Coding manuals derived post hoc from data cannot validate differences.

Evidence

This is one criticism that applies to all four content analyses; the investigators invariably examined their data in search of contrasts between prejudiced and unprejudiced subjects before they made up a scoring manual. The coding categories were defined so as to capture the contrasts that seemed to be in the data. The blind coding from the manual is simply an effort to show that differences which appear to exist when one knows whether or not a subject is prejudiced can also be found when one does not know. And also, of course, to show that the content categories can be communicated from one person to another. — John T. Jost and Jim Sidanius, Political Psychology, 601

Categorization

• XREF: Connects to the distinction between exploratory and confirmatory analysis, and to the modern debate over researcher degrees of freedom in content analysis and qualitative coding. — connects to: Survey data-file encoding artifacts can corrupt key variables. — Ilyana Kuziemko and Ebonya Washington, Why Did the Democrats Lose the South?…, 1022