choice-metrics.com

Hi all,


I built a unlabel experiment.. proposingh, and some solutions has specific attributes. I've specified 0 level as unavailable
I'd like to present only 4 solutions at any time in the choice task

Design
;alts = alt1*, alt2*, alt3*, alt4*
;rows = 360
;block= 30
;eff = (mnl,d)
;reject: (some rejections)
alt1.solut =1 and alt1.delta2 = 0, (delta2 is specific to solution1)
alt1.solut =2 and alt1.delta2 > 0, (delta2 should'nt appear for solution 2)
(.....)
;model:
U(alt1) = a1.dummy[0|0|0|0|0|0|0|0] * solut[1,2,3,4,5,6,7,8,9]
+ c1.dummy[0|0] * confort[1,2,3] + d1.dummy[0|0|0] * child01[0,1,2,3] + e1.dummy[0|0|0] * child02[0,1,2,3] + coef1 * delta1[0,-300,-200,-100] + coef2 * price[600,700,800,900,1000] + coef3 * delta2[0,100,200,300] /
U(alt2) = a1 * solut + c1 * confort + d1 * child01 + e1 * child02 + coef1 * delta1 + coef2 * price + coef3 * delta2 /
U(alt3) = a1 * solut + c1 * confort + d1 * child01 + e1 * child02 + coef1 * delta1 + coef2 * price + coef3 * delta2 /
U(alt4) = a1 * solut + c1 * confort + d1 * child01 + e1 * child02 + coef1 * delta1 + coef2 * price + coef3 * delta2 $

It's working quite fine.. Now I would like to controle attribute balance for unavailable level ..
but I'm having errors with nrows=420 and block=30 and child02[,1,2,3](2,3,3,3)
while everything is going quite fine with nrows=14 and child02[,1,2,3](2,3,3,3)

I'd like to estimate this model using Hierarchial Bayes.. how do I specify priors and model in such case?

Thanks a lot for your comments
Naji

Naji,

A quick comment... Is there a reason that you need complete attribute balance? This is not required for efficiency and actually puts on a restriction that lowers efficiency. In particular for dummy or effects coded parameters, there is no reason to demand attribute level balance, as an efficient design will automatically also guarantee some level of attribute level balance. For linear parameters, you may wish balance to assure that you could estimate nonlinear effects, but again balance is not needed.

Finding hundreds of completely balanced rows seems quite difficult, so perhaps you could search for a design that is balanced to a large extent, but not entirely? In Ngene you can specify which attributes should be level balanced and others that need not. Use the mfederov algorithm and set the upper and lower bounds for each level to appear.

For example, if you would have 9 rows, then
b1*x[1,2,3](3,3,3)
means that each level should appear 3 times exactly, while
b1*x[1,2,3](2-4,2-4,2-4)
means that each level should appear between 2 and 4 times.

Also, with hundreds of rows, Ngene will find it difficult to find a design without any rows that are repeating. Designs with many rows often cause such problems, also with dominancy. Do you really need so many rows? It is quite unusual.

I cannot comment on Hierarchical Bayes I am afraid, that is not my expertise.

Michiel

Hi Michiel


Thanks for kind answer.. In fact, I need several versions (30) for such design (14 rows) and previously I'd been adviced to use high number of rows (14x30) and to use the first 14 rows as 1st version for the design, and the next 14 rows for the 2nd version etc...
I don't need a complete balance for the design, just to minor 'Unavailable level' frequency...

I'll work on the Manuel to analyze hierarchical bayes designs

Thanks a lot
Naji