Expert,

In my industry the lot cycle time performance is measured as Cycle time at 95%tile. Is there any statistical explanation why 95%tile and not plainly maximum or average Cycle time? Thanks for your input.

Expert,

In my industry the lot cycle time performance is measured as Cycle time at 95%tile. Is there any statistical explanation why 95%tile and not plainly maximum or average Cycle time? Thanks for your input.

I'm not an expert, but...

Percentile rankings are used to describe statistical distributions. or more precisely, how an individual compares to the distribution. In your case, you say that cycle times at the 95th percentile are used as some sort of performance standard. What this means is that 95% of the cases are either greater than or less than that figure, depending on what you're measuring (I suspect that the 95th percentile represents a relatively long cycle time). It's possible to do something similar with control charts. What industry are you in?

Hi Jim, Thanks for your input:thanx: :thanx: . Understand why a percentile is used. What is not clear to me is why the number 95%, what becomes the basis of this number? can it be 97% or 98%? Is the selection of this number has something to do with the distribution shape (skewed as opposed to normal) etc etc. I'm into electronics industry.

Who chose the number? It can be anything less than 100, I suppose (there is no 100th percentile because that would mean that a point exists outside of the entire population from which the point was drawn). I've personally never heard of using percentile rankings to describe cycle-time goals. The shape of the distribution isn't relative, because percentile rankings assume a normal distribution.

I expect that 95th percentile was chosen as a conservative estimate with no particular statistical significance.

For example, suppose the average cycle time was 1.0 hr, the 95% percentile was 1.5 hr and the max was 2.1 hr. How many cycles could you complete in an 8 hr shift? On average, you could do 8 cycles. Using the 95%-tile, you could be pretty sure of completing at least 5 cycles. At the worst, you might only finish only 3 cycles.

8 cycles is a good long-term expectation.
5 cycles is a pretty good estimate of the minimum of any day.
3 cycles is a worst-case scenario.

Nothing magic about 95%; just a handy conservative estimate.

Just my $0.02


Tim F

A couple of guesses:

90% and 95% confidence intervals are very common by tradition. And the 95th percentile gives 5% in the tail, corresponding to a 90% two sided confidence interval. It also corresponds to two standard deviations from the average for a normal distribution.

To know what the 95th percentile is, you have to collect at least 20 points. If you are specifying the 99th percentile, you need to collect at least 100 points - and a lot more to have any confidence in the figure, assuming you are doing this non-parametrically.

At least the person knew not to promise the average (or the median).

At least the person knew not to promise the average (or the median).
yes. without more info from the originator, my guess would be that the person who specified using the 95 percentile knwos a little about the Theory of Constraints. promising - or using - the average cycle time dooms you to failure...covarience comes into play and once you fall behind you can never catch up (anybody remember where Herbie is?)

However, without really knowing more details, we are only guessing

I've personally never heard of using percentile rankings to describe cycle-time goals. The shape of the distribution isn't relative, because percentile rankings assume a normal distribution.

Hi Jim, Sorry for getting you back on this one rather late. Anyway, CT Goal is expressed in days such as 5,6 or 7. But the collected data of different lots is expressed in terms of percentile. The 5% is considered maverick or out of the norm, there might be some assignable causes in these lots why they goes beyond the target CT.

yes. without more info from the originator, my guess would be that the person who specified using the 95 percentile knwos a little about the Theory of Constraints. promising - or using - the average cycle time dooms you to failure...covarience comes into play and once you fall behind you can never catch up (anybody remember where Herbie is?)

However, without really knowing more details, we are only guessing

I'm getting bit of information on 95% assignment. Well, this is the industry standard now, previously it was measured in terms of average. But due to several improvement actions done over the years, what is left out is the 5% (delta from 95%) of the entire process lots would require analysis as these are being considered maverick lot (something special happened).

Ok now I got an answer on why 95%tile. But please hold on, I still have something to clarify. Is there any statistical way of expressing 95% tile other than the usual ranking of the data then finding the 95th%. Could this number could also be expressed in terms of Sigma such that 2 sigma or 2.5 sigma?

Thanks.

Ok now I got an answer on why 95%tile. But please hold on, I still have something to clarify. Is there any statistical way of expressing 95% tile other than the usual ranking of the data then finding the 95th%. Could this number could also be expressed in terms of Sigma such that 2 sigma or 2.5 sigma?

Thanks.

If you can determine the distribution of the data, or assume Normality, you could easily calculate the sigma levels or the percentiles theoretically from the distributions.

Even safer would be to just go to SPC, which is not dependent upon the distribution.