choice-metrics.com

Dear Ngene experts:
Thank you very much for your previous reply. Now I still encountered some questions, If I can get your help, I would appreciate it very much.

This is my syntax:
Design
;alts=alt1*,alt2*,alt3
;rows=24
;block=4
;eff=(mnl,d)
;model:
U(alt1)=b1[-0.0023128]*A[100,250,400,550]+b2.effects[0.0347193|0.023784|0.2419535]*B[3,2,1,0]+b3[0.0003555]*C[1000,2000,3000,4000]+b4[0.007569]*D[0,20,40,60]+b5.effects[0.3398703]*E[1,0]/
U(alt2)=b1*A+b2*B+b3*C+b4*D+b5*E/
U(alt3)=b0[0]
$

And my questions are:
1 If I increase the number of blocks, will the number of S-estimate displayed in the Ngene increase at the same time? That is means, does the number of blocks affect the number of S-estimate?

2 After I run the above syntax, the S-estimate is very large(it’s 1500), and my block is 4, which means that my sample size is 6000, which is too large, and it is very difficult to do(It's hard to find so many people). Is there any way to reduce my sample size?

I’m more thankful than I can express. Best wishes for you.

1. The S-estimates are not affected by the number of blocks, but as you say the number of respondents needed (to get statistically significant parameter estimates) can be computed by multiplying the S-estimates with the number of blocks.

2. Most of your parameters have a small S-estimate, but the effects coded coefficients that are very small (0.02 and 0.03) will be difficult to estimate. You could consider using dummy coding, which may give you better S-estimates. Note that S-estimates are only meaningful if the priors are appropriately chosen. The actual parameter estimates may turn out to be much larger, therefore your sample size requirements may be much smaller. I think that you should not worry too much about them, effects and dummy coded coefficients require a larger sample size than linear coded coefficients.

Is there a reason that your b0 coefficient is set to zero? when you use non-zero priors you should preferably use non-zero priors for all coefficients as otherwise the choice probabilities used to calculate the efficiency and S-estimates may not be consistent.

Michiel