How To Explain DOE (Design of Experiments) To Freshman

[davis007]

I have been asked to present a lecture on DOE to a class of engineering freshman. The course is a survey of what an engineer does and the instructer is trying to include a sampling of the tools used.

My question is does anyone have any ideas how to cover the topic DOE in one 55 min class at a level that will be usefull to students that have not had any statistics?

I would like to include some class activities if possible.

I am thinking to cover.

What is a Designed experiment. - how is it differnt from just running some test, what are the goals, what are the pitfalls.
When should it be used, when should it not be used.
Some general definitions- factor, response, alising, error
And would like to finish with a class activity.

Any suggestions would be very helpfull.

[Steve Prevette]

One thing I have used before, both as a student and as an instructor is a Popcorn experiment. Orville Redenbacker popcorn usually makes a claim that they either pop a higher volume, or have fewer unpopped kernels.

You can run a DOE to test that, and can be run as a 3 factor experiment:

1. Orville vs. non-Orville (choose a single brand though)
2. Buttered vs. non-buttered
3. Two different microwaves.

[Tim Folkerts]

Personally, I would try to be as "hands on" and visual as possible when teaching. The popcorn experiment sounds good (and tasty!). I used paper helicopters once. Balsa airplanes are cheap and easy, too.

For the paper helicopter, I had students brainstorm possible factors affecting the time of flight. Sometimes an example or two to hold up helps jog ideas: width of blades, length of blades, width of base, length of base, weight of paper, color of paper, add a paperclip or not, clockwise vs conterclockwise rotation, how you drop the helicopter, .... After brainstorming, I "forced" them to settle on a particular set of factors - like 7 total. Suggest that doing all of these 2^7 = 128 trials will run over into the next hour, so we want to choose a subset to try.

Mention one-factor-at-a-time, but then suggest an alternate (a 16 trial fractional factorial, but you don't even need to mention the name at this point). When you show a table with the trials listed, it is easy to see that each main factor is "high" 8 times and "low" eight times.

I then pulled out 16 helicopters that I had built (and tested!) ahead of time. I pulled out a few stopwatches. I pulled out pre-made forms for recording the data - each filled in with the factor to be tested. For example, one student gets "blade length" with a column for recording "wide" trials and a column for recording "narrow" trials. There is also a spot for dot-plots and also for the average of each data set. Emphasize that each student is recording the same data, just in different groupings.

I also introduced interactions at this point. Suggest that perhaps a paperclip works well with long blades, but no paperclip works well with short blades. So this student gets a sheet "long blade with clip OR short blade with no clip" in one column and "short blade with clip long blade with no clip" in the other.

(Make sure that you run the experiment ahead of time so that you can choose at least one interaction that is significant and one that isn't. Also, choose a pair of interactions that are confounded, so that when the data is analyzed, you can point out that the two gave identical results!)

As you drop the 16 helicopters, each student records the data on the appropriate half of his/her sheet. (I used transparencies so that I could show the results to the class more easily.). The students plot the points on a pair of axes and calculate the averages (I rounded up a couple extra calculators ahead of time!). If you want to be a little fancy, you could even do st dev or t-tests. If there are computers, I suppose you could also have Excel opened and loaded with appropriate spreadsheets.

Now you have a concrete example to discuss.
* The same data was used to analyze many different factors.
* everyone had 8 data point in each column.
* you have a significant interaction.
* you have confounded interactions.
* you can discuss how you could now explore 2-3 key factors to optimize the helicopter flight time.

I did this for an ASQ section meeting and I can't actually remember how long it took (plus I talked a little more about other ideas as well). I bet you could do all this in one hour.

Tim F


P.S. I actually found a significant interaction of paper color and paper weight (typing paper vs cardstock)! I dug up a sensitive scale and discovered that the white typing paper was heavier than the blue typing paper, but the blue cardstock was heavier than the white cardstock! Lighter weight (blue typing or white cardstock) was better than heavier weight (whte typing or blue cardstock).

[Statistical Steven]

Teach them six sigma....nobody uses DOE anymore.

[davis007]

Regarding the helocopter experiment I like the idea just not sure how much time it might take. I will do it myself first to get some ideas.

The popcorn idea is also good one.

Steven: I trust you were joking. I use DOE all the time in my current job as a quality improvement engineer and in my former position as a research engineer. In any case, quality measurement and improvement methods (including 6S) are a topic that will be covered in this class by another guest lecturer.

[Tim Folkerts]

Well Steven,

Usually I agree with you but here I think you couldn't be more wrong.

To me the heart of 6S is "Improve" and DOE is all about improving in efficient ways. 6S is the packaging; DOE is the tool. To me, your statement is like saying "Don't learn grammar, just use Word's spell-checker."

Good tools are always valuable: math skills, writing ability, DOE. The dressing that accompanies them comes and goes: slide rules, manual typewriters, 6S.


Tim F

[roland_lu]

Originally Posted by Statistical Steven

Teach them six sigma....nobody uses DOE anymore.

I am still the nobody who is using it,

[Statistical Steven]

Originally Posted by davis007

Regarding the helocopter experiment I like the idea just not sure how much time it might take. I will do it myself first to get some ideas.

The popcorn idea is also good one.

Steven: I trust you were joking. I use DOE all the time in my current job as a quality improvement engineer and in my former position as a research engineer. In any case, quality measurement and improvement methods (including 6S) are a topic that will be covered in this class by another guest lecturer.

Yes, it was a joke. I am a statistical consultant, but I am finding clients more interested in 6S and not in DOE.