Hi VK sir,

While running process capability on a certain CTQ, with 50 readings, I am not certain whether to go by normal capability analysis and use Box Cox, or to go with non-normal and do a Johnson Transformation. The data is non normal, and continous. Both analysis give me different values (Box Cox: Overall Z Bench: 7.05, Johnson Transformation: Overall  Z Bench: 7.12).

I would like to know whether my approach is wrong and I should categorize data  differently (normal/non-normal) or which test is more reliable.

Thanks!

Dear Faisal,

Your question gives no conclusion and clue of the data type of 50 readings

if the data type is contineous go for normality test if it fails go for non parametic test rather then transformation.

CAUTION: Only do transformation when mbb suggest . It can create lot of complication in data analysis in later stage.(Box cox is stronger and better transformation. if box cox fails then only go for johnson transformation.)

Suggesting to not to go for transformation untill and unless you are expert

Regards

Navin Rohilla

BB

Hi Navin,

Thanks for your reply. It was eagerly awaited. Just to reiterate, as stated in my question, the data type is contineous, and failed normality test. And thus, I was split between box cox & johnson transformation. As advised by you, you regard box cox as the stronger option. But I was told that Johnson was a much powerful transformation.

I shall follow your instructions and go with box cox, but I would really appreciate more clarity between both, in order to make better inferences and decisions.

Thanks!

 

Box cox takes your data away from reality and Johnson also takes your data more away then the Box cox transformation hence for better Analysis, Box cox transformation is better then Johnson. When Box cox some times fails to convert the data to normal,due to lamda value or if the data value is less then 1 or in negative or any other cause. so we can use Johnson transformation in these cases. so johnson transformation is stroger to transform non normal to normal but gives poor to evaluate analysis. so first choice is Box cox if it fails then go for Johnson transformation

But as said earlier would suggest you to not use any of the tranformation go for alternative tests. Transformation makes your analysis very complicated.

Navin Rohilla

BB

Max New york Life

Thanks a lot Navin! Great info.