Why These Problems Require a Method Choice
A bivariate normal problem may ask about one variable after the other is fixed, or about two threshold conditions holding at once. These are different probability questions. Conditioning on a fixed value calls for a conditional normal distribution; a joint threshold event calls for a joint probability calculation. Some problems require both steps.
The main risk is choosing a formula before translating the words into events. In particular, \(P(Y>c\mid X=x)\) is not the same quantity as \(P(Y>c\mid X> x)\). The first conditions on a fixed value and uses a conditional distribution. The second conditions on an event and generally requires a joint probability divided by a marginal probability.
For joint threshold calculations, define \(Z_X=(X-\mu_X)/\sigma_X\) and \(Z_Y=(Y-\mu_Y)/\sigma_Y\). Both have standard normal marginal distributions, and their correlation remains \(\rho\). Write \(\Phi\) for the standard normal cumulative distribution function and \(\Phi_2(a,b;\rho)=P(Z_X\leq a,\ Z_Y\leq b)\) for the standard bivariate normal lower-tail CDF.
Choosing Between Conditioning and a Joint CDF
| Question structure | Useful method | Point to check |
|---|---|---|
| \(P(Y>c\mid X=x)\) | Find the conditional normal distribution, then standardize \(c\). | Use the conditional standard deviation in the z-score. |
| \(P(X\leq a,\ Y\leq b)\) | Standardize both thresholds and evaluate \(\Phi_2\). | The CDF is a lower-tail probability. |
| \(P(X>a,\ Y>b)\) | Use complements and the joint lower-tail CDF. | Do not replace a joint upper tail with a single marginal tail. |
| \(P(Y>b\mid X>a)\) | Calculate the joint event and divide by \(P(X>a)\). | The conditioning event is an interval, not a fixed value. |
The bivariate normal conditional formula is especially useful when one coordinate is specified at a value. It does not by itself solve a problem conditioned on a range such as \(X>a\); for that, the probability distribution of \(X\) over the whole range contributes to the answer. A joint CDF or an equivalent integration is then needed.
Why the Conditional Formula Holds
The formula follows from standardizing the pair and separating the part of \(Z_Y\) explained by \(Z_X\) from the residual part. Let \(U=(X-\mu_X)/\sigma_X\) and \(V=(Y-\mu_Y)/\sigma_Y\), so both are standard normal with correlation \(\rho\). Define \(R=V-\rho U\). Then
Because \((U,R)\) is jointly normal, zero covariance implies independence. Thus, given \(U=u\), \(V=\rho u+R\) is normal with mean \(\rho u\) and variance \(1-\rho^2\). Substituting \(u=(x-\mu_X)/\sigma_X\) and converting back from \(V\) to \(Y\) gives the stated conditional mean and variance. This reasoning depends on joint normality; zero correlation alone does not give this conditional distribution for an arbitrary pair.
Worked Problems
Worked Example: A Fixed-Value Condition and an Upper Tail
A hypothetical monitoring system records two jointly normal readings \(X\) and \(Y\), in units appropriate to their respective sensors. Suppose \(\mu_X=100\), \(\sigma_X=12\), \(\mu_Y=40\), \(\sigma_Y=8\), and \(\rho=0.5\). Find \(P(Y>46\mid X=112)\). State the conditional mean and standard deviation before calculating the probability.
Method. Since \(X\) is fixed at 112, use the conditional distribution of \(Y\) given \(X=112\), rather than a joint-CDF calculation.
The conditional mean is
The conditional variance is
so the conditional standard deviation is \(\sqrt{48}=4\sqrt{3}\approx 6.9282\). Standardize the threshold:
Therefore,
The answer is approximately 0.3864. As a check, the threshold 46 is above the conditional mean 44, so the probability should be below 0.5. The common mistake is to use the conditional variance 48 as the standard deviation in the z-score; the correct standard deviation is \(\sqrt{48}\).
Worked Example: A Joint Event and a Conditional Probability
Let \(X\) and \(Y\) be jointly normal with means 70 and 20, positive standard deviations, and correlation \(\rho=-0.5\). Find \(P(Y>20\mid X>70)\). The scales of \(X\) and \(Y\) do not need to be specified because both cutoffs are their respective means.
Method. Standardize both variables. The event becomes \(Z_X>0,\ Z_Y>0\), and the requested conditional probability is the joint probability divided by \(P(Z_X>0)\).
For standard bivariate normal variables with correlation \(\rho\), the probability that both are positive is
With \(\rho=-0.5\), \(\arcsin(-0.5)=-\pi/6\). Hence
Since \(Z_X\) is standard normal, \(P(Z_X>0)=0.5\). It follows that
The answer is 0.3333. Negative correlation makes simultaneous values above both means less likely than under independence, and the conditional probability is consequently below 0.5. A common error is to report the joint probability \(1/6\) as the conditional probability; division by the conditioning probability \(0.5\) is required.
Worked Example: Nonzero Thresholds and a Joint Upper Tail
Suppose \(X\) and \(Y\) are jointly normal with \(\mu_X=500\), \(\sigma_X=40\), \(\mu_Y=30\), \(\sigma_Y=5\), and \(\rho=0.5\). Find the probability that \(X>540\) and \(Y>30\). Use a bivariate normal CDF value \(\Phi_2(1,0;0.5)\approx 0.4688\).
Method. Standardize each threshold. Then use the complement of a lower-tail event; the given CDF is not itself the probability of the requested upper-tail event.
Thus the requested probability is \(P(Z_X>1,\ Z_Y>0)\). Begin with \(P(Z_Y>0)=0.5\). From this subtract the part of that event with \(Z_X\leq1\):
The subtracted probability is the difference between the lower-tail probability for \(Z_X\leq1\) and the part where \(Z_Y\leq0\):
Therefore,
The answer is approximately 0.1275. As a check, the event is contained in \(Z_X>1\), whose probability is about 0.1587, so 0.1275 is feasible. Treating \(\Phi_2(1,0;0.5)=0.4688\) as the answer would be a tail-direction error: it is a lower-tail probability, not the upper-right probability requested.
Exam Checks That Prevent Rework
Before evaluating a probability, write the event with inequalities and identify whether the condition is a fixed value or a range. Then check each threshold’s direction after standardization. These small steps prevent errors that can otherwise survive a plausible-looking numerical answer.
- Use the right scale. Standardize \(X\) and \(Y\) separately using their own means and standard deviations. A threshold in the units of \(X\) cannot be used directly as a threshold for \(Y\).
- Keep variance and standard deviation distinct. The conditional variance is \(\sigma_Y^2(1-\rho^2)\); the z-score denominator is \(\sigma_Y\sqrt{1-\rho^2}\).
- Respect the CDF convention. \(\Phi_2(a,b;\rho)\) means both variables are at or below their thresholds. Use complements or inclusion-exclusion for upper tails.
- Check the conditioning event. A fixed-value condition uses the conditional normal formula. Conditioning on \(X>a\) instead requires \(P(X>a,Y>b)/P(X>a)\), not the fixed-value formula with \(x=a\).
- Check bounds. A joint probability cannot exceed either marginal probability, and a conditional probability must lie between 0 and 1.
Check Your Understanding
For each problem, identify the event structure before calculating. Give probabilities to four decimal places.
- Let \(X,Y\) be jointly normal with \(\mu_X=50,\ \sigma_X=8,\ \mu_Y=30,\ \sigma_Y=6,\ \rho=-0.5\). Find \(P(Y>36\mid X=42)\). (Answer: 0.2819)
- For a jointly normal pair with correlation \(-0.5\), find the probability that both variables exceed their respective means. (Answer: 0.1667)
- Let \(Z_X,Z_Y\) be standard bivariate normal with correlation \(0.4\). Find \(P(Z_X\leq0,\ Z_Y>0)\). (Answer: 0.1845)
- For standard bivariate normal variables with correlation \(0.6\), find \(P(Z_Y>0\mid Z_X>0)\). (Answer: 0.7048)
- Let \(X,Y\) be jointly normal with means 10 and 100, standard deviations 2 and 20, and correlation \(0.5\). Given \(\Phi_2(1,0;0.5)=0.4688\), find \(P(X>12,\ Y>100)\). (Answer: 0.1275)