Why the Shape of a Joint Support Matters
A joint density describes both where a pair of random variables can occur and how probability is distributed across that region. The shape of the region can sometimes settle an independence question before any detailed integration: if each variable can take values in a certain range, but some combinations of those values are impossible, the variables cannot be independent.
This is a useful diagnostic, not a complete test. A rectangular support is compatible with independence, but it does not guarantee it. And a density formula should not be checked only at points where it is positive while ignoring where the joint density is zero. The support and the density values both matter.
For a discrete pair \((X,Y)\), the joint support consists of the pairs \((x,y)\) with \(P(X=x,Y=y)>0\). For a continuous pair with joint density \(f_{X,Y}\), the support is often visualized as the closure of the region where the density is positive, with care taken that changes on sets of area zero do not affect probabilities.
Independence means that the joint distribution factors into its marginal distributions. In particular, for any events \(A\) determined by \(X\) and \(B\) determined by \(Y\), \(P(X\in A,Y\in B)=P(X\in A)P(Y\in B)\). If both marginal probabilities are positive but the joint probability is zero, independence is impossible.
The Support Test for Independence
If \(X\) and \(Y\) are independent, every value supported by \(X\) must be able to combine with every value supported by \(Y\), in the support sense. Therefore, the joint support must match the Cartesian product of the two marginal supports. For example, if each variable can take values throughout \([0,1]\), independence requires joint support throughout the square \([0,1]\times[0,1]\); a triangular joint support cannot satisfy that requirement.
The reason is the product rule for independent events. Suppose \(x\) belongs to the support of \(X\) and \(y\) belongs to the support of \(Y\). Every sufficiently small neighborhood \(U\) of \(x\) has \(P(X\in U)>0\), and every sufficiently small neighborhood \(V\) of \(y\) has \(P(Y\in V)>0\). Independence gives
Thus every neighborhood of \((x,y)\) has positive joint probability, so \((x,y)\) belongs to the joint support. Conversely, if \(x\) is outside the support of \(X\), some neighborhood \(U\) of \(x\) has zero marginal probability. Then every joint neighborhood contained in \(U\times V\) has probability zero, so no such pair belongs to the joint support. The same reasoning applies when \(y\) is outside the support of \(Y\). These two directions give the product-support result.
For a continuous model, an especially direct version of the test is to find intervals \(A\) and \(B\) for which each marginal assigns positive probability, while the joint support makes \(A\times B\) impossible. This establishes dependence without needing to calculate the full joint distribution.
Worked Example: A Uniform Distribution on a Triangle
Suppose a hypothetical pair \((X,Y)\) has joint density \(f_{X,Y}(x,y)=2\) when \(0<x<1\) and \(0<y<1-x\), and density zero elsewhere. The model is uniform on that triangular region. Determine whether \(X\) and \(Y\) are independent.
Method: check marginal probabilities against an impossible joint event. The triangle has area \(1/2\), so the constant density \(2\) integrates to \(2(1/2)=1\); the stated model is normalized.
For \(0<x<1\), integrate the joint density over the allowed values of \(y\):
By symmetry, \(f_Y(y)=2(1-y)\) for \(0<y<1\). Therefore,
and the same calculation gives \(P(Y>3/4)=1/16\). Both events have positive marginal probability. But if both occur, then \(X+Y>3/2\), which is incompatible with the support condition \(X+Y<1\). Hence,
The variables are not independent. The triangular support alone already reveals the missing combinations; the event calculation makes the contradiction explicit. A common mistake is to treat the marginal probabilities as if their product gave the joint probability automatically. That multiplication is valid only under independence, which is what must be tested.
Other Nonrectangular Supports
The same reasoning works for triangles that slope in the other direction, curved regions, or disconnected regions. What matters is not the name of the shape but whether it contains combinations that the marginal supports allow. The next example uses a triangle above a diagonal rather than below one.
Worked Example: A Diagonal Boundary Rules Out Independence
Suppose a hypothetical pair \((X,Y)\) has density \(2\) on \(0<y<x<1\) and zero elsewhere. This is a uniform distribution on a triangle of area \(1/2\). Determine whether \(X\) and \(Y\) are independent.
Method: find two positive-probability marginal events whose combination is excluded. For \(0<x<1\), the allowed values of \(y\) run from \(0\) to \(x\), so
For \(0<y<1\), the allowed values of \(x\) run from \(y\) to \(1\), so
Now calculate each marginal event probability:
The events cannot occur together: \(X<1/3\) and \(Y>2/3\) imply \(y>x\), while the support requires \(y<x\). Thus their joint probability is zero, whereas independence would require it to be \(1/9\cdot1/9=1/81\). The variables are dependent.
The trap here is to focus only on the fact that both variables range between 0 and 1. Their marginal ranges do not make every pair in the square possible. The diagonal boundary removes some combinations, which is the relevant information.
A support need not be connected to be nonrectangular. For instance, a joint distribution may place probability on two separated squares while assigning none to the cross-combinations of their coordinate ranges. Each variable can then have two separated pieces in its marginal support, but the joint support may still omit combinations from the product of those marginal supports.
Worked Example: Two Separated Squares
Suppose a hypothetical pair \((X,Y)\) has density \(1/2\) on the union of the squares \(0<x<1,\ 0<y<1\) and \(2<x<3,\ 2<y<3\), and density zero elsewhere. Each square has area 1, so the total integral of the density is \((1/2)(1+1)=1\). Determine whether the variables are independent.
Method: compare marginally possible ranges with a missing cross-combination. The probability that \(X\) lies between 0 and 1 is the density times the area of the corresponding unit square:
Likewise \(P(2<Y<3)=1/2\). But the joint support contains no point with \(0<X<1\) and \(2<Y<3\), so \(P(0<X<1,\ 2<Y<3)=0\). The product of the marginal probabilities is \(1/2\cdot1/2=1/4\), not zero. Therefore \(X\) and \(Y\) are not independent.
A common error is to see that the density has the same value on both squares and infer independence from that constant value. The location of the squares matters: the missing cross-combination prevents the joint support from being the product of the marginal supports.
Support Is a Diagnostic, Not a Complete Independence Test
A genuine mismatch between joint support and the product of marginal supports is enough to rule out independence. The converse does not hold: a rectangular support does not establish independence. A joint density can be positive throughout a rectangle and still fail to factor into the marginal densities.
To verify independence for continuous random variables, the joint density must satisfy \(f_{X,Y}(x,y)=f_X(x)f_Y(y)\) almost everywhere over the plane. “Almost everywhere” means except possibly on a set of area zero. Checking only a few density values, or checking only where a formula is written while ignoring the density’s zero region, is not a valid full test. For discrete variables, the corresponding check is \(P(X=x,Y=y)=P(X=x)P(Y=y)\) for every pair of values.
This explains why support and density checks complement one another. Support gives a fast necessary condition: independence requires the product support. If that condition passes, factorization still needs to be checked. If it fails, no calculation of conditional distributions or density values on the remaining region can restore independence.
Common Mistakes
- Calling every stated density region the support. The relevant region is where the distribution genuinely places probability, not an arbitrary description that differs only on a probability-zero set.
- Assuming a rectangle proves independence. A rectangular support is necessary for independence when the marginal supports are intervals, but the density may still vary in a dependent way across that rectangle.
- Checking only positive density values. A factorization test applies to the joint density on the whole plane, including where it is zero, and must hold almost everywhere.
- Using a single point as a density test. A density’s value can be changed at a single point without changing its distribution. A mismatch at one point alone does not disprove factorization almost everywhere.
- Multiplying probabilities before establishing independence. The product of two marginal probabilities is a comparison value in a diagnostic, not automatically the joint probability.
Check Your Understanding
For each model, use the support or stated density to decide what can be concluded about independence. All models below are hypothetical.
- A pair has density \(2\) on \(0<x<y<1\) and zero elsewhere. Compare \(P(X>2/3)\), \(P(Y<1/3)\), and their joint probability. (Answer: Each marginal probability is \(1/9\), the joint probability is 0, so the variables are dependent.)
- A pair is uniform on \(0<x<2,\ 0<y<2,\ x+y<2\), with density \(1/2\) there. Compare the events \(X>1\) and \(Y>1\). (Answer: Each marginal probability is \(1/4\), their joint probability is 0, so the variables are dependent.)
- A pair has density \(1/2\) on each of the unit squares \(0<x<1,\ 0<y<1\) and \(2<x<3,\ 2<y<3\), and zero elsewhere. What are the marginal probabilities of \(0<X<1\) and \(2<Y<3\), and their joint probability? (Answer: The marginal probabilities are \(1/2\) and \(1/2\); their joint probability is 0, so the variables are dependent.)
- A pair has density \(2x\) on \(0<x<1,\ 0<y<1\) and zero elsewhere. Is support geometry alone enough to establish independence, and does the density factorize? (Answer: The rectangular support alone is not enough, but \(f_X(x)=2x\), \(f_Y(y)=1\), and \(f_{X,Y}(x,y)=f_X(x)f_Y(y)\), so the variables are independent.)
- A pair has density \(1+(2x-1)(2y-1)\) on \(0<x<1,\ 0<y<1\), and zero elsewhere. The marginal densities are both 1. Find \(P(X<1/2,Y<1/2)\) and decide independence. (Answer: The probability is \(1/4+1/16=5/16\), not \(P(X<1/2)P(Y<1/2)=1/4\); the variables are dependent despite the rectangular support.)