iSixSigma

Regression v. DOE

Six Sigma – iSixSigma Forums Old Forums General Regression v. DOE

Viewing 14 posts - 1 through 14 (of 14 total)
  • Author
    Posts
  • #42538

    TVI
    Member

    When would you select the use of regression over DOE and vice versa….thanks!

    0
    #134340

    Mr IAM
    Participant

    Hmm.. I’ll take a stab at this one.
    I would use regression when I am trying to model, or estimate / predict an outcome based on some assocaited factors.  I.e.. like utilities use temperature to predict electricity usage.
    I would use DoE for process characterization.  When I do not know what key factors influence an outcome I would set up an experiment to determine which factors have real effect and which ones do not.  Regression is sometimes done on happenstance or historical data where true relationships between factors and outcomes can be hard to detect and / or they could be influenced by an additional factor that you where not aware of.  DoE has the advantage of collecting data real time and controlling as many factors as possible so it is easier to get a clear picture of what is going on.  DoE is even more powerful if you can add Regression as part of the analysis method.
    In analysis methods many people will start with DoE, to collect data on factors and outcomes and then use Regression, in addition to ANOVA and Factorial Plots to further analyze the results.
     I hope that helps!  Cheers!

    0
    #134342

    Robert Butler
    Participant

      The short answer is you wouldn’t select regression over DOE.  Regression is a method of data analysis, DOE is a method of data collection.  As MR Iam noted – you use regression to analyze the data gathered – that data can either be happenstance or planned (designed). 

    0
    #134347

    Mr IAM
    Participant

    I can see from your post I need to work on being more concise!
    :-)
    Cheers!

    0
    #134351

    TVI
    Member

    Great stuff….just what I needed….thanks again MR Iam and Dr Butler.

    0
    #134352

    Sigma Guppy
    Member

    Gents,
    Can you speak of DOE as causation and regression as correlation?
    Regards.

    0
    #134354

    The Lorax
    Member

    One thing I’d add:
    You can use regression rather than a DOE when there is no way that you are going to be allowed to make parts at the setting levels called for in a DOE.
    This can happen if the process being studied produces expensive items.
    Regression can be used to look at passed production and give you a proportion of the information that a DOE would have.
    However, a DOE will give a far better picture than regression ever will so if you have the choice, go with DOE.

    0
    #134356

    Mr IAM
    Participant

    I suppose you could but then correlation is it’s own statistic and sometimes that causes confusion. 

    0
    #134359

    Sigma Guppy
    Member

    The Lorax,
     
    Much Thanks.

    0
    #134366

    Robert Butler
    Participant

      I’m not trying to be offensive or dogmatic here but I do want to reiterate – One doesn’t “use regression rather than a DOE”.  As I said in the first post – regression is a method of data analysis, DOE is a method of data gathering. 
      I’m offering this repetition of thought because I have run into a number of individuals who believed DOE and regression were synonyms. When I ask these people how they analyzed the data their response is, typically, “We did the analysis. We ran the DOE and it didn’t work!”  In other words they ran DOE, looked at the numbers, declared the DOE a failure because “none of the experiments provided an optimum result”, and decide that the DOE effort was of no value.   In every instance where this occurred I’ve analyzed the data from the design using regression methods and either identified the optimum combination of variables or was able to use the results of the analysis to point in its direction.
      The regression analysis of data gathered using the methods of DOE usually results in information of a higher quality than information gleaned from the regression analysis of happenstance data (historical, retrospective, production, etc.). I say “usually” rather than “always” because if you don’t think carefully about your design and what you want to do with the data resulting from you designed gathering efforts you can wind up with a block of data of little or no value.
      As for viewing DOE as causation and regression as correlation – I wouldn’t. 

    0
    #134368

    Vilfredo
    Member

    R,Excellent reply to this thread, and agree completely. How would you classify EVOP in the scheme of things, DOE, analysis approach, method of improvement, other?V

    0
    #134372

    Craig
    Participant

    EVOP is definitely an improvement approach where you make incremental changes (in a production mode).
    It would be classified as more of a DOE approach because you are changing input factors while monitoring a response. It is also used in lieu of RSM as an optimization tool.

    0
    #134870

    Jonathon Andell
    Participant

    Regression is primarily a random effects model, where the X is “free to roam” within a range. DoE probably works best for a fixed ewffects model, where the X is set at specific values.I use the analogy: random effect models use “free range chicken” independent variables, while fixed effect models keep their chickens in coops.Hope this helps.

    0
    #134883

    Anish
    Participant

    Well said till now, I would like to add something. Basically regression and DOE are meant for modelling some phenomenon in terms of some variables and thereafter use it for analysis, improvement or optimization etc. The phenomenon could be a process or product characteristics( e.g. in robust design). Depending on from where data come for modelling we can use regression or DOE. In regression we have historical data and in DOE we generate data. Every approach has its advantages and limitations depending on resources and application… 
    I hope it gives some more clarity

    0
Viewing 14 posts - 1 through 14 (of 14 total)

The forum ‘General’ is closed to new topics and replies.