Hi,
I have a seemingly common problem but I can't find a proper way to approach
it. Let's say we have 5 samples (different size) of IC circuits coming from
5
production lines (A, B, C, D, E). We apply two different non-destructive QA
procedures to each sample, producing to sets of binary outcomes (passed:
no/yes). So, we have two groups of proportions:
QA1 QA2
no/yes no/yes
A 10/90 8/92
B 5/15 7/13
C 1/79 2/78
D 12/38 10/40
E 3/7 1/9
How would I test if the two QA procedures in question give significantly
different results, at the same time controlling for the possible production
line contribution? It looks like there are many variants of multiple
proportions tests available in R and various extra packages but none seems to
exactly fit this very simple problem. I would appreciate any advice.
Thanks,
Ivan
The information transmitted in this electronic communica...{{dropped:10}}
here is one approach:
res <- cbind( c(10, 5, 1, 12, 3, 8, 7, 2, 10, 1),
c(90,15,79,38,7,92,13,78,40,9) )
line <- gl(5,1,length=10, labels=LETTERS[1:5])
qa <- gl(2,5)
fit <- glm( res ~ line*qa, family=binomial )
summary(fit)
anova(fit, test='Chisq')
The interaction terms measure the difference between the different combinations
of QA method and production line, if they are all 0, then that means the effect
of QA is the same accross production lines and the qa main effect measures the
difference between the 2 methods (allowing for differences in the prodoction
lines), testing if that equals 0 should answer your question.
Hope this helps,
________________________________________
From: r-help-bounces at r-project.org [r-help-bounces at r-project.org] On
Behalf Of Ivan Adzhubey [iadzhubey at rics.bwh.harvard.edu]
Sent: Monday, June 09, 2008 4:28 PM
To: r-help at r-project.org
Subject: [R] Comparing two groups of proportions
Hi,
I have a seemingly common problem but I can't find a proper way to approach
it. Let's say we have 5 samples (different size) of IC circuits coming from
5
production lines (A, B, C, D, E). We apply two different non-destructive QA
procedures to each sample, producing to sets of binary outcomes (passed:
no/yes). So, we have two groups of proportions:
QA1 QA2
no/yes no/yes
A 10/90 8/92
B 5/15 7/13
C 1/79 2/78
D 12/38 10/40
E 3/7 1/9
How would I test if the two QA procedures in question give significantly
different results, at the same time controlling for the possible production
line contribution? It looks like there are many variants of multiple
proportions tests available in R and various extra packages but none seems to
exactly fit this very simple problem. I would appreciate any advice.
Thanks,
Ivan
The information transmitted in this electronic communica...{{dropped:10}}
______________________________________________
R-help at r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.
Hi Ivan, It was not clear from your original post that QA was a repeated factor. But your problem may be reframed very much like you would do with a McNemar chi-square: Just count the number of times both procedure give the same result, of each kind, and different results, again of both kinds, to get four counts by condition. These events are probably independent within your setting. You should then be able to test various binomial or Poisson models with the proper equality constraints. HTH, Yvonnick Noel, PhD University of Brittany at Rennes France> Re: [R] Comparing two groups of proportions > To: r-help at r-project.org > Message-ID: <200806100208.51742.iadzhubey at rics.bwh.harvard.edu> > Content-Type: text/plain; charset=iso-8859-1 > > Hi Rolf, > > On Monday 09 June 2008 11:16:57 pm Rolf Turner wrote: > >> > Your approach tacitly assumes --- as did the poster's question --- that >> > the probability of passing an item by one method is *independent* of >> > whether it is passed by the other method. Which makes the methods >> > effectively independent of the nature of the item being assessed! >> > > So it seems I can't just block my primary factor (QA procedure) by nuisance > one (production line) and run Cochran test to see if effects of primary > factor are identical for both its levels. > > >> > Not much actual quality being assured there! >> > > In fact, I am not interested in quality of QA procedures as much as in how > different the results are (error component). > > Thanks, > Ivan > >