choice-metrics.com

l.s.

I am new to the group, so perhaps this question belongs somewhere else. In that case please let me know if it does. I have several questions with respect to a CE design.

1. I am using percentages as an attribute, with 100% for the status quo and 80%, 50% and 20% as attribute levels. I have specified this in a pivot design as:

Design
;alts = sq, alt1, alt2
;rows = 36
;block = 6
;eff=(mnl,d)
;model:
U(sq) = LU[0] * A.ref[100] /
U(alt1) = LU[0.05] * A.piv[-20%,-50%,-80%] /
U(alt2) = LU[0.05] * A.piv[-20%,-50%,-80%] $

Is this the correct way, or would you use another specification?

2. I am doubting on the design method, i.e., either orthogonal (I have no magnitudes of priors), efficient (I do not know the signs of some prioirs), or pivot (some attributes have a clear reference point in the status quo). A few questions on this:

a. With respect to including the signs of priors in an efficient design, I have now simply used very small numbers for priors with a + or minus sign (see also syntax above). Is this the 'correct' way, or would you do this differently?

b. Although some attributes have a clear reference point, some have not. The syntax is below, LU and CON have a clear reference value in the status quo, for DUR and ALL it is less clear (the status quo is a no contract situation, the DURation and ALLocation of compensation are therefore not relevant in the status quo). I have given the syntax I am using at the moment below:

Design
;alts = sq, alt1, alt2
;rows = 36
;block = 6
;eff=(mnl,d)
;model:
U(sq) = LU[0] * A.ref[100] + CON[0] * B.ref[0] + DUR[0] * C.ref[0] + ALL[0] * D.ref[100] /
U(alt1) = LU[0.05] * A.piv[-20%,-50%,-80%] + CON[0.05] * B.piv[250,500,1000,2000] + DUR[-0.05] * C.piv[2,5,10] + ALL[-0.05] * D.piv[0%,-25%,-50%] /
U(alt2) = LU[0.05] * A.piv[-20%,-50%,-80%] + CON[0.05] * B.piv[250,500,1000,2000] + DUR[-0.05] * C.piv[2,5,10] + ALL[-0.05] * D.piv[0%,-25%,-50%] $

Ngene produces an efficient design in this case, I just do not know whether it produces a design that reflects the model. Therefore I would like to know whether this is the correct specification, given my model?

Any help is highly appreciated!

Thanks,
and best regards,
Mark

1. Which attribute values do you show to the respondents? Do you actually show percentages, or are you first asking for their reference level and then calculate the attribute value that you show? If you show percentages, then you should not use ref and piv. If each respondent has a different values for the reference attribute level, then this seems correct. I do find your utility functions a bit strange, as you only estimate a single coefficient? Or do not mean a different coefficient for sq and another one for alt1 and alt2? In that case, you should give LU a different name, for example LU in sq, and LU2 in alt1 and alt2. You provide different prior values (0 and 0.05) for them, so I assume you actually would like to estimate two coefficients?

2. If you know the sign, then Ngene is able to get rid of choice tasks with dominant alternatives. Then you can indeed simply specify very small prior values close to zero. Ngene will only check for dominant alternatives if you use ;alts = sq, alt1*, alt2* (so add the asterisk). The second syntax has the same problem as the first, that you provide different priors for the same coefficient name.

Perhaps you can provide an example of a choice task that you would like to present to a respondent in order for me to understand what you actually would like to do? I am not sure whether this is the syntax that achieves what you want.