# Thread: Penalty reward contrast analysis for identifying Kano model

1. ## Penalty reward contrast analysis for identifying Kano model

Dear all
I encounter with the tough statistic methodology as PRCA(Penalty Reward contrast Analysis) during my thesis research in identifying Kano model attribute as whether these attributes are Basic Performance or excitement factors.In the first place I intend to use multiple regression to find the derived importance(refered as regression coefficient) of each attribute regressing on the single dependent variable as shown below

CS = A1*x1 + A2*X2 + .... +An*Xn
An : Derived importance or regression coefficient
Xn : attribute performance level
CS : Customer satisfaction
But the research of N.Kano clarify the asymmetry relationship between CS and attribute so the linear multiple regression is unappropriate to verify the relationship.So the effort to identify the Kano factor come in.I have tried to use PRCA as suggested by Brandt 1985 to identify Kano factor by using dichotomous dummy variable on the regression.The data will be coded as binary bit (0,1 for high level / 1,0 for low level attribute performance)for each attribute and done the regression for all coding attribute as follows

CS = CSaverage + A11*D11+A12*D12+A21*D21+A22*D22+A31*D31+A32*D32+.. .....An1*Dn1+An2*Dn2

Aij = dummies variable coefficient :i = number of attribute:J = Number of level (1 = high,2 = Low)
Dij = dummy variable :i = number of attribute:J = Number of level (1 = high,2 = Low)

and then compare the high level dummy coeficient with low level dummy coeficient to classify Kano factor.if high > low then exciter,High = low then performance and finally high < Low then basic.

****THEN THE PROBLEM IS****
I have many attributes to verify cause high multi-collinearity between attribute so the use of regression is distorted.So the partial correlation come in.The question is whether the partial correlation make sense in clasifying Low level Performance from High level performance in PRCA method as regression?

2. ## Kruskal

Hello,

I think you should use the Kruskal Relative Importance method. It's the sum of all semi-partial correlations between every attributes ... You can also use Dominance Analysis.

Good luck!

3. Try to get rid of some regressor variables to avoid multicollinearity. Alternatively you could use the importance grid to categorize your attributes (Vavra, 1996), if you have included some measure of explicit attribute importance in your survey design.

4. ## Dear all

I want to know how to do Penalty And Reward analysis using SAS or R. Please tell me how to do this
thanks

5. Hi Watchy29,

I really want to know the progress with your research. Can I find some your paper somewhere? Or Can I find some suggestions from you with my research? Thanks.

Email: bondlee115@hotmail.com