After performing Response Surface Method for Media optimization using two component CCD, i when analyzed data i found the quadratic Model fits with 0.002 P value in next step, it also has singificant lack of fit value also with 0.002 p value. So singificant lac of fit is said to be bad. now all analysis suggest that quadratic model fits the best. how to conclude now?
If the lack of fit is significant it indicates that the error model and the experimental error or pure error are significantly different and this is not desirable. this may be due to two reasons: 1) the pure error is very small, you might check out the variation between replicate experiments and the different experiments, the variation between replicates can never be less than the variation enter experiments which are not. 2) the model really has a very big mistake, in this case must do a thorough analysis of the waste to deterctar the trouble spots or adjust other model.
Hi Vihang,
In some cases and due to replication and repeated measurement the lack-of-fit test will be significant. This is because of pure error has been underestimated. In such cases the lack of fit test is no longer be valid test and decision about the model should be made based on other statistical criteria such as PRESS, R2, adjusted R2, predicted R2, normality plots, actual versus predicted plots and rely on confirmation runs to verify the experimental results in real life applications.
As you may know lack of fit means that the values computed from your model differs significatively from the real values obtained in the experiments. Compare experimental results with the predicted ones and if the difference is really significative try reducing your model by using the toools Design Expert gives (BACKWARDS, STEP-WISE OR FORWARD) until you get the best model available.
- Badges
- Science method
- Similar topics
- Biological Science
- Ecology