Choosing a distribution and approximations
Studying this course? Keep your progress in one place.
Key facts
- Binomial: a fixed number of trials, two outcomes, independent, constant . State the conditions in context.
- The binomial fails with dependence, a changing probability (sampling without replacement from a small population) or a non-fixed .
- Normal suits a continuous, symmetrical, bell-shaped variable; it is poor for skewed or discrete data.
- If with large and near , use .
- Apply a continuity correction: .
Choosing the binomial model
When all four conditions hold; state them in context.
is suitable for a fixed number of trials with two outcomes, independent trials and constant probability . Ten seeds each germinating with probability , independently, is binomial.
State the conditions in context, for example ‘each seed germinates independently of the others, with the same probability’.
- 1
Fixed n
A fixed number of trials.
- 2
Two outcomes
Success or failure.
- 3
Independent
One trial does not affect another.
- 4
Constant p
The same probability each time.
Ten seeds each germinate independently with probability 0.45. Which model fits the number germinating?
When the binomial may not be appropriate
Name the failing condition: independence, constant , a fixed , or random selection.
Independence fails for students in one tutor group or family members sharing a habit. The probability changes when sampling without replacement from a small population (drawing 6 cards from 10). The number of trials is not fixed if you count until something happens.
To improve the model, take a random sample from the whole population.
| Independence | Constant p | Fixed n | |
|---|---|---|---|
| Fails when | Trials share a cause | Probability changes | Counting until something happens |
| Example | Same tutor group | Without replacement from 10 cards | Not a set number of trials |
Independence
- Fails when:
- Trials share a cause
- Example:
- Same tutor group
Constant p
- Fails when:
- Probability changes
- Example:
- Without replacement from 10 cards
Fixed n
- Fails when:
- Counting until something happens
- Example:
- Not a set number of trials
Six cards are drawn without replacement from a pack of ten. Why is the number of red cards not binomial?
When the Normal may not be appropriate
When the data are skewed, bounded, discrete with few values, or have more than one peak.
The Normal suits a continuous, symmetrical, bell-shaped variable such as heights or masses. It is a poor model for skewed data such as incomes, a bounded or discrete variable with only a few values, or more than one peak.
A histogram roughly symmetrical about its mean, with most values within about two standard deviations, supports a Normal model.
| Suitable | Poor model | |
|---|---|---|
| Type | Continuous | Skewed or bounded |
| Shape | Symmetrical, bell-shaped | Discrete with few values |
| Example | e.g. heights, masses | e.g. incomes, sixes in 5 rolls |
Suitable
- Type:
- Continuous
- Shape:
- Symmetrical, bell-shaped
- Example:
- e.g. heights, masses
Poor model
- Type:
- Skewed or bounded
- Shape:
- Discrete with few values
- Example:
- e.g. incomes, sixes in 5 rolls
Which variable is least suitable for a Normal model?
Normal approximation to the binomial
For large and close to , use .
The Normal approximation uses mean and variance (not its square root). For : .
It works because the binomial is roughly symmetrical when is near . It is poor when is close to 0 or 1 or is small: for it gives positive probability to negative values.
. Which Normal approximation is correct?
The continuity correction
Treat each integer as the interval from to .
A binomial takes integer values but the Normal is continuous, so apply a continuity correction. Write the integers you want first: means 30, 31, … so the boundary is .
, , , .
Worked example
. Estimate .
- 1
Mean and variance , so .
- 2
Continuity correction: .
- 3
Use the Normal cumulative distribution function with .
Using a Normal approximation, for a binomial becomes which probability?
Try an exam question
. Use a Normal approximation to estimate .
[4 marks]
- [1].
- [1]Continuity correction: .
- [1].
- [1]Probability .
That's the notes covered.
Carry on to the next subtopic.
Try a question
Which distribution models exactly?
1 mark, marked against the mark scheme
Which distribution models exactly?
1 mark, marked against the mark scheme