Univariate calibration by reversed regression of heteroscedastic data: a case study

Area of Science:

• Analytical Chemistry
• Chromatography
• Statistical Modeling

Background:

• Accurate calibration is essential for quantitative analysis using High-Performance Liquid Chromatography (HPLC).
• Understanding error sources in analytical methods is critical for reliable results.
• Traditional calibration methods may not adequately account for complex error structures.

Purpose of the Study:

• To investigate the calibration of acetaldehyde-DNPH using HPLC data.
• To identify dominant sources of uncertainty in the analytical method.
• To optimize calibration strategies by comparing statistical figures of merit.

Main Methods:

• Collection of replicate HPLC data for 33 samples across a concentration range.
• Application of weighted least squares (WLS) regression for calibration.
• Utilizing generalized least squares (GLS) for reversed regression to estimate method variance.
• Comparison of chi-squared and R-squared as figures of merit for calibration fit assessment.

Main Results:

• Data uncertainty is proportional to the signal over most of the tested range.
• Sample preparation error significantly outweighs measurement uncertainty.
• A cubic polynomial or statistically equivalent response function is required for adequate calibration.
• The method variance function includes both constant and proportional error terms.

Conclusions:

• Weighted least squares regression is necessary for accurate HPLC calibration of acetaldehyde-DNPH.
• The chi-squared statistic is a more reliable measure of calibration fit quality than R-squared.
• Analytical method development should prioritize minimizing sample preparation variability.