Point to cov if not positive definite

[rikhuijzer]

I was trying to calculate the Cronbach alpha for my dataset but got a Covariance matrix must be positive definite. error. Assuming that something was wrong with my data, I gave up and went to R's psych package only to figure out that the calculation worked there.

I asked @storopoli, and it turned out that I should use cronbachalpha(cov(Matrix(...))) which is indeed in the documentation, but well, also easy to miss... This PR makes it harder to miss the fact that you should pass a covariance matrix.

[storopoli]

I can review the PR if no maintainer is available.

[storopoli]

Might be interesting to add some tests to test/reliability.jl.

[rikhuijzer]

Might be interesting to add some tests to test/reliability.jl.

I would say it's already mostly covered by

StatsBase.jl/test/reliability.jl

Lines 40 to 43 in b0ca0b0

	# testing when Matrix is not positive-definite
	cov_not_pos = [-1 1;
	-1 1]
	@test_throws ArgumentError cronbachalpha(cov_not_pos)

[rikhuijzer]

bump @nalimilan @andreasnoack.

Could one of you take a quick look and merge or give feedback? This PR is a non-breaking and easily reversible change which may help users a lot.

[rikhuijzer]

Thanks for fixing the typo. In abf4b88, I updated the indentation because there were 9 spaces instead of 8.

[nalimilan]

Thanks!