[[email protected]]

On Fri, 20 Sep 2019 07:25:11 -0600, rbowman
wrote:

On 09/20/2019 04:32 AM, Cindy Hamilton wrote:
I recall my first job. I worked for a man studying renal adenocarcinoma
in frogs as a model for human cancer (unless you'd prefer to induce cancer
in humans for research). I typed endless statistical analyses of the
data.

When I was in college I worked summers for the state Department of
Education. They crunched a lot of numbers trying to formulate state wide
high school exams that would result in a pleasing bell curve. Sometimes
they succeeded. The crunching part was quite literal since they used
Friden square root calculators;

https://www.youtube.com/watch?v=fnmj9EzKgtg

That model isn't the square root variety but is very similar.

My memory of my statistics course was the chief use was determining how
many blivits you needed to test to insure only 5% of the blivits you
sold were junk. The 5% number varied depending on the cost of replacing
defective units and ****ing off customers.

It was a long time ago but my gut feeling is 1000 out of some 300+
million isn't a valid sample.

That is particularly true when you are just robo calling people to try
to get 1000 to talk to you. There is a big skew in your data right
there. Those are people who really think someone is going to lower the
interest rate on their credit card, the IRS takes gift cards as
payment and that microsoft found a problem in their computer.

[[email protected]]

On Fri, 20 Sep 2019 07:47:34 -0700 (PDT), Cindy Hamilton
wrote:

On Friday, September 20, 2019 at 10:22:06 AM UTC-4, rbowman wrote:
On 09/20/2019 04:32 AM, Cindy Hamilton wrote:
I recall my first job. I worked for a man studying renal adenocarcinoma
in frogs as a model for human cancer (unless you'd prefer to induce cancer
in humans for research). I typed endless statistical analyses of the
data.

When I was in college I worked summers for the state Department of
Education. They crunched a lot of numbers trying to formulate state wide
high school exams that would result in a pleasing bell curve. Sometimes
they succeeded. The crunching part was quite literal since they used
Friden square root calculators;

https://www.youtube.com/watch?v=fnmj9EzKgtg

That model isn't the square root variety but is very similar.

My memory of my statistics course was the chief use was determining how
many blivits you needed to test to insure only 5% of the blivits you
sold were junk. The 5% number varied depending on the cost of replacing
defective units and ****ing off customers.

That certainly was a very specific statistics course. Or your memory
of it was very specific.

It was a long time ago but my gut feeling is 1000 out of some 300+
million isn't a valid sample.

I can't quite recall the details. However, Scientific American was
kind enough to fill in the gaps in my memory:

https://www.scientificamerican.com/article/howcan-a-poll-of-only-100/

Cindy Hamilton

They point out the flaw in the first few paragraphs.

"You must first assume that the survey respondents have been sampled
at random from the population, meaning that people are selected one at
a time, with all persons in the U.S. being equally likely to be picked
at each point. For most polls, this is approximated by calling phone
numbers generated randomly by computer".

When a certain segment of the population will not talk to anyone on a
robo call, you are not random anymore. It is a better poll of how many
people are gullible enough to be scammed on the phone. Those people
are also more likely to take what they hear on TV as gospel.

[trader_4]

On Friday, September 20, 2019 at 12:13:57 PM UTC-4, wrote:
On Fri, 20 Sep 2019 07:47:34 -0700 (PDT), Cindy Hamilton
wrote:

On Friday, September 20, 2019 at 10:22:06 AM UTC-4, rbowman wrote:
On 09/20/2019 04:32 AM, Cindy Hamilton wrote:
I recall my first job. I worked for a man studying renal adenocarcinoma
in frogs as a model for human cancer (unless you'd prefer to induce cancer
in humans for research). I typed endless statistical analyses of the
data.

When I was in college I worked summers for the state Department of
Education. They crunched a lot of numbers trying to formulate state wide
high school exams that would result in a pleasing bell curve. Sometimes
they succeeded. The crunching part was quite literal since they used
Friden square root calculators;

https://www.youtube.com/watch?v=fnmj9EzKgtg

That model isn't the square root variety but is very similar.

My memory of my statistics course was the chief use was determining how
many blivits you needed to test to insure only 5% of the blivits you
sold were junk. The 5% number varied depending on the cost of replacing
defective units and ****ing off customers.

That certainly was a very specific statistics course. Or your memory
of it was very specific.

It was a long time ago but my gut feeling is 1000 out of some 300+
million isn't a valid sample.

I can't quite recall the details. However, Scientific American was
kind enough to fill in the gaps in my memory:

https://www.scientificamerican.com/article/howcan-a-poll-of-only-100/

Cindy Hamilton

They point out the flaw in the first few paragraphs.

"You must first assume that the survey respondents have been sampled
at random from the population, meaning that people are selected one at
a time, with all persons in the U.S. being equally likely to be picked
at each point. For most polls, this is approximated by calling phone
numbers generated randomly by computer".

When a certain segment of the population will not talk to anyone on a
robo call, you are not random anymore. It is a better poll of how many
people are gullible enough to be scammed on the phone. Those people
are also more likely to take what they hear on TV as gospel.

That's a valid point. IDK if any polsters address that. There are ways
around it, of course, but I doubt many, if any polls do it. So, if you
just rely on who will answer an unknown call, you could be biasing the
poll.

[Cindy Hamilton[_2_]]

On Friday, September 20, 2019 at 2:29:44 PM UTC-4, trader_4 wrote:
On Friday, September 20, 2019 at 12:13:57 PM UTC-4, wrote:
On Fri, 20 Sep 2019 07:47:34 -0700 (PDT), Cindy Hamilton
wrote:

On Friday, September 20, 2019 at 10:22:06 AM UTC-4, rbowman wrote:
On 09/20/2019 04:32 AM, Cindy Hamilton wrote:
I recall my first job. I worked for a man studying renal adenocarcinoma
in frogs as a model for human cancer (unless you'd prefer to induce cancer
in humans for research). I typed endless statistical analyses of the
data.

When I was in college I worked summers for the state Department of
Education. They crunched a lot of numbers trying to formulate state wide
high school exams that would result in a pleasing bell curve. Sometimes
they succeeded. The crunching part was quite literal since they used
Friden square root calculators;

https://www.youtube.com/watch?v=fnmj9EzKgtg

That model isn't the square root variety but is very similar.

My memory of my statistics course was the chief use was determining how
many blivits you needed to test to insure only 5% of the blivits you
sold were junk. The 5% number varied depending on the cost of replacing
defective units and ****ing off customers.

That certainly was a very specific statistics course. Or your memory
of it was very specific.

It was a long time ago but my gut feeling is 1000 out of some 300+
million isn't a valid sample.

I can't quite recall the details. However, Scientific American was
kind enough to fill in the gaps in my memory:

https://www.scientificamerican.com/article/howcan-a-poll-of-only-100/

Cindy Hamilton

They point out the flaw in the first few paragraphs.

"You must first assume that the survey respondents have been sampled
at random from the population, meaning that people are selected one at
a time, with all persons in the U.S. being equally likely to be picked
at each point. For most polls, this is approximated by calling phone
numbers generated randomly by computer".

When a certain segment of the population will not talk to anyone on a
robo call, you are not random anymore. It is a better poll of how many
people are gullible enough to be scammed on the phone. Those people
are also more likely to take what they hear on TV as gospel.

That's a valid point. IDK if any polsters address that. There are ways
around it, of course, but I doubt many, if any polls do it. So, if you
just rely on who will answer an unknown call, you could be biasing the
poll.

I believe fivethirtyeight.com takes the sampling methodology into account
when it rates surveys. It's got a raft of factors it considers when
judging the quality of survey results.

realclearpolitics probably does, too, but I haven't dived as deeply into
their website.

Cindy Hamilton

[trader_4]

On Friday, September 20, 2019 at 2:41:57 PM UTC-4, Cindy Hamilton wrote:
On Friday, September 20, 2019 at 2:29:44 PM UTC-4, trader_4 wrote:
On Friday, September 20, 2019 at 12:13:57 PM UTC-4, wrote:
On Fri, 20 Sep 2019 07:47:34 -0700 (PDT), Cindy Hamilton
wrote:

On Friday, September 20, 2019 at 10:22:06 AM UTC-4, rbowman wrote:
On 09/20/2019 04:32 AM, Cindy Hamilton wrote:
I recall my first job. I worked for a man studying renal adenocarcinoma
in frogs as a model for human cancer (unless you'd prefer to induce cancer
in humans for research). I typed endless statistical analyses of the
data.

When I was in college I worked summers for the state Department of
Education. They crunched a lot of numbers trying to formulate state wide
high school exams that would result in a pleasing bell curve. Sometimes
they succeeded. The crunching part was quite literal since they used
Friden square root calculators;

https://www.youtube.com/watch?v=fnmj9EzKgtg

That model isn't the square root variety but is very similar.

My memory of my statistics course was the chief use was determining how
many blivits you needed to test to insure only 5% of the blivits you
sold were junk. The 5% number varied depending on the cost of replacing
defective units and ****ing off customers.

That certainly was a very specific statistics course. Or your memory
of it was very specific.

It was a long time ago but my gut feeling is 1000 out of some 300+
million isn't a valid sample.

I can't quite recall the details. However, Scientific American was
kind enough to fill in the gaps in my memory:

https://www.scientificamerican.com/article/howcan-a-poll-of-only-100/

Cindy Hamilton

They point out the flaw in the first few paragraphs.

"You must first assume that the survey respondents have been sampled
at random from the population, meaning that people are selected one at
a time, with all persons in the U.S. being equally likely to be picked
at each point. For most polls, this is approximated by calling phone
numbers generated randomly by computer".

When a certain segment of the population will not talk to anyone on a
robo call, you are not random anymore. It is a better poll of how many
people are gullible enough to be scammed on the phone. Those people
are also more likely to take what they hear on TV as gospel.

That's a valid point. IDK if any polsters address that. There are ways
around it, of course, but I doubt many, if any polls do it. So, if you
just rely on who will answer an unknown call, you could be biasing the
poll.

I believe fivethirtyeight.com takes the sampling methodology into account
when it rates surveys. It's got a raft of factors it considers when
judging the quality of survey results.

realclearpolitics probably does, too, but I haven't dived as deeply into
their website.

Cindy Hamilton

There is how you pick the samples and then who will answer an unknown call. Idk how you could adjust for the latter. I think there are some polls that stick with a group of people they select for longer term, which would eliminate the phone answering problems, but then that has problems too.

[Hawk]

On 9/19/2019 4:30 AM, micky wrote:
In alt.home.repair, on Wed, 11 Sep 2019 11:03:12 +0100, Bod
wrote:

On 11/09/2019 10:52, Bod wrote:
A growing number of people fear a mass shooting could take place in
their community, and an overwhelming majority support background checks
for gun sales, according to a new poll.

Sixty percent of Americans said they are worried about a mass shooting
in their own community, according to the ABC News/Washington Post poll
released Monday. That's up 5 points compared to January 2013, when 55
percent of Americans said they feared a mass shooting.

https://thehill.com/homenews/news/46...community-poll

It's sadly ironic, that the very thing that you *think* makes you
safe(guns), is the very thing that brings you fear.

I guess that could have been said in WWII about the Royal Air Force and
the German air force. And maybe still today.


"There are three kinds of lies: Lies, damned lies, and statistics."

- Benjamin Disraeli -

[rbowman]

On 09/20/2019 08:47 AM, Cindy Hamilton wrote:
On Friday, September 20, 2019 at 10:22:06 AM UTC-4, rbowman wrote:
On 09/20/2019 04:32 AM, Cindy Hamilton wrote:
I recall my first job. I worked for a man studying renal adenocarcinoma
in frogs as a model for human cancer (unless you'd prefer to induce cancer
in humans for research). I typed endless statistical analyses of the
data.
When I was in college I worked summers for the state Department of
Education. They crunched a lot of numbers trying to formulate state wide
high school exams that would result in a pleasing bell curve. Sometimes
they succeeded. The crunching part was quite literal since they used
Friden square root calculators;

https://www.youtube.com/watch?v=fnmj9EzKgtg

That model isn't the square root variety but is very similar.

My memory of my statistics course was the chief use was determining how
many blivits you needed to test to insure only 5% of the blivits you
sold were junk. The 5% number varied depending on the cost of replacing
defective units and ****ing off customers.
That certainly was a very specific statistics course. Or your memory
of it was very specific.


My memory of it was that it was an 8 o'clock class taught by the aptly
named professor Dis Maly. I'm not making that up.

http://www.rpi.edu/magazine/winter20...lastthing.html

"I arrived from a small town but with vivid images of universities and
the academic life: musty buildings, clever professors, and studies of
wondrous things. My very first class filled the bill: Prof. Dis Maly was
slight of build, covered with chalk dust, and wreathed in pipe smoke.
Pleasant but somewhat distracted, he delivered lectures that were droll
but laced with wit. His eyes sparkled and danced when he spoke of
Calculus: The Mathematics of Change."

I don't remember his eyes ever sparking. Or the wit. Mrs. Maly also
taught mathematics and was very good; Dis was a cure for insomnia.

It was an engineering school hence the emphasis on engineering statistics:

https://en.wikipedia.org/wiki/Engineering_statistics

As such, most of the examples were drawn from QC applications. You do
know that the person behind the nom de plume Student was William Gossett
who worked for the Guinness Brewery. His interest was how many samples
of barley you have to pull to ensure the quality.

Most of my brushes with statistics in later life were just that, QC
applications. There is a certain degree of cynicism in mass production.
You know a percentage of the product will be defective. How much
sampling do you have to do to keep it in reasonable limits?

Another aspect is components like capacitors and resistors. Nobody sets
out to make resistors with a 20% tolerance. 5% would be nice but ****
happens. How much sampling do you do before deciding you just made a
batch of 20% components?

The 80386 chips had a similar quirk. The good ones could do 32 bit
multiplication. The bad ones were marked for 16 bit only and sold since
a lot of applications didn't require 32 bit operations at the time.
Again, how do you sort the sheep from the goats with a reasonable chance
of success?

Yes, I know there are many other uses for statistics but I also know
there have been many houses of cards built on statistics. I'm just
talking about honest procedural errors, not cooking the books to get the
answer you want.

[rbowman]

On 09/20/2019 12:41 PM, Cindy Hamilton wrote:
I believe fivethirtyeight.com takes the sampling methodology into account
when it rates surveys. It's got a raft of factors it considers when
judging the quality of survey results.

Silver had a rather red face a couple of years ago, iirc.

[rbowman]

On 09/20/2019 10:07 AM, wrote:
That is particularly true when you are just robo calling people to try
to get 1000 to talk to you. There is a big skew in your data right
there. Those are people who really think someone is going to lower the
interest rate on their credit card, the IRS takes gift cards as
payment and that microsoft found a problem in their computer.

I never answer the phone. The call goes straight to the machine and some
random time later if I notice the flashing light I'll listen to a few
seconds and delete the message.

[Cindy Hamilton[_2_]]

On Friday, September 20, 2019 at 11:13:10 PM UTC-4, rbowman wrote:
On 09/20/2019 12:41 PM, Cindy Hamilton wrote:
I believe fivethirtyeight.com takes the sampling methodology into account
when it rates surveys. It's got a raft of factors it considers when
judging the quality of survey results.

Silver had a rather red face a couple of years ago, iirc.

And he copped to it honestly.

Interpretation of statistics is where most people go wrong. Certainly
the popular press has little clue about how to present statistical
results. If I had a dollar for every time they've proclaimed some
conclusion that was within the error bars of the study, I could retire.

Cindy Hamilton

[Ralph Mowery]

In article ,
says...

I never answer the phone. The call goes straight to the machine and some
random time later if I notice the flashing light I'll listen to a few
seconds and delete the message.





One thing that has cut my robo calls is the internet phone service.
There is an option where if they know of a robo call number, my phone
will only ring one time and then I can look at the caller ID to see if I
want to call that number back. If I get nore than one call that goes
through , I can block that number.
Other unknown numbers/names to me do go to my answering machine.

---
This email has been checked for viruses by Avast antivirus software.
https://www.avast.com/antivirus

[trader_4]

On Friday, September 20, 2019 at 11:10:53 PM UTC-4, rbowman wrote:
On 09/20/2019 08:47 AM, Cindy Hamilton wrote:
On Friday, September 20, 2019 at 10:22:06 AM UTC-4, rbowman wrote:
On 09/20/2019 04:32 AM, Cindy Hamilton wrote:
I recall my first job. I worked for a man studying renal adenocarcinoma
in frogs as a model for human cancer (unless you'd prefer to induce cancer
in humans for research). I typed endless statistical analyses of the
data.
When I was in college I worked summers for the state Department of
Education. They crunched a lot of numbers trying to formulate state wide
high school exams that would result in a pleasing bell curve. Sometimes
they succeeded. The crunching part was quite literal since they used
Friden square root calculators;

https://www.youtube.com/watch?v=fnmj9EzKgtg

That model isn't the square root variety but is very similar.

My memory of my statistics course was the chief use was determining how
many blivits you needed to test to insure only 5% of the blivits you
sold were junk. The 5% number varied depending on the cost of replacing
defective units and ****ing off customers.
That certainly was a very specific statistics course. Or your memory
of it was very specific.


My memory of it was that it was an 8 o'clock class taught by the aptly
named professor Dis Maly. I'm not making that up.

http://www.rpi.edu/magazine/winter20...lastthing.html

"I arrived from a small town but with vivid images of universities and
the academic life: musty buildings, clever professors, and studies of
wondrous things. My very first class filled the bill: Prof. Dis Maly was
slight of build, covered with chalk dust, and wreathed in pipe smoke.
Pleasant but somewhat distracted, he delivered lectures that were droll
but laced with wit. His eyes sparkled and danced when he spoke of
Calculus: The Mathematics of Change."

I don't remember his eyes ever sparking. Or the wit. Mrs. Maly also
taught mathematics and was very good; Dis was a cure for insomnia.

It was an engineering school hence the emphasis on engineering statistics:

https://en.wikipedia.org/wiki/Engineering_statistics

As such, most of the examples were drawn from QC applications. You do
know that the person behind the nom de plume Student was William Gossett
who worked for the Guinness Brewery. His interest was how many samples
of barley you have to pull to ensure the quality.

Most of my brushes with statistics in later life were just that, QC
applications. There is a certain degree of cynicism in mass production.
You know a percentage of the product will be defective. How much
sampling do you have to do to keep it in reasonable limits?

Another aspect is components like capacitors and resistors. Nobody sets
out to make resistors with a 20% tolerance. 5% would be nice but ****
happens. How much sampling do you do before deciding you just made a
batch of 20% components?

The 80386 chips had a similar quirk. The good ones could do 32 bit
multiplication. The bad ones were marked for 16 bit only and sold since
a lot of applications didn't require 32 bit operations at the time.

That's all wrong. No such thing ever happened. You may be confusing
that with the 486SX, but even there it's wrong. The 386's were all
32 bit CPUs, they all did 32 bit math. Externally, the DX which came
out first, had a 32 bit bus, the SX had a 16 bit bus, but the
instruction set, registers, architecture were identical.

With the 486 there was the 486DX which came out first, which had a built-in
floating point unit added. Later the SX was introduced which did not have
the FLOATING POINT unit. Initially to get it to market quickly,
it was the same die, but it had the FPU disabled. Shortly after, the
FP was removed, reducing the die size and manufacturing cost. Both
of those had 32 bit buses and both did 32 bit math.

[Hawk]

On 9/19/2019 8:09 AM, Cindy Hamilton wrote:
On Thursday, September 19, 2019 at 7:43:38 AM UTC-4, Hawk wrote:
On 9/19/2019 4:30 AM, micky wrote:
In alt.home.repair, on Wed, 11 Sep 2019 11:03:12 +0100, Bod
wrote:

On 11/09/2019 10:52, Bod wrote:
A growing number of people fear a mass shooting could take place in
their community, and an overwhelming majority support background checks
for gun sales, according to a new poll.

Sixty percent of Americans said they are worried about a mass shooting
in their own community, according to the ABC News/Washington Post poll
released Monday. That's up 5 points compared to January 2013, when 55
percent of Americans said they feared a mass shooting.

https://thehill.com/homenews/news/46...community-poll

It's sadly ironic, that the very thing that you *think* makes you
safe(guns), is the very thing that brings you fear.

I guess that could have been said in WWII about the Royal Air Force and
the German air force. And maybe still today.


Was anyone in this group polled with that question? I never was. They
poll a small group of Americans and 55 out of 100 represent the entire
US population? Stop believing in polls. They are skewed to fit their agenda.

This group does not represent a random sample of Americans. They polled
more than 1000 people, a sufficient size for statistical significance.

Until it opposes your view.

[rbowman]

On 09/21/2019 04:02 AM, Cindy Hamilton wrote:
On Friday, September 20, 2019 at 11:13:10 PM UTC-4, rbowman wrote:
On 09/20/2019 12:41 PM, Cindy Hamilton wrote:
I believe fivethirtyeight.com takes the sampling methodology into account
when it rates surveys. It's got a raft of factors it considers when
judging the quality of survey results.

Silver had a rather red face a couple of years ago, iirc.

And he copped to it honestly.

Interpretation of statistics is where most people go wrong. Certainly
the popular press has little clue about how to present statistical
results. If I had a dollar for every time they've proclaimed some
conclusion that was within the error bars of the study, I could retire.

Sometimes I think the popular press gets led down the garden path. For
example flu season is coming. I don't want to get into a discussion of
whether or not one should get a flu shot but there is a certain vested
interest in presenting the most favorable argument even if it means
throwing out statistics that aren't quite what they seem.

Silver must be having fun this year. So far it's been like watching a
train wreck. The only Dems I had any interest in, Gabbard and Yang, got
kicked out of the clown car early.

[rbowman]

On 09/21/2019 08:01 AM, Ralph Mowery wrote:
One thing that has cut my robo calls is the internet phone service.
There is an option where if they know of a robo call number, my phone
will only ring one time and then I can look at the caller ID to see if I
want to call that number back. If I get nore than one call that goes
through , I can block that number.

The car warranty calls I get on my cell spoof a different number every
time.

[[email protected]]

On Sat, 21 Sep 2019 03:02:12 -0700 (PDT), Cindy Hamilton
wrote:

On Friday, September 20, 2019 at 11:13:10 PM UTC-4, rbowman wrote:
On 09/20/2019 12:41 PM, Cindy Hamilton wrote:
I believe fivethirtyeight.com takes the sampling methodology into account
when it rates surveys. It's got a raft of factors it considers when
judging the quality of survey results.

Silver had a rather red face a couple of years ago, iirc.

And he copped to it honestly.

Interpretation of statistics is where most people go wrong. Certainly
the popular press has little clue about how to present statistical
results. If I had a dollar for every time they've proclaimed some
conclusion that was within the error bars of the study, I could retire.

Cindy Hamilton

I am not sure if you call what I did for around 5 years at IBM
statistics but I mined big databases looking for trends and anomalies.
The first thing I found out is I can create a report that makes
something look good or bad, simply by how I structure the SQL query
and every fact is true using the same data.

[[email protected]]

On Sat, 21 Sep 2019 10:46:36 -0600, rbowman
wrote:

On 09/21/2019 08:01 AM, Ralph Mowery wrote:
One thing that has cut my robo calls is the internet phone service.
There is an option where if they know of a robo call number, my phone
will only ring one time and then I can look at the caller ID to see if I
want to call that number back. If I get nore than one call that goes
through , I can block that number.

The car warranty calls I get on my cell spoof a different number every
time.

I have asked them a few times what they can do for me on my 97 Honda.
They usually hang up once I ask them what it covers.