choice-metrics.com

Hello,

I have two categorical variables, 1 with ordering (side effects: very common, uncommon) and 1 without (injection vs infusion). I just have a few questions about which priors to use.

Code: Select all

Design
;alts = alt1*, alt2*
;rows = 12
;eff = (mnl,d)
;model:
u(alt1)=  b1[-0.00001]*infectrisk[0.4,0.5,0.8]+b2.dummy[0|0]*route [0,1,2]+b3.dummy[-0.00001|-0.00002]*sideeffect[0,1,2] /
u(alt2) = b1*infectrisk + b2.dummy * route + b3.dummy* sideeffect
$


1. I want to avoid dominant alternative so, have use the * but, the 'route' prior doesn't have an ordering, so I have used 0 priors, will this confuse Ngene? if so should i remove the *?

2. I want to check if i have the right priors for the ordered variable: level 2 has a utility of 0.00003, level 0 have a utility of -0.00001 and level 1 has a utility value of -0.00002
Risk of side-effects levels:
0=very common
1=common
2=uncommon

Thanks for your help,
Tara

1. Using zero priors will not confuse Ngene - these effectively will not count towards the dominance checks.

2. The final level specified in syntax is treated as the base level (level 2 for sideeffect). So -0.00001 and -0.00002 line up (although I might have thought very common side effects would have greater disutility). I'm not sure where you are getting a prior 0.00003 for level 2. The parameter priors (like the parameter values themselves during estimation) are relative to each other. And in the syntax you have specified, level 2 will be assigned a utility of 0 by Ngene. You might want to revise your priors with level 2 as zero, so that you have the same relative utility between the levels - level 0 becomes -0.00004 and level 1 -0.00005.

Andrew

Thanks Andrew for the clarification, that's really helpful.

Tara