Correct process for creating a linear trend line in Excel

[Onion Knight]

There has been some debate in COLA as to the correct way to create a
linear trendline in Excel. Someone in the forum kindly made a video to
show what he thought was the correct method
http://tmp.gallopinginsanity.com/LinearTrendLineCreation.mov

A couple of other people in the group say that the video skips a
number of steps but when I look at the directions from Microsoft I do
not see any they could be referring to.

Does anyone see any missing steps for the creation of a linear trend
line? I admit this is a bit over my head so any input would be welcome.

 

[Peter Köhlmann]

There has been some debate in COLA as to the correct way to create a
linear trendline in Excel. Someone in the forum kindly made a video to
show what he thought was the correct method
http://tmp.gallopinginsanity.com/LinearTrendLineCreation.mov

A couple of other people in the group say that the video skips a
number of steps but when I look at the directions from Microsoft I do
not see any they could be referring to.

Does anyone see any missing steps for the creation of a linear trend
line? I admit this is a bit over my head so any input would be welcome.

Give it up, Snit Glasser.

Nobody here believes that "Onion Knight" is anything but a sock

 

[joeu2004]

There has been some debate in COLA as to the correct way
to create a linear trendline in Excel. Someone in the forum
kindly made a video to show what he thought was the correct
method
http://tmp.gallopinginsanity.com/LinearTrendLineCreation.mov [....]
Does anyone see any missing steps for the creation of a
linear trend line?

I cannot find the original debate in comp.os.linux.advocacy (COLA); just the
thread titled "Visualizing where to draw the standard deviation line", which
refers to yet-another "a debate in COLA".

No matter; I'm not interested. I think we can answer your Excel questions
without getting into anything so esoteric.

I see no missing steps in creating the linear trendline per se using Excel.

I would also select the trendline options to display the trendline equation
and the R-squared of the trendline.

And after displaying the trendline equation, I might select the option to
Format Trendline Label in order to alter the display precision of the
coefficients. I would select Scientific with 14 decimal places if I want to
see the trendline coefficients "exactly"; that is, the most precision that
Excel will display.

But none of those "steps" affects how the linear trendline is created.

FYI, we can also use the Excel function LINEST to determine the regression
line coefficients.

If the dates are in A1:A24 and the percentages are in B1:B24, select two
horizontal cells and array-enter the following formula (press
ctrl+shift+Enter instead just Enter): =LINEST(B1:B24,A1:A24).

To compare with the "exact" trendline coefficients, I would also format the
LINEST results as Scientific with 14 decimal places.

You might notice infinitesimal differences between the trendline
coefficients and the LINEST results. But in this case, they are same up 10
or 11 decimal places in this case; "close enough for government work". As a
guess, the difference might be attributed to physically different internal
algorithms and/or to differences in internal floating-point arithmetic
anomalies.

For some insight into how the Excel LINEST and linear trendline coefficients
might be derived, see http://en.wikipedia.org/wiki/Simple_linear_regression.

Note that =COVAR(A1:A24,B1:B24)/VARP(A1:A24) in E3 and
=AVERAGE(B1:B24)-E3*AVERAGE(A1:A24) are about the same as the LINEST
results. The difference is in the 3 least-significant binary bits of the
floating-point representations in this case.

You might also take note of the wiki page section titled "Normality
assumption". That may or may not be relevant to the COLA debates.

 

[Snit]

There has been some debate in COLA as to the correct way
to create a linear trendline in Excel. Someone in the forum
kindly made a video to show what he thought was the correct
method
http://tmp.gallopinginsanity.com/LinearTrendLineCreation.mov [....]
Does anyone see any missing steps for the creation of a
linear trend line?

I cannot find the original debate in comp.os.linux.advocacy (COLA); just the
thread titled "Visualizing where to draw the standard deviation line", which
refers to yet-another "a debate in COLA".

No matter; I'm not interested. I think we can answer your Excel questions
without getting into anything so esoteric.

Really the video is all you need to see (I am the one who created it and the
other related images with trend lines)... it shows the steps and they match
what MS suggests exactly (with the exception that I use the right-click
method to get to some tools where the MS page suggests using the Ribbon).

I see no missing steps in creating the linear trendline per se using Excel.

Thank you. I think anyone who knows Excel well... or even not so well... or
even poorly but can read the steps provided by MS... can see this. But
someone in COLA kept saying I was missing steps:

cc:
------
I will gladly educate your ignorant ass, and give you step by
step instructions on how to correctly do it so you can see
all your missing steps, but first you have to stop lying.
------
So you are missing some key steps before doing the trendline.
I have said that many times now.
------

He later said he was in reference to doing more than just creating the trend
line, which is all I said I was doing - he claimed I was missing steps to do
part of a process I never claimed to do. Yes: his claims were idiotic.
This is the nature of the "advocacy" groups.

Bottom line: my steps for creating a linear trend line were flawless, which
is not to say creating such a line was the only form of analysis that could
be done on such data or that the creation of a linear trend line on data
that is clearly not linear is the best way to do an analysis.

Using the same steps, though, I was able to show a trend line with a very
good fit for the second half of 2011:
<http://tmp.gallopinginsanity.com/LinuxTrend2011-2ndhalf.png>

And also able to show how the trend changed for the first half of 2012:
<http://tmp.gallopinginsanity.com/LinuxTrendLine2012.jpg>

The fact is, a linear trend line is not going to show a very good fit with
non-linear data... or data where the trend changes.

I would also select the trendline options to display the trendline equation
and the R-squared of the trendline.

You can see where I did that in the above links... and in others that I
made... a whole bunch of them combined:
<http://tmp.gallopinginsanity.com/LinuxMultLinearTrendLines.png>

Even then my labeling is not done well... but my only claim was that I made
the linear trend line correctly, not that the graph had good labeling, etc.
I openly acknowledge it did not.

And after displaying the trendline equation, I might select the option to
Format Trendline Label in order to alter the display precision of the
coefficients. I would select Scientific with 14 decimal places if I want to
see the trendline coefficients "exactly"; that is, the most precision that
Excel will display.

Seems a bit excessive for the needs here - but I did know about those
options.

But none of those "steps" affects how the linear trendline is created.

Correct. Nor do they remove the fact there are other forms of analysis that
can be done.

FYI, we can also use the Excel function LINEST to determine the regression
line coefficients.

If the dates are in A1:A24 and the percentages are in B1:B24, select two
horizontal cells and array-enter the following formula (press
ctrl+shift+Enter instead just Enter): =LINEST(B1:B24,A1:A24).

I have used this function a couple of times... but never for any real work.

To compare with the "exact" trendline coefficients, I would also format the
LINEST results as Scientific with 14 decimal places.

You might notice infinitesimal differences between the trendline
coefficients and the LINEST results. But in this case, they are same up 10
or 11 decimal places in this case; "close enough for government work". As a
guess, the difference might be attributed to physically different internal
algorithms and/or to differences in internal floating-point arithmetic
anomalies.

For some insight into how the Excel LINEST and linear trendline coefficients
might be derived, see http://en.wikipedia.org/wiki/Simple_linear_regression.

Note that =COVAR(A1:A24,B1:B24)/VARP(A1:A24) in E3 and
=AVERAGE(B1:B24)-E3*AVERAGE(A1:A24) are about the same as the LINEST
results. The difference is in the 3 least-significant binary bits of the
floating-point representations in this case.

You might also take note of the wiki page section titled "Normality
assumption". That may or may not be relevant to the COLA debates.

Not really... though the idea of a normal distribution was also debated. A
bit of info you might or might not know: you can easily visualize where the
standard deviation (sigma) lines should be drawn on a normal distribution...
the lines are *always* at the distance from the mean to the inflection point
on the curve (where the concavity changes). I was repeatedly told by the
same person the distance was irrelevant:

cc:
-----
There'se nothing wrong with the image, other than some weird
axis labeling.
-----
Snit's so fucking stupid he thinks the sigma lines are drawn
based on distance from the mean, not area under the curve.
-----
| The sigma lines are drawn based on the area of the curve -
| which is easy to see when the images screw it up, esp. when
| they do so really badly, like in some of the ones I showed
| you.
They are not wrong.
------
LOL!!!! All of those links are fine. The first sigma lines
cover 68% of the area UNDER THE CURVE.
-----
If you would like to prove, on any single one of the links
you call incorrect, that the first sigma lines do not bound
an area that is 68.2% of the area UNDER THE CURVE, then I
would like to see it.
-----
Hahahaha your "approximate inflection points" are hilarious.
Please, post more on this subject.
------

I explained to him why he was wrong here: <

>

There are many examples of people who should know better getting this wrong.
Here are some

<http://www.udel.edu/htr/Statistics/Images/Class12/normal2.gif> From:
<http://www.udel.edu/htr/Statistics/Notes/class12.html>

Which is the example I used for showing how him can make a decent
approximation visually: <http://tmp.gallopinginsanity.com/sd.png>.

I also pointed to some other examples which at least appear incorrect to me
(though they are not as far off as the above example):

<http://www.footballguys.com/shickstandard_1_files/image009.gif> From:
<http://www.footballguys.com/shickstandard_1.htm>
Sigma lines clearly not at a far enough distance from the mean, esp. on the
graph to the right.

<http://www.gsseser.com/images/StandardDeviation2s.gif> From:
<http://www.gsseser.com/Deviation.htm>
Sigma lines clearly not at a far enough distance from the mean.

You would think that such sites would be made by people who knew better. I
openly admit I am not a math wiz but it is rather silly when sites claiming
to be teaching such things get their depictions wrong (of course, one of the
sites above is from "Footballguys"... and you might not expect them, by
stereotype, to be the most knowledgeable in such areas anyway).

Anyway, sorry to have the idiotic debate spread to other forums... though it
is sorta fun to see people who clearly know what they are talking about rip
apart those who were calling me names and insisting they "knew" I was wrong.

 

[Steve Carroll]

There has been some debate in COLA as to the correct way
to create a linear trendline in Excel. Someone in the forum
kindly made a video to show what he thought was the correct
method
http://tmp.gallopinginsanity.com/LinearTrendLineCreation.mov [....]
Does anyone see any missing steps for the creation of a
linear trend line?

I cannot find the original debate in comp.os.linux.advocacy (COLA); just the
thread titled "Visualizing where to draw the standard deviation line", which
refers to yet-another "a debate in COLA".

No matter; I'm not interested.  I think we can answer your Excel questions
without getting into anything so esoteric.

I see no missing steps in creating the linear trendline per se using Excel.

Given your reply it's pretty evident you realize that the process you
undertake depends on what your goal for creating the trend line is and
why you'd want to do the steps you mentioned in your reply. Earlier,
in another forum, Snit was talking about this and a poster cut right
to the chase by asking Snit (based on common sense and what Snit had
previously written) the two most obvious questions:

"I gather there are two questions here:

1) Is the trendline approriately fitted to that data?

2) Is the trendline useful in some way?"

Snit's reply:
"Not quite: the question was merely if the process of creating the
trend line was correct - did it follow the process of creating a
linear trend line that is supported by the build in "linear trend
line" properties of the program."

As you can see, Snit himself explains that his goal was not to see if
the "trendline approriately fitted to that data" nor to have the
trendline be "useful in some way". Personally I find this rather
strange... but Snit's goal was revealed in the same post the above
quotes came from... and it's actually even more strange. Notably,
Snit wrote about a correlation between Linux UI improvements and an
increase in market share (a thing he has been arguing in another
newsgroup):

"The point, however, was not to predict the future past where I first
created the line. I had noted that the desktop Linux distros were
clearly focusing on improving their usability - and I predicted that
an increase in usability would lead to an increase in users."

And this, from the same post:

"But the graph was initially made to show that the then-current data
fit with my past predictions - not to make any specific prediction
about the future (such as that the same level of upward trend would be
seen)."


Anyone familiar with this topic knows that, regardless of the method
he used to create his "past predictions" (or any, for that matter) the
data Snit used would not aid in supporting a correlation between UI
improvements in Linux and a rise in marketshare because it's the wrong
kind of data.

We know Snit didn't care if the line fit his data or was useful in any
way... if he's going to argue that he didn't produce the line to show
evidence of his alleged correlation (a thing he is saying up above...
but possibly without realizing it) then one would have to logically
ask:

What the hell was the purpose of his trend line?

 

[Onion Knight]

There has been some debate in COLA as to the correct way
to create a linear trendline in Excel. Someone in the forum
kindly made a video to show what he thought was the correct
method
http://tmp.gallopinginsanity.com/LinearTrendLineCreation.mov [....]
Does anyone see any missing steps for the creation of a
linear trend line?

I cannot find the original debate in comp.os.linux.advocacy (COLA); just the
thread titled "Visualizing where to draw the standard deviation line", which
refers to yet-another "a debate in COLA".

No matter; I'm not interested.  I think we can answer your Excel questions
without getting into anything so esoteric.

I see no missing steps in creating the linear trendline per se using Excel.

Thank you for your posting of this educated opinion, even though it
broke Steve. He is now running around screaming about how I and
anyone else who agreed with Snit on this must be Snit's sock puppets.
Give it a little time and he will insist you must be Snit as well.

This matter is settled. Snit was right to say his process was fine
and cc and Carroll who argued against him were wrong. Snit's linear
trendline was created without flaw. This was obvious from the start
but it is good to get a more educated and impartial stating of it. Fun
to watch Steve crumble as he reads it too.

 

[Onion Knight]

OK, but you'll be missing the best part. Here's an absolutely
hilarious exchange that shows how watered down Snit's trend line
argument has gotten (to the point where it is now literally non-
existent). In some newsgroups he cross-trolled to earlier (in the
hopes of getting an appeal to authority so he could claim a win over
cc), a poster who knows the topic cut right to the chase:

--
"I gather there are two questions here:

1) Is the trendline approriately fitted to that data?

2) Is the trendline useful in some way? "
--

Snit's outrageous reply is the funniest thing I've seen on usenet in a
long time:

--
"Not quite: the question was merely if the process of creating the
trend line was correct - did it follow the process of creating a
linear trend line that is supported by the build in "linear trend
line" properties of the program."
--

In other words, Snit knew he couldn't bullsh*t this guy so he had to
say "Not quite" to the two main reasons for even creating a trend line
in the first place!  And the reason he gave is equally as funny... he
wanted to "merely" know if his "process" in Excel was "correct" for
creating the trend line. Naturally, when you create a trend line you
have *some* goals in mind... the goals the poster stated. You can't
even talk about the process used to create the trend line until you
know what the goals for creating it are. This is just common sense of
which Snit obviously has none of... as he tossed those goals aside in
an effort to prove he was "right". Too funny

This pathetically cross-posted thread only got created because Snit
dumped the idea of using people who are good at math because they
exposed him. Now he's appealing to people who have used the trend line
function in Excel and is, like the post I just showed you, completely
dumping the reason for creating it altogether in the hopes of  saving
face. You'll notice that he is not pointing out how his goal for
creating the trend line is *not* to see if it will be "approriately
fitted to that data" or be "useful in some way" (a thing any sane
person would assume). Gee, do you think he would have received a
different answer had he made that clear? LOL!

Snit made the claim that he created a linear trendline and did so
correctly. I would love to see where you think Snit said the 2011
trend would continue into 2012, unchanged.

His 2011 trendline fit his claims well. His 2012 tendline did not. You
say that others "exposed him" on this but he never claimed otherwise.
You are just throwing a toddler tantrum because Snit was proved right
by people who know a lot more than you.

For the last decade or so you have been following him around lying
about him because your girlfriend dumped your lying ass and you blamed
him. Poor Steve dumped by an insane bitch and unable to get over it.
Yes Steve the woman who dumped your lying ass was insane and evil. If
you had half a brain you would be celebrating that she left you and
not crying about it into the next decade.

 

[Onion Knight]

There has been some debate in COLA as to the correct way
to create a linear trendline in Excel. Someone in the forum
kindly made a video to show what he thought was the correct
method
http://tmp.gallopinginsanity.com/LinearTrendLineCreation.mov [....]
Does anyone see any missing steps for the creation of a
linear trend line?

I cannot find the original debate in comp.os.linux.advocacy (COLA); just the
thread titled "Visualizing where to draw the standard deviation line", which
refers to yet-another "a debate in COLA".

No matter; I'm not interested.  I think we can answer your Excel questions
without getting into anything so esoteric.

Really the video is all you need to see (I am the one who created it and the
other related images with trend lines)... it shows the steps and they match
what MS suggests exactly (with the exception that I use the right-click
method to get to some tools where the MS page suggests using the Ribbon).

I see no missing steps in creating the linear trendline per se using Excel.

Thank you.  I think anyone who knows Excel well... or even not so well.... or
even poorly but can read the steps provided by MS... can see this.  But
someone in COLA kept saying I was missing steps:

  cc:
    ------
    I will gladly educate your ignorant ass, and give you step by
    step instructions on how to correctly do it so you can see
    all your missing steps, but first you have to stop lying.
    ------
    So you are missing some key steps before doing the trendline.
    I have said that many times now.
    ------

He later said he was in reference to doing more than just creating the trend
line, which is all I said I was doing - he claimed I was missing steps todo
part of a process I never claimed to do.  Yes: his claims were idiotic.
This is the nature of the "advocacy" groups.

Bottom line: my steps for creating a linear trend line were flawless, which
is not to say creating such a line was the only form of analysis that could
be done on such data or that the creation of a linear trend line on data
that is clearly not linear is the best way to do an analysis.

Using the same steps, though, I was able to show a trend line with a very
good fit for the second half of 2011:
<http://tmp.gallopinginsanity.com/LinuxTrend2011-2ndhalf.png>

And also able to show how the trend changed for the first half of 2012:
<http://tmp.gallopinginsanity.com/LinuxTrendLine2012.jpg>

The fact is, a linear trend line is not going to show a very good fit with
non-linear data... or data where the trend changes.

I would also select the trendline options to display the trendline equation
and the R-squared of the trendline.

You can see where I did that in the above links... and in others that I
made... a whole bunch of them combined:
<http://tmp.gallopinginsanity.com/LinuxMultLinearTrendLines.png>

Even then my labeling is not done well... but my only claim was that I made
the linear trend line correctly, not that the graph had good labeling, etc.
I openly acknowledge it did not.

And after displaying the trendline equation, I might select the option to
Format Trendline Label in order to alter the display precision of the
coefficients.  I would select Scientific with 14 decimal places if I want to
see the trendline coefficients "exactly"; that is, the most precision that
Excel will display.

Seems a bit excessive for the needs here - but I did know about those
options.

But none of those "steps" affects how the linear trendline is created.

Correct.  Nor do they remove the fact there are other forms of analysisthat
can be done.

FYI, we can also use the Excel function LINEST to determine the regression
line coefficients.

If the dates are in A1:A24 and the percentages are in B1:B24, select two
horizontal cells and array-enter the following formula (press
ctrl+shift+Enter instead just Enter):  =LINEST(B1:B24,A1:A24).

I have used this function a couple of times... but never for any real work.

To compare with the "exact" trendline coefficients, I would also formatthe
LINEST results as Scientific with 14 decimal places.

You might notice infinitesimal differences between the trendline
coefficients and the LINEST results.  But in this case, they are sameup 10
or 11 decimal places in this case; "close enough for government work".  As a
guess, the difference might be attributed to physically different internal
algorithms and/or to differences in internal floating-point arithmetic
anomalies.

For some insight into how the Excel LINEST and linear trendline coefficients
might be derived, seehttp://en.wikipedia.org/wiki/Simple_linear_regression.

Note that =COVAR(A1:A24,B1:B24)/VARP(A1:A24) in E3 and
=AVERAGE(B1:B24)-E3*AVERAGE(A1:A24) are about the same as the LINEST
results.  The difference is in the 3 least-significant binary bits ofthe
floating-point representations in this case.

You might also take note of the wiki page section titled "Normality
assumption".  That may or may not be relevant to the COLA debates.

Not really... though the idea of a normal distribution was also debated.  A
bit of info you might or might not know: you can easily visualize where the
standard deviation (sigma) lines should be drawn on a normal distribution....
the lines are *always* at the distance from the mean to the inflection point
on the curve (where the concavity changes).  I was repeatedly told by the
same person the distance was irrelevant:

  cc:
    -----
    There'se nothing wrong with the image, other than some weird
    axis labeling.
    -----
    Snit's so fucking stupid he thinks the sigma lines are drawn
    based on distance from the mean, not area under the curve.
    -----
    | The sigma lines are drawn based on the area of the curve -
    | which is easy to see when the images screw it up, esp. when
    | they do so really badly, like in some of the ones I showed
    | you.
    They are not wrong.
    ------
    LOL!!!! All of those links are fine. The first sigma lines
    cover 68% of the area UNDER THE CURVE.
    -----
    If you would like to prove, on any single one of the links
    you call incorrect, that the first sigma lines do not bound
    an area that is 68.2% of the area UNDER THE CURVE, then I
    would like to see it.
    -----
    Hahahaha your "approximate inflection points" are hilarious.
    Please, post more on this subject.
    ------

I explained to him why he was wrong here: <

>

There are many examples of people who should know better getting this wrong.
Here are some

<http://www.udel.edu/htr/Statistics/Images/Class12/normal2.gif> From:
<http://www.udel.edu/htr/Statistics/Notes/class12.html>

Which is the example I used for showing how him can make a decent
approximation visually: <http://tmp.gallopinginsanity.com/sd.png>.

I also pointed to some other examples which at least appear incorrect to me
(though they are not as far off as the above example):

<http://www.footballguys.com/shickstandard_1_files/image009.gif> From:
<http://www.footballguys.com/shickstandard_1.htm>
Sigma lines clearly not at a far enough distance from the mean, esp. on the
graph to the right.

<http://www.gsseser.com/images/StandardDeviation2s.gif> From:
<http://www.gsseser.com/Deviation.htm>
Sigma lines clearly not at a far enough distance from the mean.

You would think that such sites would be made by people who knew better.  I
openly admit I am not a math wiz but it is rather silly when sites claiming
to be teaching such things get their depictions wrong (of course, one of the
sites above is from "Footballguys"... and you might not expect them, by
stereotype, to be the most knowledgeable in such areas anyway).  

Anyway, sorry to have the idiotic debate spread to other forums... thoughit
is sorta fun to see people who clearly know what they are talking about rip
apart those who were calling me names and insisting they "knew" I was wrong.


--
The indisputable facts about that absurd debate: <http://goo.gl/2337P>
cc being proved wrong about his stats BS: <http://goo.gl/1aYrP>
7 simple questions cc will *never* answer: <http://goo.gl/cNBzu>
cc again pretends to be knowledgeable about things he is clueless about.

Funny how while Steve is throwing a toddler tantrum we have not as of
yet heard from cc. I wonder how big of a fit he will throw. All I know
is it is almost sure to be a good show. Goin' off to grab me a beer
and a big ass bag of popcorn.

 

[Onion Knight]

Again: Carroll is lying when he claims Onion Knight is my sock.  There is a
reason there is *no* support for his accusation... and a reason he is making
the accusation is clear: in addition to his paranoia, he is wanting to
change the topic from the fact I was right and he and cc were wrong.

But this is the last I will say on the subject unless there is something new
added to the subject... you know, like reason or logic or evidence or
something of value that would deserve a refutation.  Until then it is just
Carroll screaming how Onion Knight and Kaba are both my socks for no reason
other than the fact they are, at least mostly, right.

--
The indisputable facts about that absurd debate: <http://goo.gl/2337P>
cc being proved wrong about his stats BS: <http://goo.gl/1aYrP>
7 simple questions cc will *never* answer: <http://goo.gl/cNBzu>
cc again pretends to be knowledgeable about things he is clueless about.

Steve knows he has been proved wrong so he is now in major attack
mode. You and I and anyone else who notes what a lying ass he is will
all be seen as the same person. This has been his MO for a decade or
more. He is fucking insane.

 

[Snit]

Snit made the claim that he created a linear trendline and did so
correctly. I would love to see where you think Snit said the 2011
trend would continue into 2012, unchanged.

I never did... though I said it *might* continue its upward trend or it
*might* stagnate. I did not predict its significant decline.

So for 2011 I was right about my prediction. It would be an interesting
discussion to look into why things changed and I was wrong about the 2012
trend. If Steve and cc wanted to attack me for being wrong, *this* would be
the place to do so because this is the place where I was *actually* wrong.

But they are too lost in what was being discussed to figure this out (even
though I told them). And, of course, they will never admit I was right
about the my method of creating a linear trend line, nor admit I was right
about the sigma line being drawn at the distance from the mean to the
inflection point, nor admit that the examples of poorly depicted sigma lines
were in fact poorly depicted, nor will they admit that I was right to note
that the R^2 value does *not* indicate if the *process* of creating a trend
line is correct or not, etc. cc argued against me on all of these points
and he was wrong. And Steve backed cc because of his illness. And Steve
added his own lunacy in claiming that the purpose of looking at the trend
for one time period necessitates that one expects that trend to continue.
This is even more insane when one considers the linear nature of the trend
line... it is not as a percentage growth rate can continue "into infinity"
as he said. If it did, sometime in the next 100 years the total of Linux
usage would be *greater* than 100%. Maybe someday if Carroll is not being
an ass to me I will explain why this is not a likely scenario.

His 2011 trendline fit his claims well. His 2012 tendline did not. You
say that others "exposed him" on this but he never claimed otherwise.

Absolutely correct. I have been clear from the start that the 2012 data did
not fit my prediction.

You are just throwing a toddler tantrum because Snit was proved right
by people who know a lot more than you.

Carroll has had a number of very public bread-downs where he starts accusing
multiple people of being my socks (while at the same time he runs from
answering names he has used), threatens to contact my family and employer,
falsely accuses me of sexually harassing either himself or his
ex-girlfriend, and otherwise just freaks out. He then goes away for a
while, claiming to he has "found a life"... and comes back for another
round.

And this is all without me directly replying to him. Just amazing how
messed up he is.

For the last decade or so you have been following him around lying
about him because your girlfriend dumped your lying ass and you blamed
him. Poor Steve dumped by an insane bitch and unable to get over it.
Yes Steve the woman who dumped your lying ass was insane and evil. If
you had half a brain you would be celebrating that she left you and
not crying about it into the next decade.

Absolutely correct and very well stated.

 

[Snit]

There has been some debate in COLA as to the correct way
to create a linear trendline in Excel. Someone in the forum
kindly made a video to show what he thought was the correct
method
http://tmp.gallopinginsanity.com/LinearTrendLineCreation.mov [....]
Does anyone see any missing steps for the creation of a
linear trend line?

I cannot find the original debate in comp.os.linux.advocacy (COLA); just the
thread titled "Visualizing where to draw the standard deviation line", which
refers to yet-another "a debate in COLA".

No matter; I'm not interested.  I think we can answer your Excel questions
without getting into anything so esoteric.

I see no missing steps in creating the linear trendline per se using Excel.

Thank you for your posting of this educated opinion, even though it
broke Steve. He is now running around screaming about how I and
anyone else who agreed with Snit on this must be Snit's sock puppets.
Give it a little time and he will insist you must be Snit as well.

I wonder what his max has been? I am sure it has been at least three people
at once he was claiming were me. He is obsessed with me. It is whacky.

This matter is settled. Snit was right to say his process was fine
and cc and Carroll who argued against him were wrong. Snit's linear
trendline was created without flaw. This was obvious from the start
but it is good to get a more educated and impartial stating of it. Fun
to watch Steve crumble as he reads it too.

Agreed. Now let's let it drop - at least from the non-COLA groups. And,
really, even from the COLA groups.

Carroll and cc will run around making up stories about how people from all
groups on Usenet are me and will insist that these folks who know more than
they do are just wrong or missing the point or will hunt for some point to
nit pick. Meanwhile the facts will remain the same:

* My process for creating trend lines in Excel was correct.
* My prediction of an increase in usage of Linux was correct for 2011
* My prediction of Linux usage staying at that level or increasing
in 2012 has turned out to *not* be correct.

And with all the other stuff about sigma lines and incorrect depictions and
yada yada yada I was right and they were simply trolling me.

And it is easy to predict their trolling will only increase now that others
have commented on these matters. Their frenzy will be a sight to behold.

 

[Snit]

Funny how while Steve is throwing a toddler tantrum we have not as of
yet heard from cc. I wonder how big of a fit he will throw. All I know
is it is almost sure to be a good show. Goin' off to grab me a beer
and a big ass bag of popcorn.

Prediction: cc will *never* admit the depictions I linked to above are
clearly wrong. Never. He will throw fits and change topics, he will call
me names, he will forge quotes and falsely attribute them to me and to
others, he will insist he is right, he will dodge and dance and play games
to the end of time.

But he will never, ever, under any circumstance admit what has been clear
for a long time but now has even more backing - the above depictions of
sigma lines are *wrong*. And he will never admit he was ignorant of the
sigma lines being correctly depicted at the inflection points. Never. Poor
cc... he is completely predictable in his inability to admit to his own
mistakes.

And Carroll will be right their backing cc's every lie.

I think I will share some of that popcorn with you and watch their show.

 

[Snit]

Steve knows he has been proved wrong so he is now in major attack
mode. You and I and anyone else who notes what a lying ass he is will
all be seen as the same person. This has been his MO for a decade or
more. He is fucking insane.

What if Steve is right and everyone on the Internet is me... and I just have
not yet figured it out?

 

[Martin Brown]

There has been some debate in COLA as to the correct way
to create a linear trendline in Excel. Someone in the forum
kindly made a video to show what he thought was the correct
method
http://tmp.gallopinginsanity.com/LinearTrendLineCreation.mov [....]
Does anyone see any missing steps for the creation of a
linear trend line?

I cannot find the original debate in comp.os.linux.advocacy (COLA); just
the thread titled "Visualizing where to draw the standard deviation
line", which refers to yet-another "a debate in COLA".

No matter; I'm not interested. I think we can answer your Excel
questions without getting into anything so esoteric.

I see no missing steps in creating the linear trendline per se using Excel.

I would also select the trendline options to display the trendline
equation and the R-squared of the trendline.

And after displaying the trendline equation, I might select the option
to Format Trendline Label in order to alter the display precision of the
coefficients. I would select Scientific with 14 decimal places if I want
to see the trendline coefficients "exactly"; that is, the most precision
that Excel will display.

But none of those "steps" affects how the linear trendline is created.

FYI, we can also use the Excel function LINEST to determine the
regression line coefficients.

If the dates are in A1:A24 and the percentages are in B1:B24, select two
horizontal cells and array-enter the following formula (press
ctrl+shift+Enter instead just Enter): =LINEST(B1:B24,A1:A24).

To compare with the "exact" trendline coefficients, I would also format
the LINEST results as Scientific with 14 decimal places.

You might notice infinitesimal differences between the trendline
coefficients and the LINEST results. But in this case, they are same up
10 or 11 decimal places in this case; "close enough for government
work". As a guess, the difference might be attributed to physically
different internal algorithms and/or to differences in internal
floating-point arithmetic anomalies.

The difference is that (apart from in some very early versions of XL2007
where the graph regression function was wrecked to make it agree with
another well known products answers) the linear regression on XL charts
uses a well designed algorithm that gives a very good approximation to
the true least squares parameter fit to data. LINEST is demonstrably
numerically unstable although it requires at least a 3rd order
polynomial fit and moderately hostile data to break it. Examples have
been posted here or in another m.p.e group previously.
(some years back)

For linear fits the two are more or less interchangable but for higher
orders of polynomial the chart function fit is more accurate.

For some insight into how the Excel LINEST and linear trendline
coefficients might be derived, see
http://en.wikipedia.org/wiki/Simple_linear_regression.

Note that =COVAR(A1:A24,B1:B24)/VARP(A1:A24) in E3 and
=AVERAGE(B1:B24)-E3*AVERAGE(A1:A24) are about the same as the LINEST
results. The difference is in the 3 least-significant binary bits of the
floating-point representations in this case.

You might also take note of the wiki page section titled "Normality
assumption". That may or may not be relevant to the COLA debates.

What is COLA?

 

[Steve Carroll]

There has been some debate in COLA as to the correct way
to create a linear trendline in Excel. Someone in the forum
kindly made a video to show what he thought was the correct
method
http://tmp.gallopinginsanity.com/LinearTrendLineCreation.mov [....]
Does anyone see any missing steps for the creation of a
linear trend line?

I cannot find the original debate in comp.os.linux.advocacy (COLA); just
the thread titled "Visualizing where to draw the standard deviation
line", which refers to yet-another "a debate in COLA".

No matter; I'm not interested. I think we can answer your Excel
questions without getting into anything so esoteric.

I see no missing steps in creating the linear trendline per se using Excel.

I would also select the trendline options to display the trendline
equation and the R-squared of the trendline.

And after displaying the trendline equation, I might select the option
to Format Trendline Label in order to alter the display precision of the
coefficients. I would select Scientific with 14 decimal places if I want
to see the trendline coefficients "exactly"; that is, the most precision
that Excel will display.

But none of those "steps" affects how the linear trendline is created.

FYI, we can also use the Excel function LINEST to determine the
regression line coefficients.

If the dates are in A1:A24 and the percentages are in B1:B24, select two
horizontal cells and array-enter the following formula (press
ctrl+shift+Enter instead just Enter): =LINEST(B1:B24,A1:A24).

To compare with the "exact" trendline coefficients, I would also format
the LINEST results as Scientific with 14 decimal places.

You might notice infinitesimal differences between the trendline
coefficients and the LINEST results. But in this case, they are same up
10 or 11 decimal places in this case; "close enough for government
work". As a guess, the difference might be attributed to physically
different internal algorithms and/or to differences in internal
floating-point arithmetic anomalies.

The difference is that (apart from in some very early versions of XL2007
where the graph regression function was wrecked to make it agree with
another well known products answers) the linear regression on XL charts
uses a well designed algorithm that gives a very good approximation to
the true least squares parameter fit to data. LINEST is demonstrably
numerically unstable although it requires at least a 3rd order
polynomial fit and moderately hostile data to break it. Examples have
been posted here or in another m.p.e group previously.
(some years back)

For linear fits the two are more or less interchangable but for higher
orders of polynomial the chart function fit is more accurate.

For some insight into how the Excel LINEST and linear trendline
coefficients might be derived, see
http://en.wikipedia.org/wiki/Simple_linear_regression.

Note that =COVAR(A1:A24,B1:B24)/VARP(A1:A24) in E3 and
=AVERAGE(B1:B24)-E3*AVERAGE(A1:A24) are about the same as the LINEST
results. The difference is in the 3 least-significant binary bits of the
floating-point representations in this case.

You might also take note of the wiki page section titled "Normality
assumption". That may or may not be relevant to the COLA debates.

What is COLA?

The newsgroup where the troll Snit and his sock puppet Onion Knight
are lying and trolling with this topic.

 

[Onion Knight]

On 16/06/2012 00:22, joeu2004 wrote:

There has been some debate in COLA as to the correct way
to create a linear trendline in Excel. Someone in the forum
kindly made a video to show what he thought was the correct
method
http://tmp.gallopinginsanity.com/LinearTrendLineCreation.mov
[....]
Does anyone see any missing steps for the creation of a
linear trend line?
I cannot find the original debate in comp.os.linux.advocacy (COLA); just
the thread titled "Visualizing where to draw the standard deviation
line", which refers to yet-another "a debate in COLA".
No matter; I'm not interested. I think we can answer your Excel
questions without getting into anything so esoteric.
I see no missing steps in creating the linear trendline per se using Excel.
I would also select the trendline options to display the trendline
equation and the R-squared of the trendline.
And after displaying the trendline equation, I might select the option
to Format Trendline Label in order to alter the display precision of the
coefficients. I would select Scientific with 14 decimal places if I want
to see the trendline coefficients "exactly"; that is, the most precision
that Excel will display.
But none of those "steps" affects how the linear trendline is created..
FYI, we can also use the Excel function LINEST to determine the
regression line coefficients.
If the dates are in A1:A24 and the percentages are in B1:B24, select two
horizontal cells and array-enter the following formula (press
ctrl+shift+Enter instead just Enter): =LINEST(B1:B24,A1:A24).
To compare with the "exact" trendline coefficients, I would also format
the LINEST results as Scientific with 14 decimal places.
You might notice infinitesimal differences between the trendline
coefficients and the LINEST results. But in this case, they are same up
10 or 11 decimal places in this case; "close enough for government
work". As a guess, the difference might be attributed to physically
different internal algorithms and/or to differences in internal
floating-point arithmetic anomalies.

The difference is that (apart from in some very early versions of XL2007
where the graph regression function was wrecked to make it agree with
another well known products answers) the linear regression on XL charts
uses a well designed algorithm that gives a very good approximation to
the true least squares parameter fit to data. LINEST is demonstrably
numerically unstable although it requires at least a 3rd order
polynomial fit and moderately hostile data to break it. Examples have
been posted here or in another m.p.e group previously.
(some years back)

For linear fits the two are more or less interchangable but for higher
orders of polynomial the chart function fit is more accurate.

What is COLA?

The newsgroup where the troll Snit and his sock puppet Onion Knight
are lying and trolling with this topic.

What names have you used on Usenet, Steve?

How many?

Can you even list them all? No. You can't. You think everyone is
like you and is using socks.

As far as trolling that is you. Snit noted he created a linear
trendline correctly using a process exactly as Microsoft describes. He
proved this with a video where he showed his process and it was
exactly correct. Your sock puppet CC claimed Snit missed steps and you
agreed wit your sock puppet. What a surprise.

While I have not been on usenet for a long time I used to post to the
mac advocacy group years ago under a different name. Even then you
were using several sock puppets to troll Snit because you blamed him
for your girlfriend dumping you. That was about 10 years ago and you
have never stopped attacking him and lying about him.

When I came in to say Snit was right about his method of making a
linear trendline in Excel you attacked me both in usenet and by
emailing me repeatedly until I forwarded your emails to your ISP and
then you stopped.

On the original debate you can see the answer above by joeu2004, 'I
see no missing steps in creating the linear trendline per se using
Excel.' You and your sock puppet were wrong. Snit did not skip steps
as you claimed. That was all this thread was about and you turned it
into another of your avenues of attack and trolling because you could
not handle the fact you were proved wrong in both your Steve name and
your sock CC name.

In short: go **** yourself you trolling asshole.

 

[Steve Carroll]

On Jun 17, 1:57 am, Martin Brown <|||[email protected]>
wrote:

On 16/06/2012 00:22, joeu2004 wrote:
There has been some debate in COLA as to the correct way
to create a linear trendline in Excel. Someone in the forum
kindly made a video to show what he thought was the correct
method
http://tmp.gallopinginsanity.com/LinearTrendLineCreation.mov
[....]
Does anyone see any missing steps for the creation of a
linear trend line?
I cannot find the original debate in comp.os.linux.advocacy (COLA);just
the thread titled "Visualizing where to draw the standard deviation
line", which refers to yet-another "a debate in COLA".
No matter; I'm not interested. I think we can answer your Excel
questions without getting into anything so esoteric.
I see no missing steps in creating the linear trendline per se using Excel.
I would also select the trendline options to display the trendline
equation and the R-squared of the trendline.
And after displaying the trendline equation, I might select the option
to Format Trendline Label in order to alter the display precision of the
coefficients. I would select Scientific with 14 decimal places if Iwant
to see the trendline coefficients "exactly"; that is, the most precision
that Excel will display.
But none of those "steps" affects how the linear trendline is created.
FYI, we can also use the Excel function LINEST to determine the
regression line coefficients.
If the dates are in A1:A24 and the percentages are in B1:B24, select two
horizontal cells and array-enter the following formula (press
ctrl+shift+Enter instead just Enter): =LINEST(B1:B24,A1:A24).
To compare with the "exact" trendline coefficients, I would also format
the LINEST results as Scientific with 14 decimal places.
You might notice infinitesimal differences between the trendline
coefficients and the LINEST results. But in this case, they are same up
10 or 11 decimal places in this case; "close enough for government
work". As a guess, the difference might be attributed to physically
different internal algorithms and/or to differences in internal
floating-point arithmetic anomalies.
The difference is that (apart from in some very early versions of XL2007
where the graph regression function was wrecked to make it agree with
another well known products answers) the linear regression on XL charts
uses a well designed algorithm that gives a very good approximation to
the true least squares parameter fit to data. LINEST is demonstrably
numerically unstable although it requires at least a 3rd order
polynomial fit and moderately hostile data to break it. Examples have
been posted here or in another m.p.e group previously.
(some years back)
For linear fits the two are more or less interchangable but for higher
orders of polynomial the chart function fit is more accurate.
For some insight into how the Excel LINEST and linear trendline
coefficients might be derived, see
http://en.wikipedia.org/wiki/Simple_linear_regression.
Note that =COVAR(A1:A24,B1:B24)/VARP(A1:A24) in E3 and
=AVERAGE(B1:B24)-E3*AVERAGE(A1:A24) are about the same as the LINEST
results. The difference is in the 3 least-significant binary bits of the
floating-point representations in this case.
You might also take note of the wiki page section titled "Normality
assumption". That may or may not be relevant to the COLA debates.
What is COLA?

The newsgroup where the troll Snit and his sock puppet Onion Knight
are lying and trolling with this topic.

What names have you used on Usenet, Steve?

See what I mean about your trolling?

Hint: People just aren't as stupid as you need them to be.

https://groups.google.com/group/comp.os.linux.advocacy/msg/6e8fd0b61d327d78

(snip more lies/trolling)

 

[Snit]

On 6/17/12 10:23 AM, in article
(e-mail address removed), "Onion

....

What names have you used on Usenet, Steve?

Steve will *never* answer this question. He accuses you and many, many
others of being my socks (though he has no evidence for his accusations).
At the same time, when asked about his socks he runs.

Again: I am not like Carroll and do not use armies of socks.

How many?

Carroll has clearly lost track.

Can you even list them all? No. You can't. You think everyone is
like you and is using socks.

Exactly correct. But for the sake of this cross group thread, the *only*
question that is relevant is not who did what or who is good or bad or
whatever. The question is about if my process for creating linear trend
lines in Excel (and Numbers) is correct. The process. That is it... was
the process correct as I said or was it "missing steps" as they claimed?
Here is the process:

<http://tmp.gallopinginsanity.com/LinearTrendLineCreation.mov>

And using the same process, here are some of the linear trend lines that
were created:

For the latter half of 2011:
<http://tmp.gallopinginsanity.com/LinuxTrend2011-2ndhalf.png>

This supported my view that Linux usage would be increasing. I never
predicted that the trend would necessarily continue - though I did believe
the usage would continue to move up at *some* level or stay at least roughly
consistent (it would not have a strong drop).

But here is the linear trend line for the first part of 2012:
<http://tmp.gallopinginsanity.com/LinuxTrendLine2012.jpg>

An this went contrary to my prediction... but the trend line that went
contrary to my predictions (showing my prediction to be wrong) was generated
just fine - there are no "missing steps" in the process.

Back to the debate about the process - it was also claimed (by cc, with
Carroll backing his trolling), that the way he could tell I was "missing
steps" was the low R^2 value. But the R^2 value does not show if the
process is missing steps or not... notice with the latter half of 2011 my
R^2 value is very high at over 0.98% ... but using the same process I made a
number of graphs and the R^2 values are sometimes high and sometimes low:
<http://tmp.gallopinginsanity.com/LinuxMultLinearTrendLines.png>.

I did note that a linear trend line is not the best tool to analyze
non-linear data - and, it appears, the data is not linear... there was a
significant change at the start of 2012. So nobody is claiming that the
linear trend lines Excel and Numbers make are always the best for of
evaluation, and *certainly* nobody said that the trend lines I was creating
would be of great value for predicting the future... they were being used to
see if the predictions I had already made were proving to be true. For 2011
they did... for 2012 they did not.

This, of course, does not prove that the reasoning behind my predictions was
true or not - even if the data had fit my predictions perfectly (which, as
noted, it did not). It was never claimed that such tools prove any believe
in causation... but cc and Carroll repeatedly confused the concepts of the
lines *correlating* (or not) with the predictions and being a tool to prove
*causation*. They proved, repeatedly, they could not understand the
difference in these concepts.

While I have not been on usenet for a long time I used to post to the
mac advocacy group years ago under a different name. Even then you
were using several sock puppets to troll Snit because you blamed him
for your girlfriend dumping you. That was about 10 years ago and you
have never stopped attacking him and lying about him.

Correct though irrelevant to the thread (though I see why you were speaking
of this - to counter Steve's lies and help others understand why he is lying
and lashing out).

....

 

[Steve Carroll]

There has been some debate in COLA as to the correct way
to create a linear trendline in Excel. Someone in the forum
kindly made a video to show what he thought was the correct
method
http://tmp.gallopinginsanity.com/LinearTrendLineCreation.mov
[....]
Does anyone see any missing steps for the creation of a
linear trend line?
I cannot find the original debate in comp.os.linux.advocacy (COLA); just the
thread titled "Visualizing where to draw the standard deviation line", which
refers to yet-another "a debate in COLA".
No matter; I'm not interested.  I think we can answer your Excel questions
without getting into anything so esoteric.

Really the video is all you need to see (I am the one who created it and the
other related images with trend lines)... it shows the steps and they match
what MS suggests exactly (with the exception that I use the right-click
method to get to some tools where the MS page suggests using the Ribbon).

Thank you.  I think anyone who knows Excel well... or even not so well... or
even poorly but can read the steps provided by MS... can see this.  But
someone in COLA kept saying I was missing steps:

  cc:
    ------
    I will gladly educate your ignorant ass, and give you step by
    step instructions on how to correctly do it so you can see
    all your missing steps, but first you have to stop lying.
    ------
    So you are missing some key steps before doing the trendline.
    I have said that many times now.
    ------

He later said he was in reference to doing more than just creating the trend
line, which is all I said I was doing - he claimed I was missing steps to do
part of a process I never claimed to do.  Yes: his claims were idiotic.
This is the nature of the "advocacy" groups.

Bottom line: my steps for creating a linear trend line were flawless, which
is not to say creating such a line was the only form of analysis that could
be done on such data or that the creation of a linear trend line on data
that is clearly not linear is the best way to do an analysis.

Using the same steps, though, I was able to show a trend line with a very
good fit for the second half of 2011:
<http://tmp.gallopinginsanity.com/LinuxTrend2011-2ndhalf.png>

And also able to show how the trend changed for the first half of 2012:
<http://tmp.gallopinginsanity.com/LinuxTrendLine2012.jpg>

The fact is, a linear trend line is not going to show a very good fit with
non-linear data... or data where the trend changes.

You can see where I did that in the above links... and in others that I
made... a whole bunch of them combined:
<http://tmp.gallopinginsanity.com/LinuxMultLinearTrendLines.png>

Even then my labeling is not done well... but my only claim was that I made
the linear trend line correctly, not that the graph had good labeling, etc.
I openly acknowledge it did not.

Seems a bit excessive for the needs here - but I did know about those
options.

Correct.  Nor do they remove the fact there are other forms of analysis that
can be done.

I have used this function a couple of times... but never for any real work.

Not really... though the idea of a normal distribution was also debated..  A
bit of info you might or might not know: you can easily visualize wherethe
standard deviation (sigma) lines should be drawn on a normal distribution...
the lines are *always* at the distance from the mean to the inflection point
on the curve (where the concavity changes).  I was repeatedly told bythe
same person the distance was irrelevant:

  cc:
    -----
    There'se nothing wrong with the image, other than some weird
    axis labeling.
    -----
    Snit's so fucking stupid he thinks the sigma lines are drawn
    based on distance from the mean, not area under the curve.
    -----
    | The sigma lines are drawn based on the area of the curve -
    | which is easy to see when the images screw it up, esp. when
    | they do so really badly, like in some of the ones I showed
    | you.
    They are not wrong.
    ------
    LOL!!!! All of those links are fine. The first sigma lines
    cover 68% of the area UNDER THE CURVE.
    -----
    If you would like to prove, on any single one of the links
    you call incorrect, that the first sigma lines do not bound
    an area that is 68.2% of the area UNDER THE CURVE, then I
    would like to see it.
    -----
    Hahahaha your "approximate inflection points" are hilarious.
    Please, post more on this subject.
    ------

I explained to him why he was wrong here: <

>

There are many examples of people who should know better getting this wrong.
Here are some

<http://www.udel.edu/htr/Statistics/Images/Class12/normal2.gif> From:
<http://www.udel.edu/htr/Statistics/Notes/class12.html>

Which is the example I used for showing how him can make a decent
approximation visually: <http://tmp.gallopinginsanity.com/sd.png>.

I also pointed to some other examples which at least appear incorrect to me
(though they are not as far off as the above example):

<http://www.footballguys.com/shickstandard_1_files/image009.gif> From:
<http://www.footballguys.com/shickstandard_1.htm>
Sigma lines clearly not at a far enough distance from the mean, esp. onthe
graph to the right.

<http://www.gsseser.com/images/StandardDeviation2s.gif> From:
<http://www.gsseser.com/Deviation.htm>
Sigma lines clearly not at a far enough distance from the mean.

You would think that such sites would be made by people who knew better..  I
openly admit I am not a math wiz but it is rather silly when sites claiming
to be teaching such things get their depictions wrong (of course, one of the
sites above is from "Footballguys"... and you might not expect them, by
stereotype, to be the most knowledgeable in such areas anyway).  

Anyway, sorry to have the idiotic debate spread to other forums... though it
is sorta fun to see people who clearly know what they are talking aboutrip
apart those who were calling me names and insisting they "knew" I was wrong.

--
The indisputable facts about that absurd debate: <http://goo.gl/2337P>
cc being proved wrong about his stats BS: <http://goo.gl/1aYrP>
7 simple questions cc will *never* answer: <http://goo.gl/cNBzu>
cc again pretends to be knowledgeable about things he is clueless about..

Funny how while Steve is throwing a toddler tantrum

The tantrum is all Snit's... here is what he was originally arguing
before he realized his stepped on his d*ck:

https://groups.google.com/group/comp.os.linux.advocacy/msg/edd65e9dcec15bd2

 

[Onion Knight]

On 6/15/12 4:22 PM, in article [email protected], "joeu2004"
There has been some debate in COLA as to the correct way
to create a linear trendline in Excel. Someone in the forum
kindly made a video to show what he thought was the correct
method
http://tmp.gallopinginsanity.com/LinearTrendLineCreation.mov
[....]
Does anyone see any missing steps for the creation of a
linear trend line?
I cannot find the original debate in comp.os.linux.advocacy (COLA);just the
thread titled "Visualizing where to draw the standard deviation line", which
refers to yet-another "a debate in COLA".
No matter; I'm not interested.  I think we can answer your Excel questions
without getting into anything so esoteric.
Really the video is all you need to see (I am the one who created it and the
other related images with trend lines)... it shows the steps and theymatch
what MS suggests exactly (with the exception that I use the right-click
method to get to some tools where the MS page suggests using the Ribbon).
I see no missing steps in creating the linear trendline per se using Excel.
Thank you.  I think anyone who knows Excel well... or even not so well... or
even poorly but can read the steps provided by MS... can see this.  But
someone in COLA kept saying I was missing steps:
  cc:
    ------
    I will gladly educate your ignorant ass, and give you step by
    step instructions on how to correctly do it so you can see
    all your missing steps, but first you have to stop lying.
    ------
    So you are missing some key steps before doing the trendline.
    I have said that many times now.
    ------
He later said he was in reference to doing more than just creating the trend
line, which is all I said I was doing - he claimed I was missing steps to do
part of a process I never claimed to do.  Yes: his claims were idiotic.
This is the nature of the "advocacy" groups.
Bottom line: my steps for creating a linear trend line were flawless,which
is not to say creating such a line was the only form of analysis thatcould
be done on such data or that the creation of a linear trend line on data
that is clearly not linear is the best way to do an analysis.
Using the same steps, though, I was able to show a trend line with a very
good fit for the second half of 2011:
<http://tmp.gallopinginsanity.com/LinuxTrend2011-2ndhalf.png>
And also able to show how the trend changed for the first half of 2012:
<http://tmp.gallopinginsanity.com/LinuxTrendLine2012.jpg>
The fact is, a linear trend line is not going to show a very good fitwith
non-linear data... or data where the trend changes.
I would also select the trendline options to display the trendline equation
and the R-squared of the trendline.
You can see where I did that in the above links... and in others thatI
made... a whole bunch of them combined:
<http://tmp.gallopinginsanity.com/LinuxMultLinearTrendLines.png>
Even then my labeling is not done well... but my only claim was that I made
the linear trend line correctly, not that the graph had good labeling, etc.
I openly acknowledge it did not.
And after displaying the trendline equation, I might select the option to
Format Trendline Label in order to alter the display precision of the
coefficients.  I would select Scientific with 14 decimal places if I want to
see the trendline coefficients "exactly"; that is, the most precision that
Excel will display.
Seems a bit excessive for the needs here - but I did know about those
options.
But none of those "steps" affects how the linear trendline is created.
Correct.  Nor do they remove the fact there are other forms of analysis that
can be done.
FYI, we can also use the Excel function LINEST to determine the regression
line coefficients.
If the dates are in A1:A24 and the percentages are in B1:B24, select two
horizontal cells and array-enter the following formula (press
ctrl+shift+Enter instead just Enter):  =LINEST(B1:B24,A1:A24).
I have used this function a couple of times... but never for any realwork.

To compare with the "exact" trendline coefficients, I would also format the
LINEST results as Scientific with 14 decimal places.
You might notice infinitesimal differences between the trendline
coefficients and the LINEST results.  But in this case, they are same up 10
or 11 decimal places in this case; "close enough for government work".  As a
guess, the difference might be attributed to physically different internal
algorithms and/or to differences in internal floating-point arithmetic
anomalies.
For some insight into how the Excel LINEST and linear trendline coefficients
might be derived, seehttp://en.wikipedia.org/wiki/Simple_linear_regression.
Note that =COVAR(A1:A24,B1:B24)/VARP(A1:A24) in E3 and
=AVERAGE(B1:B24)-E3*AVERAGE(A1:A24) are about the same as the LINEST
results.  The difference is in the 3 least-significant binary bits of the
floating-point representations in this case.
You might also take note of the wiki page section titled "Normality
assumption".  That may or may not be relevant to the COLA debates..
Not really... though the idea of a normal distribution was also debated.  A
bit of info you might or might not know: you can easily visualize where the
standard deviation (sigma) lines should be drawn on a normal distribution...
the lines are *always* at the distance from the mean to the inflection point
on the curve (where the concavity changes).  I was repeatedly told by the
same person the distance was irrelevant:
  cc:
    -----
    There'se nothing wrong with the image, other than some weird
    axis labeling.
    -----
   Snit'sso fucking stupid he thinks the sigma lines are drawn
    based on distance from the mean, not area under the curve.
    -----
    | The sigma lines are drawn based on the area of the curve -
    | which is easy to see when the images screw it up, esp. when
    | they do so really badly, like in some of the ones I showed
    | you.
    They are not wrong.
    ------
    LOL!!!! All of those links are fine. The first sigma lines
    cover 68% of the area UNDER THE CURVE.
    -----
    If you would like to prove, on any single one of the links
    you call incorrect, that the first sigma lines do not bound
    an area that is 68.2% of the area UNDER THE CURVE, then I
    would like to see it.
    -----
    Hahahaha your "approximate inflection points" are hilarious.
    Please, post more on this subject.
    ------
I explained to him why he was wrong here: <

>
There are many examples of people who should know better getting thiswrong.
Here are some
<http://www.udel.edu/htr/Statistics/Images/Class12/normal2.gif> From:
<http://www.udel.edu/htr/Statistics/Notes/class12.html>
Which is the example I used for showing how him can make a decent
approximation visually: <http://tmp.gallopinginsanity.com/sd.png>.
I also pointed to some other examples which at least appear incorrectto me
(though they are not as far off as the above example):
<http://www.footballguys.com/shickstandard_1_files/image009.gif> From:
<http://www.footballguys.com/shickstandard_1.htm>
Sigma lines clearly not at a far enough distance from the mean, esp. on the
graph to the right.
<http://www.gsseser.com/images/StandardDeviation2s.gif> From:
<http://www.gsseser.com/Deviation.htm>
Sigma lines clearly not at a far enough distance from the mean.
You would think that such sites would be made by people who knew better.  I
openly admit I am not a math wiz but it is rather silly when sites claiming
to be teaching such things get their depictions wrong (of course, oneof the
sites above is from "Footballguys"... and you might not expect them, by
stereotype, to be the most knowledgeable in such areas anyway).  
Anyway, sorry to have the idiotic debate spread to other forums... though it
is sorta fun to see people who clearly know what they are talking about rip
apart those who were calling me names and insisting they "knew" I waswrong.

--
The indisputable facts about that absurd debate: <http://goo.gl/2337P>
cc being proved wrong about his stats BS: <http://goo.gl/1aYrP>
7 simple questions cc will *never* answer: <http://goo.gl/cNBzu>
cc again pretends to be knowledgeable about things he is clueless about.

Funny how while Steve is throwing a toddler tantrum

The tantrum is allSnit's... here is what he was originally arguing
before he realized his stepped on his d*ck:

https://groups.google.com/group/comp.os.linux.advocacy/msg/edd65e9dce...

Funny how you keep trolling Snit but can't show where he was wrong in
any place he did not point out.