Yates' correction
Studying this course? Keep your progress in one place.
Key facts
- The chi-squared statistic approximates discrete frequencies with a continuous distribution, so it tends to be too large when is small.
- Yates' correction applies only to a table, where .
- .
- Compare with the critical value for : at , at .
- The corrected statistic is always smaller, so it is more cautious about claiming an association.
Why a correction is needed
Discrete observed frequencies are approximated by a continuous distribution, which overstates the statistic when is small.
The chi-squared statistic is only an approximation: observed frequencies are discrete whereas the chi-squared distribution is continuous.
When the degrees of freedom are small this tends to make the statistic too large, so the test rejects too often.
Yates' correction (a continuity correction) reduces this effect.
| Uncorrected | With Yates' correction | |
|---|---|---|
| Statistic | Too large for small ν | Smaller |
| Rejects H0 | Too often | Less often |
Uncorrected
- Statistic:
- Too large for small ν
- Rejects H0:
- Too often
With Yates' correction
- Statistic:
- Smaller
- Rejects H0:
- Less often
Why does the uncorrected chi-squared statistic tend to be too large for small ?
When to apply it
Only for a table, where .
Apply Yates' correction when the table is a contingency table, so that . For larger tables () it is not used.
The convention that every expected frequency should be greater than 5 still applies.
| Degrees of freedom | 1 | 2 | 4 |
|---|---|---|---|
| Use Yates' correction? | Yes | No | No |
- Degrees of freedom:
- 1
- Use Yates' correction?:
- Yes
- Degrees of freedom:
- 2
- Use Yates' correction?:
- No
- Degrees of freedom:
- 4
- Use Yates' correction?:
- No
For which table is Yates' correction used?
How to apply it
Subtract from each before squaring and dividing by .
Find each expected frequency as usual, take the modulus of , subtract , then square and divide by . Compare with the critical value for : at and at .
In a table all four values of are equal, so work it out once and reuse it.
- 1
Expected frequencies
row total × column total ÷ grand total
- 2
Find |O − E|
the modulus, so it is never negative
- 3
Subtract 0.5
from each |O − E|
- 4
Square and divide by E
then sum over all four cells
- 5
Compare
with the critical value for ν = 1
Worked example
Method X: 17 pass, 8 fail. Method Y: 10 pass, 15 fail. Test at the 5% level whether the pass rate depends on method, using Yates' correction.
- 1
(pass) and (fail), for each method.
- 2
in every cell.
- 3
Corrected: .
- 4
Critical value for at 5% is , and .
In the example, what does the correction change about the conclusion?
Interpreting the result
The corrected statistic is smaller, so state the conclusion in context and note any borderline result.
Yates' correction always gives a smaller statistic than the uncorrected formula, so it is more cautious about claiming an association.
State the conclusion in context, for example "there is insufficient evidence that the pass rate depends on the training method".
If the corrected statistic is close to the critical value, say the result is borderline and that more data would help.
- Corrected statisticsmaller than uncorrected
- The 0.5 goesinside the bracket, on each |O − E|
Where is the subtracted in Yates' correction?
Try an exam question
Method X: 17 of 25 trainees pass and 8 fail. Method Y: 10 of 25 pass and 15 fail. Carry out a chi-squared test with Yates' correction at the 5% level to decide whether pass rate depends on the method. The 5% critical value for is .
[5 marks]
- [1]: pass rate is independent of method; : pass rate depends on method.
- [1]Expected frequencies: (pass) and (fail) for each method.
- [1] in every cell.
- [1].
- [1], so do not reject : insufficient evidence that pass rate depends on method.
That's the notes covered.
Carry on to the next subtopic.
Try a question
Assuming no association, find the expected number of seeds given fertiliser that germinate.
1 mark, marked against the mark scheme
Assuming no association, find the expected number of seeds given fertiliser that germinate.
1 mark, marked against the mark scheme