Closeness measures
S-measure
Computing the s
-statistic
using CausalityTools
x, y = randn(3000), randn(3000)
measure = SMeasure(dx = 3, dy = 3)
s = s_measure(measure, x, y)
0.010407848955297654
The s
statistic is larger when there is stronger coupling and smaller when there is weaker coupling. To check whether s
is significant (i.e. large enough to claim directional dependence), we can use a SurrogateTest
, like here.
H-measure
Computing the h
-statistic
using CausalityTools
x, y = randn(3000), randn(3000)
measure = HMeasure(dx = 3, dy = 3)
h = h_measure(measure, x, y)
0.12810905237231351
M-measure
Computing the m
-statistic
using CausalityTools
x, y = randn(3000), randn(3000)
measure = MMeasure(dx = 3, dy = 3)
m = m_measure(measure, x, y)
-0.02056015744486946
L-measure
Computing the l
-statistic
using CausalityTools
x, y = randn(3000), randn(3000)
measure = LMeasure(dx = 3, dy = 3)
l = l_measure(measure, x, y)
-0.0039994386909024526
Joint distance distribution
Computing the Δ
-distribution
using CausalityTools
x, y = randn(3000), randn(3000)
measure = JointDistanceDistribution(D = 3, B = 5)
Δ = jdd(measure, x, y)
5-element Vector{Float64}:
0.00033644844839855433
1.1793497199155667e-5
8.010677342822699e-6
-6.67556445235252e-7
2.225188150784354e-7
The joint distance distribution measure indicates directional coupling between x
and y
if Δ
is skewed towards positive values. We can use a JointDistanceDistributionTest
to formally check this.
test = JointDistanceDistributionTest(measure)
independence(test, x, y)
`JointDistanceDistributionTest` independence test
----------------------------------------------------------------------------------
H₀: μ(Δ) = 0 (the input variables are independent)
H₁: μ(Δ) > 0 (there is directional dependence between the input variables)
----------------------------------------------------------------------------------
Hypothetical μ(Δ): 0.0
Observed μ(Δ): 7.116151706207518e-5
p-value: 0.06376811534851834
α = 0.05: ✖ Independence cannot be rejected
α = 0.01: ✖ Independence cannot be rejected
α = 0.001: ✖ Independence cannot be rejected
The p-value is fairly low, and depending on the significance level 1 - α
, we cannot reject the null hypothesis that Δ
is not skewed towards positive values, and hence we cannot reject that the variables are independent.