Number of choice sets

[Pramisha]

Hi team and Prof. Bliemer,
Following are the attributes and attribute levels of my research:


Attribute:
Farm practice change

Attribute Levels:
1.Conventional till to conservation till
2.Conventional till to no-till
3.Conservation till to no-till
4.No cover crops to cover crops
5.No change in agricultural practice

Attribute:
Carbon Payment ($/acre)

Attribute Levels:
1.$0/acre
2.$5/acre
3.$10/acre
4.$15/acre
5.$20/acre

Attribute:
Certification

Attribute Levels:
1.None
2.Private firm
3.USDA

Attribute:
Contract Duration

Attribute Levels:
1.No contract
2.5 Years minimum contract
3.10 years minimum contract

Attribute:
Source of Carbon Payment

Attribute Levels:
1.Carbon Commodity Market
2.USDA

I am planning to make 36 choices (9 choice sets with 4 blocks) but given our attribute levels, I ain't sure if with this number, the design will be uniform or not. So, I request you to suggest to me the appropriate number of choice sets for this design and the reasons behind it?
Thank you in advance!

Regards,
Pramisha

[Michiel Bliemer]

I assume that with "uniform" you mean that you want a design that is attribute level balanced?
In that case, the number of rows needs to be divisible by the number of levels of each attribute. Since 36 cannot be divided by 5, using 36 rows will not allow an attribute level balanced design.

Instead, you could consider using 30 rows, since 30 is divisible by 2, by 3, and by 5. You could then block it is 3 blocks of 10.

Note that attribute level balance is not required, it is fine having a design that is not completely attribute level balanced, so using 36 rows would not be an issue.

Michiel

[Pramisha]

Prof. Bliemer,
Yes, I meant to ask about attribute level balance. Thank you for your response. I am clear now.

Pramisha

[Pramisha]

Hi Dr. Bliemer,
I edited the design a bit but the problem is, when runnning it says that there are many constraints, not having enough attributes or attribute levels for the number of rows required, and the use of too many scenario attributes. Can you please run this model and suggest me if I should go ahead or edit a bit?


Attribute & Attribute Levels:

A. Farm practice change

1.Conventional till to no-till
2.Conventional till to conservation till
3.Conservation till to no-till
4.No cover crops to cover crops
5.No change in agricultural practice

B. Carbon Payment ($/acre)

$0/acre
$5/acre
$10/acre
$15/acre
$20/acre

C. Contract Duration

10 years minimum contract
5 Years minimum contract
No contract

D. Source of Carbon Payment

USDA
Carbon Commodity Market
None


Dummies

Farm Practice Change
Conventional till to no-till=4, Conventional till to conservation till=3 Conservation till to no-till=2, No cover crops to cover crops=1, No change in agricultural practice=0

Contract Duration
10 years minimum contract=2, 5 years minimum contract=1, No contract= 0

Source of Carbon Payment
USDA=2, Carbon Commodity Market=1, None= 0

Prior Values: Mean (Standard deviation)

4-Conventional till to no-till (Yes): 0.6 (0.08)
3-Conventional till to conservation till (Yes): 0.55 (0.33)
2-Conservation till to no-till (Yes): 0.26093 (0.05064)
1-No cover crops to cover crops (Yes): 0.8 (0.2)

Carbon payment (Yes): 0.2 (0.04)

2- 5-years contract(Yes) : -0.86152 (0.05450)
1-10-years contract(Yes) -0.9 (0.09)

2-USDA source (Yes): -0.20278 (0.04824)
1-Carbon commodity market source (Yes): -1.04265 (0.06913)




NGene Design (D-error: )
Syntax
Design
;alts= optA*, optB*, Neither
;rows=30
;eff=(mnl,d, mean)
;block=5
;bdraws=sobol(5000)
;model:
U(optA) = b1 [(n, 0.2, 0.04)] * carbon payment[0, 5, 10, 15, 20](6-12,6-12,6-12,6-12, 6-12)
+ b2.dummy[(n, 0.6, 0.08)|(n, 0.55, 0.33)|(n, 0.26093, 0.05064)|(n, 0.8, 0.2) ] * farm practice change[4, 3, 2, 1,0]
+ b3.dummy[(n, -0.86152, 0.05450)|(n, -0.9, 0.09)]* Contract[2,1,0]
+ b4.dummy[(n, -0.20278, 0.04824)|(n, -1.04265, 0.06913)] * Source[2, 1, 0] /
U(optB) = b1 * carbon payment
+ b2 * farm practice change
+ b3 * Contract
+ b4 * Source /
U(Neither) = b0 [(n, 0.1, 0.01)]
$



• Here, we are generating three alternatives, Opt. A, Opt. B and Neither
• We are generating 30 choice sets in five blocks.
• We generated a D-efficient MNL model. We can use this model for random parameter logit (RPL) analysis too.
• We used the default-swapping algorithm.
• For the prior values, we used mean and standard deviation of attributes looking into previous studies.
• Using a design with 30 rows, we are able to estimate all possible interaction effects after data collection.

Reference for prior values:
Gramig, B. M. and N. J. O. Widmar (2017). "Farmer Preferences for Agricultural Soil Carbon Sequestration Schemes." Applied Economic Perspectives and Policy 40(3): 502-521.

[Michiel Bliemer]

The syntax runs fine for me (using Ngene version 1.2.1).

The default swapping algorithm will ensure attribute level balance so there is no need to include attribute level constraints (6-12,6-12,6-12,6-12, 6-12). Only if you want to use the modified Federov algorithm you will want to impose such attribute level constraints to ensure some degree of level balance.

Michiel

[Pramisha]

Prof. Bliemer and the team,
I used the following syntax and generated the design:

Attribute and Attribute Levels

A.Farm practice change

1.Conventional till to no-till
2.Conventional till to conservation till
3.Conservation till to no-till
4.No cover crops to cover crops
5.No change in agricultural practice

B.Carbon Payment ($/acre)

1.$0/acre
2.$5/acre
3.$10/acre
4.$15/acre
5.$20/acre

C. Contract Duration

1.10 years minimum contract
2.5 Years minimum contract
3.No contract

D. Source of Carbon Payment

1. USDA
2. Carbon Commodity Market
3. None

Dummies

Farm Practice Change
Conventional till to no-till=4, Conventional till to conservation till=3 Conservation till to no-till=2, No cover crops to cover crops=1, No change in agricultural practice=0

Contract Duration
5 years minimum contract=2, 10 years minimum contract=1, No contract= 0

Source of Carbon Payment
USDA=2, Carbon Commodity Market=1, None= 0

Prior Values: Mean (Standard deviation)

4-Conventional till to no-till (Yes): 0.6 (0.08)
3-Conventional till to conservation till (Yes): 0.55 (0.33)
2-Conservation till to no-till (Yes): 0.26093 (0.05064)
1-No cover crops to cover crops (Yes): 0.8 (0.2)

Carbon payment (Yes): 0.2 (0.04)

2- 5-years contract(Yes) : -0.86152 (0.05450)
1-10-years contract(Yes) -0.9 (0.09)

2-USDA source (Yes): -0.20278 (0.04824)
1-Carbon commodity market source (Yes): -1.04265 (0.06913)

NGene Design (D-error: )
Syntax
Design
;alts= optA*, optB*, Neither
;rows=30
;eff=(mnl,d, mean)
;block=5
;bdraws=sobol(5000)
;model:
U(optA) = b1 [(n, 0.2, 0.04)] * carbon payment[0, 5, 10, 15, 20
+ b2.dummy[(n, 0.6, 0.08)|(n, 0.55, 0.33)|(n, 0.26093, 0.05064)|(n, 0.8, 0.2) ] * farm practice change[4, 3, 2, 1,0]
+ b3.dummy[(n, -0.86152, 0.05450)|(n, -0.9, 0.09)]* Contract[2,1,0]
+ b4.dummy[(n, -0.20278, 0.04824)|(n, -1.04265, 0.06913)] * Source[2, 1, 0] /
U(optB) = b1 * carbon payment
+ b2 * farm practice change
+ b3 * Contract
+ b4 * Source /
U(Neither) = b0 [(n, 0.1, 0.01)]
$


· Here, we are generating three alternatives, Opt. A, Opt. B and Neither
· We are generating 30 choice sets in five blocks.
· We generated a D-efficient MNL model. We can use this model for random parameter logit (RPL) analysis too.
· We used the default-swapping algorithm.
· For the prior values, we used mean and standard deviation of attributes looking into previous studies.
· Using a design with 30 rows, we are able to estimate all possible interaction effects after data collection.


However, I couldn't get all the prices included in a block, like there was no $ 0 option in any of the choice sets of Block 1 and similar stuff happened in the other two blocks. Can you please explain to me what could be the reason behind it?

Moreover, I am thinking of keeping restrictions in the design, with the following combinations:
No change in practice should always go with $0
No change in practice should always go with No source of carbon payment (None)
No change in practice should always go with no contract duration.
$0 should always go with No source of carbon payment (None)
Positive $ amount should never go with no source of payment.

Can you please guide me on how to go forward regarding this?

Thank you!

[Michiel Bliemer]

The syntax looks good (although a "]" is missing after the carbon payment levels). Further, like with any scripts or programming language, you should not use spaces in variable names as that can create unexpected results, especially when using constraints. Please use carbon_payment, etc.

To answer your first question, it is not possible to get each level to appear in each block in an efficient design, this is only possible with orthogonal design. Blocking is formally not defined for efficient designs, but in Ngene we decided to let users block designs by minimising correlations between the blocking column and the other attributes (in orthogonal designs this correlation is zero). Blocking can only be done perfectly in an orthogonal design (if one exists).

Regarding constraints, the default swapping algorithm allows conditional constraints. For example for your last constraint:

;cond:
if(alt1.carbon_payment > 0, alt1.source > 0),
if(alt2.carbon_payment > 0, alt2.source > 0)

Michiel