Why are polls always so off? How to read Political Polls!
0I decided to explain my process for predicting winners in political elections in battleground states. It's a process but takes effort. This first method only works when there are competing polls that favor either individual (scroll to the bottom for notes on ties and sweeps):
If you ever wondered why so many reporters FAIL at reading polls correctly, I explain why. They aren't educated in reading polls.
The process can be boiled down to the following rules:
1. Do not trust any individual polling agency that has wild results week from week, unless there is a major incident and numerous agencies swing as wildly.
2. Look for and trust polls which seem to remain steady, unless there is a major incident and numerous agencies swing as wildly.
3. Adjust or ignore the values of every poll which has a hard leaning factor.
4. Ensure the polling companies properly segregate the polling information and respondents - yes segregation in social engineering is a bad word, but in marketing it is a fact on how to read different parts of the population and their total voting power in an election.
5. Know the voting power and opinions of the unpolled electorate in fraction format. 1/2, 2/3rds, 3/5ths and 3/4ths.
6. To measure most recent accurate polling preduction Use this formula:
Add together all of the most similar most recent polls - minimum of 3 sources.
Calculate the differences for each candidate.
Add the fractional values of the unpolled electorate after you decimalize them. Wisconsin unpolled electorate historically had a 2/3rds majority favoring Biden based on past information.
Compare the final results.
In my example the most recent polling would be a total questionable of 8.
Biden has 3 (questionable).
Trump has 5.
Biden has a .375 share
Trump has a .625
Calculate the unpolled electorate into the averaged polled electorate:
Trump's .62 +.33 (drop the highest thousandths place .005 and drop the other thousandths place if it's below 5 for simpler math) *
Biden's .375 + .66 (drop the highest thousandths place .006 for simpler math, the remaining thousandths place is 5, so keep it) *
*Thousandths places are highly unreliable. This tends to give a more accurate number.
Biden has 1.035
Trump has .95
1.035
- 0.95
=====
0.085
Difference 8.5%
Even if you retain the thousandths place:
Biden at 1.041
Trump at 0.958
=============
Difference 8.3%
For polls that are virtual ties, or for polls that solely have 1 candidate leading, the ONLY thing you CAN do is examine the unpolled electorate.
"Dewey Beats Truman" is a primary example of the Press being too incompetent to understand the unpolled electorate where the unpolled electorate massively favored Truman over the polled electorate's favorite of Dewey.
As long as the polls are right on, and the unpolled electorate are properly evaluated with the polled electorate, there should be no issues in predicting a winner.
For ties - well sometimes the unpolled electorate is a tie in and of itself. Do not make predictions on polls with ties unless you are certain the unpolled electorate have a history of favoring one candidate over another.