Support and Normalization Checks for Conditional Laws

Why Conditional Laws Need Their Own Checks

A conditional probability mass function or density is built from a joint law and a marginal probability or density. A small error in the marginalization step can therefore make the conditional formula wrong even when the joint model looks plausible. Checking the support, denominator, nonnegativity, and total probability provides a direct way to catch such errors.

The checks depend on whether the variables are discrete or continuous. In either case, the key is to use the values that are actually possible under the joint distribution. A conditional law cannot put probability on an impossible value, and its total probability over its conditional support must equal one.

Definition: For discrete random variables $X$ and $Y$, if $P(Y=y)>0$, the conditional pmf is $$ p_{X\mid Y}(x\mid y)=\frac{P(X=x,Y=y)}{P(Y=y)}. $$ Its support consists of the values $x$ for which the joint probability $P(X=x,Y=y)$ is positive. For continuous random variables with joint density $f_{X,Y}$, if $f_Y(y)>0$, the conditional density is $$ f_{X\mid Y}(x\mid y)=\frac{f_{X,Y}(x,y)}{f_Y(y)}, \qquad f_Y(y)=\int_{-\infty}^{\infty} f_{X,Y}(x,y)\,dx. $$ The conditional density is zero outside the slice of the joint support at that $y$.

A Systematic Validity Check

A proposed conditional law should pass four checks. First, identify the conditional support: fix the conditioning value and determine which values of the other variable remain possible. Second, calculate the marginal denominator by summing or integrating the joint law over that same slice. Third, verify that the conditional probabilities or density values are nonnegative. Finally, sum or integrate the conditional law over its support and confirm that the result is one.

For a discrete conditional pmf, the normalization check is a sum over all possible $x$. For a continuous conditional density, it is an integral over the conditional support. A density's value need not be at most one; only its integral must equal one. Also, if $P(Y=y)=0$ in a discrete model, the ratio defining the conditional pmf is unavailable. In a continuous model, a conditional density obtained by a ratio is defined where $f_Y(y)>0$; values at conditioning points with zero marginal density cannot be obtained from that ratio.

Fix the conditioning value.
Use the joint support to determine exactly which values of the variable being conditioned on remain possible.

Find the marginal denominator.
Sum or integrate the joint pmf or density over the full slice allowed by the support.

Form the conditional law.
Divide the joint probability or density by the corresponding positive marginal probability or density.

Check normalization and sign.
Confirm nonnegativity and that the conditional probabilities sum to one or the conditional density integrates to one.

Discrete Conditional Laws

For a discrete joint pmf, the marginal probability of the conditioning value is the sum of the joint probabilities along the corresponding row or column. Which one is appropriate depends on which variable is being conditioned on. The conditional pmf then divides each joint cell in that row or column by the marginal total.

Worked Example: Checking a Conditional PMF From a Joint Table

A hypothetical quality study classifies a component by its inspection score $X\in\{0,1,2\}$ and whether it is sent for secondary review, $Y\in\{0,1\}$. The joint probabilities are:

$X$	$P(X=x,Y=0)$	$P(X=x,Y=1)$
0	0.10	0.15
1	0.20	0.25
2	0.10	0.20

Find and check the conditional pmf of $X$ given $Y=1$. The method is to sum the $Y=1$ column, divide each entry in that column by its sum, and check the resulting probabilities.

Joint-law check. The six entries sum to $0.10+0.15+0.20+0.25+0.10+0.20=1.00$, and all are nonnegative. Thus the table is a valid joint pmf.

Denominator and support. The conditioning probability is the column total $P(Y=1)=0.15+0.25+0.20=0.60$. All three $X$-values have positive joint probability in this column, so the conditional support is $\{0,1,2\}$.

Conditional probabilities. $P(X=0\mid Y=1)=0.15/0.60=0.25$; $P(X=1\mid Y=1)=0.25/0.60=5/12$; and $P(X=2\mid Y=1)=0.20/0.60=1/3$.

Normalization check. The sum is $0.25+5/12+1/3=3/12+5/12+4/12=1$. Each value is nonnegative, so this is a valid conditional pmf. A common mistake is to divide by the row total for $X=1$, which is $0.45$. That would use $P(X=1)$, not the required $P(Y=1)$, and would not produce the conditional distribution of $X$ given $Y=1$.

Continuous Conditional Densities and Their Slices

For a continuous joint density, marginalization means integrating across a slice of the support while holding the conditioning variable fixed. The limits of integration are not automatically the full range of the variable: they must come from the joint support. This is a frequent source of errors on triangular or curved regions.

Worked Example: A Triangular Support and a Conditional Density

Suppose $X$ and $Y$ have joint density $f_{X,Y}(x,y)=1$ for $0<x<1$ and $0<y<2(1-x)$, and zero elsewhere. Find $f_Y(y)$, then find and check $f_{X\mid Y}(x\mid 0.5)$. The method is to derive the horizontal slice at a fixed $y$, marginalize over that slice, and take the density ratio.

First check the joint density. For fixed $x$, the allowed length of the $y$-interval is $2(1-x)$, so the total integral is $\int_0^1 2(1-x)\,dx=[2x-x^2]_0^1=1$. The density is nonnegative and integrates to one.

Find the marginal density of $Y$. The support condition $y<2(1-x)$ rearranges to $x<1-y/2$. Thus $0<y<2$, and at a fixed such $y$, $x$ ranges from $0$ to $1-y/2$. Therefore $f_Y(y)=\int_0^{1-y/2}1\,dx=1-y/2$ for $0<y<2$, and zero otherwise. As a check, $\int_0^2(1-y/2)\,dy=[y-y^2/4]_0^2=1$.

Condition on $Y=0.5$. The slice has $0<x<1-0.5/2=0.75$. The marginal denominator is $f_Y(0.5)=1-0.5/2=0.75$. Hence $f_{X\mid Y}(x\mid 0.5)=1/0.75=4/3$ on $0<x<0.75$, and zero elsewhere.

Normalization check. The integral is $(4/3)(0.75)=1$, and the density is nonnegative. The common mistake is to use a conditional density of $2/3$ on this interval: its integral would be $(2/3)(0.75)=0.5$, not one. Dividing the joint density $1$ by the correct marginal $0.75$ gives $4/3$.

How Incorrect Marginalization Breaks a Conditional Law

A marginal density is calculated by integrating over the values that are possible at the fixed conditioning value. If the bounds ignore the joint support, the resulting denominator may be too large or may even describe a different function. Dividing by that incorrect value can produce a formula that is nonnegative but does not integrate to one on its proposed conditional support.

Worked Example: Detecting an Incorrect Slice

Suppose a hypothetical pair of measurements has joint density $f_{X,Y}(x,y)=2(x+y)$ on $0<x<y<1$, and zero elsewhere. Find the conditional density of $X$ given $Y=y$, and show why integrating over the wrong bounds gives an invalid result.

Check the joint density. The density is nonnegative on its support. Its total integral is $\int_0^1\int_0^y2(x+y)\,dx\,dy$. The inner integral is $[x^2+2yx]_0^y=y^2+2y^2=3y^2$, so the total is $\int_0^1 3y^2\,dy=1$. Thus it is a valid joint density.

Marginalize with the support bounds. At fixed $y$, the condition $0<x<y$ gives $0<x<y$, with $0<y<1$. Therefore $f_Y(y)=\int_0^y2(x+y)\,dx=3y^2$ for $0<y<1$. Its integral is $\int_0^1 3y^2\,dy=1$.

Form the conditional density. For $0<y<1$, the denominator $3y^2$ is positive, and $f_{X\mid Y}(x\mid y)=2(x+y)/(3y^2)$ for $0<x<y$, and zero otherwise. For every such $y$, its integral is $\int_0^y2(x+y)/(3y^2)\,dx=(3y^2)/(3y^2)=1$.

Check a specific value. At $y=0.5$, the support is $0<x<0.5$, and $f_{X\mid Y}(x\mid 0.5)=2(x+0.5)/0.75=(8/3)(x+0.5)$. Its integral is $(8/3)[x^2/2+0.5x]_0^{0.5}=(8/3)(0.125+0.25)=1$.

The incorrect calculation. If one mistakenly integrates from $x=0$ to $x=1$, the alleged marginal becomes $\int_0^1 2(x+y)\,dx=1+2y$. This includes values $x>y$, which are outside the joint support. At $y=0.5$, it gives denominator $2$, so the resulting proposed conditional density is $(x+0.5)$ on $0<x<0.5$. Its integral is $0.375$, not one. The failed normalization exposes the incorrect marginalization; the correct slice is $0<x<y$.

Why the Ratio Formula Works

The conditional density ratio is the continuous counterpart of dividing a joint probability by the probability of the conditioning event. Its normalization follows directly from marginalization, provided the conditioning value is in the region where its marginal density is positive.

Justification: Fix $y$ with $f_Y(y)>0$, and let $B$ be a set of possible $x$-values. The conditional density must assign probability $$ P(X\in B\mid Y=y)=\int_B f_{X\mid Y}(x\mid y)\,dx. $$ Using $f_{X\mid Y}(x\mid y)=f_{X,Y}(x,y)/f_Y(y)$, its integral over the full conditional support is $$ \int \frac{f_{X,Y}(x,y)}{f_Y(y)}\,dx =\frac{1}{f_Y(y)}\int f_{X,Y}(x,y)\,dx =\frac{f_Y(y)}{f_Y(y)}=1. $$ The final equality uses the definition of the marginal density. The same argument also explains why the integration limits must cover exactly the allowed slice of the joint support.

Exam Checks and Common Traps

When a proposed conditional law fails, locate the first failed check rather than trying to adjust the final formula by inspection. An incorrect support often causes an incorrect marginal; an incorrect marginal then causes the ratio to fail normalization. A density greater than one is not by itself an error, as the triangular example illustrates. Conversely, a nonnegative formula is not automatically a valid conditional density.

Using the wrong marginal. To find $X$ given $Y=y$, the denominator must be $P(Y=y)$ in the discrete case or $f_Y(y)$ in the continuous case.
Ignoring the joint support. Integration limits must follow from the support after fixing the conditioning value.
Checking only the formula's sign. Nonnegativity is necessary but not sufficient; total conditional probability must also equal one.
Requiring a density to be at most one. Densities can exceed one on short intervals. Their integrals, not their heights, are constrained to equal one.
Dividing by zero. The ratio requires a positive discrete conditioning probability or positive marginal density at the conditioning value.

Key takeaway: A reliable conditional-law check uses the joint support twice: first to determine the conditional support, and again to set the marginalization limits. Then verify the denominator is positive, the conditional law is nonnegative, and its total probability is one.

Check Your Understanding

For each problem, identify the relevant conditional support and verify the proposed or calculated law by checking its total probability.

A joint pmf has $P(X=0,Y=0)=0.1$, $P(X=1,Y=0)=0.2$, $P(X=0,Y=1)=0.3$, and $P(X=1,Y=1)=0.4$. Find $P(Y=0\mid X=1)$ and $P(Y=1\mid X=1)$, and check their sum. (Answer: The conditional probabilities are $1/3$ and $2/3$, respectively; their sum is $1$.)
A joint density is $2$ on $0<y<x<1$ and zero elsewhere. Find $f_Y(y)$ and the support and density of $X\mid Y=0.25$. Check its integral. (Answer: $f_Y(y)=2(1-y)$ for $0<y<1$; the conditional density is $4/3$ on $0.25<x<1$, and its integral is $1$.)
A proposed joint density has the form $c x$ on $0<x<y<1$, and zero elsewhere. Find $c$, then find $f_Y(y)$ and $f_{X\mid Y}(x\mid y)$. Verify normalization of both densities. (Answer: $c=6$; $f_Y(y)=3y^2$ on $0<y<1$; $f_{X\mid Y}(x\mid y)=2x/y^2$ on $0<x<y$, whose integral is $1$.)
A joint density is $1$ on $0<x<1,\ 0<y<2(1-x)$, and zero elsewhere. A candidate claims $f_{X\mid Y}(x\mid 0.5)=1$ for $0<x<0.75$. Find the correct conditional density and explain the candidate's normalization error. (Answer: The marginal is $f_Y(0.5)=0.75$, so the correct conditional density is $4/3$ on $0<x<0.75$; the candidate's integral is $0.75$, not $1$.)
A joint density is proportional to $x+y$ on $0<x<y<1$. Find its normalizing constant and the conditional density of $X$ given $Y=y$. A proposed marginal integrates $x+y$ from $0$ to $1$ at fixed $y$; explain why those bounds fail. (Answer: The normalizing constant is $2$; $f_{X\mid Y}(x\mid y)=2(x+y)/(3y^2)$ for $0<x<y<1$; integrating to $1$ includes impossible values $x>y$.)

What You'll Learn