Journalists and statistics – an unfortunate marriage.
It has long been know that many journalist have serious numeracy issues – not the least when it comes to reporting so called research findings, or political survey results.
As a result, statistics get a bad name, as in this report. For survey results or research results to have validity, two things are essential.
The sample use needs to be of an adequate size – this will affect the margin of error. For example a sample of 100 will have a margin of error of around 10%, whereas a sample of 1000 will have a margin of error of around 3.2% (95% confidence level, for a large population).
The sample used must be representative of the “population” of interest. This is the hard bit. Getting such a sample is no easy matter.
When journos report their results, they should be obliged to give the sample size and the margin of error.
An example: A survey of 500 voters suggests that 43% would vote for the Wacko party and 39% would vote for the Bonkers party. What does this mean?
500 in the sample would indicate a margin of error of around 4.5%. This is important. It actually means that in terms of statistical significance, there is no difference in voter preference.
This is because the margins of error overlap.
39% >>>>>>> 34.5 ……………….. 43.5
43% >>>>>>> >>>>>>>38.5 ………………….47.5
The margin of error is roughly 1divided by square root of sample size. Then convert this decimal to a percentage
e.g. 1divided by Square root 500 = 0.0447 which is 4.47%
So journos! Here is where to go to get help with your numeracy!
No comments:
Post a Comment