P chart vs. IMR chart
Six Sigma – iSixSigma › Forums › Old Forums › General › P chart vs. IMR chart
- This topic has 11 replies, 6 voices, and was last updated 13 years, 9 months ago by
Ron.
-
AuthorPosts
-
September 15, 2008 at 1:17 pm #50939
If i use P chart to plot % defective I get a lot of points out of control.
If i use IMR chart I mayby get one point or none.
I have read that IMR is OK to use, but i get two very different conclusions, P – out of control, IMR – in control.
Is P charts more sensitive ?
Jan0September 15, 2008 at 3:32 pm #175764
JenniferAParticipant@JenniferAInclude @JenniferA in your post and this person will
be notified via email.Is the data you are plotting binomial?
0September 15, 2008 at 4:46 pm #175766
T NiebankMember@T-NiebankInclude @T-Niebank in your post and this person will
be notified via email.Jan,
IMR and P charts are used for two different types of data. If your data is Attribute data use the P chart. If it is Continuous data use the IMR.0September 15, 2008 at 5:20 pm #175768Produced xxxx parts, yy bad parts = z% bad parts.
So yes the data are binomial, but mayby the proces is not ;-)0September 15, 2008 at 5:24 pm #175769I get that.
But even Wheeler in his SPC book states that it is OK to use IMR on some types of atribute data.
He even states that in many cases binominal data dont come from binominal processes, thus IMR would be prefereable.
Jan0September 15, 2008 at 5:58 pm #175773
Ken FeldmanParticipant@DarthInclude @Darth in your post and this person will
be notified via email.As usual, you are getting erroneous information. Yes, Wheeler says you can switch to an I/MR chart if the p>=5. If your sample size is getting too large, you start to violate some of the assumptions of the binomial and the large n in the denominator are giving you a false tightening of the control limits. If p>=5 then an assumption of continuous/normal can be made and thus the use of the I/MR.
0September 15, 2008 at 6:49 pm #175776
JenniferAParticipant@JenniferAInclude @JenniferA in your post and this person will
be notified via email.Look at page 252 of Wheeler’s Advanced Topics in SPC.
Usually the two charts are very similar, so your question is interesting. I’m guessing that you have something going on in the data. Does the sample size from sample-to-sample vary considerably (more than about 25%)? The I chart won’t take that into account, but the p-chart will. Plot the sample size vs. %defectives to see what that looks like.0September 15, 2008 at 6:52 pm #175777
JenniferAParticipant@JenniferAInclude @JenniferA in your post and this person will
be notified via email.Darth,
Do you mean np?
Darth stated the sample size concern better than I did, sorry about that.0September 15, 2008 at 7:14 pm #175780
Ken FeldmanParticipant@DarthInclude @Darth in your post and this person will
be notified via email.Good catch, np>5.
0September 15, 2008 at 10:53 pm #175790
Bower ChielParticipant@Bower-ChielInclude @Bower-Chiel in your post and this person will
be notified via email.Hi JanI’ve been involved in some charting of data on the process of care of stroke patients. It was a state requirement that monthly proportions of patients having a CT brain scan within 48 of admission to hospital be reported. We decided not to use p-charts for this data because the probability of having a scan within 48 hours is not constant from patient to patient since, for example, the provision of radiographers differs (sadly) between week-ends and week days. Thus we could not assume a binomial distribution which underpins the p-chart. We therefore used an Individuals (X)chart. Apparently justification for this approach may be found in the Western Electric Quality Control Handbook from 1956.You might find it of interest to read Improved Control Charts for Attributes by David Laney available in full and free at http://www.quadrantonegroup.com/pdf/laneydoc.pdf. He introduces the p’ chart and argues it does a better job than the X chart when binomial assumptions are not met. I’ve written a crude Mintab macro to produce a basic p’chart which I’ll happily pass on to you if you contact me at [email protected] WishesBower Chiel
0September 16, 2008 at 6:26 am #175796Yes sample seize varys from 1.000 to 5.000.
Maby I get “punished” for the large sample sizes, making my CL to narrow.0September 16, 2008 at 11:29 am #175799Jan,
Dr. Donald Wheeler as always is a great mentor I highly recommend his book “Understanding Statistical Process Control”.
P charts are not the chart of choice in most cases.
0 -
AuthorPosts
The forum ‘General’ is closed to new topics and replies.