Monday, November 21, 2011

Understanding political polls: margin of error

The poll showed 46 percent supporting Romney and 41 percent backing Obama if the 2012 presidential election were held now. The difference is just past the poll's 4 percent margin of sampling error. Thirteen percent are undecided

This is incorrect. In fact, Obama and Romney would be statistically tied.

Proper use of Margin in error in political horse race

For each direction, there is a 4% margin of error.  This means that if the poll were taken 100 timeswith the same sample:

Obama's support would be 37-45
Romney's support would be 42-50

95% of the time.

5% of the time, the results would be something even more different.( outliers)

So the same poll could be done tomorrow and the result come up Obama 45 Romney 42.
Or it could be Romney 50 Obama 37.

In politics, we like to write stories about momentum.  Also, we want a definitive storyline.
But due to money restrictions, custom, and simply not wanting to deal with reality, polling organizations don't constantly poll the same population regularly.

Now, if 10 straight polls with a 4% margin of error show Romney with a 46-41 lead, then there would be reason to make the above statement because the margin of error would be much smaller.

Guess when do pollsters actually sample the same population daily? Tracking polls usually a week or so from the election.  Guess what they also do? They average the day's results with the past 2 or 3 days results.  Because if they didn't, they'd report a wildly changing number that people wouldn't believe.  Remember that when you see other polls.

No comments: