Many websites compile averages of public polling. However, the FiveThirtyEight.com averages distinguish themselves in several ways, which are designed to provide for a more scientific outlook on polling and election outcomes.

Firstly, FiveThirtyEight.com does not treat all polls equally. Instead, polls are weighted, with more reliable polls having more influence on the averages. The polls are weighted based on three factors:

(i) The pollster’s historical track record of accuracy. We determine the amount of error that particular pollsters introduce based on a rigorous analysis of more than 160 elections dating back to 2000. For a fuller description of our methodology, click here, and for our current pollster ratings click here. Rasmussen Reports is presently rated third out of 32 public pollsters.

(ii) The poll’s sample size. Larger sample sizes are not a cure-all for poor methodology. In fact, polls begin to encounter diminishing returns fairly quickly as they increase their sample size. All else being equal, however, a larger sample will give you a more reliable result, and so larger sample sizes are weighted more heavily by an appropriate amount.

(iii) The recentness of the poll. Instead of arbitrarily deciding which polls qualify as “recent”, our methodology instead assigns an exponential decay function to each poll, such the reliability of a poll will decrease gradually with each passing day.

The second way in which the FiveThirtyEight.com averages are distinct is that we combine the polling results with a demographic model based on regression analysis. This analysis considers as many as 16 variables such as race, religion, age, socioeconomic status, and the partisan affiliation of each state. In states where there is abundant polling data, the regression analysis will receive relatively little weight. However, in states that have not been polled for a long time, the regression analysis may be the primary component in the polling averages. The regression estimates are updated continuously based on any new polling data that comes in.

Finally, FiveThirtyEight.com simulates the election 10,000 times each day based on the current polling averages. The simulation recognizes that the amount of uncertainty is higher the further we are from Election Day Months before an election, a state may be considered a toss-up if a candidate holds a 5-point lead in a given state. If the same 5-point lead were held on the morning of the election, however, the candidate’s odds of winning that state would be very good. In addition, the simulations recognize that the polling in one state is not necessarily independent of the polling in other states. Oftentimes, if a candidate receives a bounce in his polling, it will be reflected nationwide.. For a fuller discussion of our simulation methodology, please see the FiveThirtyEight.com.

FiveThirtyEight.com is not affiliated with Rasmussen Reports. We have been asked to provide our polling averages as a service to Rasmussen Reports. For any additional questions and inquiries, we may be reached: mailto:here538dotcom@gmail.com.