Must the events occur in time order?

No. The rule factors a joint probability; either valid ordering can describe the same events.

Does dependence invalidate multiplication?

No. Dependence is handled through the conditional factor. Multiplying unconditional probabilities is the special independence case.

They are closely related but distinct. Equating the two product-rule factorizations and dividing by a positive marginal yields Bayes' rule for reversing a conditional probability.

AI Glossary / Graphs and symbolic AI / Intermediate

Product rule

Definition

The product rule in probability calculates the chance that two events both occur by multiplying the probability of one event by the conditional probability of the other given the first. In symbols, P(A and B) = P(A) × P(B given A), provided P(A) is positive. This entry concerns probability, rather than the product rule for differentiation.

What it means

The word “given” does essential work. If 40% of submitted files are images and 25% of those images need resizing, then 10% of all submitted files are images needing resizing. The second percentage refers to the image group, not to every submitted file.
The rule is valid for dependent events; independence is not required. If independence is justified, P(B given A) equals P(B), and the expression simplifies to P(A) × P(B). Without that justification, multiplying two overall percentages can produce the wrong answer. The rule therefore provides a basic way to combine uncertain conditions while respecting their relationship.

How it works

Start by defining both events within the same population and time frame. Identify a marginal probability for the first event and a conditional probability for the second within that first event. Multiply them to obtain the joint probability.
The order can be reversed: P(A and B) = P(B) × P(A given B), when the relevant conditioning probability is positive. These two factorizations describe the same joint event. Repeated application gives the chain rule for several events. In a discrete three-event case, P(A and B and C) = P(A) × P(B given A) × P(C given A and B). Simplifying the conditioning requires additional assumptions.

Three practical examples

An operations analyst can combine the proportion of uploaded image files with the proportion of images requiring a specific processing step.
A test designer can estimate the joint frequency of two observed software conditions using the appropriate conditional rate.
A probabilistic model can factor a joint distribution into conditional components, making dependencies explicit instead of treating every input as unrelated.

Think of it this way

Think of selecting a tray containing one group of items and then selecting a marked subset within that tray. The marked subset's share of the entire collection is the tray's share multiplied by the marked share inside it. The analogy is limited because events need not happen in a physical sequence, and one event does not have to cause the other.

Worked example

Consider an invented set of 1,000 uploaded files. Of these, 400 are images. Among the images, 100 require resizing. Let I mean “is an image” and R mean “requires resizing.” Then P(I) = 400/1,000 = 0.4 and P(R given I) = 100/400 = 0.25. Their joint probability is 0.4 × 0.25 = 0.1, matching 100/1,000.
Suppose 60 of the 600 non-image files also require resizing under this system's broader definition, such as adjusting document pages. Then P(R) = 160/1,000 = 0.16. Multiplying the marginals would give 0.4 × 0.16 = 0.064, which is not the observed joint probability. I and R are not independent in this dataset.
Reversing the factorization works: P(I given R) = 100/160 = 0.625, so 0.16 × 0.625 = 0.1. If 80% of the 100 image files requiring resizing also need a manual preview, the three-condition joint proportion is 0.4 × 0.25 × 0.8 = 0.08, or 80 files.

Illustrative case study

Fictional case: Birch Media predicts how many uploads will enter a special handling queue. A draft spreadsheet multiplies the overall image share by the overall resizing share. A reviewer notices that resizing rates differ between images and other files.
The team replaces the second factor with the resizing rate among images and checks the result against the underlying count table. It also labels every denominator in the report. This fictional example illustrates a common analytical correction; the numbers are invented, and no real company outcome is implied.

Common misconceptions and mistakes

Do not multiply marginal probabilities merely because a question contains the word “and.” That shortcut needs independence. Do not swap P(A given B) with P(B given A); these generally differ, even though their appropriately weighted products agree.
The joint event cannot be more probable than either of its constituent events. Use that as a basic reasonableness check. For elementary event conditioning, P(B given A) is undefined when P(A) = 0, although P(A and B) is then zero. Also, a product-rule calculation does not establish causality, and proportions from one observed dataset do not automatically become stable forecasts for another population.

Business use

The product rule is useful whenever a business estimate involves successive filters, nested groups, or combinations of conditions. It helps teams translate a percentage statement into an expected count without losing track of the relevant population.
Ask for the denominator behind every input: all files, all image files, or image files already needing resizing. If the goal is forecasting, investigate whether the future mix and conditional rates resemble those used in the calculation. Preserve a full count table when possible so that another analyst can reproduce the marginals, conditional rates, and joint totals rather than trusting a chain of unlabeled percentages.

Check your understanding

In the worked example, what is the probability that a file is not an image and requires resizing?

Answer: The non-image share is 0.6, and the resizing rate among non-images is 60/600 = 0.1. The product is 0.06, corresponding to 60 of the 1,000 files. Adding this to the image-and-resizing probability 0.1 gives P(R) = 0.16.

Related terms

Learn next

Learn Bayes' rule and the chain rule next. Reconstruct all four cells for image/non-image and resizing/no-resizing from the example. Check that they sum to 1, then compute the probability that a file is an image given that it does not require resizing.

Chain rule

Sources

Stuart Russell and Peter Norvig, AIMA Chapter 13 teaching slides, conditional probability, product rule and chain rule: https://aima.cs.berkeley.edu/slides-pdf/chapter13.pdf

Previous: Product of experts Next: Production (computer science)