A categorical model for continous variable

[sukunta]

Dear Michiel Bliemer,
The model specification of my pilot study suitable for a categorical model for a continuous variable (Time) (pseudo R2 of the continuous model less than categorical model 0.3906, 0.9308, respectively). Should I design the syntax for Time is categorical for the main survey?

Sincerely yours,
Sukunta

[Michiel Bliemer]

Yes you can. But it depends on what your study objective is. If you want to obtain willingness-to-pay then having dummy/effects coded attributes will not give you single WTP value and you cannot easily use it for forecasting with interpolated values. For example, if you estimate dummy coded coefficients for levels 10 and 20 (relative to 30), then you cannot compute the choice probability if time is 15.

If there are nonlinearities in a quantitative continuous variable, then you can consider applying a nonlinear transformation. For example, you can estimate a model with beta * ln(time), where ln is the natural logarithm. This way you can always interpolate (i.e., compute probabilities for level 15) although your WTP value will now again not be single value but rather a function of time. To estimate a model with a transformation, you simply transform your data to ln(time) and put that in the estimation tool. In Ngene you would to the same, you would put values ln(10), ln(20) and ln(30) as levels of the time attribute.

Michiel

[sukunta]

Dear Michiel Bliemer,
Thank you so much. From your recommendation, then my syntax is

Code: Select all

design
;alts=alt1*,alt2*,alt3
;rows=12
;block=2
;eff =(mnl,d,mean)
;con
;bdraws = gauss(3)
;model:
U(alt1)=b0 [(n, 2.501704, 1.141856)]+b1.dummy[(n,-0.5077506, 0.2668998)]*private[1,0]+b2.dummy[(n,0.4270188, 0.1971011)]*open[1,0]+b3.dummy[(n, 0.2385852, 0.3606243) |(n, 0.2457456, 0.2188252)]*time[1.30103
, 1.477121,1] + b4.dummy[(n, 1.035963, 0.3080124) |(n, 0.740783, 0.2492737)]*channel[1,2,0]/
U(alt2)= b0 [(n, 2.501704, 1.141856)]+b1* private +b2* open +b3* time +b4* channel
$


and in the questionnaire, I should converse to 10,20,30 min.
Sincerely yours,
Sukunta

[Michiel Bliemer]

In that syntax you now have both dummy coding and a transformation. You should choose. OR dummy coding, OR a transformation, but not both. In both cases, you replace the dummy variable or the transform with 10, 20 and 30 in the survey.

[sukunta]

Dear Michiel Bliemer,
Thank you so much. My syntax was corrected follows your recommendation is

Code: Select all

design
;alts=alt1*,alt2*,alt3
;rows=12
;block=2
;eff =(mnl,d,mean)
;con
;bdraws = gauss(3)
;model:
U(alt1)=b0 [(n, 2.501704, 1.141856)]+b1.dummy[(n,-0.5077506, 0.2668998)]*private[1,0]+b2.dummy[(n,0.4270188, 0.1971011)]*open[1,0]+b3[(n, 0.0121578, 0.0109568)]*time[1,1.477121,1.30103]+ b4.dummy[(n, 1.035963, 0.3080124) |(n, 0.740783, 0.2492737)]*channel[1,2,0]/
U(alt2)= b0 [(n, 2.501704, 1.141856)]+b1* private +b2* open +b3* time +b4* channel
$


Is it correct?
Sincelery yours,
Sukunta

[Michiel Bliemer]

If you want to use ln(10), ln(20), and ln(30) then the syntax should be

b3[...] * time[2.30, 2.99, 3.40]

and you will have to estimate the b3 prior in your pilot study by using the same transformation.

[sukunta]

Dear Michiel Bliemer,
Thank you so much for your kindness.
Sincerely yours,