On this page:

EVALUATION OF THE COLINTON ALL POSTAL VOTE BY-ELECTION PILOT SCHEME

« Previous | Contents | Next »

Listen

APPENDIX 2: GUIDE TO STATISTICAL RELIABILITY

STATISTICAL RELIABILITY

The respondents to the questionnaire are only a sample of the total 'population'. It is not certain therefore that the figures obtained are exactly those we would have if everybody had been interviewed (the 'true' values). However, we can predict the variation between the sample results and the 'true' values from a knowledge of the size of the samples on which the results are based and the number of times that a particular answer is given.

The confidence with which we can make this prediction is usually chosen to be 95% - that is, the chances are 19 in 20 that the 'true' value will fall within a specified range. The table below illustrates the predicted ranges for different sample sizes and percentages results at the '95% confidence interval', based on a random sample.

Size of sample on which survey result is based

Approximate sampling tolerances applicable to percentages at or near these levels

10% or 90%
+

30% or 70%
+

50%
+

100 interviews

6

9

10

200 interviews

4

6

7

300 interviews

3

5

6

400 interviews

3

4

5

500 interviews

3

4

4

Source: MORI

For example, on a question where 50% of the people in a sample of 500 respond with a particular answer, the chances are 95 in 100 that this result would not vary by more than four percentage points, plus or minus from a complete coverage of the entire population using the same procedures. However, while it is true to conclude that the "actual" result (95 times out of 100) lies anywhere between 46% and 54%, it is proportionately more likely to be closer to the centre of this band ( i.e. at 50%).

When results are compared between separate groups within a sample, different results may be obtained. The difference may be "real", or it may occur by chance (because not everyone completed a questionnaire). To test if the difference is a real one - i.e. if it is "statistically significant" - we again have to know the size of the samples, the percentages giving a certain answer and the degree of confidence chosen. If we assume "95% confidence interval", the difference between two sample results must be greater than the values given in the tables below:

Actual Sample Size

10% or 90%
+

30% or 70%
+

50%
+

Overall (500) vs:

Sub-groups of:

100

5

8

9

200

3

5*

5

300

2

3

4

400

1

2

2

*For example, if 30% of the total sample (500) give a particular answer, and 34% of respondents in a sub-group of 200 give the same answer, there is not a statistically significant difference between the responses of the two groups.

Looking at the third column of the above table shows that there needs to be a difference of +5 percentage points between the two results in order for the difference to be statistically significant.

Therefore, if 36% of the latter group give the same answer, then this is a statistically significant difference (since there is more than a 5 point difference between the two).

« Previous | Contents | Next »

Page updated: Tuesday, May 31, 2005