I spend a lot of time dissecting polls. It is sort of a nerd hobby of mine. I am not alone, but there are a lot of people who feel their brain start to swell and their eyes glaze over when they contemplate the demographics and “internals” within the polls.
I will try to give you an easy primer as to what the polls mean, how they work and what things are good and what things are bad.
First, there are several important things to know about polling. Some polls are online, but most actually call people. Many try to call both land line and cell phones to get a good sample. And they ask questions to determine what demographic you may fall into.
The goal of a good poll is to measure the opinion of the people and be as accurate as possible. Unfortunately, there are also polls out there that seek to influence opinion rather than measure it. And there are a lot of them.
I am going to try to help you understand how you can tell the difference. There are a lot of factors that can increase accuracy of polls, and a lot that can “skew” the poll towards one outcome or another.
I am not going to get too deep in the mechanics of this, so this post is obviously not intended for political junkies and math majors who absolutely love this stuff. So the “experts” will find this boring.
And I am going to take a look at real data from today and show you how you can easily learn to decipher the polls.
Below is a snapshot of the polling as it stands this morning at Real Clear Politics:
As you can see, the average of the polls taken between July 5 and July 20 shows Clinton with a +2.7% lead. But what you may not know is that all polls are not created equal. And while averaging the polls is probably a better measure than looking at each poll individually, I tend to get a better feel by looking at the polls individually.
One of the first things a pollster has to do is to figure out their sample demographics. For instance, if you wanted to take a poll in a city to determine which school was the best in the opinion of the students, you wouldn’t want to ask just the students at one school. You would want a sample of all of the schools to get an accurate measure. On the other hand, if you wanted to make sure that one school came out on top, you would poll more people at one school. (Hint – that poll is worthless.)
So how do you get a good measure of the status of the race between Donald Trump and Hillary Clinton? Well, you need to use some sort of basic formula to make sure you poll a good sample of all the people that will be voting. Obviously, most Democrats will say Hillary and most Republicans will say Trump. So how do you decide what ratio of Republicans, Democrats and Independents you use? Well, a lot of it is guess work. Most times you can look at the last presidential election and use a mix that actually happened then.
So if we look at 2012, 38% of the voters were Democrats, 32% were Republicans and 29% were Independents. And even if we go back to 2008, the R’s, D’s and I’s voted in almost the same ratio, so most pollsters should be using a mix similar to that if the turnout was expected to be similar. However, logic dictates that the enthusiasm for Hillary is not going to be anywhere near what Barack Obama saw. In 2004 blacks were 11% of the electorate. In 2008 and 2012, they were 13%. It is doubtful blacks will turn out for Hillary like they did for Obama. Twice. And it is also doubtful Democrats will turn out like 2012 or 2008. In 2004, the Bush – Kerry election, Democrats and Republicans made up 37% each. We are most likely looking at something closer to 2004 than 2008 or 2012. Sorry for all the numbers, but it is important to look at what happened in the past. And we all know Obama was a phenomenon unlike any other recently. Love him or hate him he got people to the polls to vote for him. That isn’t going to happen this year.
So any poll that samples far more Democrats than Republicans is inflating Hillary’s numbers. But this appears to be case in the majority of the polls. At least the ones that show their demographics, which is another problem with polls.
But let’s look at the RCP average above. The first column is the poll is the polling outfit that took the poll. If you click on the pollster (click here for the actual RCP page) you can sometimes find the demographics of the poll, but you have to look very hard with many of them. Presently, Rasmussen is the first poll and shows Trump with a 1% lead in the general election. Rasmussen only allows Platinum (paid) members access to the demographics. So unless you join, you can’t see that the Rasmussen poll questioned 33% Republicans, 37% Democrats and 30% Independents. So it looks like Rasmussen has subtracted a point from the 2012 Democrat turnout and added it to the 2012 Republican turnout. This reflects the lack of enthusiasm we are seeing with Democrats and the increased enthusiasm we are seeing among Republicans. So 2012 was +6% Democrats and Rasmussen is using +4% Democrats for 2016.
The next poll is Reuters/Ipsos. If you click on the second column you will see the numbers of D’s, R’s and I’s. I did the math (so you don’t have to) and the poll questioned 48% Democrats, 37% Republicans and 15% Independents. So we can see that this poll samples 11% more Democrats than Republicans, predicting almost double the 6% turnout ratio for Hillary that Obama actually received. So naturally, the results are going to be heavily in favor of the Democrats. Does anyone actually believe that Hillary is going to get a larger turnout than Obama? So when Rasmussen samples 4% more Democrats then Republicans they find Trump with a 1% lead. And Reuters adds 7% more Democrats and has a 4% lead for Clinton.
So let’s look at another poll with a 4% lead for Hillary. Go down to the Economist/YouGov poll. There you will find they sampled 35% Democrats, 25% Republicans and 40% Independents. So this one is Democrats +10 (4% more than Obama received) and shows Hillary with a 4% lead.
So are you beginning to see the way the polls work?
Low and behold, the more Democrats you have in the sample, the more Hillary leads.
Some other factors to understand. After the date the table shows the sample size. The larger the sample, the more accurate the poll. the next column is MoE or Margin of Error. You will notice the error decreases as the samples get larger. Of course, a lot of other factors come into play. But a margin of error simply means that they are plus or minus the MoE number. For instance, the Economist/YouGov poll has a MoE of 4.5%. What that means is that even though they show a +4% lead for Clinton, 45% to 41% for Trump, the numbers could also be 45.5% Trump and 41.5% Clinton, 4 point lead for Trump. (That would be best case for Trump.) And that is due pretty much entirely to the number of people polled.
Also notice beside the sample size there are two letters. RV and LV. RV is a registered voter and LV stands for Likely Voter. They ask questions to determine your likelihood of voting in this election. Some ask you directly, others ask questions to determine your voting pattern. You will sometimes see the letter “A”. This means “All” people. Whoever answered the phone basically.
Polling Registered voters is considered less accurate than polling the ones that will most likely vote. I also notice that polling Registered Voters seems to skew more towards Democrats by a point or two. We only have 2 showing LV’s and they show either a tie or +1 Trump. In order to get a good sample size (I like 1000+) you have to call a lot more people because the ones not likely to vote are tossed out. And that is more expensive. Usually around Labor Day, when more people start paying attention, the polls will shift from RV’s to LV’s. And as LV’s tend to move towards the Republicans, you will also see the +10 and +11 Democrat polls shrink down to something more realistic. The media often reports that the polls are “tightening” the closer we get to election day. But the truth is, the pollsters know that they will never be called out for a bad poll weeks away form the election and that you are judged on the last poll or two of every election.
The first week of November the polls will start getting pretty accurate. Funny how that happens.
So when a poll shows a big lead for the Democrat, the only way that will actually happen is if the Democrats register a lot of dead people and illegal aliens. Which does happen. But hopefully the voter fraud will be minimal and the American people will decide.
So take the polls with a grain of salt. They are not always measuring opinion, they are sometimes seeking to shape opinion. And quite a few of the pollsters are Democrats.
A word about “weighting”. Polls sometimes use formulae to correct the demographics. That way they can “fix” the poll if they sample too many of one demographic or another. There are not too many polls that are open with this “art”. In fact, there are a lot of polls that only give the results and you have no idea what they did. If you read a poll and it does not tell you exactly how they manipulated the numbers, I really can’t rely on that poll.