Continuous Random Variables: From Point Probabilities to Areas
A continuous random variable measures a quantity that can take any value in a range, including fractional values: waiting time, fill volume, temperature, or distance between arrivals. A discrete variable counts distinct outcomes, such as the number of defects. This difference changes how probability is calculated: a continuous model assigns probability to intervals through area, not to individual points.
Let denote the random variable and a possible value. Its probability density function (PDF) is nonnegative and has total area 1. The height is a density, not the probability of exactly .
For an interval, probability is the area under the density curve:
A single point has zero width and therefore zero area: . For a continuous variable, including or excluding an interval endpoint does not change its probability. Integration plays the role that summation plays for a discrete probability mass function; tables or software can provide the result without integrating from first principles.
The cumulative distribution function (CDF) accumulates probability up to a value, while the tail probability captures values above it:
The lower limit is understood to be the minimum possible value when the variable has a restricted range. The expected value is the probability-weighted center and the variance measures squared spread around it:
| Question | Probability expression |
|---|---|
| At most | |
| Greater than | |
| Between and | |
| Exactly |
Key takeaways
- A continuous variable can take fractional values; probability concerns intervals.
- A PDF is a density; area under it gives probability.
- The full area is 1, and any single point has probability 0.
- CDFs, tails, expectation, and variance extend familiar discrete ideas using integration.
Uniform Distribution
A uniform random variable assigns equal density throughout a finite interval . Since the density is constant, probability is proportional to interval length: an interval twice as long is twice as likely.
For , the CDF and upper tail are:
Therefore, for :
Uniform Distribution
Driving time example
A trip takes between 100 and 140 minutes and is modeled uniformly. Thus , , and the mean is minutes. The density is per minute.
- At most 130 minutes:
- More than 105 minutes:
- Exactly 120 minutes: , despite the positive PDF height of . The density is not a point probability.
Phone battery example
Battery time to recharge ranges uniformly from 8 to 12 hours. Here , , and mean time is hours.
- Recharge needed by 9 hours:
- Battery lasts at least 11 hours:
- 80th percentile: solve to get hours. Thus 80% of battery durations are at most 11.2 hours.
The discrete counterpart is a fair die: its six distinct outcomes each have probability . The continuous uniform model instead spreads probability evenly over an interval.
Key takeaways
- Uniform probability is interval length divided by total range.
- The mean is the midpoint; variance depends on squared interval width.
- A percentile is found by setting the CDF equal to its probability and solving for .
- A density value such as is not the probability of one exact time.
Exponential Distribution: Time Between Random Events
When events occur randomly at a constant average rate , the exponential distribution often models the waiting time between consecutive events. It measures time or distance between events, not the event count. Its support is nonnegative, and its mean is .
For :
The mean and standard deviation are equal. The density is right-skewed. Common applications include manual task duration, call duration, equipment lifetime until repair, and waiting in a queue, when the constant-rate model is appropriate.
Exponential Distribution: Time Between Random Events
Customer-service wait example
Average wait is 2 minutes, so and per minute.
- Less than 1 minute:
- More than 3 minutes:
- Between 1 and 3 minutes:
- 90th percentile: solve , so and minutes.
Memoryless property and bus waiting
The exponential distribution is memoryless: if an event has not happened by time , the additional wait has the same exponential distribution as the original wait. The elapsed wait does not change the remaining-wait distribution.
A bus wait averages 20 minutes, hence per minute.
- Wait 15 minutes or less: .
- Wait more than 30 minutes: .
- Wait between 15 and 30 minutes: .
- Having already waited 10 minutes, wait at least 10 more: .
- Having already waited 20 minutes, wait at least 10 more: again.
Connection to the Poisson process
A Poisson process counts random events over time at a constant rate. If gaps between successive events are independent exponential random variables with rate , the count during a time interval of length has a Poisson distribution with mean . When gaps are deterministic, the count is deterministic; when gaps are exponential, the count is random.
For highway defects averaging 3 per kilometre, the mean distance between defects is km, and the exponential rate is 3 per kilometre.
- At least 0.5 km with no defect: .
- At least 2 km with no defect: (about 0.002 using the displayed table precision).
Oil-change time and counts
An oil change takes an exponentially distributed time with mean 20 minutes, so per minute.
- Finish within 20 minutes: . This is the probability relevant to a 20-minute guarantee.
- Finish between 15 and 30 minutes: .
The corresponding number completed in a fixed time follows a Poisson model, with mean equal to the rate times elapsed time. In a 4-hour (240-minute) shift, . Thus more than 15 means at least 16:
In an 8-hour (480-minute) day, , and the probability of more than 30 (at least 31) completions is approximately as reported by the Poisson table.
Key takeaways
- Exponential models waiting time; Poisson models event count over a fixed interval.
- Convert mean wait to rate using , keeping units consistent.
- Memorylessness means the additional wait distribution is unchanged by elapsed time.
- For a Poisson process, expected count over duration is .
Normal Distribution and Standard Normal Calculations
The normal distribution, also called the Gaussian or bell-shaped distribution, models many measurements and is central to statistical inference. Examples include heights, weights, test scores, scientific measurements and rainfall. A normal variable is described by its mean and standard deviation :
It is symmetric around , where mean, median and mode coincide. Half the area lies on each side; skewness is zero. The tails extend indefinitely, approaching but never touching the horizontal axis. A larger makes the curve wider and flatter; a smaller makes it narrower and taller. The empirical coverage rule is approximately:
| Range around the mean | Probability |
|---|---|
| 0.683 | |
| 0.954 | |
| 0.997 |
The standard normal variable has mean 0 and standard deviation 1. Its density is ; its CDF is often written . Any normal variable can be standardized:
The resulting -score counts standard deviations above (positive) or below (negative) the mean. Tables/software provide CDFs, tails, interval probabilities, and inverse percentiles. For example, if , then maps to , and maps to ; hence .
For a normal model with mean 10 and standard deviation 3, the CDF at 13 is 0.841, the upper tail at 13 is 0.159, and the interval from 7 to 13 has probability 0.683. In Excel, NORM.DIST(x, mean, standard_dev, TRUE) returns the CDF and FALSE returns density; NORM.INV(probability, mean, standard_dev) returns the value at a cumulative probability. Inverse upper-tail questions use left cumulative probability .
Normal Distribution and Standard Normal Calculations
Standard-normal probability practice
CDF and tail results below are approximate table values. Since is continuous, no probability is double-counted at a shared endpoint.
| Quantity | Result |
|---|---|
| 0.159 | |
| 0.841 | |
| 0.933 | |
| 0.994 | |
| 0.499 | |
| 0.288 | |
| 0.433 | |
| 0.345 | |
| 0.579 | |
| 0.885 | |
| 0.242 |
Inverse examples: left area 0.21 corresponds to ; central area 0.90 is between about and ; central area 0.20 is between about and ; left area 0.955 corresponds to (the transcript gives 2.58, which does not match this cumulative probability); and right area 0.692 corresponds to .
A two-sided critical value leaves area in each tail, so the central area between and is .
| Central area | Tail area | (approx.) |
|---|---|---|
| 0.68 | 0.16 | 1.00 |
| 0.80 | 0.10 | 1.28 |
| 0.90 | 0.05 | 1.64 |
| 0.95 | 0.025 | 1.96 |
| 0.99 | 0.005 | 2.58 |
Sketching the bell curve and shading the requested area helps distinguish left tail, right tail and central intervals before consulting a table.
Investment-return example
Annual returns have a normal model with and percentage points.
- Unsatisfactory, below 5%: , giving .
- Excellent, above 10%: , so the right tail is approximately 0.748.
- Moderate, between 5% and 10%: .
- 90th percentile: . A return above this is in the top 10%.
- A 90% upper-tail probability corresponds to the 10th percentile, about 8.1%. A central 90% interval is about 7% to 17%.
Screen-time example
Daily screen time has hours and hours.
- More than 4 hours: , so .
- Between 6 and 12 hours: approximately 0.757.
- Top 20% threshold is the 80th percentile: hours; more time places a child in the top fifth.
- Bottom 20% threshold is the 20th percentile: hours. Equivalently, 80% exceed about 6.3 hours.
- A central 80% interval is about 5.2 to 11.6 hours.
Exam-completion times
Completion time has minutes, minutes, for 60 students.
- At most 2 hours (120 minutes): ; probability 0.067. Expected number: students.
- Between 2 and 3 hours (120 to 180 minutes): probability 0.866; expected number .
- At least 3 hours (180 minutes): probability 0.067; expected number about 4. The equal tails follow from symmetry. Convert hours to minutes before using the model.
Choosing between filling machines
Packet weights are normal. Compare their probabilities against the target range rather than judging by mean alone.
| Machine | Mean, SD (kg) | Below 1 kg | Between 1 and 1.05 kg | Central 90% interval (kg) |
|---|---|---|---|---|
| 1 | 1.01, 0.02 | 0.309 | 0.669 | 0.97 to 1.04 |
| 2 | 1.03, 0.04 | 0.227 | 0.465 | 0.96 to 1.09 |
For machine 1, the 90% interval is centered on 1.01 and has half-width approximately , yielding about 0.98 to 1.04 by direct calculation; the transcript/template reports 0.97 to 1.04. For machine 2, the reported interval is 0.96 to 1.09. Neither machine puts 90% of packets between 0.99 and 1.01 kg. Machine 2 has a lower underweight probability in this comparison, but fewer packets in the 1 to 1.05 kg band and greater spread; the choice depends on the quality criterion.
Glass-sheet thickness
Thickness is normal with mm and mm.
- Below 2.9 mm: , so probability is 0.202.
- Above 3.2 mm: , so upper-tail probability is 0.048.
- Central 95% interval: approximately , or 2.76 to 3.24 mm.
Key takeaways
- A normal model is set by and ; symmetry makes the two halves equal.
- Standardization expresses distance from the mean in standard deviations.
- Left area, right tail, interval, and inverse-percentile questions are distinct; sketch the requested region.
- Keep units consistent, especially for time and percentiles.
Normal Approximation to the Binomial Distribution
The binomial distribution is discrete, while the normal is continuous. When the binomial shape is sufficiently bell-like, the normal can approximate its probabilities, avoiding cumbersome factorial calculations. A common rule in this module is and , where . Match the binomial mean and variance:
Because a continuous variable has zero probability at a point, use a continuity correction to include the discrete integer outcomes in a half-unit-wide interval:
| Binomial event | Normal approximation event |
|---|---|
Normal Approximation to the Binomial Distribution
Sixteen fair-coin tosses
For , , so , variance , and .
- At most 5 successes, exactly: .
- With continuity correction, for is approximately 0.1056.
- Between 8 and 11 successes inclusive, exact binomial probability is 0.5598. Approximate with .
Invoice errors
Among invoices, error probability is ; . Mean , variance , and standard deviation . Conditions hold: and .
- Exactly 12 errors: binomial probability is about 0.09. Approximate by the normal area from 11.5 to 12.5:
- At most 13 errors: correction gives ; exact binomial value is about 0.876.
These are approximations, and accuracy generally improves when both and are sufficiently large.
Startup funding example
Of 200 seed-funded startups, 20% receive the next funding level. Let count successful transitions: , , .
Both conditions hold: and exceed 5.
- At least 45: apply lower-bound correction, . Standardize: . The upper tail is about 0.2148, or 21%.
- At most 50: upper-bound correction gives . Here , giving about 0.9686, or 97%.
Normal approximations are useful computationally, but remain estimates of the exact binomial probabilities. The Poisson distribution also has mean and variance both equal to its rate parameter; the closing passage on approximating Poisson with binomial is internally inconsistent, so no additional approximation rule is inferred here.
Key takeaways
- Approximate a binomial with when both expected successes and failures are sufficiently large.
- Preserve the discrete event using a half-unit continuity correction.
- Match the mean and variance before standardizing.
- State results as approximations and retain exact binomial values when available.
Additional Worked Examples
Battery life under two usage patterns
Talk-time battery life averages 20 hours; internet-use battery life averages 7 hours. Model each as a separate exponential variable, and , with different rates.
- Talk time: . No more than 15 hours:
- Talk time: more than 20 hours:
- Internet use: . No more than 5 hours:
Identify the correct usage-specific random variable and mean before calculating; the exponential rate is the reciprocal of the mean.
Website visits: counts and gaps
A website receives an average of 6 visits per minute. The mean gap is seconds, so interarrival time is exponential with mean 10 seconds and rate per second. Counts are Poisson with mean rate times duration.
- Mean time between visits: seconds.
- No visits in one minute: for ,
- At most 10 visits in two minutes: , so the table gives .
- Gap at most 15 seconds: for exponential with per second,
- Gap more than 20 seconds:
This example switches between the Poisson count and exponential gap while keeping the same event rate and consistent time units.
Investment-fund returns
Three-year returns have mean 20% and standard deviation 4 percentage points. Let .
- Return above 25%: . The CDF at 1.25 is 0.894, so
- Return at most 10%: , so .
- Threshold for top 10%: left cumulative probability is 0.90, corresponding to . Convert back:
Thus the reported threshold is about 25.1%. Standardize first, use the standard-normal table, then translate back to return units.
Diesel prices in two cities
Jamnagar price has mean ₹90 and standard deviation ₹0.5; Surendranagar price has mean ₹92 and standard deviation ₹0.8.
- Surendranagar at most ₹91: . The left-tail probability is 0.1056, about 10%.
- Jamnagar above ₹91: . The standard-normal CDF is 0.9772; therefore the upper tail is , or 2.28%.
- Jamnagar above ₹92: . The upper tail is very close to zero; the source describes this as roughly no stations.
Seed funding and normal approximation
This example is worked in the preceding approximation section: for 200 companies and a 20% transition rate, and . With continuity correction, at least 45 is approximated by , and at most 50 by . The count is binomial; the matched normal supplies approximate areas.
Key takeaways
- Each usage mode, city, or time window can define a different random variable or parameter set.
- Poisson counts and exponential gaps are two views of the same constant-rate arrival process.
- Standardize normal quantities with their own mean and standard deviation before reading a table.
- Continuity correction is needed when a discrete count is approximated by a continuous normal curve.
Test yourself on this module
Questions from these notes, one at a time, with feedback. Your weak topics are saved.