What is your predicted R-sq? In your case, pure error is very low when compared to residual. In that case, lack of fit will be significant. It shows that there are some issues in your model. Check if the quality of your results are right. If you are confident, then try removing the outliers. 

thanks

Lack of fit techniques are applied to check the models are adequate. It should be insignificant for the model to fit well in the experimental design. If pure error is less the residual error then Lack of fit is significant. Try to use narrow range of variables if you want  the insignificant lack of fit. 

The lack-of-fit test is used as support test for adequacy of the fitted model. It is a measure of the significance of the lack-of-fit error which is its contribution to the residual error; a statistic made up of lack-of-fit and

pure error. The pure error is a reflection of the variability of the

observations of each treatment, the sum of squares of which is the sum

of the squared deviations of the responses from the mean response in

each set of replicates. The lack-of-fit error, on the other hand, measures

the error arising from a deficiency in the model. In a situation where a

model fits the data well, the mean square of the lack-of-fit error will be

very close to that of the pure error and the resulting ratio, the Fstatistic,

will be small (close to 1). If the error probability, P of the small

lack-of-fit F-statistic is large compared to the significance level, then

the lack-of-fit is not significant and the model adequately explains data

in the region of experimentation. Conversely, if the F-statistic of the

lack-of-fit error is large (indicating that the bulk of the error of the

residuals is due to lack-of-fit) and its corresponding error probability, P

is small, then the lack-of-fit test is significant and suggests the

inadequacy of the regression model to explain the data variations. Please visit the link Article Response surface modeling and optimization of major medium v...

Hi

For the best model, p-value of model should be significant (p0.05)

If lack of fit comes significant then these three methods should be tried: 1- Central points should be repeated 2- Higher grade models should be selected 3-The transfer function must be used for the data

Best regards,

Please help me in this.

1. If lack of fit is not mentioned in ANOVA of all the response, RSM, Design expert then is that kind of optimization is correct.

2. If p value of one factor out of three factors is not significant then is that optimization is correct