Skew

Skew, in layman’s terms, means that data is distorted. The data points don’t fall evenly around the center of a distribution.

Consider this example. Assume ten people are in a room, and you want to know what their average household income is. If this was a typical cross section of America, the number would be $73,298, as of 2014 US Census data.

Now imagine Microsoft’s Bill Gates was one of those people. He would skew the results. The average income would skyrocket by a few hundred million.

In statistical terms, “skew” has a specific meaning. A skewed distribution has a longer tail on one side of distribution curve. A positive skew has a longer right tail; a negative skew has a longer left tail. Skewness is the measure of the asymmetry of the distribution.

For laypeople, though, most people use the term “skewed” to mean that the data is distorted or has outliers. Simply identifying that the data has a skew to it is often enough to get an observer to dive deeper into the problem. Identifying that asymmetry when an even result is expected is a red flag. The obvious next step is to dive into the root cause analysis to identify the factors that skewed the output.

Time Studies and Skewness

Not all data will be symmetrical in manufacturing. You might see symmetry in the data describing a hole size, but in Lean, you are often looking at time studies.

Those hardly ever show symmetry. That’s because when things go right, there is very little fluctuation, but when things go wrong, the right tail grows. There is much more likelihood of something extending the time than reducing it, and the size of those deviations in time is much greater when things go wrong than when they go better than expected.

Much of the work in Lean is devoted to shortening that long tail and getting more consistency in the time a process takes to complete.