Automatic versus manual data analysis

Posted to Statistics  |  Tags:  |  Nathan Yau

Hilary and Roger touch on some interesting topics in the most recent Not So Standard Deviations, specifically on scalable and automated data analysis.

At the surface, it can seem like computers should be able to do the bulk of any analysis. Plug in the data, crunch the numbers in an algorithmic black box, and presto change-o you get a list of actionable insights. From that point-of-view, you should be able to build software that does almost everything for you. That’s almost never the case, and you realize it quickly once you dig into the data yourself.

It’s the same deal with visualization.

You see the end result, and it’s easy to imagine applying the same chart to another dataset. Geometry and color are easy to make with a couple lines of code. The chart should be generalizable, right? Sure, but the challenge is getting to that final point. There are various paths you can take when you start with a dataset — what it means, the questions you want to ask — along with various decisions along the way.

Automating the process. That’s the hard part.


The Changing American Diet

See what we ate on an average day, for the past several decades.

Causes of Death

There are many ways to die. Cancer. Infection. Mental. External. This is how different groups of people died over the past 10 years, visualized by age.

A Day in the Life of Americans

I wanted to see how daily patterns emerge at the individual level and how a person’s entire day plays out. So I simulated 1,000 of them.

How You Will Die

So far we’ve seen when you will die and how other people tend to die. Now let’s put the two together to see how and when you will die, given your sex, race, and age.