# sample for control chart…

Six Sigma – iSixSigma › Forums › Old Forums › General › sample for control chart…

- This topic has 7 replies, 5 voices, and was last updated 14 years, 2 months ago by jimmie65.

- AuthorPosts
- November 29, 2005 at 6:34 am #41554
Hello..

I need help to determine what is the best way(or any better way,like strata and so on…already got confuse.. again) to do sampling for control chart. Thank you.0November 29, 2005 at 7:51 am #130426Hi ,

What kind of Data are you working on, Also are you aware of the total population.

There are a few ways to generate the sample you are looking for but it depends on the total population size and the margin of error you can accept.

0November 29, 2005 at 8:29 am #130431

spacefrogMember@spacefrog**Include @spacefrog in your post and this person will**

be notified via email.The size of a population being studied has direct impact on the size of a sample needed to better understand that population. You need to know how much you should sample based on the precision and confidence level necessary to meet your business objectives.

Points to consider1.Estimated Standard Deviation:For Continuous Data: You need to have an estimate of this already…see if there is some data that already exists, or alternatively or take a sample of ~30 and calculate the standard deviation. This can be calculated in Minitab or in Excel.

2.Error or Allowable uncertainityEnter the level of uncertainty in your estimate that you are willing to accept. For example, if you are estimating cycle time for deliveries, how precisely do you need to estimate the average? Within +/- 1 day, +/- 1 hour, +/- 1 minute? The resulting 95% confidence interval will be of width 2*E (since E is stated as a +/-). The more precisely you need to estimate, the larger the sample size will be! Note that this is a practical consideration, not a statistical one. For continuous data, remember to enter this in the same units as the average so, if your unit is time, and the average will be estimated in hours, enter a acceptable uncertainty of 4 hours as a “4” in the “E” cell. For Discrete data, E is the error you are willing to accept formatted as a percentage.

3.Proportion Defective:For Discrete Data: This is Discrete data it is the percentage defective that we are estimating. Enter an estimated 10% defective rate as 10%. If you don’t have a “ballpark estimate”, you can use 50% for worst case scenario, i.e., it will give you a sample size that will work for any proportion defective, but will err on the high side. Alternatively, gather ~100 data points to get a preliminary estimate. Discrete data requires a higher number of samples than continuous.

4.Population Size:When the population of interest is finite, and of known size, consider using the finite population sample size. However, we only need to be concerned about finite populations when we could realistically sample more than about 20% of the entire population. For example, we may have 500 employees in a facility, and want to interview a sample to estimate the proportion of employees who have been properly trained in dealing with a specific compliance issue. This gives us a finite total population of 500. If we could realistically sample more than about 100 people, we should use the finite population formula, since this is over 20% of the population. We would enter 500 for N.0November 29, 2005 at 8:45 am #130432Hi Hanna,

Spacefrog’s response was one of the most comprehensive replies I have read on the forum and should solve your problem.

Thanks Spacefrog, even I got to pickup certain points out of it.

Cheers!

Sean0November 29, 2005 at 9:02 am #130433Hi,

Thanks to both of you spacefrog and sean for helping me. I’ll try to follow the guideline that all of you’ve gave to me. And I hope that it will succeed. Thanks again.=)0November 30, 2005 at 5:50 am #130485Hi spacefrog,

May I ask you some question..I need some advice or a guideline.Thank you.

Rite now I’m monitoring a daily inspection process. Each day there would be a different number of part that should be going into inspection. So I want to monitor the process whether it’s in spec or vice versa. The part comes in batches. A batch contains 10 pcs.Can I just take a first 10 batches for sampling and represent the whole process for a certain day?

FIY, my previous peers who handle the same process used all incoming data,(without doing sampling) to form a control chart.Can it be that way?(liked what have been done by my peers..)

Please help me and thank you..0November 30, 2005 at 12:40 pm #130491

quality1Participant@quality1**Include @quality1 in your post and this person will**

be notified via email.Hanna !

I’m not professional, but my thoughts:Why you want additional sampling ? Is it needed ? Do you have quality problems, or just a new chart ?

What kind of incoming data you have ? What is the incomind data colection frequency?

Why not to trust ? If you can’t trust, why you don’t improve that ?

How stable (or how much variation) is your process in one shift ?( a lunch break or time spent on work, I consider as variation if humans are involved)

Don’t take the first 10 batches, and qualify the whole shift. take samples, at the beginning and at least at the end of the shift. This way you can avoid at least uncertaneity between shifts.

For determining the sample size you can use batch or shift volume, and considering your process stability.

Good luck. I’m more a practical, then theorethical, person, from production.

quality1

0November 30, 2005 at 1:56 pm #130497

jimmie65Participant@jimmie65**Include @jimmie65 in your post and this person will**

be notified via email.Spacefrog –

Great answer. One question, though. When you say “enter 500 for N”, are you referring to a calcualtor or a formula? And can you post the formula (or link)?

Jimmie0 - AuthorPosts

The forum ‘General’ is closed to new topics and replies.