Help with understanding correlation ...

Karabiner

TS Contributor
It is statistically significant (p < 0.001). That means, based on the sample data,
you assume that in the population from which the sample was drawn the
correlation is different from (larger than) r=0.00000

With kind regards

Karabiner

Drilon

New Member
It is statistically significant (p < 0.001). That means, based on the sample data,
you assume that in the population from which the sample was drawn the
correlation is different from (larger than) r=0.00000

With kind regards

Karabiner

Thank you very much!

So I can say that correlation exists, higher the X, higher Y gets. But, what about .28? is it to low?

Karabiner

TS Contributor
This is a sample coefficient. It is not the population coefficient. The sample coefficient is larger
or smaller than the true population coefficient, due to sampling error. But in the present case,
the sampling error is small, because the sample size is large. So maybe the true coefficient is
near 0.28. Whether this large or small, depends on the field of study. In psychology, sociology,
medicine, biology r=0.3 is often considered medium sized. But if you had reason to assume
a coefficient of size r=0.8 or so, then 0.28 would obviously be small.

With kind regards

Karabiner

Last edited:

priyac1987

New Member
The correlation is too low. It should be at least 0.5 to proceed with the regression test

Karabiner

TS Contributor
The correlation is too low. It should be at least 0.5 to proceed with the regression test
This makes not much sense for me, I'm afraid. Could you elaborate a bit?

With kind regards

Karabiner

Karabiner

TS Contributor
Well, first of all, 0.5 refers to R² in that document, not to r. And a general claim that only R² values above 0.5
(i.e. r > 0.7!) are worth considering, would plainly be silly. It contradicts most of the empirical work done in the
social and life sciences. The examples in that document make that claim even more dubious. For example, they
discuss an R² of 0.7 (i.e. r = 0.83!) when explaining happiness; given that happiness measures have a reliability of
0.8 at best, that would mean that they want to explain nearly the complete non-error variance by k=1 other
variable. Such goals are more than...heroic.

With kind regards

Karabiner

priyac1987

New Member
It should be more than 0.5 to be considered effective. However, it’s not always a compulsion as R square just states the proportion of variance.
So your conclusion about generalisation is incorrect. Moreover i have provided references in that article in case you doubt my approach.

Last edited:

Miner

TS Contributor
r is an indicator of how strong the correlation is. Specifically, how strong the signal is relative to the noise. The "cutoff" for how good an r value needs to be depends on your needs and the purpose of your study. I work in industrial statistics, and an r of 0.5 would be of little practical use to me because there is too much noise. However, if I were in another field, and wanted to better understand the underlying factors, an r of 0.3 might be acceptable. It depends on your particular needs.

Karabiner

TS Contributor
r is an indicator of how strong the correlation is. Specifically, how strong the signal is relative to the noise. The "cutoff" for how good an r value needs to be depends on your needs and the purpose of your study. I work in industrial statistics, and an r of 0.5 would be of little practical use to me because there is too much noise. However, if I were in another field, and wanted to better understand the underlying factors, an r of 0.3 might be acceptable. It depends on your particular needs.
Absolutely. And because of this, claiming that only R² > 0.5 is worth considering in any case and under any circumstances, would simply be silly.