Use the CDF when you need the probability that a value falls below a threshold or within a range. Use the PDF when you need to visualize the distribution’s shape, compare the relative likelihood of specific values, or describe how data is spread. The core distinction: the CDF gives you actual probabilities, while the PDF gives you probability density, which only becomes a probability when you calculate the area over an interval.
Why the PDF Doesn’t Give You a Probability Directly
The probability density function (PDF) of a normal distribution is the familiar bell curve. Its y-axis represents density, not probability. This trips up a lot of people. If you plug a value into the PDF formula and get 0.24, that doesn’t mean there’s a 24% chance of observing that exact value.
For any continuous distribution, the probability of landing on one exact value is zero. Think of it this way: between any two numbers on a number line, there are infinitely many possible values. The chance of hitting precisely 3.00000… rather than 3.00001 is effectively 1 divided by infinity. The area under a single point on a curve has no width, so it has no area, and probability equals area under the curve.
What the PDF height does tell you is relative likelihood. A point where the PDF is tall (near the mean) is more densely packed with probability than a point where the PDF is short (out in the tails). This makes the PDF useful for understanding the shape and center of a distribution, but not for answering questions like “what’s the probability of scoring below 85?”
What the CDF Actually Calculates
The cumulative distribution function answers a specific question: what is the probability that a value is less than or equal to x? It takes the entire area under the PDF curve from negative infinity up to x and returns that as a single number between 0 and 1. Mathematically, the CDF is the integral of the PDF. In practical terms, it’s a running total of probability as you move left to right across possible values.
Visually, the CDF looks like an S-shaped curve (sometimes called a sigmoid). It starts near 0 on the far left, climbs steeply through the middle of the distribution, and levels off near 1 on the far right. At the mean of a standard normal distribution, the CDF equals 0.5, because exactly half the probability lies below the mean.
For example, the CDF of the standard normal distribution at x = 1.53 is about 0.937. That means there’s a 93.7% probability of observing a value at or below 1.53 standard deviations above the mean. This is the kind of number you can use directly in a calculation or decision.
Use the CDF for Range and Threshold Problems
Any time your question involves “what’s the probability that…” followed by an inequality, you want the CDF. These are the most common scenarios:
- Below a cutoff: “What percentage of test scores fall below 600?” This is a direct CDF lookup. Convert 600 to a z-score, then read the CDF value.
- Above a cutoff: “What’s the probability of waiting more than 10 minutes?” Calculate 1 minus the CDF at 10, since the CDF gives you the area to the left and you want the area to the right.
- Between two values: “What fraction of parts measure between 4.8 cm and 5.2 cm?” Subtract the CDF at the lower bound from the CDF at the upper bound. So P(4.8 ≤ X ≤ 5.2) = CDF(5.2) minus CDF(4.8).
- Finding percentiles: “What score puts a student in the 90th percentile?” Use the inverse CDF (sometimes called the quantile function) to find which x-value corresponds to a cumulative probability of 0.90.
- P-values in hypothesis testing: P-values are tail areas under the curve, which are CDF calculations.
The subtraction trick for ranges is worth emphasizing, because it’s the single most common application in coursework and real analysis. The CDF at point b gives you all the probability to the left of b. The CDF at point a gives you all the probability to the left of a. Subtracting removes the portion you don’t want, leaving only the probability between a and b.
Use the PDF for Shape, Comparison, and Modeling
The PDF is the right tool when you’re not trying to calculate a specific probability but instead need to understand or communicate the distribution’s behavior:
- Visualizing how data is spread: Plotting the PDF shows you where values concentrate and how spread out the tails are. A narrow, tall bell curve means low variance. A wide, flat one means high variance.
- Comparing distributions: Overlaying two PDFs on the same graph instantly shows how two groups differ in center and spread. You can’t get this visual comparison from a CDF nearly as intuitively.
- Identifying the most likely region: The peak of the PDF marks the mode (which equals the mean in a normal distribution). The height at any point tells you how concentrated the data is there relative to other points.
- Likelihood calculations: In maximum likelihood estimation and Bayesian statistics, you evaluate the PDF at observed data points to assess how well a model fits. The density values aren’t probabilities, but their relative sizes tell you which parameter values make the observed data most plausible.
- Fitting curves to histograms: When you plot a histogram of your data with density on the y-axis (rather than raw counts), you can overlay the PDF to check whether a normal distribution fits your data well.
How They Connect Mathematically
The CDF is the integral of the PDF. If you know calculus, this means the CDF at any point x equals the total area under the PDF curve from negative infinity to x. Going the other direction, the PDF is the derivative of the CDF. Where the CDF is climbing steeply (near the mean), the PDF is tall. Where the CDF flattens out (in the tails), the PDF is near zero.
This relationship is why you’ll sometimes hear that the PDF “generates” the CDF. Every CDF value is built by adding up all the PDF’s density from the far left to that point. In software, functions like dnorm in R return PDF values (density), while pnorm returns CDF values (cumulative probability). In Python’s SciPy, the equivalent methods are .pdf() and .cdf() on a distribution object. If you use a TI calculator, normalcdf computes the area between two bounds (a CDF operation), while normalpdf returns the density height at a point.
A Quick Decision Rule
If your answer should be a probability (a number between 0 and 1 that you could report as a percentage), use the CDF. If your goal is to describe, visualize, or compare the shape of a distribution, use the PDF. In a stats course, most homework problems asking “find the probability that…” are CDF problems. The PDF matters when you’re plotting, fitting models, or explaining why a distribution looks the way it does.
One common source of confusion: z-tables printed in textbooks are CDF tables. Each entry is the cumulative area to the left of a z-score. When you look up z = 1.53 and find 0.937, you’ve used the CDF. The bell curve drawn at the top of those tables is the PDF, serving as a visual reference to show which area you’re calculating.