choice-metrics.com

Dear team members,
I am new to the Ngene software and so to the Bayesian priors.
Currently, I am working on irrigation water management which has a CE objective.
To this end, I have developed and ran the following design but it provides a D-error of 1.6 and the S-estimate requires more than 500 intervies which is by far greater than what it is supposed to be.
So, is there any problem to this design? Can you please show me a way out from this problem?
Design
;alts = alt1*, alt2*, sq

;rows = 12

;eff = (mnl,d,mean)
; alg = mfederov(stop = total(100000 iterations))
;require:
sq.PRDUR = 1, sq.ALOM = 0, sq.IRGM = 0, sq.POLUNCER = 0 , sq.WTRPRI = 37.87

;bdraws = gauss(1)

;model:

U(alt1) = b2[(n,0.213649,0.057338)] * PRDUR[1,2,5] + b3 .dummy[(n,0.421583,0.108026)] * ALOM[0,1] +

b4 .dummy [(n,-0.087957,0.017974)|(n,-0.090775292,0.028759)] * IRGM[0,1,2] + b5 .dummy [(n,0.617929607,0.1824397)|(n, 0.699192869,0.2880091)] * POLUNCER[0,1,2] +

b6[(n,-0.023486492, 0.0098606)] * WTRPRI[18.94,37.87,75.74,113.61] /
U(alt2) = b2 * PRDUR +
b3 * ALOM +
b4 * IRGM +
b5 * POLUNCER +
b6 * WTRPRI /

U(sq) = b7[0] +
b2 * PRDUR +
b3 * ALOM +
b4 * IRGM +
b5 * POLUNCER +
b6 * WTRPRI
$

When I run the syntax it creates a D-error of 0.19, which sounds fine to me.

The S-estimates are only meaningful if your priors are reasonable. In your case, b4 with dummy coded attribute IRGM seems problematic. Levels of dummy coded variables are 0 or 1 and you state that the coefficients are -0.08 and -0.09. If that is true, then IRGM is not really a relevant attribute in the choice task. If you believe that it should be, then your coefficients are likely not reasonable.

You state that 500 respondents is far greater than what it is supposed to be. What do you base this on? Sample size requirements are case specific and including dummy coded attributes, especially attributes that are deemed not that important, are typically difficult to estimate and may require large sample sizes.

Michiel

Dear
Michiel Bliemer,

thank you very much.
I have revisited my design and the following is the revised one after making some adjustments on the b4 dummy coded attribute.
There are improvments but still I have a feeling that whether this design is acceptable.
So, can you help please?

Design
;alts = alt1*, alt2*, sq

;rows = 12

;eff = (mnl,d,mean)
; alg = mfederov(stop = total(10000000 iterations))
;require:
sq.PRDUR = 1, sq.ALOM = 0, sq.IRGM = 0, sq.POLUNCER = 0 , sq.WTRPRI = 37.87

;bdraws = gauss(1)

;model:

U(alt1) = b2[(n,0.209265,0.086001)] * PRDUR[1,2,5] + b3 .dummy[(n,0.199165,0.108026)] * ALOM[0,1] +

b4 .dummy [(n,-0.361222,0.007969)|(n,-0.492981469,0.0579373)] * IRGM[0,1,2] + b5 .dummy [(n,0.395512527,0.1824397)|(n, 0.476775789,0.2880091)] * POLUNCER[0,1,2] +

b6[(n,-0.023486492, 0.0098606)] * WTRPRI[18.94,37.87,75.74,113.61] /
U(alt2) = b2 * PRDUR +
b3 * ALOM +
b4 * IRGM +
b5 * POLUNCER +
b6 * WTRPRI /

U(sq) = b7[0] +
b2 * PRDUR +
b3 * ALOM +
b4 * IRGM +
b5 * POLUNCER +
b6 * WTRPRI
$

The syntax and output seem fine to me.

Michiel

Dear professor Michiel Bliemer,

thank you very much for your help.
I just want to know whether there are no dominant alternatives in my design.
From this forum posts and the responses, I realised that using both ;orth and ; eff=(mnl,d) can solve the problem; am I correct?
On the other hand, Is there any way of making level balances correct and the choice tasks relevant in the design?

I do not know the study but you can verify easily yourself whether there is dominance by checking whether there exist choice tasks in your survey where in one alternative all attribute levels are favourable compared to other alternatives.

Orthogonal designs cannot remove dominant alternatives, this can only be done in efficient design because they are more flexible.

You have specified a * after alt1 and alt2, which automatically removes dominant alternatives in case you provide parameter priors (which you have).

Michiel

Dear prof Michiel Bliemer,
thank you very much, it is really helpful.
The studyy is on the economics of irrigation water management which has a CE objective where we ask farmers about alternative irrigation water managmenet strategies.
To this end, we found from the public consultancy that duration of water use rights(PRDUR), water allocation methods(ALOM), irrigation methods(IRGM), probability of policy effectivenes(POLUNCER) and price of irrigation water( WTRPRI) use for crop production are the possible attributes chosen.

The other question is , how to correct/check whether the level balance is correct and choice tasks are relevant?

That is unfortunately a subjective rather than objective question. You can manually go through each set and see if they make sense. You can also pilot the tasks generated and ask people who complete them what they think. Finally, you can examine the choice probabilities for each task and see if any of the alternatives have extreme choice probabilities (close to 0 or 1). You can easily obtain the choice probabilities from the output generated in Ngene - they are provided under the design properties tab of the output file.

John

Dear prof johnr,
thank you very much for the constructive comments.

Dear prof Michiel and others,
I have run the following design. I wonder why am I getting too many extreme values in my prices?, for example the price of 228 is supposed to be the choke price which many people are not expected to choose but it appears many times.
Can you please lend me a hand?

Design
;alts = alt1*, alt2*, sq
;rows = 12
;eff = (mnl,d,mean)
; alg = mfederov(stop = total(10000000 iterations))
;require:
sq.PRDUR = 1, sq.ALOM = 0, sq.IRGM = 0, sq.POLUNCER = 0 , sq.WTRPRI = 38
;bdraws = gauss(1)
;model:
U(alt1) = b2[(n,0.209265,0.0086001)] * PRDUR[2,5,1] + b3 .dummy[(n,0.199165,0.0108026)] * ALOM[1,0] +

b4 .dummy [(n,-0.361222,0.007969)]* IRGM[1,0] + b5 .dummy [(n,0.395512527,0.01824397)|(n, 0.476775789,0.02880091)] * POLUNCER[1,2,0] +

b6[(n,-0.0018, 0.0038606)] * WTRPRI[20,38,76,114,152,228] /

U(alt2) = b2 * PRDUR +

b3 * ALOM +

b4 * IRGM +

b5 * POLUNCER +

b6 * WTRPRI /

U(sq) = b7[0] +

b2 * PRDUR +

b3 * ALOM +

b4 * IRGM +

b5 * POLUNCER +

b6 * WTRPRI
$