Real Clear Politics is my favorite place to go to look up how candidates are doing in the nation and in Iowa. They list a number of polling companies’ results and then provide their own average, but upon closer inspection, the polls that they present are not very straightforward. Take a look at their latest results in Iowa, for example:
|Iowa Democratic Caucus|
Thursday, January 3 | Delegates at Stake: 45
|RCP Average||12/20 to 12/28||–||29.8||26.5||26.3||5.3||4.8||Clinton +3.3|
|American Res. Group||12/26 – 12/28||600 LV||31||24||24||5||5||Clinton +7.0|
|Strategic Vision (R)||12/26 – 12/27||600 LV||29||28||30||2||5||Obama +1.0|
|Quad City Times||12/26 – 12/27||500 LV||28||29||29||7||3||Tie|
|LA Times/Bloomberg||12/20 – 12/26||389 LV||31||25||22||7||6||Clinton +6.0|
These results can be grouped into two pairs – one that shows a dead heat, and another that shows a significant lead for Clinton.
Also note that the spread of Clinton’s results is only 3% (28 is the lowest, 31 the highest). Edwards has a spread of 5%, and Obama has a spread of 8%, from 22 to 30. I can only guess that this would tell us that Clinton is most likely to finish where the polls predict, but that the other candidates’ results will be much harder to gauge.
Also, I’d agree that the RCP averages split the difference and reduce the risk of being totally wrong, but when 4 polling outfits return 2 types of results, you have to wonder at their methods and accuracy. I think its more likely that the truth lies at one of the two results than between them.
For more on polling and its statistical complications, including the effect of margins of error, the normal distribution, and sample size, check this primer.