Answer this Question

sachin_01: Add as a Colleague
Hi Moin, There are other tools which you can use instead of ANOVA and DOE. Cominig to your Question , ANOVA is test of means, now when you say your data is Attribute, you need to know the Y data type as well i.e , your problem metric is it continous ?, thumb rule says, when your Y is Continous and X is Attribute your can use ANOVA. I hope I have answered your ANOVA Que. Now the DOE, Yes you can do DOE,but it is a very lengthy process to conduct DOE, Before Conducting DOE you may have to understand why DOE, what are you going to infere out of DOE, so my suggestion is that ask your project advisor as to why should I use DOE in Improve phase. Regards, Sachin

09/07/2009 12:54:12 AM EDT

putnam: Add as a Colleague
01/12/2010 11:02:35 AM EST

ANOVA is used to determine if the means (of the continuous Y's) of at least two nominal/ordinal X's are different. The null hypothesis is that they are NOT different. Since you have NOMINAL (aka binomial/attribute) data, you can apply a similar approach but the mathematics are different. You should use a Chi-squared test to determine if the pass/fail rates are different. You can use such approaches in the ANALYZE phase to identify potential causes. You can use this approach in the IMPROVE phase to confirm your improvements are different (BETTER) than the baseline condition. A DOE could be used to test different designs or methods and the Chi-squared to compare them. Ideally, however, you have better information if you can identify a continuous variable to test as a response instead of the pass/fail data.

sachin_01: Add as a Colleague
12/14/2009 4:45:44 AM EST

Hi Aramos, If the data is Attribute you cannot use non parametric test, coz we use non parametric test when data is non normal, and in attribute data we do not plot NPD, we plot Binomial or poisson Distribution, hence we cannot use Non Parametric test, if I am wrong kinldy explain, which Non parametric test we can use and ho ? Regards, Sachin

KKH: Add as a Colleague
12/04/2009 12:13:27 AM EST

For Attribute data (analysis of categorical data) I would use "goodness-of-fit" test. Chi square statistic.

aramos33@netzero.net: Add as a Colleague
11/15/2009 11:15:13 AM EST

If you have attribute data you must use non-parametric analysis....

ghuntzinger: Add as a Colleague
11/09/2009 9:05:15 PM EST

Moin: I think I remenber the attribute AnoVa requires a minimum of 15% of the parts in the study to be out of specification. The operators involved must of course pass/fail correctly each part in the study. The parts are not identifiable to anyone but the person doing the documenting. Specific details can be found on the web for attribute AnOVa to properly set up your analysis. Yes the DOE will look at interaction of the variables. This is not only a difficult test to set up, and also depending on the number of variables that can potentially interact will take a very long time to complete with each results dictating each next analysis. You will need an SME in DOE to correctly prioritize the potential interacting variables, structure of the data to be analyzed, and the correct variable exchange. My advise is get with your Prof. quickly for how involved their expectation is for the minimum number of variables required for the DOE. You will be very much pressed for adequate time. Good luck. Gary

Navin.rohilla@yahoo.co.in: Add as a Colleague
11/03/2009 10:07:07 AM EST

Dear, 1st thing is that ANOVA can not be performed for discrete attribute data. If the data is discrete Nominal then check the Normality of the data and test for equal variance. If both test passes then you can select anova other wise suggest you to go for kruskal wallis test or moods median test. If the discrete nominal data is average or sum then it can behave like contineous if the normality test is passed by it. Anova is hypothesis test to validate that the two or more sample has same or differnt means or not. Suppose you have 3 different method to prepare the item. which method gives you the best mean and is it significantly better then the other two can be detected by comparing there means of defects count. DOE is the worlds best tools if learnt and adopted carefully. it checks the combination and there mutual interaction of various factor to give best product but cumbersom to perform. kindly check the cost involve and feasability of the DOE in your project Black belt Navin Rohilla