After posting my recent blog on **hypothesis testing** I received a request to write about regression in a similar manner. As I am always focused on the “voice of the customer” I will take on regression tonight. This will be part 1 of this series.

**Many Faces of Regression**

First of all when the word regression is used it can mean a few things. There are different types of regression such as simple linear regression, non linear regression, and logistic regression.

Don’t sweat it though. The statistics geeks may try to confuse you with all their fancy words and formulas but in the end the concepts are quite simple. If knuckle heads like me can understand it, anyone can.

**Simple Linear Regression**

Now then, let us focus in on simple linear regression this evening. We use linear regression when we have a variable input (X) and variable output (Y) and want to see if there is any correlation between the two.

For example, say you wanted to know if there was any relationship between the temperature (variable data) of your injection molding machine and the weight (variable data) of the part it produced.

To test this, you could simply produce some parts at different temperatures, starting at lower temps and working yourself to higher temps. Once the parts were made you could weigh them (on an approved measurement system of course).

With this data, you could perform regression analysis to determine how strong the relationship is. As always, a nice software package like **SigmaXL** makes this a snap but you can also program spreadsheets to do the math for you.

**Make Predictions**

The coolest thing about regression is the fact that we can build statistical models with the data which allows us to predict what the output (Y) will do given a specific input (X).

In part 2 of this series on regression I will introduce a few more tricks such as R-sq, fits, and residuals. Stay tuned!

**Subscribe to LSS Academy**

If you enjoyed this article please consider subscribing to our **full feed RSS**. You can also **subscribe by email** and have new articles sent directly to your inbox.

## 6 Comments

## robert

March 4, 2007 - 11:52 pmRon – perhaps you’d like to comment on sample sizes? The sensitivity of the correlation coefficient to random variations should be considered. With small samples chance variations can easily give the appearance of correlation when in fact none exists. I often find that data is collected, then I’m asked to analyse it. I often dissapoint when I suggest that more is required, especially when it may be difficult or expensive to acquire.

Rob

## Ron Pereira

March 5, 2007 - 8:41 amGood idea Rob. Thanks!