Influence of Data Quality on Regression

Convergence can be slow, especially if the minimum exists in a long narrow valley in parameter space. Press et al. describe several more efficient multidimension optimization strategies. 

Regression problems in impedance spectroscopy may become ill-conditioned due to improper selection of measurement frequencies, excessive stochastic errors (noise) in the measured values, excessive bias errors in the measured values, and incomplete frequency ranges. The influences of stochastic errors and foequency range on regression are demonstrated by examples in this section. The issue of bias errors in impedance measurement is discussed in Chapter 22. The origin of stochastic errors in impedance measurements is presented in Chapter 21. 

---