Appentency analysis

Question

I have a webcast on Thursday with a pet food company.

They gave me two XLS files. One is about DoE analysis. Quite simple in JMP.

The other one is less trivial for me. Perhaps someone could have an idea of the better way to analyze it. You will find it in attachment.

They want to have a classification of the 4 products with pairwise comparison.

An additional question would be to see if the Day variable has an impact.

The way of how data are stored are "nested".

all suggestion will be helpful, Thank you!

Florence

testdappetance.jmp

brady_brady · Answer

Here are Melinda's thoughts; I have attached her data table with the Choice scripts. I've altered the table to obfuscate the context a bit, as the original table was undisguised customer data.

Cheers,

Brady

Hi, Brady!

·        You are up the right tree (Ha! I see what you did there.) Choice is probably a better alternative to Matched Pairs. The only issue is that the Choice platform isn’t smart enough to use the data as-is. I had to transpose it so there was a row for each choice offered with an indicator to tell me which one was the “winner”.

o  That table is attached.

o  It has three scripts in it for the different models I tried. Product 3 appears to be the clear favorite, no mater how you slice it.

·        If the amount eaten is important, they could try weighting each Choice by ounces eaten. That’s kind of ad-hoc, and I don’t recommend it. If it’s really important that we try that for them, I can do some math to see what happens.

·        They could make day a “Subject” variable (or leave it out, since it doesn’t seem to be important). They do this way because the day is constant for each choice pairing. That tells the Choice platform to only estimate an intercept and not try to estimate a main effect for each one. I have a script that uses Day in the attached table. The tests say it doesn’t matter.

Transposed Data.jmp

brady_brady · Answer

I’ve got a question in to Melinda to get her thoughts on analyzing this as a choice design. In any event, it seems pretty apparent visually that 3 > (1,4) > 2. The graph in the attached file shows the proportion of food (type corresponding to the first word in the column name), eaten. I.e., if food 2 and 3 are offered, and the subject eats 10g of food 2 and 30g of food 3, plot 0.25 on the graph for column named “product 2 product 3”. Hence, in addition to the y-values you’ve got to look at the column name itself in this graph. For example, the upper right pane shows that product 1 is preferred to product 2. We can also see that “Day” doesn’t seem to impact things much. CAVEAT: once assumption I am making (which could be illegitimate) is that there is enough food for the subject to each as much of either food as it wishes. This seems legitimate to me, but consider a subject that wants 100g of food, when there is only 25g of "preferred" food. Once this is gone, they'd be forced to eat 75g of the other, and we'd get a ratio of 0.25, grossly misleading us as to the subject's preference. There is also a column in the data table called "first choice". In most cases, but not all, this agrees with the food of which the subject ate more. NOTE: I've changed the text above and the attached table to disguise context a bit, as the original data was undisguised customer data. Cheers,Brady Disclaimer: Pensez à l'environnement avant d'imprimer ce mail - Please consider the environment before printing this email 
Les informations contenues dans ce message et ses pié ces attaché eséventuelles sont confidentielles et destinées à la seule ou aux seules personnes ci-dessus désignées. 
This message and attachment are confidential, privileged and intended solely for the addressee(s). Any unauthorized publication, use, dissemination, forwarding, printing or copying of the email and its attachment is prohibited.

Compare Choices.jmp