How to prohibit pairs of levels from showing together

Conjointly lets you stop specific pairs of levels from showing together. Open the Advanced settings tab of your experiment and find the Prohibited pairs of levels module. It has two types of prohibitions:

• Within a single alternative: two levels never appear on the same product concept.
• Across alternatives in a choice set: if one level appears in one alternative, the other level cannot appear in any other alternative of the same screen.

Both types change only the experimental design (which concepts respondents see), not the way the report is calculated. They can only be set or changed before the experiment is launched. For best results, use prohibitions sparingly and click Check these prohibitions after each change.

Within a single alternative

Use this when a combination of levels on one product would look unrealistic to consumers. For example, a particular price ($5) might be incompatible with a particular material (recycled). With the pair “$5 + Recycled” prohibited, recycled can still show at other prices, but a concept like Option 3 below is never shown:

Keep in mind that you only need to prohibit combinations that are unrealistic in the eyes of the consumer, rather than those that are not feasible for the company to make. For example, if consumers would not be surprised to see $5 recycled tissue paper, but it is not feasible for the company, you should not prohibit this pair in the design of your experiment. Instead, you should simply ignore product concepts with this combination when you look at the results of the analysis.

Across alternatives in a choice set

Use this when two levels should not compete against each other in the same question (i.e. levels X and Y never appear in different alternatives of the same choice set). It does not stop X and Y appearing together on one alternative; add a within-alternative prohibition as well if you need that.

• X and Y can be levels of different attributes, or of the same attribute (for example, two brands that should never be compared directly).
• X and Y can even be the same level. For example, prohibiting “Brand A + Brand A” means Brand A never appears twice in one choice set.

For example, with the pair “Brand A + Brand B” prohibited across alternatives, Brand A can compete with other brands, but a choice set like this is never shown:

This type of prohibition works with every conjoint type Conjointly offers, such as Generic, Brand-Specific and their MaxDiff variants, Claims Test, Brand-Price Trade-Off. It also works with all level types (features, brands/SKUs and prices), and Partial Profile Conjoint. You can add up to five prohibited pairs across alternatives, but we recommend no more than three.

Tips for keeping the number of prohibitions low

If you find that you need more than four prohibitions within a single alternative, or more than three across alternatives, we strongly encourage you to think through these options:

1. If you are using Generic Conjoint, you might need Brand-Specific Conjoint where you can set which levels apply to which brands.

2. Remove prohibitions of any pairs that are at least marginally plausible in the eyes of the consumer.

3. Remove prohibitions of pairs that are plausible to the consumer but not feasible for the company to supply (as discussed above, they should not be prohibited in the design).

4. Make sure to set other advanced settings to “Automatic”.

For more specific restrictions, you can also customise the experimental design, in which case you will need to check for conflation yourself.

Why is the number of prohibited pairs limited?

The limits differ for the two types of prohibitions:

• Within a single alternative: the maximum depends on the number of attributes and levels in your study.
• Across alternatives in a choice set: the maximum is five pairs, whatever the size of your study.

Without limitation, the analysis would be imprecise (and sometimes impossible).

The system will not stop you from launching a study with a conflated design, but it may not be able to produce a report for it. Before running the analysis, Conjointly checks the design and shows an error on the report page if prohibitions or customisations have removed too many combinations of levels for preferences to be estimated. Because prohibitions cannot be changed after launch, the fix at that point is to duplicate the experiment, remove or relax the prohibitions in the copy, and collect new responses. To avoid this, click Check these prohibitions before you launch.

For more tips on attributes and levels, read the guide on specifying attributes and levels in conjoint analysis.

An example of conflation

Too many prohibitions can make it impossible to run the modelling because they cause conflation: two attributes become so entangled that their effects can no longer be told apart.

Consider a study with two attributes:

• Colour: yellow, blue
• Price: $1, $2, $3, $4, $5

Now suppose we add the following five prohibited pairs:

1. Yellow with $1
2. Yellow with $2
3. Yellow with $3
4. Blue with $4
5. Blue with $5

These prohibitions mean that blue can only ever appear with $1, $2, or $3, and yellow can only ever appear with $4 or $5. Colour and price are now perfectly aligned: every low-priced option is blue, and every high-priced option is yellow.

As a result, when respondents dislike the higher-priced options, we cannot empirically tell whether it is because they dislike higher prices or because they dislike yellow. The two explanations are indistinguishable in the data, so the model usually will not run.

Because we use a Bayesian modelling approach that estimates preferences using a numerical method rather than an algebraic one, it will sometimes (on some runs) still try to impute preferences even when there is conflation. It is purely down to luck whether the model will or will not run in those cases. If the model fails, you can try resetting the report manually to make it run again. But there is no guarantee that the analysis will work, and even if it does, it will be flawed.