## Regression Trees

While multiple regression is a great tool for analyzing data, most clients (sadly) have a hard time “getting” it.  Over the years of working with corporate executives I have found that regression trees are much easier for non-stat people to understand.

Below is a tree from SYSTAT that was created from data collected by employees in annual employee satisfaction study.  Putting the tree together and making it “pretty” helped C-level executives understand what we needed to work on.  Please be sure you transform your data before running your analtyics.  Employee satisfaction data nearly always has negative skewness.

Employee satisfaction is a fascinating topic.  A decent introduction and review  can be found in this book:

Here’s a great book to get ideas on conveying data visually

## Report Generation

In market research I would very frequently have to crank out a ton of different versions of reports for a given study.  In the corporate world, like it or not, PowerPoint is still king.  🙁   By building a generic template I was able to drastically speed-up my report generation.

Setting up the template takes a little time at first however, in the long run, it will save you an amazing amount of time!

There is a lot more I automated however having a flexible template is always a great place to start!

## Psychological affects of the slope

I’ve been on a (much needed) diet since mid January.  While my progress is going really well, I was looking at the graph I’d made and was surprised how much my attitude changed when I simply changed the ratio of one axis.  I have been working in stats for ~20 years and I’m well aware of how people purposely try and lie/mislead with graphs so I didn’t think changing the ratio would affect my mood but damn was I wrong!

In the below two graphs they both have the exact same slope & R-square however, the one the left is wider (making it seem like my weight-loss is going slower), than the one on the right.

Which graph would you rather be using to see your progress?   And, yes, I know I truncated the graph at 180.  While weight is a “scale” variable (excuse the pun) I have an ideal weight I’m aiming for.  Besides, nobody wan’ts to see me at 100 lbs.

Psychological affects of the slope

This experiment reminded me of Dan Ariely’s books. This one is exceptionally well written and thought provoking.

## Correlation Analysis

A few years back I fielded a national study on the frequency of visiting restaurant chains.  In SPSS I did correlation analysis however, as is always the case, showing a correlation matrix to the client is not an option (if I want them to understand anything.)

So I took the matrix and imported into SYSTAT then computed an Additive tree.  Additive trees examine the response patterns across variables and group them, according to their similarities, in the shape of a two-dimensional tree. The closer items appear to each other, the higher the correlation between them.  (The color coding is subjective and is just added to aide interpretation)

The Additive Tree below is much easier to evaluate as clear patterns can be seen in how consumers “see” the chains.

Here’s a great book to get ideas on conveying data visually

And here is a book utilizing SYSTAT by one of the programmers. I learned the vast majority of my statistics from this great book!