Lingo for layfolk?

>Hi all,
>
>I am preparing a course for math layfolk, but need
>to communicate the concepts of 'continuous' and
>'discrete' variables. Users of a software package
>will need to use these concepts in order to select
>the proper analysis methods for whatever data
>they need to proces.
>
>I can't use those terms, or participants will likely
>flee the course. So I have tried to come up with
>simpler terms:
>
>Continous variable = measurable variable
>Discrete variable = countable variable
>
>Does this make sense to others than me?
>
>Rune

If you are talking about variables discrete in amplitude I find your terms
more confusing that the originals. I can't count 1.125 of something, but
its a nice discrete value in a fixed point binary word.

If you are talking about variables discrete in time your terms don't convey
the idea at all.

Steve

On 6 Sep, 08:57, Fred Marshall <fmarshallxremove_th...@acm.org> wrote:
> On 9/5/2011 10:44 PM, Rune Allnor wrote:
>
>
>
>
>
> > On Sep 5, 10:01 pm, Fred Marshall<fmarshallxremove_th...@acm.org>
> > wrote:
> >> On 9/5/2011 1:30 AM, Rune Allnor wrote:
>
> >>> I'n not sure. The measurements in this application
> >>> will be done at discrete points in time; the trick is
> >>> to select the correct representations of the various
> >>> types of measurements:
>
> >>> - physical dimensions of a produced item is a continuously
> >>> &#4294967295; &#4294967295; distributed quantity
> >>> - the number of defects or flaws associated with the same
> >>> &#4294967295; &#4294967295; item is a discretely distributed quantity.
>
> >>> The objective is to track such quantities over time
> >>> in order to diagnosticize troublespots and improve
> >>> production.
>
> >>> The mechanics of doing the anslyses are the same in
> >>> each case, but the underlying sw functionality differs.
> >>> The users need to understand the difference well
> >>> enough to select the correct options.
>
> >>> Rune
>
> >> Oh, well then I didn't understand the objective very well I guess.
>
> >> But now you have me curious:
>
> >> The physical dimensions of a produced item, or should one say: the
> >> physical dimensions of produced items?, is a population of discrete
> >> events it seems to me. &#4294967295;That there may be the possibility of any value
> >> down to the limits of measurement being recorded doesn't make it any
> >> less discrete does it? &#4294967295;Well, I do understand that these values would
> >> appear to have some physical character that suggests the values come
> >> from a possbily continuous distribution.
>
> > One measures some items, and attempt to infer
> > conclusions about the population. The main
> > differences between SPC and 'regular' statistics
> > are that
>
> > - SPC uses some rather simplified computations
> > - SPC deliberately aims to track temporal developments
> > &#4294967295; &#4294967295;on a sample-to-sample basis, whereas 'regular'
> > &#4294967295; &#4294967295;statistics usually is based on stationary properties
> > &#4294967295; &#4294967295;across the sample.
>
> >> Of course the number of defects or flaws is an integer. &#4294967295;So, it's a
> >> matter of counting.
>
> >> I'd have to wonder if putting the dimensions in a computer using 32-bit
> >> binary, suitably scaled, is any different than counting? &#4294967295;They seem the
> >> same to me but perhaps there's a philosophical difference?
>
> > Computer number systems have a limited number
> > of bits and so a finite number of states. The bit pattern
> > points to the represented number value, depending
> > on encoding.
>
> > Rune
>
> Yeah, I understand about computer number systems. &#4294967295;I was trying to
> relate that to the counting flaws example. &#4294967295;Both result in "integers" if
> you like.
>
> Seems to me that SPC isn't any different. &#4294967295;Maybe you calculate mean and
> variance of dimensions over some termporal window and then compare
> results window-to-window. &#4294967295;Isn't that essentially what a control chart
> does with mean values?

Yes and no: SPC is not *one* technique, but a collection
of similar techniques that can be used for the same reasons,
but on different collections of data.

Control charts (Shewhart charts) can work on means of samples,
or on individual data points. They can work on discrete or
continuous data. The purpose and user interaction is similar
in all cases; the technical details in the computations
somewhat differnt.

Rune

On 6 Sep, 09:22, "steveu" <steveu@n_o_s_p_a_m.coppice.org> wrote:
> >Hi all,
>
> >I am preparing a course for math layfolk, but need
> >to communicate the concepts of 'continuous' and
> >'discrete' variables. Users of a software package
> >will need to use these concepts in order to select
> >the proper analysis methods for whatever data
> >they need to proces.
>
> >I can't use those terms, or participants will likely
> >flee the course. So I have tried to come up with
> >simpler terms:
>
> >Continous variable = measurable variable
> >Discrete variable = countable variable
>
> >Does this make sense to others than me?
>
> >Rune
>
> If you are talking about variables discrete in amplitude I find your terms
> more confusing that the originals. I can't count 1.125 of something, but
> its a nice discrete value in a fixed point binary word.

It is. But in this application I am confined to the
universe of natural numbers; 0,1,2,...

> If you are talking about variables discrete in time your terms don't convey
> the idea at all.

In poth cases the measurements are made at discrete points
in time. But one case can produce a real number, whereas
the other is limited to natural numbers.

Rune

On 9/6/2011 7:13 AM, Rune Allnor wrote:
>
>>
>> Yeah, I understand about computer number systems.  I was trying to
>> relate that to the counting flaws example.  Both result in "integers" if
>> you like.
>>
>> Seems to me that SPC isn't any different.  Maybe you calculate mean and
>> variance of dimensions over some termporal window and then compare
>> results window-to-window.  Isn't that essentially what a control chart
>> does with mean values?
>
> Yes and no: SPC is not *one* technique, but a collection
> of similar techniques that can be used for the same reasons,
> but on different collections of data.
>
> Control charts (Shewhart charts) can work on means of samples,
> or on individual data points. They can work on discrete or
> continuous data. The purpose and user interaction is similar
> in all cases; the technical details in the computations
> somewhat differnt.
>
> Rune

Rune,

Well then, might you be more specific?  I don't know what technical 
details you're referring to.  And the point here is to help with 
terminology.

Fred

On 6 Sep, 17:29, Fred Marshall <fmarshallxremove_th...@acm.org> wrote:
> On 9/6/2011 7:13 AM, Rune Allnor wrote:
>
>
>
>
>
>
>
> >> Yeah, I understand about computer number systems. &#4294967295;I was trying to
> >> relate that to the counting flaws example. &#4294967295;Both result in "integers" if
> >> you like.
>
> >> Seems to me that SPC isn't any different. &#4294967295;Maybe you calculate mean and
> >> variance of dimensions over some termporal window and then compare
> >> results window-to-window. &#4294967295;Isn't that essentially what a control chart
> >> does with mean values?
>
> > Yes and no: SPC is not *one* technique, but a collection
> > of similar techniques that can be used for the same reasons,
> > but on different collections of data.
>
> > Control charts (Shewhart charts) can work on means of samples,
> > or on individual data points. They can work on discrete or
> > continuous data. The purpose and user interaction is similar
> > in all cases; the technical details in the computations
> > somewhat differnt.
>
> > Rune
>
> Rune,
>
> Well then, might you be more specific? &#4294967295;I don't know what technical
> details you're referring to. &#4294967295;And the point here is to help with
> terminology.

It's stuff like how to compute estimates for certain
parameters that go into the computations: If the
data are better described by a Poisson distribution
the estimate is computed in thus way; it the binomial
distribution applies it is computed in that way; with
a Gaussian distribution it is computed in yet another way.

Again, the user only needs to know how to identify or
describe the situation at hand, and check the correct
boxes, and these things will be taken care of under
the hood of the SW package.

Rune

On 9/6/2011 2:57 AM, Fred Marshall wrote:

   ...

> Seems to me that SPC isn't any different. Maybe you calculate mean and
> variance of dimensions over some termporal window and then compare
> results window-to-window. Isn't that essentially what a control chart
> does with mean values?

SPC isn't necessarily about flaws. Suppose a spec on a screw-machine 
part is 1' +- 1/64 (+-.015625) and parts have been consistently between 
.994 and .996. Now I begin to see some parts at .998 and none as small 
as .995. These are actually closer to spec, yet I might have cause for 
worry. SPC can allay my unease or prompt action. (The fact is, the tool 
bit is wearing.)

Jerry
-- 
Engineering is the art of making what you want from things you can get.

On 9/6/2011 12:52 PM, Rune Allnor wrote:

   ...

> Again, the user only needs to know how to identify or
> describe the situation at hand, and check the correct
> boxes, and these things will be taken care of under
> the hood of the SW package.

Automatic computation is a GOOD THING. It reduces the work and certainly 
also computational errors. Automatic selection of details is also 
helpful, but it creates the risk that the package will be used by people 
who have no understanding of what it does, let alone how. Then when some 
ridiculous outcome turns up, people will thing that because a /computer/ 
says so, it must be right. People who use  and rely on the program need 
to be able to do without it, at least in principle. I think you're on 
the right track. (Moroney wrote for people like your students.)

Jerry
-- 
Engineering is the art of making what you want from things you can get.

On 9/6/2011 10:57 AM, Jerry Avins wrote:
> On 9/6/2011 2:57 AM, Fred Marshall wrote:
>
> ...
>
>> Seems to me that SPC isn't any different. Maybe you calculate mean and
>> variance of dimensions over some termporal window and then compare
>> results window-to-window. Isn't that essentially what a control chart
>> does with mean values?
>
> SPC isn't necessarily about flaws. Suppose a spec on a screw-machine
> part is 1' +- 1/64 (+-.015625) and parts have been consistently between
> .994 and .996. Now I begin to see some parts at .998 and none as small
> as .995. These are actually closer to spec, yet I might have cause for
> worry. SPC can allay my unease or prompt action. (The fact is, the tool
> bit is wearing.)
>
> Jerry

There is a story that tells about the drawings/spec's for an existing 
American internal combustion engine like for a Ford or a Chevy that was 
in production.  The drawings and spec's were taken to Japan and they 
made the parts to spec .. using tighter tolerances on the dimensions 
than the Americans had done.

The parts wouldn't assemble.

It seems that the drawings were probably incorrect in the sense that the 
parts wouldn't fit together if perfect.  But there was enough slop in 
the production process to allow them to fit anyway .. perhaps with a bit 
of parts selection.  Maybe not a good thing but verrrry interesting!

It was Rune who mentioned counting flaws.....

I'm still struggling with the context here.  I don't get where 
continuous vs. discrete comes in at all really.

Fred

On 9/6/11 11:00 PM, Fred Marshall wrote:
> On 9/6/2011 10:57 AM, Jerry Avins wrote:
...
>> Suppose a spec on a screw-machine
>> part is 1' +- 1/64 (+-.015625) and parts have been consistently between
>> .994 and .996. Now I begin to see some parts at .998 and none as small
>> as .995. These are actually closer to spec, yet I might have cause for
>> worry.
>
> There is a story that tells about the drawings/spec's for an existing
> American internal combustion engine like for a Ford or a Chevy that was
> in production. The drawings and spec's were taken to Japan and they made
> the parts to spec .. using tighter tolerances on the dimensions than the
> Americans had done.
>
> The parts wouldn't assemble.
>
> It seems that the drawings were probably incorrect in the sense that the
> parts wouldn't fit together if perfect. But there was enough slop in the
> production process to allow them to fit anyway...

but they would have to have slop in a particular direction (some 
dimensions bigger than the nominal spec and some smaller) to fit if they 
didn't fit when perfect.

you would think that when making the prototype or very first run of the 
engine, the parts would have been machined in the engineering shop to 
*very* good specs with very little slop (as good as the Japanese parts).

so, something about this story remains hard for me to understand.

-- 

r b-j                  rbj@audioimagination.com

"Imagination is more important than knowledge."

robert bristow-johnson <rbj@audioimagination.com> wrote:

(snip, someone wrote)
>> There is a story that tells about the drawings/spec's for an existing
>> American internal combustion engine like for a Ford or a Chevy that was
>> in production. The drawings and spec's were taken to Japan and they made
>> the parts to spec .. using tighter tolerances on the dimensions than the
>> Americans had done.

>> The parts wouldn't assemble.

>> It seems that the drawings were probably incorrect in the sense that the
>> parts wouldn't fit together if perfect. But there was enough slop in the
>> production process to allow them to fit anyway...

I suppose that sounds right if the machines tended to overcut.
Holes would be too big, and things that fit into them too small.
Maybe the original designers knew about the properties of
the machines that they were designing for, but that didn't
get into the drawings.

> but they would have to have slop in a particular direction (some 
> dimensions bigger than the nominal spec and some smaller) to fit if they 
> didn't fit when perfect.

Not so obvious, but it reminds me of a story from many years ago,
about the tolerance and distribution of resistor values.  There
are (or were) commonly resistors with 5% and 10% tolerance, and
one might expect the 10% to have values somewhat distributed around
the nominal value, some near, some not so near.  The actual case was
that the 10% resistors were between 5% and 10% away (one side or
the other), as the closer ones were selected out as the 5% resistors.

Given a value and uncertainty, and no reason to believe that the
value is Gaussian distributed around the mean, you might be
surprised as to how they actually come out.

> you would think that when making the prototype or very first run of the 
> engine, the parts would have been machined in the engineering shop to 
> *very* good specs with very little slop (as good as the Japanese parts).

I don't know the process enough to say.

-- glen