Ok, as long as it's interpreted carefully (mostly the bad experiences, fewer of the good experiences) should be useful.

Of course we know a priori that the results will be grossly biased. People unhappy with their situation take part in such surveys w/ a much higher probability than those that are happy.

dplyr band_instruments isn't loaded with library(dplyr). But then data(band_instruments) also doesn't work and you need
> data(band_instruments, package = 'dplyr')

I'm pretty sure something has changed.

Has this behaviour changed at some point in the last ten years? Now seems almost no datasets need you to do data(the_dataset)

Great work @freerangestats.info It'll be interesting to see how and why this is wrong in multiple analyses and over multiple years.

I'll have to work out what the regions being used were. But it just can't be true in any meaningful sense. The least ridiculous is a region with life expectancy of 40 (crazy low) and another with 108 (crazy high).

And pretty sure USA doesn't have a 68 year life expectancy gap.

Avoid anything that has external points of failure. Applying for date, or emailing and asking for date are a problem. A two week delay is not a problem in a PhD or postdoc, but pretty massive in a three month MSc project.

That's all I've got!

They may be crap, be aware and accepting of that.
Plan a project that builds in steps. Do "a pass" worth of analysis, then try to extend to merit/distinction.

Lots of projects are a fail until you've completed 95%, but some students won't manage that.

I now realise that's what other people were saying. oh well

Only happens for Mersenne-twister, which probably means something.

set.seed(0, kind = "Wichmann-Hill")
x0 <- runif(5)
set.seed(1, kind = "Wichmann-Hill")
x1 <- runif(5)

x <- cbind(x0,x1)
x
x0 x1
[1,] 0.4625532 0.1297134
[2,] 0.2658268 0.9822407
[3,] 0.5772108 0.8267184

The above typos are to prove I'm human.