An ordered (ordinal) dependent variableLogistic (logit) regression
Christer Thrane · 2019
Regression models for numerical, binary and unordered categorical y-variables are now among the tools in regression kit. The unknown distance between the values and answer categories of ordered variables makes linear regression a less than optimal modeling strategy, and the more appropriate procedure is the ordinal regression model. The challenge, since there are five answer categories for the dependent variable, is to convey as much information as possible with the smallest amount of numbers. The predicted probability approach is one viable means in this regard; another approach is the use of marginal effects. The benefit of using the marginal effect approach is its ease of interpretation; it tells what happens to y in percentage points when x increases by one unit, holding controls constant at some predefined, fixed level.