Experimental Design

Experimental Designs are the engines which drive a choice experiment. Below are some workhorse designs we routinely use. For custom Experimental Designs contact us.

Best-Worst Designs

These are used in Best-Worst experiments when we only want to know the rank ordering of a list of features. For more information on Best-Worst click here.

31 items

Used when we want the rank order of up up to 31 items. The respondent is faced with the simple task of choosing the best and the worst from a subset of 3.

set size: each respondent sees 10 statements.

minimum sample size: 620 respondents per segment

file: bibd_31x310.txt

21 items

Used when we want the rank order of up to 21 statements. The respondent is faced with the simple task of choosing the best and the worst from a subset of 3.

set size: each respondent sees 10 statements.

minimum sample size: 420 respondents per segment

file: bibd_21x210.txt

16 items

Used when we want to rank up to 16 items but with a smaller sample size. Here the respondent must select the best, second-best, worst and second-worst.

set size: each respondents sees 8 statements

minimum sample size:40 respondents

file: bibd_16x20_4alt.txt

Choice Experiment Designs

Used for more sophisticated experiments where we want to directly value the levels of product attributes and are only interested in the main effects. For more information on Choice Modelling click here.

Item has 7 attributes with 2 levels

Used for simple product packaging models where we are testing the inclusion/exclusion of specific features or benefits.

set size:each respondent sees 8 sets

minimum sample size: 20 respondents*

coverage: 128 unique product combinations

file: 2^7_8.txt

Item has 7 attributes with 4 levels

Used for simple product packaging models where each product attributes could have different levels. e.g. price ($10,$20,$30,$40), colour (Red, Green, Blue, Yellow) etc.

suggested set size:each respondent sees 8 sets

minimum sample size: 80 respondents

coverage: 16,384 unique product combinations

file: 4^7_32.txt

Item has 17 attributes with 8 levels

A good general purpose design for more complex products that has many attributes (17), for example credit cards or computer hardware.

set size:each respondent sees 8 sets

minimum sample size: 320 respondents

coverage: 2,000,000,000,000,000 + combinations

file: 8^17_128.txt

Item has 17 attributes with 16 levels

A heavy duty design that can model fine detail over many attributes, or can be refigured to increase the number of attributes at a small penalty to level detail.

set size:each respondent sees 8 sets

minimum sample size: 640 respondents

coverage: 300,000,000,000,000,000,000 combinations

file: 16^17_256.txt

Note:

a) Suggested sample sizes are based on a minimum of 20 observations per treatment per segment. They are the minimum for for the statistical integrity of the models, not for sample representability.

b) Each design should be row and column randomised prior to use.