Snit
Posts:
25
Registered:
6/15/12
|
|
Re: Visualizing where to draw the standard deviation line
Posted:
Jun 15, 2012 3:52 PM
|
|
On 6/15/12 12:38 PM, in article
f76505dc-792a-405c-96a4-9b52f6bc183b@n33g2000vbi.googlegroups.com, "Onion
Knight" <onionknightgot@gmail.com> wrote:
> On Jun 15, 7:20 pm, Steve Carroll <fretwiz...@gmail.com> wrote:
>> On Jun 15, 10:52 am, Snit <use...@gallopinginsanity.com> wrote:
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>> On 6/15/12 9:23 AM, in article jrfnh3$17...@news.cc.tut.fi, "Kaba"
>>
>>> <k...@nowhere.com> wrote:
>>>> 15.6.2012 19:01, Onion Knight kirjoitti:
>>>>> Do you know much about linear trendlines? The same debate included the
>>>>> drawing of those in Excel. The same person who spoke of the standard
>>>>> deviation also showed how to make a linear trendline,
>>>>> http://tmp.gallopinginsanity.com/LinearTrendLineCreation.mov
>>
>>>>> Others in the same group insisted he was missing steps but if you look
>>>>> at the Microsoft site it offers instructions and it seems his process
>>>>> is just fine. What steps if any did he skip? Would love to get some
>>>>> other input from people who are not involved.
>>
>>>> I gather there are two questions here:
>>
>>>> 1) Is the trendline approriately fitted to that data?
>>
>>>> 2) Is the trendline useful in some way?
>>
>>> Not quite:
>>
>> Not quite?! The obvious question: Why bother to produce a trend line
>> at all, then?
>>
>>> 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.
>>
>>> I am well aware that there are other forms of analysis... and also aware
>>> that their is a great deal of volatility in the stats, in part because we
>>> are talking such a small percentage of a whole and there is bound to be
>>> plenty of "noise" that enters the data, but also because the usage itself
>>> seems to go up and down quite a bit... though mostly staying between 1% and
>>> 2% of desktop usage.
>>
>>> The question was more about *had* there been an increase and less about if
>>> the increase will continue.
>>
>> No, the question was about whether or not the increase could be
>> attributed to UI improvements (as you admit to below).
>>
>>> I did not expect such an increase to
>>> continue... so the fact that it did not was not contrary to my views...
>>
>> Your "view" here was based on something that had no way to be
>> supported using the given data. As for this being contrary to your
>> view, it most certainly was contrary. Your prediction, which you said
>> was supported by your allegation of a correlation (that doesn't
>> provably exist), was that UI improvements lead to an increase in
>> marketshare. That any increase went back down only serves to teach you
>> not to make absolute statements like this (a thing you have a penchant
>> for doing).
>>
>>> though the drop in 2012 was something I did not predict and did go against
>>> my predictions. But at no time did I say the upward trend of the latter
>>> half of 2011 (or 20111 as a whole) was likely to continue.
>>
>> But it proves cc's point, there really was no increase to speak of and
>> all this talk of your "predictions" is BS.
>>
>>> As far as the questions asked in the video, *clearly* the trend line, as
>>> created with the built in linear trend line function of Excel, was going up
>>> - but nobody said this meant that you could follow such a line to make good
>>> future predictions. Just looking at the last couple of months of data
>>> showed that trying to use the trend line for that purpose was not likely to
>>> be successful. :)
>>
>> As cc pointed out to you:
>>
>> "If there was a correlation between focusing on usability issues and
>> an increase in Linux usage, then it would a clear roadmap on how to
>> get more users to Linux." - cc
>>
>> And let's recall some of the unsubstantiated garbage that you wrote...
>> in a thread entitled "Disturbing Anomalies in Desktop Linux
>> Statistics" you stated:
>>
>> "Hopefully as Ubuntu and others continue to focus on usability which
>> will lead to this trend continuing, but let us not pretend that there
>> is huge news now. There is, however, a good sign - and yet another
>> example of where the herd has been clueless as they disagree with me."
>> - Snit
>>
>> Notably, you're talking about 'clueless herds' of people who
>> rightfully pointed out how your fat-headed BS *was* just BS.
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>>> Answering question 1, there are many ways to a line to data. These go
>>>> under the umbrella term of linear regression:
>>
>>>> http://en.wikipedia.org/wiki/Linear_regression
>>
>>>> It is possible to choose it differently how to measure the goodness of a
>>>> fit. No doubt in the video the lines are fitted using ordinary least
>>>> squares. Least absolute deviations, for example, would give something
>>>> similar.
>>
>>>> Answering question 2, no. The trendline is attempting to build a model
>>>> for the data. Models are created to summarize and to predict the future
>>>> (think of physics). In this case the line has no summarization value,
>>>> since _all_ of the data can be seen at glance. Since there is so small
>>>> amount of data, and so much fluctuation, the line probably does not have
>>>> any prediction value either.
>>
>>> Part of the challenge is that the trend changed from the first half of 2011
>>> to the second half and then again at the start of 2012. I show this here:
>>
>>> 2011: <http://tmp.gallopinginsanity.com/LinuxTrendLine2011.png>
>>> 2012: <http://tmp.gallopinginsanity.com/LinuxTrendLine2012.jpg>
>>
>>> And all sorts of other breakdowns here:
>>> <http://tmp.gallopinginsanity.com/LinuxMultLinearTrendLines.png>
>>
>>> 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.
>>
>> Yup... and as is the same for predictive purposes... there is no way
>> the data you used was ever going to show a correlation between
>> improved UI and marketshare (which, as you finally admit here below,
>> is the reason you produced the trend line. This story now has an
>> ending).
>>
>>> For 2011, esp. the latter half, this was shown to be true - based on the
>>> data we have. While the data set is small (only six months / data points),
>>> the trend fit very well there:
>>> <http://tmp.gallopinginsanity.com/LinuxTrend2011-2ndhalf.png>. So it at
>>> least seemed as though the real world data was fitting with my prediction
>>
>> No, it didn't seem that way, you only mistakenly believed it. That
>> data could have been based on *any* number of reasons, none of which
>> necessarily have to do with UI improvements. That's a fact and, until
>> you look a *much* better data you're flying blind.
>>
>>> (which, to be clear, does not prove cause and effect).
>>
>> Exactly right. So how do you claim above that it 'seemed' otherwise?
>> You're obviously very confused.
>>
>>> In 2012, as shown in the above link, the trend reversed itself and my
>>> prediction of an increase of users has been shown to *not* be the case, at
>>> least based on the data we have.
>>
>>> But the graph was initially made to show that the then-current data fit with
>>> my past predictions
>>
>> Which you just talked about above:
>>
>> "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." - Snit
>>
>> And you created your trend line to show evidence of your prediction
>> "that an increase in usability would lead to an increase in users."
>>
>> And you were unable to show this correlation with the data that you
>> used (as expected).
>>
>> (note to anyone reading: If all this isn't proof that Snit is a TOTAL
>> whack job... nothing is)
>>
>>> - not to make any specific prediction about the future
>>> (such as that the same level of upward trend would be seen). The fact that
>>> the trend reversed itself and usage went *down* and did not at least stay
>>> stable,
>>
>> Fact: There *was* no "trend" that was unequivocally attributable to
>> your alleged "correlation".
>>
>>> however, is contrary to my claims that there would be an increase in
>>> users (I think it is fair to say there was an unstated assumption that the
>>> increase would remain - not increasing forever but the number of users would
>>> not drop).
>>
>> To be more fair, as you *finally* admit to here up above, you created
>> your media to show a correlation that you were never going to be able
>> to show using the data you used.
>>
>>> All of this is confounded by the stats being based off of only one source
>>> (there are many that measure such things and not all agree) and that this is
>>> based on web usage
>>
>> Not really as that has little to do with why your correlation stood no
>> chance of being supported. It's the limitation of the data as based
>> solely on percentage rise and fall over time (regardless of where you
>> obtained, unless it was a UI 'study').
>>
>>> - which while it serves as a reasonable approximate of
>>> usage (and is likely the best we have) it is not the same. Also, the
>>> percentages include both desktop and non-desktop usage, so to try to be more
>>> accurate in terms of desktop usage, things such as Android and iOS would
>>> have to be removed from the stats and the percentages would have to be
>>> adjusted accordingly (though at this time the difference is so small as to
>>> likely be statistically insignificant... I suspect that will change with
>>> time).
>>
>> No, limiting the data to rising and falling percentage over time won't
>> ever produce the kind of correlation you've suggested, it's simply the
>> wrong kind of data for that.
>
> My apologies for any action of mine that lead to Steve posting such
> trash to the non Linux groups. Also sorry if you just got a similar
> post as it seems Google Groups ate my first one. Maybe it is just
> taking time?
I did not see an earlier post so I think it was just lost. Odd.
In any case, Carroll will follow me around from forum to forum - not your
fault. In any case, above and in his other post Carroll shows he does not
understand:
* I had predicted that the improvements in Linux usability would lead to an
increase in users. This was supported by the upward trend at the latter
half of 2011: <http://tmp.gallopinginsanity.com/LinuxTrend2011-2ndhalf.png>.
The trend for 2012, though, has changed and the usage is now dropping. Had
it even stayed the same it would have fit my rather vague predictions. I
never predicted the trend could continue to move upwards like that.
* Steve shows he does not understand the difference between causation and a
correlation. I noted the data supported my prediction; I never said it
proved my reasoning was correct. It correlating well with my vague
prediction does not mean that this proves causation, and I have been very
clear with this. I also have been very clear that the 2012 data has gone
contrary to my prediction.
* When Carroll says there was no increase at all he is wrong. For the
latter half of 2011 there clearly was... and this whole silliness began
shortly after that time. Again, there is no disagreement it has since
dropped in 2012.
In short: I made a rather vague prediction and the data from the latter half
of 2011 supported the prediction well (correlated with it - it did not show
anything about the cause and effect I talked about... nor did I ever say it
did). Had the usage stayed the same or continued to go up then my initial
prediction would have been shown to be correct (though, again, the cause and
effect relationship would not be shown by the data). A continuation of the
upward trend, however, was not required for my prediction to have been shown
to be correct. My prediction was shown to be incorrect, however, with the
drop of usage. So I was wrong to make the prediction that it would roughly
stay at that higher level or go up... but the fact I was wrong is not
sufficient for Carroll - he feels the need to lie about me and my views.
And that is the last I will say about him and his BS in this thread unless a
non-COLA person asks or looks to be taking anything my stalker (Carroll)
says with any level of credibility. He has been following me around from
forum to forum trolling me with his own name and socks since 2004 for
reasons which are completely irrelevant to any of this.
--
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.
|
|
