choice-metrics.com

Dear Prof.Michiel,
I generate syntax and the Sp-estimate maximum is b3=37.9.

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[10,20,30]+ 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
$

The 200 respondents of the main survey were estimated model with mixed logit. The result found b1 and b3 were not significant. I should discuss this result that it was not significant because the sample size was not enough or it is no significant difference following the estimated model.
Sincerly yours,
Sukunta

It can be both. It is not possible to say whether statistical insignificance means that the sample size is too small or that the corresponding attributes are not relevant in choice behaviour. Note that with a large enough sample size, ANY attribute will become statistically significant. All you can conclude now is that it seems that PRIVATE AND CHANNEL are not the most important attributes, but may play a minor role (you cannot reject the null hypothesis, but you should not accept it either).

Your sample size estimates are only valid if your priors are correct. Your parameter estimates deviated from the priors so therefore your sample size estimates are not accuriate. You can update the sample size estimates by updating the priors in the Ngene syntax with the parameter estimates obtained in the main study if you like. This does not change anything, but it will give you a more accurate sample size estimate.

Michiel

Dear Prof. Michiel,
Thank you so much for your reply. I generated the new model by Baysient efficient design with coefficient and SD from the main survey (200 respondents).
I found Sp estimate value b1(d0)=41.04 b3(d1) =3659.199. Could I conclude that the b1 are not relevant in choice behavior if the sample size is representative of the population?
Sincerely yours,
Sukunta

It is not possible that the Sp-estimate indicates that you need 41 respondents for b1 while it is not significant in the survey with 200 respondents, both need to be consistent. It sounds like you are generating a new Bayesian design, you should of course evaluate the design you used in the survey.

You should load the design you used for your survey in Ngene and then use the following property in the syntax while you use the parameter estimates as fixed (not Bayesian) priors.

;eval = design_name.ngd

Michiel

Dear Prof.Michiel,
I update the sample size by updating the priors in Ngene syntax from the main survey (200respondents)

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, 5.458635, 3.014236)]+b1.dummy[(n,-0.266547,1.424629 )]*private[1,0]+b2.dummy[(n,0.427396, 1.301125)]*open[1,0]+b3[(n, 0.001597, 0.01134)]*time[20,30,10] + b4.dummy[(n,0.692211, 0.65329) |(n, 0.905447, 0.903167)]*channel[1,2,0]/
U(alt2)= b0+b1* private +b2* open +b3* time +b4* channel
$

How do I interpret the Sp estimate result?
Sincerely yours,
Sukunta

See my previous comment. This syntax generates a NEW experimental design, this is not the experimental design you used in survey and its Sp estimates have no meaning for your study since it belongs to a design that you did not use for your study.

I believe that you are putting too much emphasis on sample size estimates. They are only useful during the experimental design phase but should be ignored during model estimation phase. You can verify that the sample size estimates are consistent with your model estimates by evaluating your design, see my previous comment, but other than that they are not important, you should focus your explanations on the parameters, standard errors, and t-ratios that come out of model estimation.

Michiel

Dear Michiel,
Thank you so much for your kindness in the recommendation.
Sincerely yours,
Sukunta