Dear Researcher,
I went through a lot of research articles using RSM for optimization of experimental conditions.
I am wondering why in many research articles the optimization carried out have the Lack-of-Fit (LOF) significant?
While it is clearly stated that for best the fitting, the model should be significant but LOF should not.
Can we still say that the model applicable for the process and is correct?
Please help.
Regards
Pankaj
Dear Pankaj,
A value for the Lack-of-Fit may not be given because it becomes insignificant in one of the case where the repeated data are similar and if one of the values of the repeated central point is different from the average.
For more details and information about this subject i suggest you to see links in topic.
-Modeling and optimization of fermentation variables for enhanced ...
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5491433/
-Response surface modeling and optimization of ultrasound-assisted ...
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3732422/
-One-Factor RSM - Stat-Ease
https://www.statease.com/productattachments/index/download?id...
-Simple Response Surface Methodology: Investigation on ... - MDPI
www.mdpi.com/1996-1944/8/1/339/pdf
Best regards
Are you saying there are published papers with significant lack-of-fit? That would make me worried. Lack-of-fit tests are not entirely reliable and there is considerable debate about which (if any) are good, but a small p-value for one or several lack of fit tests must be taken seriously and a paper reporting it should not go through peer review without a good explanation for the model misfit.
Dear Jonas Nyman,
Yes here is an example you can read:
Article Application of Response Surface Methodology (RSM) for Optimi...
For each response the value of LOF is significant even at
Thank you. Yes, they have a 3^3 experiment, 27 combinations in a full experiment. They do a fractional experiment and measure 15 of those 27. You simply cannot get around the fact that you have to do a lot of experiments...
Dear Jonas Nyman, Overall is it possible to present LOF value significant?
A significant LOF indicates that the model does not fit the data, i.e. it is wrong, and I would not expect such models to be published. However, LOF tests are known to be unreliable, so I cannot exclude the possibility that it may be justifiable in some circumstances. But then, the LOF should be carefully explained or commented on in the paper, and the model perhaps justified by some other means.
Dear, Significant value of lack of fit indicates that the model does not fit the data, which means the your model is wrong. I would recommend you to change your model or parameters' range because with significant lack of fit it would not expect the work to be published.
Dear Jabir Mumtaz, Thank you very much for your answer. But you should check the paper that I mentioned above in reply to Jonas Nyman.
Please give your opinion on this.
Yes dear Pankaj Kumar , I agree with you, but this is very rare. This is because sometimes the actual experiments can show significant LOF, but it should be insignificant. you can see in majority of latest research articles.
Pankaj Kumar you can check this article too to get an idea.
Article INVESTIGATION OF ELECTRIC DISHARGE MACHINING PARAMETERS TO M...
Is there any solution to lack of quadratic effect (insignificant order of 2) in CCD response surface model with good R-Squared and insignificant lack of fit to enable optimization?
2FI model also fit perfectly well with insignificant lack of fit but have a doubt in using it for the optimization of my response.
Thanks.
Dear Bashir Adelodun as experts said above, you should try to check the data of central points (0 level), I successfully, bring LoF insignificant as one or more of the data at central point was not correctly place or were of similar kind. try to check if the data is of same level or mean.
Thanks.
Pankaj Kumar Thank you for your response. I do have a good LoF (insignificant p-value) and a very good R-squared value but where I have problem is quadratic effect. My quadratic effect is not significant in the second order model. For the optimization, the second order model is highly recommended unlike the 2FI model.
Bashir Adelodun could you please share the model summary table here, or privately in message?
- Badges
- Science topic
- Similar topics
- Engineering
- Modeling