Researchers often encounter ordinal measures that they wish to examine as dependent variables in their research—variables where the categories are ordered (running from high to low or low to high), but the distance between the categories is unknown. For example, respondents might be asked if they strongly disagree, disagree, agree, or strongly agree with a statement. Or, rather than give an exact value for their years of education, respondents might be asked whether they had no education, some grade school, grade school graduate, some high school, and so on. While it might be tempting to treat ordinal dependent variables as though they were continuous and use techniques like ordinary least squares regression, this can result in misleading estimates of independent variable effects and incorrect tests of statistical significance. Ordinal regression models are therefore preferred under these circumstances—but there are many ordinal models to choose from. This entry begins with a detailed discussion of perhaps the most popular choice, the ordered logit model (also called the proportional odds model). The discussion will cover when the model might be appropriate, the major assumptions of the model (and how they can be violated), and how to interpret model results. However, in many cases, other ordinal models and methods will be more powerful or appropriate. This entry therefore also discusses the ordered probit model, the generalized ordered logit model, interval regression, scoring methods, heterogeneous choice/location scale models, stereotype models, stage models, and the rank-ordered logit model—as well as briefly explains when and why each might be preferred.
By: Richard A. Williams & Christopher Quiroz | Edited by: Paul Atkinson, Sara Delamont, Alexandru Cernat, Joseph W. Sakshaug & Richard A. Williams Published: 2020 | Length: 10 | DOI: http://dx.doi.org/10.4135/9781526421036885901 |