Sample size
Example: Confidence Levels and Sample Size
For a survey to project results confidently (95% confidence, plus or minus 5%) to the total population, the responding sample used must be large enough. But fortunately a fairly small sample can project rather well to a very large population as shown in the table below. In general, 400 respondents will project to a very large population (but the sample must be random or scientifically selected, and you must consider the smallest subgroup that you want to analyze as the basis for figuring your sample). If you start with a random or scientifically selected sample... To project to You need this many responses
| To project to | You need this many responses |
| 1,000,000 | 384 |
| 100,000 | 382 |
| 10,000 | 369 |
| 5,000 | 356 |
| 2,500 | 332 |
| 2,000 | 322 |
| 1,000 | 277 |
| 500 | 216 |
| 250 | 151 |
| 200 | 131 |
| 100 | 79 |
For example, if you estimate your audience to be 100,000, you will need to have roughly 380 respondents to project with high confidence (95%, + or - 5%). If you use a disk-by-mail survey and expect to achieve a 40% response rate, you will need to mail out approximately 1,000 disks in the mail. If you have subgroups to consider you will likely need a larger sample to mail to.
You can see from the above chart, there isn't much difference between the sample size required for 1,000,000 and the sample size required for 5,000. One further qualifying note is required here: at 40% response rate there is a chance that there may be some non-response bias in the data.
