Categorical Variables in Regression Analysis: A Comparison of Dummy and Effect Coding

Abstract

<span style="font-family: 宋体; font-size: small;"> </span><p style="margin: 2.5pt 0cm; mso-pagination: widow-orphan; tab-stops: 17.05pt center 207.65pt;" class="MsoNormal"><span lang="EN-US" style="font-size: 12pt; mso-font-kerning: 0pt; mso-fareast-language: EN-US; mso-bidi-language: AR-OM;"><span style="font-family: Times New Roman;">The use of categorical variables in regression involves the application of coding methods. The purpose of this paper is to describe how categorical independent variables can be incorporated into regression by virtue of two coding methods: dummy and effect coding. The paper discusses the uses, interpretations, and underlying assumptions of each method. In general, overall results of the regression are unaffected by the methods used for coding the categorical independent variables. In any of the methods, the analysis tests whether group membership is related to the dependent variables. Both methods yield identical <em>R</em>²and <em>F</em>. However, the interpretations of the intercept and regression coefficients depend on what coding method has been applied and whether the groups have equal sample sizes.</span></span></p><span style="font-family: 宋体; font-size: small;"> </span

Factors influencing the length of hospital stay of patients with anorexia nervosa – results of a prospective multi-center study

“…We could not reach normal distributions using common transformations (log, exponential, square root) and hence decided to dichotomize these variables using a median split. All categorical and ordinal variables were dummy-coded [ 31 , 32 ]. In detail, we included the following predictors: median-split age (24 years and younger; 25 years and older), mean-centered age of onset, living alone (yes; no), partner (yes; no), education (base level: primary education; contrasts: secondary education, university degree, other), previous hospitalizations due to a mental disorder (yes; no), AN subtype (restrictive; purging), mean-centered admission BMI, dummy variables for the centers (base level: BB; contrasts: HH or P), admission PHQ-9 score, admission PHQ-15 score, admission GAD-7 score, median-split FEVER (sum score of 42 and below vs. above 42), and all EDI-2 subscales.…”

Characterizing the Occurrence of Dockerfile Smells in Open-Source Software: An Empirical Study

“…Projects with different programming languages are likely to have different development paradigms [26]. Effect coding [38] provides one way of using categorical predictor variables in the linear regression model. With effect coding, the experimental effect is analyzed as a set of contrasts that opposes all but one experimental condition to one given experimental condition.…”

Livelihoods, precarity, and disaster vulnerability: Nicaragua and Hurricane Mitch

“…The diversified livelihood category serves as a reference category. The use of effects coding instead of dummy variable coding is beneficial because it provides (i) estimates in relation to a mean livelihood effect, as opposed to comparison versus a reference category, and (ii) estimates of a main effect of damage when incorporating interaction effects (Alkharusi, 2012).…”