# How to Use Statistics to Understand Poll Results

The Math Dude: Quick & Dirty Tips to Make Math Simpler

Math Dude available on QuickAndDirtyTips.com " data-pin-do="buttonBookmark">

More episodes of the Math Dude available on QuickAndDirtyTips.com Image: Quick & Dirty Tips

Scientific American presents Math Dude by Quick & Dirty Tips. Scientific American and Quick & Dirty Tips are both Macmillan companies.

In today’s article, we’re wrapping-up our introductory series on fundamental statistics by talking about how knowledge of statistical quantities like the mean and standard deviation can help you understand the significance of the latest political polling results.

Should You Believe the Results of All Political Polls?
Should you believe the results of every political poll you see reported in the news? The short and simple answer is: “no.” For me, there are two reasons for this. First, being skeptically natured, I tend not to believe a lot of what I see until I can verify it for myself. Some people have agendas that they’d like to steer you and me towards, and I generally try to ensure that that doesn’t happen to me blindly. The second reason is a bit less conspiratorial in nature: even though most polls are conducted properly, the results are often reported improperly. That usually has to do with not understanding the statistical nature of the poll on the part of the reporter. But, since you’ve learned how to calculate mean values and can answer the question what are the range and standard deviation? you now know everything you need to decipher poll results and to decide for yourself whether or not you believe them.

Why and How are Polls Conducted
Let’s start by talking briefly about how polls are conducted and why they’re taken in the first place. Polls are used to figure out the opinions and preferences of the entire population without having to ask every single person what they think. In other words, the goal is to poll a subset of the entire population (this subset is called a sample) and come up with an answer that is representative of what the population as a whole believes. The most important factor in creating an accurate poll is to come up with a sample that represents the diversity of the entire population. It must be chosen carefully so as not to overrepresent any one group.

View
1. 1. zyzzyva57 12:32 PM 4/24/13

How is Margin of Error determined?

2. 2. SteveO in reply to N a g n o s t i c 06:27 PM 4/25/13

N a g n o s t i c, you are not correct. Or rather you are kind of correct, but for the wrong reasons. :) In the real world, samples are far more accurate than attempts at total population counts. (Compare the Census' actual count versus its sample results - no one in their right mind would use actual counts for policy decisions. The Founding Fathers didn't really get statistics...)

The real trick with political polls has nothing to do with statistics, but with predictive models that use those statistics. In the case of political polls, imagine a case where each pollster gets exactly the same statistics. Each one could very well predict a totally different outcome because they are using different models to try to predict how the people who actually answered their survey will vote (or whether they will vote at all) on Election Day.

That is one big place where hidden biases creep in. (Question design, sampling plans, etc are other areas.) A Republican pollster could look at the data and say, "Well in my expert opinion, the Republican base is riled up, so I am going to arbitrarily decide to assign self-identified conservative Republicans a higher probability of voting then a self-identified liberal Democrat." His Democratic colleague might say, "Well I think that my landline poll under-represents Democrats, since they are more likely to have no landline and just a cell phone, so I'll add a couple of points to the actual results to account for that."

Now of course that is a gross simplification of how those response and voter turnout models work, but you can easily see how personal biases of the researcher can easily affect *predictions* from the same exact response statistics.

Nate Silver used a bunch of probabilistic models to use aggregate poll results to make a pretty accurate prediction. That too is less about statistics than about modeling. Predicting highly non-linear stochastic systems (e.g. humans) is, so far, not very much like inferential statistics at all.

zyzzyva57, so-called "margins of error" are related to the statistical term "confidence interval." Basically, you can rephrase these something like this: "My best guess as to what percentage of people are going to vote for Romney is 52%, but I didn't sample every single person, so I acknowledge that the *real* percentage could be different. In fact in looking at my data and my sample size, I am 95% confident that the real percentage is between 50 and 54%." How you arrive at the actual number is explained adequately well in Wikipedia.

3. 3. gmperkins 05:23 PM 4/27/13

You are basically/essentially correct but the main problem is that the typical person has such a poor understanding of statitistics. Even trained 'scientists' misuse statistics on a regular basis. I feel like before anyone can post/write/publish anything supported by statistics that it needs to be reviewed by statisticians. Clearly costly/time consuming but almost worth it due to the constant abuse I see almost daily.

Next comes 'proper use' but 'improper conclusions'. But that is a whole nuther pie to cut up.

You must sign in or register as a ScientificAmerican.com member to submit a comment.
Click one of the buttons below to register using an existing Social Account.

## More from Scientific American

• News | 5 hours ago | 2

### Bacterium Reverses Autism-Like Behavior in Mice

• News | 5 hours ago

### Dyslexia Linked to Brain Communication Breakdown

• TechMediaNetwork | 6 hours ago

### Sharks Do Get Cancer: Tumor Found in Great White

• Extinction Countdown | 9 hours ago

### Can You Guess Which Country Has the Most Endangered Species?

• News | 10 hours ago

More »

## Latest from SA Blog Network

• ### Bill Nye's Open Letter to Barack Obama

PsiVid | 3 hours ago
• ### You ever dance with the Devil's Fossils...

History of Geology | 7 hours ago
• ### The Formula for Kick Starting U.S. Manufacturing Begins with Technology

Observations | 8 hours ago
• ### Can You Guess Which Country Has the Most Endangered Species?

Extinction Countdown | 9 hours ago
• ### Advert Informs that Dmitri Mendeleev Knew the Science of Perfect Vodka

PsiVid | 12 hours ago

## Science Jobs of the Week

How to Use Statistics to Understand Poll Results

X

Give a 1 year subscription as low as \$14.99

X

X

###### Welcome, . Do you have an existing ScientificAmerican.com account?

No, I would like to create a new account with my profile information.

X

Are you sure?

X