Variance regression models in experiments with few replications

✍ Scribed by P. A. Barbetta; J. L. D. Ribeiro; R. W. Samohyl

• Publisher: John Wiley and Sons
• Year: 2000
• Tongue: English
• Weight: 168 KB
• Volume: 16
• Category: Article
• ISSN: 0748-8017
• DOI: 10.1002/1099-1638(200009/10)16:5<397::aid-qre348>3.0.co;2-x

✦ Synopsis

Variance models are highly important in developing robust products and processes. These models can be employed in process robustness studies through the use of response surface methodology. In most of the applications the models are constructed in terms of the logarithm of the sample variance or the logarithm of squared residuals. This paper presents an alternative to the standard logarithmic transformation and a procedure for aggregating sample variances with squared residuals. In experiments with few replications, these procedures result in the least squares method producing more accurate and robust estimates of the response model, according to assessments made by Monte Carlo simulations.