Thursday 17 February 2011

Academics - a blessing and a curse

There are some disadvantages to having academics as friends - one of them is that they quite happily give you a good finger-wagging over the least wee thing.  Here's a message I received from a friend of mine who obviously likes statistics even more than I do.  I've taken her name off the bottom in case she hassles me again and I can't reproduce the graph properly, I can't make it clear (I'm sure I'll get pelters for that, too), I guess I just failed my exam!

C

Calum,
I do not like the 2nd paragraph of your blog (well, what you are saying is fine but it is horribly covered in percentages). What you need is a nice pretty graph, like the one I have attached for you.


Also, if you don’t already have it (but you probably do and apologies if so), the basic formula for calculating a 95% confidence interval for any proportion in Excel is:
Lower 95% CI limit: =B2-(1.96*(SQRT(B2*(1-B2)/C1)))
Upper 95% CI limit: =B2+(1.96*(SQRT(B2*(1-B2)/C1)))
where (for example) ‘B2’ is the cell you’ve put the proportion of interest (e.g. Alex Salmond’s approval rating) and ‘C1’ is the cell you’ve put the sample size in (e.g. 1019).

When comparing proportions (for example Salmond’s approval vs Gray’s approval) if the confidence intervals overlap, they lie within the margin of error and there is no statistical difference between them. If they do not overlap then Salmond’s approval is likely to be significantly higher or lower (statistically) than Gray’s.

What you should see from the Mori data is that Salmond’s approval rating is significantly higher than that of all the other party leaders (and Tavish Scott’s is significantly lower), that there is no statistical difference in dissatisfaction towards any of the party leaders, and that significantly more people have an opinion on Alex Salmond than any of the other party leaders (and significantly fewer people have an opinion on Tavish Scott than either Salmond or Gray, but not Goldie).

NB: Should you like this formula and want to use it but in future for whatever reason you need a wider margin of error, you could take a 99% confidence interval by substituting the 1.96 figure for 2.58.

Don’t say I’m not good to you.



3 comments:

cynicalHighlander said...

Or you could get a star with 'Could try harder' comment

Calum Cashley said...

Nah, she's a hard marker!

BrianSJ said...

People who have naive faith in statistical significance are not statisticians and can be ignored with 99% confidence. A wee dram will get you the last percentage if you need it.