Yes and No.
Deleting an old version of MATLAB will not interfere with executing newer versions. The versions do not share anything (except the parts you load in through the Add-On Explorer: those gets shared.) And you can certainly save a lot of disk space by deleting old versions.
Factors to consider, though:
- Mathworks only keeps documentation for the last 5 years online. After that, you are expected to use the built-in documentation
- sometimes you encounter third-party toolboxes that insist on being run in a particular version or range of versions. In some cases it means that there is some incompatibility and they will not work with newer versions; in some cases it just means that they did not test with a newer release, and don't want to be stuck trying to debug with newer versions
- There are some device drivers that will only work up to R2015b on Windows 32. If you need (or might need) one of those devices then you do not have much choice
- as you develop software, you might encounter cases where "this used to work!". Sometimes the failure is due to source code changes you forgot you made or which did not seem relevant to the failure. Sometimes the failure is due to operating system changes. Sometimes the failure is due to hardware changes. Sometimes the failure is due to you having updated MATLAB and the new version of MATLAB has bugs that the older one did not have; sometimes the failure is due to you having updated MATLAB and the new version does better error checking and catches an error in your previous code. Sometimes the failure is due to you having updated MATLAB and something works differently for documented reasons. Sometimes the failure is due to you having updated MATLAB and the new version withdrew functions you were using. And... most of those cases, a good debugging step is to go back and run your software in a previous version of MATLAB and see whether the problem sticks with the system (fails the same way in the older version) or depends upon the MATLAB release.
- Sometimes when you use a newer release, the software just gives different results than you are expecting, such as if the high performance math libraries were improved and the changes cascade. In such cases, you can need the older versions on hand to test extensively against the newer version to figure out which version is the "better" version for your purposes. Notice I do not say which version is "right": It is fairly common for versions to turn out different but both to be "valid". For example, it turns out that when you sum() arrays, the answer you get can depend upon how many CPUs you have
- Sometimes you will end up interacting with someone who has older releases. When they report seeing something different than you see, there can be all kinds of reasons (some of which I list above), and it helps a lot to be able to test in the same release they have, to rule out (or not) differences between the versions.
