Fitting with Serial Correlation

If the y X, values are definitely nonlinear, a transformation must be performed. Let us see how this is performed in Example 3.3. In a drug-dosing pilot study, the blood levels of an antidepressant drug, R-0515-6, showed the drug elimination profile for five human subjects, presented in Table 3.19. 

Clearly, the rate of eliminating the dmg from the blood is not linear, as it begins declining at an increasing rate 6 h after dosing. The regression analysis for the non transformed data is presented in Table 3.20. 

Let us compute the Durbin-Watson test for a lag of 1, using the six-step procedure. 

P 0 (serial correlation is significant and positive over time). 

FIGURE 3.12 Residual vs. time logic y transformation. Example 3.3. 

---

Several commercially available mixes of pesticide compounds were used for calibration of the system by serial dilution and external standard techniques. A least squares regression was applied to curve fit the calibration data. Calibration curves for organochlorine pesticides (OCPs) were linear ranging from below 0.01 to lOng/pL with correlation coefficients ranging from 0.994 to 0.999 (Tables 4.12 and 4.13). 

---