Why Supports Matter in an Independence Check
A proposed joint distribution may look complicated, but its support can provide a quick structural check. If \(X\) and \(Y\) are independent, their possible values cannot be restricted by a joint constraint that excludes part of the product of their marginal possibilities. For example, if each marginal can take values throughout an interval, independence cannot make the pair occupy only a triangular part of the corresponding rectangle.
A support check is a necessary test, not a complete test: a dependent distribution can have exactly the same support as the product of its marginal supports. The key is to use support to rule out a proposed factorization when the geometry makes it impossible, and then use the distribution itself when the supports agree.
The closure matters. A point may have probability zero itself and still belong to the support because every neighborhood around it has positive probability. For a continuous variable uniform on \((0,1)\), for instance, the support is \([0,1]\): every neighborhood of 0 or 1 intersects the interval in a set of positive probability. The endpoints are not positive-probability outcomes, but they are support points.
The Support of an Independent Pair
Write \(\operatorname{supp}(X)\) for the support of \(X\), and \(\operatorname{supp}(X,Y)\) for the support of the pair. The product of marginal supports is the set of all pairs whose first coordinate lies in the support of \(X\) and whose second coordinate lies in the support of \(Y\).
Here is why the equality holds. Suppose first that \(x\) and \(y\) belong to their respective marginal supports. Every neighborhood of \(x\) has positive probability under \(X\), and every neighborhood of \(y\) has positive probability under \(Y\). A sufficiently small neighborhood of \((x,y)\) contains a rectangle \(U\times V\), where \(U\) and \(V\) are neighborhoods of \(x\) and \(y\). Independence gives \(P(X\in U,Y\in V)=P(X\in U)P(Y\in V)>0\). Thus every neighborhood of \((x,y)\) has positive probability, so \((x,y)\) is in the joint support.
For the reverse inclusion, if \(x\) is outside the support of \(X\), some neighborhood \(U\) of \(x\) has \(P(X\in U)=0\). A small neighborhood of any pair \((x,y)\) lies inside a strip \(U\times\mathbb{R}\), which has joint probability zero. Therefore \((x,y)\) cannot be in the joint support. The same argument applies if \(y\) is outside the support of \(Y\). This proves that the joint support is contained in the product of the marginal supports, completing the justification.
The theorem applies to discrete and continuous distributions. In density problems, independence is also expressed by \(f_{X,Y}(x,y)=f_X(x)f_Y(y)\) almost everywhere. The support conclusion is about the distributions, so it does not depend on how a density is defined at individual points or on a boundary of probability zero.
Worked Example: A Triangular Density Cannot Factor
A hypothetical pair of measurements has joint density \(f_{X,Y}(x,y)=1\) when \(0<x<1\) and \(0<y<2x\), and density 0 elsewhere. Find the marginal supports and determine whether \(X\) and \(Y\) are independent. The method is to compute the marginals, take support closures, and compare the joint support with their product. This avoids treating the positive-density region as the support itself.
Check the model. Its total probability is \(\int_0^1\int_0^{2x}1\,dy\,dx=\int_0^1 2x\,dx=[x^2]_0^1=1\), so this is a valid density.
Find the marginal densities. For \(0<x<1\), \(f_X(x)=\int_0^{2x}1\,dy=2x\); it is zero outside that interval. For \(0<y<2\), the constraint \(y<2x\) requires \(y/2<x<1\), so \(f_Y(y)=\int_{y/2}^1 1\,dx=1-y/2\); it is zero outside that interval. Their integrals are \(\int_0^1 2x\,dx=1\) and \(\int_0^2(1-y/2)\,dy=[y-y^2/4]_0^2=1\).
Determine the supports. Every neighborhood of either endpoint of the \(X\)-interval has positive marginal probability, so \(\operatorname{supp}(X)=[0,1]\). Similarly, \(\operatorname{supp}(Y)=[0,2]\). Their product is the closed rectangle \([0,1]\times[0,2]\). The joint support is the closure of the region \(0<x<1,\ 0<y<2x\), namely \(\{(x,y):0\le x\le1,\ 0\le y\le2x\}\). In particular, the line \(y=2x\) belongs to the joint support.
Compare and conclude. The point \((1/4,3/2)\) belongs to the product of the marginal supports but lies outside the joint support, because \(3/2>2(1/4)=1/2\). At that point the product of the marginal density formulas is \(f_X(1/4)f_Y(3/2)=(1/2)(1/4)=1/8\), while the joint density is zero. More importantly than this single-point comparison, the joint density is zero throughout an open region with \(0<x<1\) and \(y>2x\), while both marginal densities are positive there. The joint density therefore cannot equal the product of the marginal densities almost everywhere. The variables are dependent. The support mismatch gives the same conclusion directly.
Using Supports in Discrete Models
For a finite-valued random variable, the support is just the set of values with positive probability. A joint table makes the product-support check especially direct: list the values appearing in each marginal, form every pair of those values, and compare that set with the pairs having positive joint probability. If a pair is missing from the joint support but appears in the product support, independence is impossible.
Worked Example: A Missing Pair in a Joint Probability Table
A hypothetical inspection records two binary indicators, \(X\) and \(Y\). The joint probabilities are \(P(X=0,Y=0)=1/2\), \(P(X=0,Y=1)=1/4\), \(P(X=1,Y=1)=1/4\), and \(P(X=1,Y=0)=0\). Find the marginal supports and test independence. The method is to construct the product support and check a missing pair against the product of its marginal probabilities; this avoids assuming that all binary pairs must occur jointly.
Check the total. The four probabilities sum to \(1/2+1/4+1/4+0=1\).
Compute the marginals. For \(X\), \(P(X=0)=1/2+1/4=3/4\) and \(P(X=1)=0+1/4=1/4\). For \(Y\), \(P(Y=0)=1/2+0=1/2\) and \(P(Y=1)=1/4+1/4=1/2\). Thus \(\operatorname{supp}(X)=\{0,1\}\) and \(\operatorname{supp}(Y)=\{0,1\}\), so the product support contains all four ordered pairs.
Test the missing pair. The pair \((1,0)\) has joint probability 0, but the product of its marginal probabilities is \(P(X=1)P(Y=0)=(1/4)(1/2)=1/8\). Independence would require these probabilities to be equal. They are not, so \(X\) and \(Y\) are dependent. The joint support omits \((1,0)\), while the product support includes it.
Matching Supports Does Not Prove Independence
The support theorem provides a one-way diagnostic. Independence forces the joint support to equal the product of the marginal supports, but the equality of those sets does not force independence. Support records where probability occurs, not how probability is distributed across that region. To establish independence, the joint distribution must still factor into its marginal distributions.
Worked Example: A Full Rectangular Support with Dependence
Let a hypothetical continuous pair have density \(g(x,y)=1+\tfrac12(2x-1)(2y-1)\) on \(0\le x\le1,\ 0\le y\le1\), and density zero elsewhere. Decide whether the variables are independent even though the joint support is rectangular. The method is to verify the density, calculate both marginals, and test factorization at interior points. This avoids treating matching supports as proof of independence.
Verify the density. On the square, each factor \(2x-1\) and \(2y-1\) lies between \(-1\) and \(1\). Their product lies between \(-1\) and \(1\), so \(g(x,y)\ge1-1/2=1/2\). Also, \(\int_0^1(2x-1)\,dx=[x^2-x]_0^1=0\). Consequently, the double integral of \(g\) over the square is \(1+\tfrac12(0)(0)=1\).
Calculate the marginals. For each \(x\in[0,1]\), \(g_X(x)=\int_0^1g(x,y)\,dy=1+\tfrac12(2x-1)\int_0^1(2y-1)\,dy=1\). Likewise, \(g_Y(y)=1\) for \(y\in[0,1]\). Each marginal is uniform on \([0,1]\), and each marginal support is \([0,1]\). Since the joint density is at least \(1/2\) throughout the square, every neighborhood of every point of the square has positive probability. Thus the joint support is \([0,1]\times[0,1]\), exactly the product of the marginal supports.
Test factorization. At the interior point \((1/4,3/4)\), the marginal density product is \(1\cdot1=1\), but \(g(1/4,3/4)=1+\tfrac12(-1/2)(1/2)=1-1/8=7/8\). The density formula is continuous, so this difference persists on a neighborhood of that point; it is not just a discrepancy at an isolated point. The density does not factor almost everywhere, and \(X\) and \(Y\) are dependent despite having a product support.
Common Support Traps
The distinction between a positive-density region and a support is a frequent source of wrong conclusions. In the triangular example, one might incorrectly report the marginal supports as \((0,1)\) and \((0,2)\), and the joint support as the strictly bounded triangle. Those are the respective regions where the displayed densities are positive, not the supports under the neighborhood definition. The correct marginal supports are \([0,1]\) and \([0,2]\), and the joint support is the closed triangle \(0\le x\le1,\ 0\le y\le2x\). Boundary points belong because every neighborhood of each boundary point intersects the positive-probability region.
A second trap is to conclude “independent” merely because the joint support has the shape of the product of the marginal supports. The final example rules that out: its support is the full square, but its joint density does not equal the product of its marginal densities. A support comparison can disprove independence when the supports conflict; it cannot replace a factorization check when they agree.
Check Your Understanding
For each problem, use support closures where appropriate. The answers follow each question so you can check your work.
- A discrete pair has positive probabilities at \((0,0),(0,1),(1,0)\), and zero probability at \((1,1)\). Both marginals assign positive probability to 0 and 1. Can the variables be independent? (Answer: No; the product of the marginal supports includes \((1,1)\), but the joint support does not.)
- A joint density is constant on \(0<x<1,\ 0<y<x\), and zero elsewhere. Find the constant and state the marginal supports and joint support. (Answer: The constant is \(2\); the marginal supports are \([0,1]\) for both variables, and the joint support is \(\{(x,y):0\le y\le x\le1\}\). Independence is ruled out.)
- Suppose \(X\) and \(Y\) are independent, with supports \([-2,3]\) and \([1,4]\), respectively. What must the joint support be? (Answer: \([-2,3]\times[1,4]\).)
- On the unit square, let \(h(x,y)=1+\tfrac14(2x-1)(2y-1)\), with density zero outside. Find the marginal densities and determine whether support equality establishes independence. (Answer: Both marginal densities are 1 on \([0,1]\); the joint support equals the product support, but the variables are dependent, since \(h(1/4,1/4)=1+1/16=17/16\ne1\).)
- A proposed joint density is zero outside the unit square and positive throughout its interior. Each marginal has support \([0,1]\). Is independence established by these supports, or what further check is needed? (Answer: Independence is not established; check whether the joint density equals the product of the two marginal densities almost everywhere.)