One could say there is a natural order to things. In survey research, when we think of order we are thinking about ordinal data. The ability to leverage the respondent’s willingness to rank their choices is what differentiates ordinal from nominal data. This steps our analytical options up a notch from measuring simple percentages and modal values.
Ordinal data is literally data that can be placed into an order. This can be applied to baseball teams, horse racing, consumer preferences, or your place in the checkout line. A word of caution, ordinal data implies that we can specify an order, but we cannot speak accurately to the distance between those ordered items. The consistent distance between intervals is a function of interval data, which will be covered in the next post.
Asking participants to rank-order a series of companies, products or candidates is a common question type found in market research or political polling. For example:
Please state your preference for the following candidates, with one (1) being least preferred and five (5) being most preferred.
As always your list should be randomized to prevent order bias. From the analysis standpoint, you can look at each contender’s median rank accompanied by the percentage of times they were ranked most preferred. It would not be appropriate to use averages, as we cannot say with certainty the distance between one position and another is equal.
The ‘winner’ or most preferred would be the brand that has the highest median rank (closest to five in this example). If you incorporate percentages in your analysis then you can say, for example, that twice as many respondents ranked Bugs Bunny as their most preferred candidate.
Survey questions that work with ordinal data, such as the scenario above, force respondents to make a choice. As survey author, you must decide whether or not to allow for ties (the same rank being applied to more than one category).
Most multivariate statistical routines require continuous data, but there are routines that can work with ordinal data such as logistic regression. This technique allows you to incorporate multiple variables to predict an ordinal variable.