linear regression simpson's paradox

82 Views Asked by At

I need advice on this problem. It is related to Simpson's Paradox.

Consider three binary variables $X, Y, Z$ and all taking values in $\{0, 1\}.$

Consider the following inequalities:

$P(X = 1) > P(Y = 1)\tag 1$

$P(X = 1 \land Z = 1) < P(Y = 1 \land Z = 1)\tag 2$

$P(X = 1 \land Z = 0) < P(Y = 1 \land Z = 0) \tag 3$

Assumed all inequalities, (1) (2) and (3), are true, which of the following claims are true?:

a) If I perform a linear regression of X on Z, the $e^Z$ must be negative.

b) If I perform a linear regression of Y on Z, the $e^Z$ must be positive.

c) If I perform a linear regression of X − Y on Z. the $e^Z$ must be negative.

Thanks for your advice in advance.