Video

Ok, then the design run may be influencial in terms of the model or the individual coefficients. One should at least go over the experimental log book and double-check that everything went fine with the experiment. These plots are tools to uncover influential design runs. Best wishes, Erik

@@erikvanhatalo7 Ok, thanks. I will double check. Just a bit unsure about these plots. Does it mean that in order for a design to be accepted, all runs must lie within the limit of these two plots?

Glad you found it valuable. 👍

I've been thinking about it, and I think info from higher order effects is less likely to be statistically significant and more difficult to interpret. Is that the reason why they are selected?

Yes, the fact that higher order interactions are less likely to accurr (a priori) makes them good candidates to "sacrifice" to blocks. Higher order interactions do occur but have been shown to occur much more seldom than main effects and low order interctions in real life. Best wishes! Erik

@@erikvanhatalo7 thanks

I am glad you liked it! :)

The constant value k in the figures from the software is a constant that may be added to the ressponse value. For example, there may be instances if we have responses with value 0 (zero). Then the logarithm ln(0) is not defined and the conctant k can help in that problem.

Hi! Thanks for the nice comment. I am not familiar enough with JMP or Minitab to know if they have some special functionality such as for the POE - propagation of error. But it should definitely be possible to do the analysis as a combined array approach and just keep track of which variables are "noise" variables and which are design variables. Sorry that I do not know the details. Best wishes, Erik

Thanks! I am glad you find some useful videos here! It's a rather dry topic, but I like it :)

I am glad it was of some value! 👍

Just spotted that too. Mistake in the book I think

Thanks Dennis, good of you to notice that! You are correct in that the pressure variable in the natural units does not match the coded pattern. This image is borrowed from the book companion slides so it is a typo in Montgomery's book. Nonetheless, the mistake was also mine in not noticing that. Best wishes, Erik

Thanks! Glad it was of value! 👍

Glad it was of value! 🙂

@@erikvanhatalo7 Dear sir, Very useful content for preparing a thesis.

Thanks, cheers!

Thanks! Glad it was useful! :)

Contact info: www.ltu.se/staff/e/erivan-1.81005?l=en

Thanks Erik

Thanks! Do you mean the 2(7-4) design? after full fold over? Trying to remember from last year. I do not think it may be explained easily in this comment. But alias patters can be calculated from the complete design, by hand. In this case I think I took it from the Design Expert software.

@@erikvanhatalo7 Thanks for your reply Erik. I managed to figure out how to get the alias pattern for res III + full fold over. I’d like to ask two questions if I may 1. If you add a full fold over to a 2(3-1)III you get a 2^3 full factorial and you can analyse this using method of contrasts, Yates method or matrix methods. However, if we take the first 8 run res III (the 2(5-2)III) then add a fold over you don’t get a full factorial. What would you call this design? Does it even have a name? I can see how to calculate effects using the method of contrasts or matrix methods but I don’t think Yates method will work here. For 2 level full and fractional factorial designs Yates can be used to calculate effects but not for other designs (such as fractional factorial with full fold over added). Does what I’ve said make sense? Sorry for such a long rambling message.

Hi! I will not provide pdf material of my teaching slides publically unfortunately. If you need to go in depth in this area I would recommend a textbook such as Montgomery's Design and analysis of experiments.

Kk sir

Hi! Since the factors are categorical these dummy variables A[1], A[2], etc are used to indicate effects for certain settings and combinations of the categorical factors. This is why it looks a bit strange.

Thanks! Glad you liked it. Now in order for you to fit second order terms, such as A^2, without confounding you typically need a second order design. You may want to chck out central composite designs or Box-Behnken designs. These are examples of common designs in response surface methodology. Cheers!

Thanks! Glad you liked it. I never made a video just on rotatability alone. May come in the future. For the CCD you can make the design rotatable throuh the aloha value. Rotatable designs have equal prediction variance at all positions that have equal distance from the design center. Many times reasonable choice.

Thanks for your reply Erik. It’s the whole idea of why you’d pick a certain value and why a particular value of alpha gives equal prediction variance that I don’t understand. I don’t even understand what equal prediction variance means in the context of a certain response surface eg parabola, stationary ridge, rising ridge or saddle (col or minimax). Going to hit the books on this subject over Xmas