Independence From Joint-Marginal Factorization

Why Joint-Marginal Factorization Matters

When two random variables are independent, learning the value of one does not change the probability distribution of the other. A practical way to assess this property is to compare their joint distribution with the distributions of the variables separately. For discrete variables, compare joint probability masses with products of marginal probabilities. For continuous variables with a joint density, compare the joint density with the product of the marginal densities.

The order of work matters: first find both marginals from the joint model, then compare the joint model with their product. A plausible-looking table or density is not enough. Independence requires the factorization to hold throughout the possible values, or almost everywhere for densities.

Definition: Suppose $X$ and $Y$ have a joint probability mass function $p_{X,Y}(x,y)$ and marginal pmfs $p_X(x)$ and $p_Y(y)$. They are independent if $$ p_{X,Y}(x,y)=p_X(x)p_Y(y) $$ for every pair $(x,y)$. If they have a joint probability density function $f_{X,Y}(x,y)$ and marginal densities $f_X(x)$ and $f_Y(y)$, they are independent if $$ f_{X,Y}(x,y)=f_X(x)f_Y(y) $$ for almost every $(x,y)$. For a density, “almost every” allows differences on sets of area zero, which do not affect probabilities.

In the discrete case, the marginal pmfs are obtained by summing the joint pmf over the other variable: $p_X(x)=\sum_y p_{X,Y}(x,y)$ and $p_Y(y)=\sum_x p_{X,Y}(x,y)$. In the continuous case, the marginal densities are obtained by integrating: $f_X(x)=\int_{-\infty}^{\infty}f_{X,Y}(x,y)\,dy$ and $f_Y(y)=\int_{-\infty}^{\infty}f_{X,Y}(x,y)\,dx$. The integrals can be restricted to the joint density’s support. In either setting, the marginal distributions must be derived from the given joint model; they are not chosen to make factorization work.

What Factorization Tells Us

The product on the right side of the criterion is the probability mass or density that would result from combining the two marginal distributions independently. One short justification shows how this product behaves for discrete variables. If the joint pmf factors, then for any sets of values $A$ and $B$,

\begin{aligned} P(X\in A,\ Y\in B) &=\sum_{x\in A}\sum_{y\in B}p_{X,Y}(x,y)\\ &=\sum_{x\in A}\sum_{y\in B}p_X(x)p_Y(y)\\ &=\left(\sum_{x\in A}p_X(x)\right) \left(\sum_{y\in B}p_Y(y)\right)\\ &=P(X\in A)P(Y\in B). \end{aligned}

The third line follows because the summand is a product with one factor depending only on $x$ and the other only on $y$. The same idea applies to continuous variables: when the joint density factors, integrating over a rectangular region separates into the product of two one-dimensional integrals. Thus factorization makes probabilities for separate restrictions on $X$ and $Y$ multiply. The criterion is a way to assess independence from the joint model; it is not enough to check one selected pair of values or one event.

Worked Example: Checking Every Cell of a Discrete Joint Model

A hypothetical monitoring system records $X$, the number of alerts on a shift, and $Y$, whether the shift includes a scheduled equipment check. The joint pmf is given below. Determine whether $X$ and $Y$ are independent.

$X\backslash Y$	$Y=0$	$Y=1$
$X=0$	0.08	0.12
$X=1$	0.12	0.18
$X=2$	0.20	0.30

Method. Sum across rows and columns to find the marginal pmfs, then compare each joint entry with the corresponding product. This avoids the mistake of deciding independence from only one cell.

Step 1: Check the joint probabilities. Their sum is $0.08+0.12+0.12+0.18+0.20+0.30=1.00$, as required for a pmf.

Step 2: Find the marginal pmf of $X$. The row sums are $p_X(0)=0.08+0.12=0.20$, $p_X(1)=0.12+0.18=0.30$, and $p_X(2)=0.20+0.30=0.50$.

Step 3: Find the marginal pmf of $Y$. The column sums are $p_Y(0)=0.08+0.12+0.20=0.40$ and $p_Y(1)=0.12+0.18+0.30=0.60$.

Step 4: Compare all six cells. For $X=0$, the products are $0.20(0.40)=0.08$ and $0.20(0.60)=0.12$. For $X=1$, they are $0.30(0.40)=0.12$ and $0.30(0.60)=0.18$. For $X=2$, they are $0.50(0.40)=0.20$ and $0.50(0.60)=0.30$. Each product matches its joint entry.

Therefore $p_{X,Y}(x,y)=p_X(x)p_Y(y)$ at every possible pair, so $X$ and $Y$ are independent. As a check, $P(X=2,Y=1)=0.30$, and the marginal product is $P(X=2)P(Y=1)=0.50(0.60)=0.30$.

Applying the Density Criterion

For continuous variables, a density’s value at a single point is not itself a probability. The comparison is between the joint density and the product of the marginal densities over the model’s domain, with equality required almost everywhere. A practical check is to derive both marginals and then look for a region of positive area where the two expressions differ. Such a mismatch rules out independence.

Worked Example: A Joint Density That Factors

Suppose a hypothetical pair of measurements has joint density $f_{X,Y}(x,y)=(x+1)/4$ for $0\le x\le2$ and $0\le y\le1$, and zero elsewhere. Determine whether the variables are independent and find $P(X>1,\ Y<0.5)$.

Method. Integrate out each variable to obtain the marginals, verify that the joint density equals their product, and use the factorization to calculate the event probability. This avoids treating the joint density as though it were already a marginal density.

Step 1: Check normalization. The integral of the joint density is $\int_0^2\int_0^1 (x+1)/4\,dy\,dx =\int_0^2 (x+1)/4\,dx =[x^2/8+x/4]_0^2=1$.

Step 2: Find $f_X(x)$. For $0\le x\le2$, $f_X(x)=\int_0^1(x+1)/4\,dy=(x+1)/4$; it is zero otherwise.

Step 3: Find $f_Y(y)$. For $0\le y\le1$, $f_Y(y)=\int_0^2(x+1)/4\,dx=[x^2/8+x/4]_0^2=1$; it is zero otherwise.

Step 4: Test factorization. On the rectangle $0\le x\le2,\ 0\le y\le1$, $f_X(x)f_Y(y)=((x+1)/4)(1)=(x+1)/4=f_{X,Y}(x,y)$. Outside the rectangle, the joint density and the product of the marginal densities are both zero. Thus the variables are independent.

Step 5: Calculate the requested probability. Independence gives $P(X>1,Y<0.5)=P(X>1)P(Y<0.5)$. The two marginal probabilities are $P(X>1)=\int_1^2(x+1)/4\,dx=[x^2/8+x/4]_1^2=1-(1/8+1/4)=5/8$ and $P(Y<0.5)=\int_0^{0.5}1\,dy=1/2$. Their product is $(5/8)(1/2)=5/16=0.3125$.

Check. Direct integration gives $\int_1^2\int_0^{0.5}(x+1)/4\,dy\,dx =(1/2)\int_1^2(x+1)/4\,dx=(1/2)(5/8)=5/16$, matching the result.

Worked Example: Uniform Marginals Do Not Guarantee Independence

Let the joint density be $f_{X,Y}(x,y)=1+\tfrac12(2x-1)(2y-1)$ for $0\le x\le1,\ 0\le y\le1$, and zero elsewhere. Determine whether $X$ and $Y$ are independent, and calculate $P(X>0.5,\ Y>0.5)$.

Method. Find both marginal densities, compare their product with the joint density, and calculate an event probability as a check. The common mistake here is to assume that uniform marginals alone establish independence.

Step 1: Check that the function is a valid density. Since each factor $2x-1$ and $2y-1$ lies between $-1$ and $1$, the density is at least $1-\tfrac12=0.5$, so it is nonnegative. Also, $\int_0^1(2x-1)\,dx=[x^2-x]_0^1=0$, and the corresponding integral over $y$ is also zero. Therefore the integral of the constant term over the unit square is $1$, while the extra term integrates to zero. The total integral is $1$.

Step 2: Find the marginals. For $0\le x\le1$, $f_X(x)=\int_0^1[1+\tfrac12(2x-1)(2y-1)]\,dy=1$, since $\int_0^1(2y-1)\,dy=0$. Likewise $f_Y(y)=1$ for $0\le y\le1$. Both marginals are uniform on $[0,1]$.

Step 3: Compare the joint density with the marginal product. The product is $f_X(x)f_Y(y)=1$ on the unit square. In contrast, for $x>0.5$ and $y>0.5$, both $2x-1$ and $2y-1$ are positive, so $f_{X,Y}(x,y)=1+\tfrac12(2x-1)(2y-1)>1$. This difference occurs on a region of positive area, not merely at an isolated point. The joint density therefore does not equal the marginal product almost everywhere, and the variables are not independent.

Step 4: Check with an event probability. The product of the marginal probabilities for the upper halves is $P(X>0.5)P(Y>0.5)=(0.5)(0.5)=0.25$. The joint probability is

\begin{aligned} P(X>0.5,Y>0.5) &=\int_{0.5}^1\int_{0.5}^1 \left[1+\tfrac12(2x-1)(2y-1)\right]\,dy\,dx\\ &=(0.5)(0.5)+\tfrac12 \left(\int_{0.5}^1(2x-1)\,dx\right) \left(\int_{0.5}^1(2y-1)\,dy\right)\\ &=0.25+\tfrac12(0.25)(0.25)\\ &=0.28125. \end{aligned}

The joint probability $0.28125$ differs from the product $0.25$, consistent with the failed density factorization.

Common Errors and a Reliable Check

Factorization is a comparison of complete distributions, not a shortcut based on a visual impression or a single summary. Keep the model type clear: use pmfs for discrete pairs and densities for continuous pairs. For a continuous model, compare densities almost everywhere; do not infer independence from equality at a point or from equality of a few integrals.

Using the joint entry as a marginal. A marginal requires summing or integrating over the other variable first.
Checking only one cell or point. A match at one pair does not establish factorization everywhere. In a discrete model, check every possible pair; with densities, a mismatch on a positive-area region is enough to rule out independence.
Assuming separate marginals settle the question. The third example has uniform marginals, yet its joint density does not factor.
Ignoring the domain. A density is zero outside its support. Include that domain when comparing the joint density and the product of marginals.

A disciplined calculation is: derive the two marginals, form their product, compare with the joint pmf or density on its possible values, and state the conclusion. This approach is aligned with the SOA Exam P (Probability) syllabus and helps prevent conclusions based on incomplete checks.

Check Your Understanding

For each problem, find the relevant marginal distributions before deciding whether the pair is independent.

A discrete pair has joint probabilities $p(0,0)=0.15$, $p(0,1)=0.35$, $p(1,0)=0.15$, and $p(1,1)=0.35$, with all other probabilities zero. Are $X$ and $Y$ independent? (Answer: Yes; $p_X(0)=p_X(1)=0.50$, $p_Y(0)=0.30$, $p_Y(1)=0.70$, and each joint probability is the corresponding product.)
A joint density is $f_{X,Y}(x,y)=2x$ for $0\le x\le1,\ 0\le y\le1$, and zero elsewhere. Are the variables independent, and what is $P(X>0.5,Y<0.25)$? (Answer: Yes; $f_X(x)=2x$, $f_Y(y)=1$, and the probability is $(3/4)(1/4)=0.1875$.)
A discrete pair has $p(0,0)=0.30$, $p(0,1)=0.20$, $p(1,0)=0.10$, and $p(1,1)=0.40$. Are the variables independent? (Answer: No; $p_X(0)=p_X(1)=0.50$, $p_Y(0)=0.40$, $p_Y(1)=0.60$, but $p(0,0)=0.30\ne(0.50)(0.40)=0.20$.)
On the unit square, let $f_{X,Y}(x,y)=1+\tfrac14(2x-1)(2y-1)$, and let the density be zero elsewhere. Are the variables independent, and what is $P(X>0.5,Y>0.5)$? (Answer: No; both marginals are uniform, but the joint probability is $0.25+\tfrac14(0.25)(0.25)=0.265625$, not $0.25$.)

\(X\backslash Y\)	\(Y=0\)	\(Y=1\)
\(X=0\)	0.08	0.12
\(X=1\)	0.12	0.18
\(X=2\)	0.20	0.30

What You'll Learn