All language subtitles for 06 - Identify and minimize sources of error

- [Instructor] Statistical analysis might seem like

an exact science, but if you've ever tried

to apply statistics to real life,

you know that, in fact, it is not.

In this movie, I'd like to review

a number of potential sources of error

that might creep into your own analyses,

because once you know what they are,

you can do your best to avoid them.

Not using random samples can be a huge source of error.

One famous example comes from the 1936

U.S. Presidential election,

where a telephone poll of "Literary Digest" subscribers

projected that Alfred Landon

would beat Franklin Delano Roosevelt by a wide margin.

In fact, Roosevelt won nearly 2/3 of the popular vote.

The error came from two sources.

"Literary Digest" was a conservative publication,

which biased the results,

and the poll was conducted by telephone.

In 1936, only the financially well-off

had telephones in their homes,

so that biased the results, as well.

You can also run into investigator bias.

It's easy to anticipate what your data will tell you,

that's normal, but you shouldn't let those expectations

affect your judgment.

Many interesting discoveries come from the moment

when you look at your data and think, that's strange,

because the results don't fit your preconceived notions.

You can also run into trouble working with old data.

The world changes and just because your customers

might have been ready for a product two years ago,

doesn't mean they are now.

And finally, you can run into trouble basing your policy

on a survey or experiment with a small sample.

The more data you can get, the better,

and the greater variety of people that you ask,

again, selected randomly, the better your analysis will be.

Hopefully, pointing out these sources of error

will help you in your own analysis

and you can use random data,

eliminate your own personal bias,

work with the newest data that you have,

and have an adequate sample size.