Limitations on the Use of Filters within NIPO Conjoint Measurement

NIPO ODIN Version 5.17

Limitations on the Use of Filters within NIPO Conjoint Measurement

First of all, when creating filters in the NIPO Conjoint Measurement, it is important to emphasize that Conjoint Measurement is about hypothetical or virtual products.

The use of hypothetical products is a necessity for independently estimating the effects (weights and importances). A filter harms this independence. It makes those effects confounded, blended or mixed. The filter's influence should therefore be reduced to a minimum.

In practice it means that a filter will block only a limited number of level combinations. For example in a conjoint survey on houses with the attributes surface of ground floor (50m², 70m² or 90m²) and number of bedrooms (1,3 or 5), the combination of 50m² and 5 bedrooms and 90m² and 1 bedroom may be filtered out, but no more. Otherwise the parameter influences will be mixed.

As a guideline: do not filter out more then 1 level in a list of 3. For 4 levels there is also a maximum of 1, for 5 level attributes a maximum of two levels can be filtered out, for 6 also only 2, 7 approximately 3 et cetera. The more attributes are involved the bigger the mix effect will be.

Moreover, if a filter becomes larger, and the number of possible combinations decreases, the system is more likely to show non-preferred combinations anyway. And if the filter is too complex, the filter will be ignored completely.

Furthermore, there are other ways to prevent impossible combination from showing, for example by combining the impossible combinations in one attribute. In the example of the conjoint on houses, make one attribute with 7 levels: 50m² - 1 bedroom, 70m² - 1 bedroom, 50m² - 2 bedrooms, 70m² - 2 bedrooms, 90m² - 2 bedrooms, 70m² - 3 bedrooms, 90m² - 3 bedrooms.

It is recommended that prices are not included as part of the level descriptions of more then one attribute. The utility values should aid the decision to perceived value rather than guide a respondent towards a logically consistent cost. This is because the intention of conjoint is to estimate utility for a certain level, not for that level and a price. The inclusion would complicate the measurements in the sense that it mixes two levels. The complication is not only for the analyst but also for the respondents: they have to add up all the different prices.