Previous: Codebook

Up: Develop Coding Procedures

Next: Coder Training and Evaluation

Coding Models

1. Categorical Classification: Either mutually exclusive coding where coders will select the single best classification for the unit of analysis, or non-exclusive coding where coders will select all categories relevant to the unit of analysis
2. Dimension / Affect Coding: Coders will score each unit of analysis on a relevant dimension (scale). Scores are specific to the dimensions on which they are evaluated. Any numerical value can be used to establish the affect scale, but it is essential to carefully define what each discrete score represents. For example, on an affect scale {$-2$ , $-1$ , 0, 1, 2} coders should be provided with specific examples of what constitutes a $-1$ as opposed to a $-2$ score, and so on. Coders should be given a single dimension, a choice of dimensions, or multiple dimensions on which to code each document. Each dimension should be a single, exhaustive scale containing all relevant positions. Some examples of dimensions: ``Economic liberalism / conservatism,'' ``Leadership ability,'' ``Hawk/Dove,'' etc. In single dimension coding, coders are only given one dimension on which to code, so each unit of analysis will be coded on that dimension or coded as irrelevant. In multiple dimension coding, coders are given a list of dimensions, and code the unit of analysis on all relevant dimensions.
3. Develop coding infrastructure: Consider whether hand coding done with computer assistance would be helpful, or whether keeping track in a spreadsheet or on paper is sufficient. Commercial options include programs like Atlas.ti, Nud*ist, Xsight, or EZ-text. Some projects develop their own computer programs for collecting coding data. The selection of a coding interface can have a dramatic effect on the results. For example, see discussion in Kwon, Shulman & Hovy (2006).