Q and Displayr have a special Question Type or Variable Set Structure called Ranking that is specifically designed for the analysis of ranking data. This means that statistical models designed for the analysis of ranking data are automatically applied whenever the Question Type or Structure is set to Ranking.
This setting can be used for:
- Repeated Measures Single Factor Experiments, in which respondents have been asked to rank alternatives, including situations where there are tied rankings.
- Blocked ranking experiments, including MaxDiff and more exotic variants such as anchored MaxDiff.
Where an experiment involves rankings but the objects being ranked are combinations of attributes, these are modeled using the methods described in Multifactor Experiments.
This outputs the rank as a Probability % statistic based on the logic that the highest value is the most preferred. This statistic is derived from a logit transformation to the Coefficient statistic, which itself is obtained using a model called Rank-Ordered Logit With Ties.
Note, when handling expected missing data, such as respondents are only meant to rank 5 out of 15 brands, for example, the missing data category should be set to a value of -1. In this model this setting indicates that any option with that value was less preferable than the last item the respondent chose.
Additionally, as this is an iterative model which searches for the optimal solution each time it calculates, changing the order of categories may result in very small differences in the lower decimal places.
See also
- MaxDiff for an overview of key MaxDiff concepts and resources.
- Experiments
- Statistical Tests for Ranking Questions
- Latent Class Analysis and Mixture Models
Further reading: MaxDiff software.