"flipper" wrote [below and...]

See the next post I made to Phil. You have made false assumptions
about what I assumed.

Whatever the distribution of single values, multiplying the number
of resistors will make it more bell-shaped than it was.

If 1% tolerance means that all the resistors are individually
measured to be within 1%, then my statement that a compound resistor
has a better chance of being within tolerance would be incorrect.

But my main point, that compound resistors will be more tightly
clustered around a nominal value and sparse at the edges, is true.

Furthermore, I have said that resistors aren't manufactured by
chance. That is, they don't make random values and then sort them.
They actually target a particular nominal value for each batch. We
should expect some kind of bell-like distribution, therefore.
Looking at a few pages on the net, for example:

http://www.circuitree.com/CDA/Articl...108386,00.html

http://www.mcm.dupont.com/MCM/en_US/...pass_tech5.pdf

you can see that this is true.

The question to which I can't find an answer is, do they chop off
the tails of the distribution? The only way that could be done would
be by measuring every resistor.

Looks like they do.

If Phil is right when he says that normal service life is included,
then they absolutely definitely *cannot* chop off the tails of the
distribution.

In any case, my main point about *chance* is correct: a compound
resistor will probably be closer to its nominal value than the
individual resistors it is made of.

More points below in context:

"Phil Allison" wrote
That'd be like a 1W 2% resistor


** WRONG !!!

Even if *both* resistors were 1% high, the series value is also
1% high.

Using resistors in series or parallel IMPROVES the % accuracy
of
the final value.

Phil, those two statements appear contradictory, because your use
of
"%" changes from the first to the second. Or you are wrong...

The part that's incorrect is thinking it improves the % accuracy,
as
the first, and correct, statement shows. If they're 1% tolerance
they
can be 1% off. That's what it means. And 2 of them together can be
1%
off as well. The tolerance is 1%.


What improves is the *chance* that a resulting value will be
within
tolerance.

The "chance" of the pair being within tolerance is 100%. Well, if
the
individuals are within tolerance, as specified, that is.

Ok, if they chop off the tails, which it seems they do (but still
not certain...)

Since quoted tolerance assumes a particular distribution
of values around the quoted value, improving that chance is the
same
thing as narrowing the tolerance.

Neither of those is correct. The tolerance number assumes nothing;
they'll simply be within 1% and if they're all 1% off it's just as
valid as if only one of them is off 1% off.. It is you who are
apparently assuming a particular distribution.

No, I am only assuming that it has a distribution, which it must
have. Second clause of my statement is wrong only if they chop off
the tails...

Second, 'chance' has nothing to do with the tolerance unless you
are
'hoping' to get something other than what's specified. But, in any
case, the tolerance is neither 'improved' nor lessened. You can
still
end up with 1% off and that's all a 1% tolerance says: you can't
*depend* on them being any closer than 1% and it doesn't tell you
'how
much' you can or can't depend on it. And it 'might be' anywhere
in-between. The *only* thing known is they'll be within 1%

No, not the only thing. You know they are more likely to be closer,
and you can calculate exactly by how much. You are getting carried
away with your point now. You were better off keeping it simple.

Which gets back to the distribution and the dangers of assuming
one.

There is bound to be a distribution. How could there not be?

What you are probably thinking of is the classic bell curve
distribution and assuming that distribution applies but it may
very
well not depending on how they're made and selected.

I did not say that was my assumption, and my point does not require
it. Actually it *is* bell shaped to begin with. The only question
was
whether they chop off the tails...

For example, take
a resistor line that includes .1%, .5%, and 1% tolerance ranges.
It is
not uncommon for those to be selected from the same manufacturing
process so the .1% resistors are culled out first, then the .5%,
and
then the remaining are 1%. Which means the .1% resistors may very
well
have a classic bell curve distribution but the .5% values will
have a
.1% 'hole' around the .1% value and the 1% resistors will have a
.5%
'hole' (might be multiple holes too). So your 'chances' of finding
a
right spot on 1% resistor might be virtually 0, unless there's not
enough demand for .1% and .5% values so that some of the 'better'
ones
get the lower tolerance rating anyway to fill the larger demand in
that market. Which means you can't even 'depend' on the dern
'hole'
being there.

This is more useful stuff, if it is true. But here you suggest that
the 1% resistors do *not* have the distribution tails chopped by the
selection process, so you have rather shot yourself in the foot.

Once again, compound resistors, *regardless of the distribution of
their components* will concentrate values around the nominal, and
fewer will be found at the extremes. Also, "holes" near nominal will
be filled in more than those at the extremes.

Assuming distributions when none is specified can lead to
disappointing surprises.

You've said that so many times it's stopped making sense.

Now, if you were hand selecting a tighter tolerance from a lower
tolerance batch then pairing them might improve your yield because
there might be more 'good' combinations of offsetting pairs than
solely single values, if the unknown distribution allowed it. But,
then, you can't necessarily count on that. (the manufacturer can
because he knows the distribution)

Yes you can count on it. Show me *any* plausible distribution where
it would not be true. Now you are just making things up.

Lastly, to cover the subsequent message, tolerance is the initial
tolerance, at 25C, as delivered to your doorstep and before you
begin
torturing the thing. Everything else, solder, lead stress,
temperature
coefficient, long term drift, surge, voltage stress, temperature
cycling, etc. are in addition to the initial tolerance.

What subsequent message? Anyway, if you are correct here
(evidence?), then it does at least mean that it is *possible* they
chop the tails. Altogether you have been quite convincing on that
issue...but where is your evidence?

cheers, Ian

PS More evidence for your chopping...

http://www.keithley.com/data?asset=3871