Selected Reader Feedback to the Article, "A Wonderful Day for a Boy & His Dog," Stat-Teaser, April 2006

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

From: Harry F. Schiestel, The Game of Golf — 100 Methods to Improve™, http://www.myGolfDNA.com/

Hi Jon Kraber
I just read page 3 of the April 2006 issue of Stat-Teaser and WOW was I impressed.  What a wonderful project with your dog Jasper.

Jon, may I encourage you to continue these fun DOE type of experiments.  Who says a kid cannot learn what adults think is beyond themselves.

I taught both my sons DOE when they were in grade 7 so you beat them both by 1 year.  They were both medalists at Canada Wide Science Fair.

My oldest son is now 23 and he has a rich cause and effect reasoning which transcends to other things he does, both leisure and in business.

You will not today comprehend what your mom is teaching you, but you are learning a new way of thinking that will serve you your whole life.

If I can offer 1 piece of advice, it is to develop a comprehensive Cause and Effect Diagram before your form a single hypothesis to test out.

I will send you our “The Game of Golf – 100 Methods to Improve™”.  It is a C&E diagram that distills the whole game of golf down to 1 page.

When done correctly it forms the foundation for all subsequent research aimed at improvement.  I will send it out tomorrow to your mom Shari.

Also since you like dogs check out Tanner our Consumer Affairs Officer at http://www.mygolfdna.com/team.htm .

He is a great dog who is a lifelong learner.  In closing Jon, keep up the fun rich learning environment with your mom and you will do well in life.

To an elite player, golf should not be viewed as a sport but rather a system.  I hope you enjoy the signed Golf Print that is coming your way.

Hi Shari
I applaud your efforts to make a profound impact on the life of your son Jon.  It doesn't get much better than what you have displayed with him.

Some will criticize you for helping Jon too much.  Few people truly learn anything of significance that is not first mentored by someone else.

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

From: Dr. Steve L. Aprahamian, Senior Scientist, Business Development, Insulation Group, Bayer MaterialScience LLC

Dear Shari,

I read the article "A Wonderful Day for a Boy and his Dog" in the April 2006 "Stat-Teaser" and I wanted to make a few comments posed by your questions and comments.

I wanted to preface this by saying that while I have some experience with Experimental Design and their analyses I am not an expert by any means and my statistics background is informal. I have worked in the Chemical Industry for 20 years, over half of that doing process improvement in Bayer's manufacturing facility in New Martinsville, WV (I am currently involved with application development in rigid polyurethane foam insulation systems). In my work in the plant we did many small experiments in the production units. Since production expected to be able to use everything manufactured, even during our changes, our parameter settings were not broad and we tended to not make large changes in the experimental results so we do not always see changes that were statistically significant.

Our plans were to look at the what the data suggested rather than what it proved. If something was not seen to be any worse, but could be better, we may do additional runs of this setting to see. The worst is that there is no change (as it was not significant), but the plus would be we made an improvement.

My experience has been that sometimes the statistical model of assume no change unless it can be demonstrated is not always the way to go, in many settings. The averages do show the difference you are asking about in the data (1.44 vs. 1.61) it is just not large enough to be significant. This non-significance can be from various reasons:
1)There is really no difference and what you are seeing is just random variation
2)There is a difference but it is too small for the precision of the testing
3)There is a difference but it is masked by other factors.

The info you provided suggests that:
1) Jasper does not like Scooby Snacks (this is also your primary conclusion)
2) Jasper likes Pup-peroni and Beggin Strips (also one of your conclusions)
3) Jasper prefers BS over PP (also seems to be your gut feeling)
4) Even though not significant in your analysis, I think there is a positional factor that is at play. I think part of what masks this is how much he disliked the SS. I am not sure if you analyzed for interactions. If you ignore what he did not like, it seems he ate first more often whatever is in the LEFT position. Whether this was closer to the door, or he searched starting from the left is unclear. It just seems that he came to this first and if he liked it (whether there was something he preferred or not) he ate this first. The middle is where you really seem to see a difference in the BS and PP.

It may be of interest to look at the various raw data to see what pattern for selecting was to see if anything stands out. I kind of view this as a slightly flawed design if you are trying to determine preferences. It would be akin to checking what people grab at a dessert buffet. You start at the left and you take items of interest. People don't go thru to see what is there and then take in the order of preference, they go thru and take what they like. Preference comes in only with choice: they are only allowed to take one or they are full and only want one: the dog does not have this decision.

Even the order of eating may not be an indication of preference, some people save the best for last, others eat the best first.

Also, did the dog taste all the items before the test started? It seems from the data that he had not, which means further testing will yield different results. If he knew he did not like the SS before the test, he never would have taken them. Now that he knows he does not like them, the taking will be different. This adds a "time factor" where the earlier results are more random than true selection. It may have been better to allow him to smell each item and then taste it and do this several times so that he knew the items before he was tested.

Also I would think to not have all three items in every test may be better. Perhaps put a random sampling of 5 items down and check the 1st 3 grabbed for example, perhaps even stopping him from getting any more than 3 items (which adds part of the decision in for the dog). If he knows that he can get any of them, there is no need for a preference, if you force a choice, you can start seeing preference. (I see the like/dislike as one test, but preference as another)

An excellent article (kudos to your son). I like the designs and analyses which can make me think of further work and designs and also make me think of what the work suggests is really going on.

Answer from Shari Kraber:

Thank you for your extensive comments. We did our best not to impose a position bias by letting Jasper approach the test area dead-on — not from either the left or right. However, that doesn’t mean that he doesn’t already have a built-in tendency to choose from left to right. (It’s hard to get him to talk to me — rather like my middle-school boys!)
 
You offer some nice suggestions to consider, especially when dealing with human sensory data.
 
Thanks again for the feedback!

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

From: Wendell Smith, Eastman Kodak

From: Joseph G. Voelkel, PhD, Associate Professor and Chair, Graduate Statistics Program, Center for Quality and Applied Statistics, Kate Gleason College of Engineering, Rochester Institute of Technology

Well, you guys always keep the entertainment level up in your newsletter. 

I think your analysis is OK (even it it’s not theoretically mathematically justifiable). This may answer your question: if you wanted to improve it, you could treat factors B, C, and reps as blocks and use Friedman’s rank test (assuming no interaction between A and B or A and C). (There is a recent article on rank tests for factorial experiments but I don’t recall where it was — in my simple way of thinking, rank tests in a full analysis of factorial designs are a tough sell.) This approach would also make more sense because factor A is a within-small-time factor while factors B, C, and reps are across-larger-time factors so they should technically be treated differently. But here I don’t think you could naturally distinguish between “ate last” and “did not eat.” It must be pretty bad for a dog not to eat it! 

Answer from Shari Kraber:

I actually did analyze the data a few different ways, including various blocking schemes. For this particular data set, the answers were always similar, but the question has been thought-provoking for some of our readers. 

My final conclusion (after the vet confirmed that Jasper had indeed gained weight) was that Jasper refuses anything liver-flavored. This was confirmed by offering other treats with that smell. Bacon and sausage, on the other hand, must be carefully guarded when placed on the kitchen table! We recently put Jasper on some liver-flavored glucosamine-chondroitin chewables for his arthritis and we have to coat them with peanut-butter before he will eat them. Picky, picky.