Binomial, normal and Poisson models
Introduit| English | Sens |
|---|---|
| Poisson distribution/ˈpɔɪsn ˌdɪstrɪˈbjuːʃn/ | A probability model for event counts in a fixed interval when events occur independently at a constant average rate. |
| expected value/ekˈspektɪd ˈvæljuː/ | espérance mathématique |
A board of pins, 1889
- In 1889, Francis Galton described his "quincunx" in the book Natural Inheritance: balls fall through rows of pins, bouncing left or right at each pin with equal chance. The number of right-bounces for each ball follows a binomial distribution, and with many rows the pile of balls takes the bell shape of the normal distribution. Galton used it to explain why heights cluster around an average.
- S2 models counts with the binomial and Poisson distribution 泊松分布, and approximates them when $n$ or the mean is large.
When the binomial fits
- $X \sim \mathrm{B}(n, p)$ needs a fixed number $n$ of trials, two outcomes, independent trials and a constant probability $p$.
- $\mathrm{P}(X = r) = \dbinom{n}{r}p^r(1 - p)^{n - r}$, with expected value 期望值 $\mathrm{E}(X) = np$ and variance $np(1 - p)$.
Which situations can be modelled by a binomial distribution?
Difficulté 2/5 · 140 pièces + un petit bonus lors de la première réussite aujourd'hui
Without replacement changes $p$; time is continuous.
$X \sim \mathrm{B}(40, 0.25)$. Find $\mathrm{E}(X)$.
Difficulté 2/5 · 140 pièces + un petit bonus lors de la première réussite aujourd'hui
$np = 40 \times 0.25$.
Cumulative probabilities
- $\mathrm{P}(X \le r)$ adds the probabilities from 0 to $r$; a calculator gives it directly.
- $\mathrm{P}(X \ge r) = 1 - \mathrm{P}(X \le r - 1)$ and $\mathrm{P}(X > r) = 1 - \mathrm{P}(X \le r)$.
Binomial, normal and Poisson models
A binomial model needs fixed n, independent trials, two outcomes and constant p
Compare the model with the worked case and explain one change.
Which expression equals $\mathrm{P}(X \ge 4)$ for a binomial $X$?
Difficulté 1/5 · 100 pièces + un petit bonus lors de la première réussite aujourd'hui
At least 4 is the complement of at most 3.
Worked example: a binomial count
30% of seeds fail to grow. In a packet of 10, find the probability that at most 2 fail.
- Model: $X \sim \mathrm{B}(10, 0.3)$, assuming seeds fail independently.
- Terms: $\mathrm{P}(X = 0) = 0.7^{10} = 0.0282$, $\mathrm{P}(X = 1) = 10(0.3)(0.7^9) = 0.1211$, $\mathrm{P}(X = 2) = 45(0.3^2)(0.7^8) = 0.2335$.
- Add: $\mathrm{P}(X \le 2) = 0.3828$ (4 d.p.).
$X \sim \mathrm{B}(6, 0.5)$. Find $\mathrm{P}(X = 3)$ as a decimal.
Difficulté 2/5 · 140 pièces + un petit bonus lors de la première réussite aujourd'hui
$\binom{6}{3} \times 0.5^6 = \dfrac{20}{64}$.
The Poisson distribution
- $X \sim \mathrm{Po}(\lambda)$ counts events that happen singly, independently and at a constant average rate $\lambda$: $\mathrm{P}(X = r) = \dfrac{e^{-\lambda}\lambda^r}{r!}$, with mean and variance both $\lambda$.
- The rate scales with the interval: 3 calls per hour means $\mathrm{Po}(6)$ for 2 hours. With $\lambda = 3$, $\mathrm{P}(X = 2) = \dfrac{e^{-3}3^2}{2} = 0.224$.
Calls arrive at a rate of 2 per hour. Find the probability of fewer than 2 calls in 2 hours, to 4 d.p.
Difficulté 2/5 · 140 pièces + un petit bonus lors de la première réussite aujourd'hui
$X \sim \mathrm{Po}(4)$: $\mathrm{P}(X \le 1) = e^{-4}(1 + 4) = 0.0916$.
Approximations
- If $n$ is large and $p$ is small, $\mathrm{B}(n, p) \approx \mathrm{Po}(np)$. If $n$ is large and $p$ is close to 0.5, $\mathrm{B}(n, p) \approx \mathrm{N}(np, np(1 - p))$; for large $\lambda$, $\mathrm{Po}(\lambda) \approx \mathrm{N}(\lambda, \lambda)$.
- A normal approximation needs a continuity correction: $\mathrm{P}(X \le 45)$ becomes $\mathrm{P}(Y < 45.5)$.
Using a normal approximation $Y$ with a continuity correction, $\mathrm{P}(X \ge 30)$ becomes
Difficulté 1/5 · 100 pièces + un petit bonus lors de la première réussite aujourd'hui
30 is included, so start from 29.5.
A binomial model needs a constant $p$. Drawing 5 cards from a pack without replacement changes $p$ after every draw, so it is not binomial. Check all four conditions before using $\mathrm{B}(n, p)$.
$\mathrm{B}(n, p)$ counts successes in $n$ independent trials with constant $p$: $\mathrm{P}(X = r) = \tbinom{n}{r}p^r(1 - p)^{n - r}$, mean $np$. $\mathrm{Po}(\lambda)$ counts random events at a constant rate, with mean and variance $\lambda$. Approximate by a Poisson when $p$ is small, or by a normal with a continuity correction when $n$ or $\lambda$ is large.
Faults in a cloth occur at random at a rate of 2.5 per 10 m. Find the probability of no faults in 10 m, and of at most 3 faults in 20 m. State one assumption.
Scale the rate, then add the terms.