BookmarkSubscribeSubscribe to RSS Feed

Experimental Design for Various Advertisment concepts

A while back, I read an article in Harvard Business Review about how a simple experimental design can be built to identify the most important attributes in a web banner. The design discussed used an experimental design and was focuses finding the design that would provide the highest ROI.

As an example, a company might run a few banner advertisments that have several different colors, several different placements, several different sizes, several different promotional messages, and have several different costs (due to the placements). After a few experimental advertisements, the company was able to determine the most optimal strategy.

How did they do this?
JMP has the capability to create designs of experiments (DOEs) which are fundamentally intelligent ways of picking which experiments to run to get the greatest amount of information out of the fewest number of tests. In your case, the ROI of the ads is a function of all of those different factors you listed. You could try every combination of colors, placements, sizes, etc., but depending on how many different settings there are for each of those variables, there could be thousands of combinations. A DOE can cut down the number of combinations needed if you can assume certain things about how those variables interact to result in a given ROI. So as long as you have a way to change these website banner layouts and a way to measure the ROI, JMP can help you generate the set of ads you need to run.

There is a DOE guide included with the reference materials in JMP that has a lot of information. You can also try searching for "design of experiments" online to learn more about them.

Community Trekker


Jun 23, 2011

If you can get a group of volunteers or a focus group, you may also be interested in JMP's choice experiment platform new to JMP 8. I believe it does exactly what you're describing, and allows you to ask each person a small subset of the choices to get a good result.

There was a blog post about it awhile back:
The first option under DOE is Custom Design. This would be simple using categorical factors. There is a tutorial for the custom designer under Help>Tutorials>DOE Tutorial



Jun 23, 2011

There is a recent book published in 2007 called Testing 1-2-3 Experimental Design with Applications in Marketing and Service Operations by Johannes Ledolter and Arthur J. Swersey. You may want to check it out.