There’s a few things that are significant when it comes to polling that people don’t understand.Polls are conducted by statisticians with the intention of being interpreted by people with some background in statistics, but as the number of firms conducting polling increased and the number of people following the “horse race” part of politics also increased, there’s been a lot of misunderstanding regarding interpretation.
So in essence, you’ve got it wrong.
If you think that the President has re-election sewn up because of some of the latest numbers, you’re way off the mark.
Here’s what you need to know about polling, and here’s why your candidate may be doing way better — or worse — than you think:
Margin of Error
Here’s the part most people miss, and even others fail to report. The Margin of Error is a built in range with the poll. Most people believe that when a candidate is at 48 points in the polls, the candidate is exactly at 48% in the state. That’s inaccurate.
There is error in the poll by statistical default, and if the announced margin of error is, for example ±4, then that means that the pollster is very sure that — in the real world — between 44% and 52% of the people actually support the candidate.
This is obviously enough to push a candidate over or under the threshold for victory, and means that even with a six point lead in a poll with a margin of error of ±4 a favoured candidate still may very well lose or landslide.
One more thing on this possible range: The range — in our example, 44% to 52% — is normally distributed, so the likelihood that a poll is four points off is much lower than the likelihood it is a little bit off.
There are two kinds of bias you need to know about to understand polling.
Photo: Eric Thayer / Getty Images
The first is more akin to in-house effects, or the partisan skew that is consistently built into some polls. These firms don’t need to be listed here, but they are on both sides of the aisle and I’m sure the commenters will be kind enough to point out a few.One way that in-house effect can be used to learn about the real world is that, while the polls may not be accurate (i.e. correct) they may be precise (a consistent and correctable amount of wrong).
The other kind of bias is statistical bias, and to understand that you have to understand how polls are carried out. When a pollster sets out to do a poll, the pollster does not say to herself “I’ll get 50 Democrats and 20 Independents and 40 Republicans.”
Instead, the poll contacts a random sample, usually drawn from a phone bank, with a random geographic distribution in the desired region. They then ask the person who picks up the phone their party affiliation.
So a poll with 40% Democrats and 30% Republicans (usually) isn’t cooking its books, it just so happened that either (a) more Democrats picked up the phone or (b) The state has inherently more Democrats or even (c) that Republican voters are identifying as independents more, or (d) the pollster didn’t get a statistically large enough sample.
One of the most important parts: Sampling
The third and final thing that most people misunderstand about polling is that all polls are not created equal.
Some polls contact people via cell phone as well as by landline. Some are carried out on the internet. Others, even, are pooled from volunteers, while others are sampled randomly. The way a poll is carried out has measurable effects on the outcome.
When a poll is strictly land line, it skews older and traditionally to the right. When a poll is mobile phone only, the opposite effect is seen.
This is why many firms try a hybrid model, calling both cell phones and landlines. Internet polls are either entirely useless or require a large grain of salt at the moment, but the medium is rapidly maturing as a source of credible polls.
Even more, look at where the final number comes from. If it’s from registered voters, that is better than “adults of voting age” but is still imperfect.
A poll that reports Likely voters — as in, people who said “yes” when asked if they were planning to vote — is a much more compelling picture of how election day will go.
Polls now are switching to likely voter reports at this point in the cycle, and it’s worth noting that Republicans get an edge when likely voters are polled rather than merely registered voters.
Here’s why you need to know all of this:
Any way you cut it, polls are going to change, and it’s really easy to see how you could make a compelling case for either candidate based on recent polling — Therefore Romney is undervalued.
The polls we’ve seen for the past few weeks have been almost exclusively registered voter polls, and polling firms are just starting to ask about likely voters. Mitt Romney should expect a bump from that, maybe in the area of one to two points. That’s the difference between a victory and a loss in a few states.
Even more, the thesis that Democrats are oversampled has validity, and the cause has to be investigated. It’s almost certainly not because of across the board deliberate house effects, but something about either the method of sampling (landlines, cell phones) is off a bit or Republican voters outside of the mainstream, particularly the Tea Party and Libertarian wings, are increasingly identifying themselves as “independent” to pollsters for some reason.
Plus, Romney is well within the margin of error in several states. So it’s statistically possible that he’s ahead and we just haven’t seen it yet.
Either way, oversampling democrats mean that the reality is more in Romney’s camp.
In short: stop freaking out if your candidate is a little behind, and resume freaking out if your candidate isn’t that far ahead in one of these polls.
NOW WATCH: Briefing videos
Business Insider Emails & Alerts
Site highlights each day to your inbox.