bkrosenblatt

[Tommaso Salvitti]

I have to calculate the correlation between two variables that are neither linear, nor normally distributed, nor monotonic. I would like to use "bkrosenblatt". Is it correct according to you???

[Nick Cox]

bkrosenblatt (from SSC, as you are asked to explain) is about testing for dependence, and not at all about measuring correlation.

Every once in a while someone comes up with a new measure of correlation (or dependence, or whatever) that supposedly generalizes beyond monotonic relationships. Usually initial interest and enthusiasm is tempered by scepticism and a flurry of counter-arguments and counter-examples.

https://www.statalist.org/forums/for...of-correlation is perhaps the most prominent recent example.

Relationships can be complicated. but that is one major reason for models. On the whole I'd take the line that modelling the relationship with some appropriate functional form is likely to be immensely more interesting and useful than putting a number on closely they are related in some extenuated sense.

To that end, why not

* post the data

* show a scatter plot

* tell us a bit about what you expect from disciplinary knowledge or even theory

???

[Tommaso Salvitti]

bkrosenblatt is designed to analyze the dependency between two variables and also works well when: The variables do not have a monotonic relationship: There is no need for one variable to constantly increase or decrease relative to the other. Captures more general relationships, including nonlinear and nonmonotonic ones. The variables do not follow a normal distribution: It does not require any assumptions about the distribution of the data. It is robust to non-Gaussian data, with skews or other particular characteristics.

[Tommaso Salvitti]

• 
  Use bkrosenblatt if: Variables exhibit complex or irregular relationships that do not fit into linear or monotonic models. You cannot assume normality of distributions (e.g., data is highly skewed, multimodal, or contains outliers).

[Nick Cox]

I don't know what #3 and #4 are quoting or (more importantly) how they add to your question. If you are using AI, that probably explains something about your posts.

I wrote the command for fun, but have never used it for real.

[Tommaso Salvitti]

for me it 's correct

[Nick Cox]

I can't see that you're answering my questions, but it's good if you feel positive.

[Tommaso Salvitti]

ok i give you and example and can you tell me with these data if bkrosenblatt is correct?

* Example generated by -dataex-. For more info, type help dataex
clear
input byte(EMOTIV ados_todtoddler_totale)
50  .
50  .
50  .
50  .
50  .
50  3
55  7
50  7
50  1
50 13
51  .
51  .
51  7
50  6
50  6
50  5
50  0
50 12
50  1
50  6
50  5
51  5
50  .
68 14
55  6
50 19
50  1
50 10
51  8
50  6
51 20
52  6
55  7
50  6
50 12
55  7
51  6
50  1
50  3
50  7
50 11
50  3
50 14
50  2
51  0
51  2
50  2
51  6
59 15
50  4
50  2
55 21
59  6
50  1
55  2
50 22
50  3
55 21
50  7
50  8
50 18
50  2
50  2
51 11
51  1
50  1
51 12
50  0
50  2
50  7
50  .
50  4
50  8
50  5
67 10
50  1
50  8
50  6
50  4
67 14
50  2
51  6
51  8
51  6
50  4
51  2
50  1
50  4
50  8
50  .
50  3
51 13
50  5
50  5
50  9
50 14
50  4
50 14
50  3
50  0
end

------------------ copy up to and including the previous line ------------------

[Tommaso Salvitti]

or which kind of correlation coeffcient could i use?

[Tommaso Salvitti]

i m waiting your answer Joseph Coveney
And Happy xmsa

[Tommaso Salvitti]

Test di kendall is it correct?

[Tommaso Salvitti]

ktau var1 var2 is it more appropriate for these kind of variable??

[Nick Cox]

You can edit a post within 1 hour of first posting. That would cut down on these one-liners.

If you don't think your relationship is even monotonic (#1) then none of the usual correlations can help (#9 #11 #12).

You've answered the first question in #2. Thanks for that. It would be foolish of me to try more advice without knowing what these variables are and why EMOTIV has such a peculiar distribution.

[Tommaso Salvitti]

Thanks a lottare.. and sorry

[Tommaso Salvitti]

Sorry the touch screen...thank you so much