non-even space of attribute level

[wxy008]

Dear Ngene experts,

I have a general question of using Ngene for experimental design. Can I use non-even level space of attribute in the OOD, efficient, or Bayesian design? For example, Can I use a price attribute with non even levels, saying 1,2,5,8,? Will this cause any problems? Do I have to use even level space, such as 1,2,3,4?

Thanks for your help in advance.

Wei

[Michiel Bliemer]

No that is not a problem. If those levels make more sense than equidistant levels, then please do so. In case you use an OOD, the data will no longer be orthogonal, but that is not an issue anyway.

Michiel

[wxy008]

Thanks so much for your help, Michael.
I have another price level question. I used 4 price levels to do OOD design, say $1,$2,$3,$4, and collect a pilot data. But right now, I would like to change the price levels, say changing to $2.5, $2.75, $3,$3.25. Can I still use the results of the pilot data with original price levels as prior information to do Bayesian design with new price levels?

Thanks,

Wei

[Michiel Bliemer]

Yes that should be fine.

Note that a range of $2.50 - $3.25 is quite narrow, meaning that the coefficient for price will have a much larger standard error then when using a range of $1.00-$4.00.

Michiel

[wxy008]

Thanks so much. If I change to wider range of price, then the coefficient of price will have smaller Stander error. Does this mean changing to wider price range will be better than changing to narrower price range, if we create Bayesian design using pilot results with original price range?

[Michiel Bliemer]

Attribute level range is probably the most important factor to reduce standard errors, as it increases trade-offs and therefore makes parameters easier to estimate. Of course, the level range needs to be realistic and not lead to dominant alternatives. But generally, the more variation in attribute levels (i.e., the wider the range), the better.

Michiel

[wxy008]

Thanks so much, Michiel. I really appreciate all your help.

Wei