What is the difference between preference scores and distribution?
In a conjoint study such as Generic Conjoint, the partworth utilities are used to calculate and derive the subsequent outputs such as average preferences for levels and distribution of preferences for levels. Although these outputs are derived from the same set of partworth utilities, they view the utilities at different perspectives and involve different calculations. Specifically:
The average preferences for levels (or level partworths) looks at the average rating of preference towards each attribute level across respondents. The calculation involves taking the average partworth utilities across respondents before performing scores rescaling. An attribute level with higher score indicates that overall it is strongly preferred by respondents.
The distribution of preferences for levels looks at the preference distribution among respondents, where it shows the proportion of respondents that will choose a certain level within an attribute. An attribute level with higher percentages means more respondents are more likely to choose this level over the other levels.
Due to the differences above, sometimes these outputs will reveal slightly different outcomes in terms of the more preferred levels. The results are not necessarily contradicting to each other, and still can be interpreted through the following possible explanations:
An attribute level has the highest average preference, but does not have the highest distribution of preferences: It is likely that this level is highly preferred by a small subset of respondents, while moderately preferred by the others. As a result, the average preference is inflated by the minority group, but the majority of respondents will still go to the other levels instead.
An attribute level has the highest distribution of preferences, but does not have the highest average preference: It is likely that this level has a hit-or-miss characteristics. It has gained more supporters to choose it over the other levels, but the remaining respondents also don’t find it relevant or favourable to have high utility scores. As a result, the average preference is even out instead.
We recommend looking at both the average preferences and the distribution of preferences to have a better view of your consumers’ preference as they provide different angles to understand the preferences.