Ordinal Attribute Study – Variance To Standard

Six Sigma – iSixSigma Forums General Forums General Ordinal Attribute Study – Variance To Standard

Viewing 6 posts - 1 through 6 (of 6 total)
  • Author
  • #53552


    Hi To All,

    I am glad to see that the “wave of contentiousness” seems to have passed, as this site is a valuable tool for learning.

    I have a problem or a study of Ordinal Attribute Data, wherein there are three evaluators, who are given parts (thirty in total) to evaluate on a scale of 1-10, 1 worst, 10 best. The same thirty parts were evaluated by a team of experts; they assigned or established “standard or master ratings” to these parts.

    My prof has indirectly suggested that the parts should be blocked, which makes sense as we are interested in evaluator variance (performance) compared to the accepted standard and compared to other evaluators.

    My inclination is to (Minitab) run an Attribute Agreement Analysis, and interpret the results. Yet this option does not afford you the option of blocking…..what the heck am I missing? Is another method preferable? If so, what is the technical or procedural rationale?

    Thank you for your assistance…




    Based on your description of the problem, blocking does not seem to be required. The difference between evaluators as well as evaluators versus the standard is part of the analysis. Evaluator is not a blocking variable.

    In general, blocking is required when some degree of randomization is not possible. Is there something about the problem that suggests a lack of randomization?



    Hi Thanks for the response.

    It seems as though the prof is indirectly suggesting that another method may or can be used to evaluate variance to an established standard or reference.

    The second issue regards interpretation of DOE results:
    Given – I ran a fractional factorial, resolution IV experiment six factors at two levels each, three replications, alpha = 0.05.

    The Minitab results are below; they reveal one of the rare occasions when, three of six main effects are significant, one two-way BD is significant (one factor B is not significant the other factor D is), also ABF a three-way is significant yet none of the factors is significant. The results seems kind of odd, although I am still prety new at this. (note A=belt Speed, B =Jet oven temp…etc)

    Estimated Effects and Coefficients for Rating (coded units)

    Term Effect Coef SE Coef T P
    Constant 6.438 0.07803 82.50 0.000
    Block 1 -0.125 0.11035 -1.13 0.265
    Block 2 0.000 0.11035 0.00 1.000
    Belt Speed 0.208 0.104 0.07803 1.34 0.190
    Jet Oven Temp -0.125 -0.063 0.07803 -0.80 0.428
    Pot Life -1.458 -0.729 0.07803 -9.35 0.000
    Oven Temp -2.542 -1.271 0.07803 -16.29 0.000
    Oven Time -1.625 -0.813 0.07803 -10.41 0.000
    Mesh 0.125 0.062 0.07803 0.80 0.428
    Jet Oven Temp*Oven Temp 1.125 0.563 0.07803 7.21 0.000
    Belt Speed*Jet Oven Temp*Mesh -0.375 -0.188 0.07803 -2.40 0.021

    S = 0.540583 PRESS = 18.1972
    R-Sq = 93.40% R-Sq(pred) = 88.89% R-Sq(adj) = 91.62%

    Analysis of Variance for Rating (coded units)

    Source DF Seq SS Adj SS Adj MS F P
    Blocks 2 0.500 0.500 0.2500 0.86 0.433
    Main Effects 6 135.625 135.625 22.6042 77.35 0.000
    2-Way Interactions 1 15.188 15.188 15.1875 51.97 0.000
    3-Way Interactions 1 1.688 1.688 1.6875 5.77 0.021
    Residual Error 37 10.812 10.812 0.2922
    Total 47 163.813

    Any constructive thoughts on this?

    Thanks for your help….





    A little clarification regarding the Attribute Agreement Analysis;

    We were given the following hint which leads me to beleive I should block on “parts”
    “(Hint: the 30 parts were run under different conditions and present various degrees of delamination. The focus is on determining if any – or all – of the operators rate the parts significantly different from the master ratings or from each other. Also, should the parts be blocked?).” The thinking that “parts” is a difficult or uncontrollable variable, manufactured under varying condtions at different times etc.

    Yet the easiest way to run the test, at least in Mintab 15 is an Attribute Agreement Analysis, yet I can’t determine how to block on parts with this option.

    Thanks for your help.




    It’s OK if the parts are different (it’s actually important that they are) as long as the differences (delamination differences?) are representative of the differences that are the basis for the rating.

    The key need for randomization is in the presentation order of the parts to the raters. This needs to be randomized to eliminate bias by rating order (e.g. ratings may change after seeing many parts compared to seeing the first parts)

    If parts are going to need to be uniformly rated regardless of what conditions they were run under then I don’t see a need for blocking.



    Hi thanks for the response, blocking on parts is necessary because the introduction states that they are run or assembled under varying conditions..that are difficult to control or uncontrollable.

    I believe I solved the dilemma, by Minitab>>Stat>>ANOVA>>Two-Way>> input Parts in the “Row Factor” window, Measurement in the “Response” and operator in the “Column Factor”

Viewing 6 posts - 1 through 6 (of 6 total)

You must be logged in to reply to this topic.