Dear Gentlemen,

When I supplied the example I was worried that you would miss the point what
was emphasized at the last sentence:
"Scope of this example was only to present different methods for calculating
the average resistance of a sample in SIR-testing -nothing more ore less."

I agree that the results are not normally distributed. Also I agree that
they maybe are if the sample size is bigger.

But my point was that with extreme values it makes a difference how you
calculate the average.
Arithmetic mean of resistances was 6.58 GOhm and geometric mean of
resistances was 5.36 GOhm. So what is the "correct" average. I also
calculate the averages of currents (Arithmetic mean of currents is 3.51 nA,
Geometric mean of currents is 2.80 nA) and the equivalent resistances: 4.27
GOhm for arithmetic and 5.36 GOhm for geometric mean. And that, in my
opinion -when compared to averages calculated from the resistances-
indicates that geometric mean is correct method because it gaves us same
result in both ways (calculating the individual resistances and the
calculating the average OR calculating the average of currents and the
calculating the equivalent resistance.)

The average gives us only the estimate of average of population. And also
the distribution (SD, minimum and maximum) should also reported.
If the distribution is a problem I would suggests the center limit theorem.

B.R.
Jarmo Kiiski

> -----Original Message-----
> From: Graham Naisbitt [SMTP:[log in to unmask]]
> Sent: 1. syyskuuta 2000 20:32
> To:   [log in to unmask]
> Subject:      [TN] SIR Data - a late offering
>
> Hello folks
>
> Here is a rather belated observational input on the recent exchanges re
> SIR test results. I've been kinda pre-occupied with urgent stuff.
>
> Anyhow, Jarmo Kiiski asked about the statistical evaluation of data and I
> must point out that recent research data has proven that existing flux
> characterisation tests are....well, here is what has been put out by the
> British National Physical Research Laboratory:
>
> "The present standard method for evaluating board reliability and for flux
> qualification uses inappropriate test conditions, and hence provides
> unrealistic test data often with exceptionally high and misleading SIR
> values. This is the far reaching conclusion of a landmark European
> collaborative programme led by NPL which has now proposed a credible
> replacement test method"... (sic) that is "using finer pitches on the test
> coupons, lower voltages reflecting the actual device field strength and
> frequent monitoring to detect transient effects such as dendrite
> formation, as the key elements for a new test method. Another benefit is
> that test times can be reduced from 7 days to 3 days."
>
> I believe that what Jarmo, Cara and Amanda may be seeing/have seen, is
> "blips" in the occasional readings they have taken which may actually be
> dendrites forming and/or collapsing but largely missed because the
> measurements have been taken too infrequently - or did I miss something?
>
> Brian certainly hits the nail on the head by suggesting that the sample
> size is perhaps too small - he hit a few other things too, I guess! My
> view is that your measurements are insufficient to yield you the results
> you seek i.e. choosing the right process material. I certainly could not
> contribute to the debate on statistical interpretations.
>
> However, I would also suggest that unless you make a test of all process
> materials on a coupon that is more representative of your full production
> process you will continue to get unreliable results from both your
> material tests and the end product in the field. If its reliability
> matters, then you really should take note.
>
> Regards
>
> Graham Naisbitt
> [log in to unmask] <mailto:[log in to unmask]>
> www.concoat.co.uk <http://www.concoat.co.uk>
> Concoat Ltd
> Alasan House, Albany Park
> CAMBERLEY GU15 2PL UK
> Tel: +44(0)1276691100
> Fax: +44(0)1276691227

##############################################################
TechNet Mail List provided as a free service by IPC using LISTSERV 1.8d
##############################################################
To subscribe/unsubscribe, send a message to [log in to unmask] with following text in
the body:
To subscribe:   SUBSCRIBE TECHNET <your full name>
To unsubscribe:   SIGNOFF TECHNET
##############################################################
Please visit IPC web site (http://www.ipc.org/html/forum.htm) for additional
information.
If you need assistance - contact Keach Sasamori at [log in to unmask] or
847-509-9700 ext.5315
##############################################################