Coding, Fixed

Fixed coding refers to coding approaches that involve predefined coding categories that are collectively exhaustive and mutually exclusive. These categories are logically derived by researchers based on a priori empirical knowledge or preexisting theories. It is mostly used in quantitative research, where data coding and analysis are deductively driven. The goal is to test researchers’ preexisting propositions or theories against patterns and relationships as shown in the empirical data. It can also be a way of testing the emergent categories from qualitative research such as thematic analysis. Developing a fixed coding scheme requires that researchers know prior to analysis what they are looking to test in the study. This entry examines the two requirements for creating coding categories in fixed coding (collectively exhaustive and mutually exclusive), discusses the importance of unitizing, provides some examples of content analysis that lends itself to fixed coding, and concludes with an exploration of coding reliability.

After a sample is obtained and units of coding are determined, coding categories are defined. There are two critical requirements for designing coding categories in fixed coding: The categories must be collectively exhaustive and mutually exclusive. To put it simply, every unit of data should be placed into one and only one category. The first requirement—being collectively exhaustive—means that the categories for a variable should cover the entire range of possibilities, so that every unit of data can find one descriptive category that applies. To make sure the list is exhaustive, oftentimes fail-safe categories such as “not applicable,” “none of the above,” “do not know,” or “other” (often followed by a text box to describe the scenarios) are included. For example, if a question asks respondents to report which is their favorite platform of social media, the categories may include “Facebook,” “Twitter,” “Instagram,” “YouTube,” “Google Hangout,” and “Foursquare” as the major options specified for the question. There are, however, other social networking sites or tools that the researcher may not be interested in listing in the survey or ones for which the researcher simply may not be aware. For situations like this, an option “Other: Please specify” should be an added category to ensure every respondent has a category to select. As another example, to code for the sentiment of a tweet, it is not enough to have “overall positive” and “overall negative”; there should also be a category “neutral or balanced.”

The second requirement—being mutually exclusive—refers to having clear boundaries between categories so that every unit belongs unambiguously to only one category. There should [Page 146]be no overlaps between categories. For example, for a question that asks “Which of the following best describes your age group?” a list of categories like “a) under 21, b) 21 to 25, c) 25 to 35, d) 35 to 45, e) 45 to 55, f) 55 to 65, g) 65 or older” are problematic. A respondent of age 25, for example, would find himself or herself belonging to both b) and c); a similar issue exists for someone age 35, 45, 55, or 65. These response categories, therefore, are not mutually exclusive. A list of mutually exclusive categories would be “a) under 21, b) 21 to 24, c) 25 to 34, d) 35 to 44, e) 45 to 54, f) 55 to 64, g) 65 or older.” These two requirements—collectively exhaustive and mutually exclusive—guarantee that the universe of the given phenomenon is fully and cleanly partitioned into the coding categories. They are critical to data coding in both survey design and quantitative content analysis.

...