Year: 2021
Type: preprint
Abstract: Years of education is a commonly used outcome variable in many lifecourse studies. We argue that such studies may derive additional insights from analysis of years of education as an ordinal rather than continuous outcome. The conservative approach that we advocate for---the cumulative link model---produces estimates with acceptable properties when not the true model unlike the standard linear model, which experiences failures when inappropriately used. We illustrate this via simulation. In particular, studies of interactions can yield high levels of false discovery when the linear model is used for analysis of data from the cumulative link model; the cumulative link model avoids this problem. Further, estimates from the cumulative link model can be tested to determine whether the linear model is appropriate. We use empirical data to illustrate additional insights that are readily derived from application of the ordinal model and offer a suggested workflow for future analysis.
Institutions: Stanford University, University of Wisconsin–Madison
Language: English
Subfield: Statistics and Probability
Field: Mathematics
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W4250971671