Comparison of Logistic Regression and Linear Regression in Modeling Percentage Data

• DOI: 10.1128/aem.67.5.2129-2135.2001
• Publication Date: 2002-07-27T10:00:58Z
• Authors: Lihui Zhao; Yuhuan Chen; Donald W. Schaffner
• Keywords: Goodness of fit; Data set; Regression diagnostic

ABSTRACT Percentage is widely used to describe different results in food microbiology, e.g., probability of microbial growth, percent inactivated, and positive samples. Four sets percentage data, percent-growth-positive, germination extent, for one cell grow, maximum fraction tubes, were obtained from our own experiments the literature. These data modeled using linear logistic regression. Five methods compare goodness fit two models: predictions closer observations, range differences (predicted value minus observed value), deviation model, regression between predicted values, bias accuracy factors. Logistic was a better predictor at least 78% observations all four sets. In cases, models much smaller. The correlation always stronger. Validation (accomplished part set) also demonstrated that model more accurate predicting new points. Bias factors found be less informative when evaluating developed since neither these indices can zero. Model simplification with set. simplified as powerful making full it gave clearer insight determining key experimental