As we know that the Spearman correlation is suitable measure under the assumption of monotonic relationship.

I would like to know that is there any standard method to figure out mathematical rather graphically (by assessing scatter diagrams)?

I am interested to identify variables that have non-monotonic relationship.

I also read the following How to test for a monotonic relationship between two variables, without assuming a specific functional model?


Your understanding of Spearman's rank-correlation measure seems wrong. There is no monotonicity assumption in applying it. In fact it was designed exactly for the purpose of measuring how monotonic the relationship is: if it is 1 (resp. -1), then a higher x value means a higher (resp. lower) y value.

The closer the value is at 0, the more you can say "the relationship (if any) is non-monotonic".

Now if you want to distinguish a weak non-monotonic relationship from a strong non-monotonic relationship, we need to get a little bit creative: One option is to compare the square of Spearman's rank correlation with the R-squared from a rank-regression with non-linear parts such as squares or splines: $$ \text{rank}(y_i) = \alpha + \beta_1 \text{rank}(x_i) + \beta_2 \text{rank}(x_i)^2 + \varepsilon_i $$ A low squared Spearman's rank correlation but high R-squared from such regression indicates a strong non-monotonic relationship.

A small example for illustration:

x <- seq(0, pi, by = 0.01)
y <- sin(x)
plot(y ~ x)

cor(rank(y), rank(x))^2 # almost 0
summary(lm(rank(y) ~ poly(rank(x), 2))) # Multiple R-squared:  0.9375

# More safe (works in almost any case):
summary(lm(rank(y) ~ ns(rank(x), df = 6))) # Multiple R-squared:  0.9986

So here, we would speak of a strong non-linear relationship. A quick look at the scatter plot: enter image description here

| cite | improve this answer | |

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.