Binomial, normal and Poisson models
Introducido| English | Significado |
|---|---|
| binomial distribution/baɪˈnəʊmɪəl ˌdɪstrɪˈbjuːʃn/ | A probability model for the number of successes in a fixed number of independent trials with the same success probability. |
| normal distribution/ˈnɔːml ˌdɪstrɪˈbjuːʃn/ | A continuous, symmetric bell-shaped probability distribution specified by its mean and standard deviation. |
| expected value/ekˈspektɪd ˈvæljuː/ | valor esperado |
Deaths by horse kick, 1898
- In 1898 the statistician Ladislaus Bortkiewicz studied 20 years of records from Prussian cavalry corps: soldiers killed by horse kicks. Most corps-years had none, some had one, a few had two or more. The counts matched a simple model almost exactly.
- Random counts can be predictable in the long run. The binomial distribution 二项分布, the normal distribution 正态分布 and, in Applications and Interpretation HL, the Poisson distribution model them; each has an expected value 期望值.
Bortkiewicz found that rare random counts can follow a predictable model.
Dificultad 1/5 · 100 monedas + un pequeño bono en el primer intento de hoy
Horse-kick deaths in Prussian cavalry corps.
The binomial model
- $X \sim B(n, p)$ needs a fixed number $n$ of independent trials, each a success with the same probability $p$.
- $P(X = k) = \dbinom{n}{k}p^k(1 - p)^{n-k}$; $E(X) = np$; $\text{Var}(X) = np(1 - p)$.
Which conditions does a binomial model need?
Dificultad 2/5 · 140 monedas + un pequeño bono en el primer intento de hoy
And two outcomes per trial.
X ~ B(5, 0.2). Find P(X = 0) to 4 d.p.
Dificultad 2/5 · 140 monedas + un pequeño bono en el primer intento de hoy
0.8⁵ = 0.32768.
X ~ B(20, 0.3). Find Var(X).
Dificultad 2/5 · 140 monedas + un pequeño bono en el primer intento de hoy
20 × 0.3 × 0.7.
The normal model
- A continuous, symmetric, bell-shaped model with mean $\mu$ and standard deviation $\sigma$. Standardise with $z = \dfrac{x - \mu}{\sigma}$.
- About 68%, 95% and 99.7% of values lie within 1, 2 and 3 standard deviations of the mean.
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.
From a raw value to a normal-model area
$z=(x-\mu)/\sigma$
Connect the lesson’s raw values to standardised bounds and the correct shaded region. Changing a model does not show that real data follow it.
Heights have μ = 170 cm and σ = 8 cm. Find the z-score of 182 cm.
Dificultad 2/5 · 140 monedas + un pequeño bono en el primer intento de hoy
(182 − 170)/8.
About what percentage of a normal population lies within 2σ of the mean?
Dificultad 1/5 · 100 monedas + un pequeño bono en el primer intento de hoy
The 68–95–99.7 rule.
The Poisson model (AI HL)
- Counts of events in a fixed interval at a constant average rate $m$: $P(X = k) = \dfrac{e^{-m}m^k}{k!}$, with mean and variance both $m$.
- Rates scale: 3 per hour means $m = 6$ for two hours.
Worked example: three models
- Binomial: $X \sim B(5, 0.2)$. $P(X = 0) = 0.8^5 = 0.32768$; $P(X \ge 1) = 0.67232$; $E(X) = 1$, $\text{Var}(X) = 0.8$.
- Normal: IQ scores, $\mu = 100$, $\sigma = 15$: a score of 130 has $z = 2$, so about 2.5% score higher (more exactly $P(X > 130) \approx 0.0228$).
- Poisson (AI HL): horse-kick deaths with $m = 0.7$ per corps-year: $P(X = 0) = e^{-0.7} \approx 0.497$.
Checking assumptions
- Binomial fails if trials are dependent or $p$ changes, such as drawing without replacement from a small group.
- For a continuous variable, $P(X = a) = 0$: ask for intervals.
Not every count is binomial. Counting red cards drawn from a small pack without replacement changes $p$ at each draw. Check all four conditions before using $B(n, p)$.
Drawing 5 cards without replacement from 10 and counting hearts is binomial.
Dificultad 1/5 · 100 monedas + un pequeño bono en el primer intento de hoy
p changes with each draw.
Use $B(n, p)$ for a fixed number of independent trials with constant $p$: $E = np$, $\text{Var} = np(1-p)$. Use the normal model for continuous symmetric data via $z$-scores, and (AI HL) Poisson for counts at a constant rate. State and check the assumptions.
A factory's bolts have lengths normally distributed with μ = 50 mm and σ = 0.4 mm. A bolt is rejected if outside 49.2–50.8 mm. Estimate the proportion rejected and, if bolts are packed in tens, the probability that a box has no rejects.
Normal, then binomial.