How to Check a Regression Fit
podcast episode with speaker Fred Schenkelberg
When fitting a line or curve to data, it’s a model. When modeling, it is worth remembering the George Box quote, “Essentially all models are wrong, but some are useful.” Yet, how do we separate a useful model from one that isn’t useful? One step in finding a helpful regression model is to consider the residuals. BTW: Weibull analysis is another term for regression analysis.
Residuals in regression analysis are the differences between the data and the model-predicted values. When the regression ‘fits’ the data, the residuals should represent the naturally occurring experimental (measurement) errors. These should be well-behaved differences that tend to fit a normal distribution centered on zero.
Let’s explore what residuals are, where they come from, and how to evaluate them to detect whether the fitted line (model) is adequate. Not checking or using a poor model is a recipe for major errors when making decisions. Checking residuals is one step to validating a model, yet it’s quick and easy to accomplish.
This Accendo Reliability webinar was originally broadcast on 10 May 2022.
Related Content
Regression Metrics episode
How do I Control a Process That Trends Naturally Due to Tool Wear? article
Are You a Keyboard Jockey? article
Reliability Engineering Statistics course
Starting with Mathematical Foundations with Fred Schenkelberg episode
Leave a Reply