The software says the best distribution to analyze my set of data is Johnson family distribution.
One thing I found in both Johnson family analysis (one with no LSL & the other with zero as LSL) is the P value remains same.
Yes, this is true because the p-value is not related to the spec limits - only the curve.
Assuming the limiting factor for P value is 0.05 my process is well with in normal distribution in both the cases. So can I say my process is normal?
I do not look for a "limiting factor", I look for the
highest p value. You can use the [Select Distribution] button in the software to
force the software to calculate the statistics for any of the distributions in its list. Force it to calculate the normal distribution, and compare it to the best fit distribution. In order to accept that the normal distribution is "close enough" of an estimation of the data, either it should be the best fit (it is not, the Johnson Family is) or the p-values should be relatively close (for example, if the best fit p value was .50 and the normal distribution fit was .45, I would call it "close enough" to normal as a fit) You might say that if the best fit is less than .05, then I would not trust that
any of the curves is a good fit.
Also I see there is no Pp value indicated if we do not specify LSL (in this case the Ppk is good), can you explain me why it happens?
The capability is only being evaluated by its distance from the USL. The distance from 0 is not relevant.
The Ppk value becomes worse if we specify zero as LSL, so you have suggested not to go by this method, Pls explain me how you arrived at this conclusion.
Because, by specifying 0 as a LCL, the calculations
assume you are trying to center the process in the middle of LCL and UCL. It cuts everything in half. But the target is not the "middle",
it is 0! That is why it does not make sense to use 0 as an LCL.
Hope this helps!! By the way, you will
love this software! A lot easier to use than Mintab, etc.! I use it all of the time.