Imagine my surprise when reading the March 28, 2016 issue of BusinessWeek and stumbling across the article titled “Lies, Damned Lies, and More Statistics.” In the article, "BusinessWeek" warned readers to beware of “p-hacking,” which is the statistical practice of tweaking data in ways that generate low p-values but actually undermine the test (see p-value definition below). One of the results of “p-hacking” is that absurd results can be made to pass the p-value test, and important findings can be overlooked. For example…

A study from the Pennington Biomedical Research Center in Baton Rouge[1] followed 17,000 Canadians over 12 years and found that those who sat for most of the day were 54% more likely to die of heart attacks that those that didn’t.

54%!? Yikes, that’s a scary fact. Proof that sitting kills you by heart attack. As a person who spends a lot of time sitting behind a desk, or on an airplane, or at sporting events, this “54% more likely to die of heart attacks” fact is very concerning.  Can I cheat certain death by throwing out my current desk and buying one of those expensive “stand up” work desks? Sounds like a bargain.

But the BusinessWeek article concludes with the statement “… hold findings to a higher standard if they conflict with common sense.” Bottom-line: think!

Unfortunately, people have a tendency to blindly trust a claim from any source that they deem credible, even if it completely conflicts with their own experiences, or common sense.

It only takes a couple stats and lack of common sense to make a dangerous conclusion and claim it’s a fact. It’s harder to buy a gun in Illinois than most other states. Gun-related murder rates are higher in Illinois than most other states. So… we can conclude that stiffer gun laws cause murder. Right? No, we can’t conclude that from those stats.

But I started to think, and challenge the assumption that there is some sort of causality between sitting and heart attacks. Some questions that immediately popped to mind included:

• Are there other variables, like lack of exercise or eating habits or age, which might be the cause of the heart attacks?
• Was a control group used to test the validity of the study results?
• Is there something about Canadians that makes them more susceptible to sitting and heart attacks?
• Who sponsored this study? Maybe the manufacturer of these new expensive “stand up and work”-type desks?

One needs to be a bit skeptical when they hear these sorts of “factoids.” We should know better than to just believe these sorts of claims blindly.

Let’s use this to remind ourselves to think before jumping to conclusion. And this is a great opportunity to employ our “thinking like a data scientist” techniques to identify what other variables might contribute to this “54% more likely to die of heart attacks” observation. In particular, this is an opportunity to test the “By Analysis” to explore what other variables we might want to consider. To perform the “By Analysis,” let’s craft the statement against which we want to apply this technique:

“I want understand details on each of the study’s participants by…”

Here are some of the variables and metrics that we could test to see if they might be predictors of heart attacks:

Upon further analysis, we start to see groupings of related variables and metrics around specific use cases including:

• Lifestyle (age, gender, Body Mass Index, cholesterol levels, blood pressure, naps, hours of sleep, etc.)
• Diet (calories, fat intake, sugar intake, alcohol, smoking, amount of fish, organic foods, etc.)
• Exercise (frequency of exercise, recency of exercise, type of exercise, level of effort, heart rate, etc.)
• Work (hours of work, days of work per week, stress levels, amount of airline travel, managing people, etc.)
• Vacation (recency, frequency, days of vacation, location of vacation, vacation activities)
• Environmental (number of sunny days, number of rainy days, number of grey days, range of temperatures, range of humidity, local traffic congestion, population density, amount of local green space, local entertainment, etc.)

And there are likely others that we would want to identify and test with respect to those variables and metrics ability to predict heart attacks.

The goal is to use these techniques to identify which metrics and variables are the best predictors of performance, and if you are doing that, you’re thinking like a data scientist!

If you are interested in learning more about this “Thinking Like A Data Scientist” process, join me at EMC World in Las Vegas (that should set off your stress alerts), on Monday, May 2, 12:00 PM – 1:00 PM. I promise not to stress you out!

Optional Reading: Hypothesis Testing, Null Hypothesis and p-values
Nerd Warning! I’m going to try to explain the p-value, but to do that I also need to explain the concepts of hypothesis testing and null hypothesis.

A hypothesis test is a statistical test to determine whether there is enough evidence in a sample of data to infer that a certain condition is true for the entire population. For example, the hypothesis to test is whether test group A had better cancer recovery results than test group B due to medication X. A hypothesis test examines two opposing hypotheses about a population: the null hypothesis and the alternative hypothesis.

The null hypothesis is the hypothesis that there is “no effect” or “no difference” between the test groups. The null hypothesis is a test that there is no significant difference between the test groups, and that any observed difference between the test groups is due to sampling or experimental error.

The alternative hypothesis (that there exists an effect or a difference between the test groups) is the hypothesis you want to be able to conclude is true.

You use a p-value to make the determination to reject the null hypothesis. If the p-value is less than or equal to the level of significance, which is a cut-off point that you define, then you can reject the null hypothesis. The smaller the p-value, the stronger the evidence to reject the null hypothesis (i.e., no significant difference between test groups) in favor of the alternative hypothesis (i.e., there is significant difference between test groups).

A common misconception is that statistical hypothesis tests are designed to select the more likely of two hypotheses. Instead, a hypothesis test only tests whether to reject the null hypothesis.

By the way, even if we “fail to reject” the null hypothesis, it does not mean the null hypothesis is true. That’s because a hypothesis test does not determine which hypothesis is true; it only assesses whether available evidence exists to reject the null hypothesis[2].

Confusing? Yea, that’s what I think as well.

The Minitab Blog was a great source for much of the above data (http://blog.minitab.com)

[1] Sources: “Cutting Daily Sitting Time to Under 3 Hours Might Extend Life by Two Years; Watching TV for Less Than 2 Hours a Day Might Add Extra 1.4 Years”, July 10, 2012 and “Sedentary behaviour and life expectancy in the USA: a cause-deleted life table analysis” BJ Open Accessible Medical Research

[2] Check out “Bewildering Things Statisticians Say: “Failure to Reject the Null Hypothesis” for more details on failing to reject the null hypothesis.

The post The “Thinking” Part of “Thinking Like A Data Scientist” appeared first on InFocus.