UPDATE:
It’s all done – The Poll Cruncher
—–
In preparation for the next election, I’m planning on building up an arsenal of handy political resources masquerading around as spiffy toys – the Electorate Demographic Profiles being the first. The next cab of the rank is a little app that pulls apart a given poll. You’d be surprised at the number of emails I get that ask questions over the margin or error of a given poll – things like “If the poll was 55, what’s the probability that the real result is between 55 and 57?“.
Well now, you can find that out.
Simply enter the poll result and the sample size into the app under “Poll 1″ and the margin of error and a probability function of the poll will be automatically generated. Further, if you want to know the probability that the true level of public opinion is between any two values on the basis of a given poll, simply enter those two values in the “Maximum” and “Minimum” spots and it will tell you that as well – just make sure that the largest number is entered in the Maximum box and the smallest number in the Minimum box. So, for instance, if we had a Newspoll with a sample size of 1100 that came out as a TPP of 55% to the ALP, it will tell you that the Margin of Error is 2.94% and chart the probability function as a graph. If you want to know the probability that the true value of ALP support is actually between the values of 55% and 57%, simply enter 57 as “Maximum”, 55 as “Minimum” and it will tell you that the probability is 40.88%.
But!
We’ve also got some room left on the app – conveniently marked 
– so what do you want included in it? At the moment, the data for a second poll creatively called “Poll 2″ can be entered and it will chart the probability of that poll with the first poll – so you can see how the dispersion of the probabilities of the true value of public opinion change as sample size changes. But since this is an app for your use, it’s probably a good idea to ask what you want in it. So folks, what will we fill it up with? What additional thing would you like to see in this little resource? All suggestions welcome – although I can’t promise that I will actually be able to build what you might like, but we’ll see how we go. For those that don’t want flash clogging up the front page, you’ll have to click through to see it.
3 Comments
Massive love for this tool. Perhaps you can implement Gibbs sampling in the remaining box :p
Spiffy indeed and we need something like this to rub the noses of lazy journalists ignorant of MOE issues into. Bit puzzled about why the range in the bottom part of Poll 1 is in the order Maximum then Minimum since I would nearly always express such polling ranges minimum first. I’ve also noticed that when Poll 2 adds to the graph, the x-axis isn’t rescaled if poll 2’s range is partly off to either side.
My pick for something to put in the space would be a comparison of the Poll 1 and Poll 2 results that says whether or not one is significantly larger than another, with appropriate p-value.
Gibbs sampling – there’s always one comedian
That’s a good idea changing the Max and Min around – have no idea why I put them back to front. The scaling of the axis in the charts is based on Poll 1 rather than on both polls, but I can widen the parameters of the axis values for poll 1 which should help a bit – but probably won’t completely solve the problem.
I could build a thing that let’s people put in values (like under Poll 1), but which returns the probability that the true value is between those inputs taking into account the change in the poll results between Poll 1 and Poll 2?
The significance of the change is a good idea – might do it as mean change, upper limit and lower limit, and significance – if it will all fit.