choice-metrics.com

Dear Ngene team,

We are planning a choice experiment for insulation boards. The alternatives, attributes and their planned levels are:
Alternatives: Cellulose board (cellu), Stone wool board (stone), Wood fiber board (wood), Reed fiber board (reed)
Attributes and levels: Thermal resistance (ther) (0,1,2), Domestic/ imported (dome) (0,1), Fire safety (fire) (0,1,2), Environmental friendliness (env) (0,1,2) and Price €/m2 (price) (8,12,16,21,32,43)

Some alternatives can have only certain levels due to their physical characteristics (such as stone wool that does not burn-> fire=2). The possible attribute levels by alternatives are:
cellu: ther (0,1,2); dome (0,1); fire (0,1) env (1,2); price (8,12,16,21)
stone: ther (1,2); dome (0,1); fire (2) env (0,1); price (8,12,16,21)
wood : ther (0,1,2); dome (0,1); fire (0,1) env (1,2); price (8,12,16,21)
reed : ther (0,1,2); dome (0,1); fire (0,1) env (1,2); price (16,21,32,43)

We have drafted a following syntax:

Design
;alts = cellu, stone, wood, reed, none
;rows = 36
;eff = (mnl,d)
;alg=mfederov
;reject:
stone.ther < 1 ,
wood.fire > 1 ,
stone fire< 2 ,
reed.fire > 1 ,
cellu.fire > 1 ,
cellu.env < 1 ,
wood.env < 1 ,
reed.env < 1 ,
stone.env > 1 ,
cellu.price > 22 ,
stone.price > 22 ,
wood.price > 22 ,
reed.price < 16

;model:
U(cellu) = b1[0.001] +
b2.dummy[0.001|0.002] * ther[1,2,0] +
b3.dummy[0.001] * dome[1,0] +
b4.dummy[0.001|0.002] * fire[1,2,0] +
b5.dummy[0.001|0.002] * env[1,2,0] +
b6[-0.001] * price[8,12,16,21,32,43] /

U(stone) =b7[-0.001] +
b2.dummy * ther +
b3.dummy * dome +
b4.dummy * fire +
b5.dummy * env +
b6 * price /

U(wood) = b8[0.001] +
b2.dummy * ther +
b3.dummy * dome +
b4.dummy * fire +
b5.dummy * env +
b6 * price /

U(reed) = b9[0.001] +
b2.dummy * ther +
b3.dummy * dome +
b4.dummy * fire +
b5.dummy * env +
b6 * price
$

It seems that we have a problem in the attribute levels and syntax, and Ngene is not able to provide us a design. We assume that the issue is that in the case of stone wool attribute “fire” mixes with the constant for stone wool. What do you think? What would you recommend for us concerning the design and how your recommendation would look like in a syntax?

Kind regards
Eija

Correct, your constant for stone cannot be estimated if fire has a fixed level. Since level 2 for fire is unique to stone, you need to remove this level from fire whereby safety level 2 for stone is captured by the constant. This results in the Ngene script below.

Code: Select all

Design
;alts = cellu, stone, wood, reed, none
;rows = 36
;eff = (mnl,d)
;alg=mfederov
;reject:
stone.ther < 1 ,
cellu.env < 1 ,
wood.env < 1 ,
reed.env < 1 ,
stone.env > 1 ,
cellu.price > 22 ,
stone.price > 22 ,
wood.price > 22 ,
reed.price < 16

;model:
U(cellu) = b1[0.001] +
b2.dummy[0.001|0.002] * ther[1,2,0] +
b3.dummy[0.001] * dome[1,0] +
b4.dummy[0.001] * fire[1,0] +
b5.dummy[0.001|0.002] * env[1,2,0] +
b6[-0.001] * price[8,12,16,21,32,43] /

U(stone) = b7[-0.001] +
b2.dummy * ther +
b3.dummy * dome +
b5.dummy * env +
b6 * price /

U(wood) = b8[0.001] +
b2.dummy * ther +
b3.dummy * dome +
b4.dummy * fire +
b5.dummy * env +
b6 * price /

U(reed) = b9[0.001] +
b2.dummy * ther +
b3.dummy * dome +
b4.dummy * fire +
b5.dummy * env +
b6 * price

$

Michiel

Many thanks Michiel! It seems to work well.
Eija