, especially considering available computers these days. I have worked with matrices of 1.3Mx250 in R with no problem (in a Linux environment, mind you).

If you don't find what you want in R, develop it yourself, it is not as difficult as you may think. Besides imagine the amount of people that would benefit from your effort!

We, modern researchers, have received a lot from other people, it is time for us to repay the debt! That is the philosophy of open source.

Regards,

Augusto

Adding here, Mathworks point of view..

https://www.mathworks.com/discovery/matlab-vs-r.html

For learning, I think both of the languages are not really difficult to learn.

To me, the differences come when I need to reuse someone else's code for debugging and re-productivity. In this matter, I can say that Matlab is ways better. I have been doing programming in 10 years, working with both Matlab and R. But working with someone else's R code is always a nightmare for me.

Put it frankly, for a project in which I need to reuse someone else's code, or running a new package/functionality, it only takes me up to 2 hours to completely understand/debug/reuse the Matlab code. But for R, it takes at least 8 hours, including installing package, figuring out error during installation, understanding the code and data structure, poor documentation, etc. 6 hours difference, well people may think 'not a big deal'. However, let's convert to price: my salary is approximately $30 per hour. So 6 hour means $180. That's more than enough to buy Matlab for 2 years on campus. 3 projects means ... $540 lost due to R. Matlab 'permanent' license is just $500, plus a small amount of 'maintenance and update'. I have at least 4 projects from small to big per month! So it is not worthy for me to use some cheap R code, compared to the well-documented and structured Matlab.

R may be faster to produce a new thing. But I never like it.

I am afraid I have to pour water into the wine, that everybody is drinking about R. My main criticism is about the poor documentation. By poor I don’t speak about the wealth of documentation, but its quality. Very often, examples are overly simplistic, not providing the clue to come to a more sophisticated application of the documented function. In addition, functions are often documented without specification of the package, were it can be found. Best example is the legendre() function in matlab style, which I intended to use over the ‘R-style-legendre()-function’. You find a documentation page somewhere on the web, but it doesn’t tell you in which package the function is located. I am a newbie to R and I have spent hours in searching. Reason for my wish to use the matlab-style legendre()-function is the very bad documentation of all legendre-type functions in the ‘orthopolynomial’ package, which is hardly more than providing tabular data of function names. No return types are documented. How can productive work be based on that kind of documentation? Ok, someone else has had the same problem as I have and provided a package that helps you locate specific functions from all CRAN-packages. But if I try to install this package, it turns out that it is outdated and is no longer running on more modern R-versions. The recommend search on rseek.org brings me just back to where I started.

Things like that happen all the time. R is far from being a tool for enabling rapid productivity. Sorry for this