You are viewing a single comment's thread.

view the rest of the comments →


[–] Mathurin1911 0 points 2 points (+2|-0) ago 

Meh, statistical adjusting is complicated. Its part of why I dont trust the more complicated statistical measures, but adjustment can make sense in some scenarios.

Think about it like this. Lets say you sample 1000 people on their enjoyment of waving sticks at fences to keep kids off their lawn. You find that 45% of those 1000 people very much enjoyed the activity. However, you also find that you somehow managed to accidently select a sample that was 65% over the age of 80. When using the sample you might find it reasonable to weight the choices of the sample subjects to make the outcome more representative of the population. IE, placing greater weight on the younger subjects to make the poll better reflect what it would have been if you hadnt gone to a retirement village to conduct your poll.

Why not just rerun the sample somewhere else, somewhere more representative of the whole population? Because sampling is the expensive part.

This is not a defense of tactics used to generate polls, just one of the idea of adjustment. I personally abhor the "Fun with numbers" tactics that all politicians seem happy to use.


[–] 1HepCat 0 points 2 points (+2|-0) ago 

In my work, we usually call it 'smoothing'.

Given a population of 1000, pollsters A, B, C, D and E each sample 50 people at random. Pollster E finds results that differ somewhat significantly from the findings of A, B, C and D so he reduces confidence in his own sample because he's only got 50 data points whereas the other polls have ~200 data points combined. This might be reasonable because the relative weakness of E's sample makes it more sensitive to outliers.