interpret the probability of an outcome of a continuous random variable as an area under its density curve, and compute probabilities for uniform and exponential distributions.
Instruction
In Statistics A, random variables took separate values: 0, 1, 2 rainy days. Many quantities are continuous: a waiting time, a temperature, an amount of rain. A continuous random variable can take any value in an interval, so we cannot list its values with probabilities. Instead we use a density curve.
Density curves (PS.4). A density curve lies on or above the horizontal axis and has a total area of exactly 1 underneath it. The probability that X falls between a and b is the area under the curve between a and b. Two consequences follow. First, P(X = a) = 0 for any single value, because a line has no area, so P(X < a) and P(X ≤ a) are equal. Second, the height of the curve is not a probability. It is a density: probability per unit of x.
Think of a density curve as the limit of a relative-frequency histogram. As you collect more data and narrow the bins, a histogram scaled so its total area is 1 settles toward a smooth curve.
The uniform distribution. A bus comes every 12 minutes, and you arrive at a random moment. Your wait W is equally likely to be anywhere from 0 to 12 minutes. Its density is a rectangle of width 12. For the area to be 1, the height must be 1/12. Then:
- P(W > 9) = (12 − 9)(1/12) = 3/12 = 0.25
- P(3 < W < 5) = 2/12 ≈ 0.167
- The mean is the balance point, 6 minutes.
The exponential distribution (PS.7). Some waiting times are better described by a curve that starts high and decays: the time until the next call to a help line, or the next customer at a counter, when events happen at random at a steady average rate. The exponential density with mean μ is
f(x) = (1/μ) e−x/μ, x ≥ 0, so P(X > t) = e−t/μ
(This area formula comes from calculus. You can verify it numerically by adding thin rectangles in a spreadsheet.) Illustrative example: if calls to a help line arrive at random, on average every 4 minutes, then the wait X for the next call is exponential with μ = 4:
- P(X > 6) = e−6/4 = e−1.5 ≈ 0.223
- P(X < 2) = 1 − e−0.5 ≈ 0.393
- The median m solves e−m/4 = 0.5, so m = 4 ln 2 ≈ 2.77 minutes, less than the mean of 4. The long right tail pulls the mean up, just as with skewed data in Statistics A.
The exponential has a surprising property. If you have already waited 2 minutes, the chance of waiting more than 6 more is still e−1.5. It has no memory. This fits truly random arrivals, but not a bus on a schedule, which is why the bus used the uniform model.
Choosing a model. A density curve is a model of a process, not a fact about it. You choose one from the context (is every value equally likely? do events arrive at random?) and then check it against data. The most important continuous model, the normal distribution, is next.
📖 Read for this lesson
The reading opens over the lesson; close it to come back to exactly where you were. The checks below draw on it.
Formative check
Work through these before moving on. They are not graded — they tell you, and your teacher, whether the standard below has landed yet.
For a continuous random variable X, what is P(X = 5)?
A wait uniformly distributed from 0 to 12 minutes has density height ; P(wait > 9) = .
For an exponential waiting time with mean 4 minutes, the median wait is also 4 minutes.
Sketch the exponential density with mean 4 and shade P(2 < X < 6). Compute it with the formula (e^(−2/4) − e^(−6/4)). Then check it in a spreadsheet by adding the areas of 40 thin rectangles of width 0.1 from 2 to 6 under f(x) = 0.25e^(−x/4). Explain in two sentences why the rectangle sum is close but not exact.
How confident are you that you can read probabilities as areas under a density curve and compute them for uniform and exponential models?
Practice
Work these on paper or in your notebook, then open Check your answer. Aim for all of Fluency and Application; try at least one Challenge.
Printable version: this unit’s practice workbook (PDF)
Fluency
Build speed and accuracy with the core skill.
- A wait W is uniformly distributed from 0 to 20 minutes. Find the height of the density, P(W > 15), P(5 < W < 8) and the mean wait.
Check your answer
Answer: Height 1/20; P(W > 15) = 0.25; P(5 < W < 8) = 0.15; mean 10 minutes.Areas of rectangles. - For the same uniform wait, find P(W = 7).
Check your answer
Answer: 0.A single value has no area. - A waiting time is exponential with mean 5 minutes. Find P(X > 5).
Check your answer
Answer: e^(−1) ≈ 0.368.P(X > t) = e^(−t/μ). - For the same exponential wait (mean 5), find P(X < 2).
Check your answer
Answer: 1 − e^(−0.4) ≈ 0.330. - Find the median of an exponential distribution with mean 5.
Check your answer
Answer: 5 ln 2 ≈ 3.47 minutes.Solve e^(−m/5) = 0.5. - An exponential wait has mean 10 minutes. Find P(X > 20).
Check your answer
Answer: e^(−2) ≈ 0.135. - Can a density curve have height greater than 1 anywhere? Explain.
Check your answer
Answer: Yes. Height is density, not probability; only the total area must be 1 (for example, uniform on 0 to 0.5 has height 2). - A density curve is symmetric about 12. What is P(X < 12)?
Check your answer
Answer: 0.5.
Application
Use the skill in context. Show your reasoning.
- (Illustrative) Calls to a school’s attendance line arrive at random, on average every 3 minutes. What is the probability of waiting more than 6 minutes for the next call? More than 9 minutes, given that 3 minutes have already passed?
Check your answer
Answer: P(X > 6) = e^(−2) ≈ 0.135. Given 3 minutes have passed, P(X > 9 | X > 3) = P(X > 6) ≈ 0.135, by the memoryless property. - A light-rail train arrives exactly every 15 minutes, and you arrive at a random time. Which model fits your wait, uniform or exponential? Find P(wait > 10).
Check your answer
Answer: Uniform on 0-15 (arrivals are scheduled, not random); P(wait > 10) = 5/15 ≈ 0.333.The exponential fits random arrivals, not scheduled ones. - For an exponential wait with mean 4 minutes, find P(2 < X < 6) and explain why the mean exceeds the median.
Check your answer
Answer: e^(−0.5) − e^(−1.5) ≈ 0.383. The distribution is skewed right; the long right tail pulls the mean (4) above the median (≈ 2.77).
Challenge
Stretch problems. Expect to think before you write.
- The density f(x) = cx on 0 ≤ x ≤ 2 (and 0 elsewhere) is a valid density curve for one value of c. Find c, then find P(X < 1).
Check your answer
Answer: The area is a triangle: ½ · 2 · 2c = 2c = 1, so c = 0.5. P(X < 1) = ½ · 1 · 0.5 = 0.25.Area under a straight-line density is a triangle.
Review
Keep earlier skills sharp.
- X ~ Binomial(n = 15, p = 0.3). Find P(X = 2), P(X ≤ 2), and the mean and standard deviation of X.
Check your answer
Answer: P(X = 2) ≈ 0.0916; P(X ≤ 2) ≈ 0.1268; mean 4.50; SD ≈ 1.775P(X = k) = C(15, k) pk(1 − p)n−k. - A game pays −10 with probability 4/13; 5 with probability 5/13; 50 with probability 4/13. Find the expected value.
Check your answer
Answer: 185/13 ≈ 14.23Multiply each value by its probability and add.
Two Real Quantities from the Earthquake Catalog: One Uniform, One Not Exponential
USGS Earthquake Catalog (ComCat) via the FDSN event web service: earthquakes of magnitude 5.5 and above, latitude 32.5-42.0 N, longitude 124.5-114.1 W, 1990-01-01 to 2025-12-31; the rectangle-sum table is SLS original; retrieved September 23, 2026 · Public domain (works of the United States government), with an SLS original worked table
Why this reading. The lesson ends with a rule — “a density curve is a model of a process, not a fact about it; choose one from the context and then check it against data” — and then gives you no data to check anything against. Its bus and help-line examples are invented. Below are two real quantities from the U.S. Geological Survey’s earthquake catalog: one that a uniform model describes well, and one that an exponential model describes badly, for a reason the lesson’s “no memory” paragraph predicts. The rectangle sum the open question asks for is also worked, so it can be done without a spreadsheet.
Table 1. Earthquakes of magnitude 6.0 or greater, California and western Nevada, 1990–2025
USGS Earthquake Catalog (ComCat), searched September 23, 2026 for events of magnitude 6.0 and above, latitude 32.5° to 42.0°N, longitude 124.5° to 114.1°W, January 1, 1990 through 2025, type “earthquake”. Magnitudes are as cataloged (a mix of magnitude types), rounded to one decimal. The box is a rectangle, so it takes in western Nevada and misses nothing of California.
| # | Date (UTC) | Magnitude | Place, as cataloged | Days since previous |
|---|---|---|---|---|
| 1 | 1991-08-17 | 6.0 | 30 km SSW of Scotia, CA | — |
| 2 | 1992-04-23 | 6.1 | The 1992 Joshua Tree, CA Earthquake | 249.4 |
| 3 | 1992-04-25 | 7.2 | The 1992 Petrolia, CA Earthquake | 2.6 |
| 4 | 1992-06-28 | 7.3 | The 1992 Landers, CA Earthquake | 63.7 |
| 5 | 1992-06-28 | 6.3 | The 1992 Big Bear, CA Earthquake | 0.1 |
| 6 | 1993-05-17 | 6.1 | 46km E of Big Pine, CA | 323.3 |
| 7 | 1994-01-17 | 6.7 | Northridge, CA, earthquake | 244.5 |
| 8 | 1994-09-12 | 6.1 | 4 km S of Carter Springs, NV | 238.0 |
| 9 | 1999-10-16 | 7.1 | The 1999 Hector Mine, CA Earthquake | 1859.9 |
| 10 | 2003-12-22 | 6.5 | 10 km NE of San Simeon, CA | 1528.4 |
| 11 | 2014-08-24 | 6.0 | The 2014 South Napa, CA Earthquake | 3897.6 |
| 12 | 2019-07-04 | 6.4 | Ridgecrest Earthquake Sequence | 1775.3 |
| 13 | 2019-07-06 | 7.1 | Ridgecrest Earthquake Sequence | 1.4 |
| 14 | 2020-05-15 | 6.5 | Monte Cristo Range, NV Earthquake | 314.3 |
| 15 | 2021-07-08 | 6.0 | Antelope Valley, CA | 419.5 |
| 16 | 2021-12-20 | 6.2 | 7km N of Petrolia, CA | 164.9 |
| 17 | 2022-12-20 | 6.4 | 15km WSW of Ferndale, CA | 364.6 |
Seventeen events give 16 waiting times. Mean gap 715.5 days, median 281.9 days, standard deviation 1054.2 days. Mean far above median: right skew, as an exponential model would want. So try one.
Check the exponential model with mean μ = 715.5 days
For an exponential distribution, P(X > t) = e−t/μ. Compare with the fraction of the 16 observed gaps that exceed t:
| t (days) | Model P(gap > t) = e−t/715.5 | Observed fraction of the 16 gaps above t |
|---|---|---|
| 30 | 0.959 | 13/16 = 0.812 |
| 180 | 0.778 | 11/16 = 0.688 |
| 365 | 0.600 | 5/16 = 0.312 |
| 730 | 0.360 | 4/16 = 0.250 |
| 1460 | 0.130 | 4/16 = 0.250 |
| 2920 | 0.017 | 1/16 = 0.062 |
The model’s median would be μ ln 2 = 496 days; the observed median is 282. Too many short gaps and too many very long ones. Look at the short ones: 0.1 days (Landers to Big Bear, the same afternoon), 1.4 days (the two Ridgecrest shocks), 2.6 days (Joshua Tree to Petrolia, which is a coincidence, not an aftershock), 63.7 days. Large earthquakes trigger others nearby. That is memory: once a magnitude-7 has happened, the chance of another large event in the next few days is far higher than the exponential’s constant rate allows. The exponential is the right model for arrivals that are independent of each other, and aftershocks are the opposite of independent. Seismologists model catalogs like this one with a clustering process instead, and they “decluster” a catalog before fitting an exponential to the mainshocks. Sixteen gaps is also a small sample; say so before you say the model fails.
Table 2. Minute past the hour, all 47 events of magnitude 5.5 or greater in the same search
Same catalog query with the magnitude floor lowered to 5.5 (the 17 events above are among them). The minute of the hour at which an earthquake occurs has no reason to prefer any value: a uniform model on 0 to 59 says every minute is equally likely.
| Events 1–24 | 43 | 15 | 43 | 29 | 50 | 6 | 57 | 0 | 1 | 43 | 5 | 14 | 8 | 14 | 20 | 30 | 31 | 33 | 23 | 27 | 17 | 22 | 6 | 46 |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Events 25–47 | 59 | 59 | 15 | 15 | 16 | 26 | 7 | 47 | 20 | 18 | 22 | 13 | 53 | 33 | 19 | 47 | 3 | 32 | 40 | 49 | 10 | 34 | 8 |
A uniform distribution on the integers 0 to 59 has mean 29.5 and standard deviation √((602 − 1)/12) = 17.3. The 47 minutes have mean 26.1 and standard deviation 16.9. By quarter hour, 0–14, 15–29, 30–44 and 45–59 minutes: 13, 15, 10, 9, against 11.75 expected in each. No quarter is far off, which is what a real uniform looks like with 47 draws: not perfectly flat. (Lesson 4.1 gives you the formal test for “far off”.) With a continuous uniform on [0, 60), P(a random event falls in the first 9 minutes) = 9/60 = 0.15; the data give 9/47 = 0.19.
The rectangle sum, worked
The open question: check P(2 < X < 6) for the exponential with μ = 4 by adding 40 rectangles of width 0.1 under f(x) = 0.25e−x/4. The exact area is e−0.5 − e−1.5 = 0.60653 − 0.22313 = 0.38340.
| Rectangle | 1 | 2 | 3 | 4 | 5 | … | 40 | Sum of all 40 |
|---|---|---|---|---|---|---|---|---|
| Left edge x | 2.0 | 2.1 | 2.2 | 2.3 | 2.4 | … | 5.9 | |
| Left height f(x) | 0.15163 | 0.14789 | 0.14424 | 0.14068 | 0.13720 | … | 0.05719 | |
| Left area, width 0.1 | 0.015163 | 0.014789 | 0.014424 | 0.014068 | 0.013720 | … | 0.005719 | 0.38821 |
| Right-edge area | 0.014789 | 0.014424 | 0.014068 | 0.013720 | 0.013382 | … | 0.005578 | 0.37863 |
| Midpoint area | 0.014975 | 0.014605 | 0.014245 | 0.013893 | 0.013550 | … | 0.005648 | 0.38339 |
The left-edge sum overshoots and the right-edge sum undershoots, because the curve is falling: a rectangle drawn from its left edge sticks up above the curve, one drawn from its right edge leaves a sliver uncovered. The exact area is trapped between them, and their average, 0.38342, is already very close; the midpoint rule lands on 0.38339, one unit off in the fifth decimal. The sum is close but not exact because each rectangle replaces a curved top by a flat one; make the rectangles thinner and the error shrinks toward zero, which is the idea calculus makes precise.
Sources: USGS Earthquake Catalog search (earthquake.usgs.gov); USGS FDSN Event Web Service (the query used here) (earthquake.usgs.gov)