Statisticians_group : Messages : 3274-3303 of 4110

try searching
www.gigapedia.org

On Thu, Feb 26, 2009 at 3:25 PM, Neethu Thomas <gtg634t@...> wrote:

1. Brockwell, P.J. and Davis, R.A. (1991), Introduction to Time Series
and Forecasting, 2nd
edition, Springer-Verlag, New York.
2. Tsay, R.S. (2005), Analysis of Financial Time Series, 2nd Edition,
Wiley Series in Probability
and Statistics.
Other recommended books:
1. . Fan, J., Yao, Q. (2003). Nonlinear Time Series. Springer-Verlag,
New York.
2. Box, G.E.P., Jenkins, G.M. and Reinsel, G.C. (1994). Time Series
Analysis: Forecasting
and Control, 3rd Edition, Prentice Hall, New Jersey.
3. Chat¯eld, C. (1996). The Analysis of Time Series, 5th edition,
Chapman and Hall, New
York.
4. Shumway, R.H., Sto®er, D.S. (2006), Time Series Analysis and Its
Applications (with R
examples), Springer-Verlag, New York.

#3302

From: Andrew Hartley <khahstats@...>
Date:: Sun Mar 1, 2009 1:05 am
Subject:: Re: Re: Non parametric ANCOVA

khahstats

Offline

Claire,
with regard to testing for goodness of fit (to, e.g., normality): Remember that no variable (whether on the original scale or a transformed one) is of exactly a normal (or of any other parametric) distribution. All our distributions are approximations which we hope will be close to the truth. So, testing for goodness of fit doesn't help you decide whether a variable is being generated according to a particular process; it only tells you whether your sample size is large enough to detect the deviations from that process that we (should) know are there all along.

--- On Sat, 2/28/09, lexie75@... <lexie75@...> wrote:

From: lexie75@... <lexie75@...>
Subject: [Statisticians_group] Re: Non parametric ANCOVA
To: Statisticians_group@...
Date: Saturday, February 28, 2009, 1:06 PM

Hello
Thank you for the reply, it's appreciated. I have already transformed the variables using
square root, log10 and inverse but each of these transformed variables violate the levene
and box tests. I want to test for differences in gender between 6 different delinquency
variables. Because of the nature of the variables, it was not surprising they would be
positively skewed, but I can't get them acceptable for parametric tests. So ideally I would
have run a MANCOVA if my variables had transformed the way I wanted them to. So I feel
bit stuck as to what to do. All I could think of was to run a set of non parametric
ANCOVAs and adjust them using a bonferroni correction. I have tried work out if it is possible to add a covariate to the Kruskal-Wallis Test but I can;t work out how.
Claire

--- In Statisticians_ group@yahoogroup s.co.in, Andrew Hartley <khahstats@. ..> wrote:
>
> Clair,
> yes that's the basic approach. However, why would you want to do it? Nonparametric
ANCOVA takes away your ability to estimate mean differences in the outcome (dependent
variable) between levels of class variables (& a whole lot more). If you have a
customer/client & they want to test whether 2 levels of a class variable are associated with
different mean responses, well, you already know they are associated with different
means, so what's the point?
>
> Even if such testing was inherently meaningful, the ranks don't come even close to being
symmetrically distributed, as long as there's much overlap between the responses
associated with the levels of the class variable. If the responses for Level A of a class
variable are shifted to the right, say, of those of Level B, then the responses for Level A will
have a long tail to the left, & those for Level B will have such a tail to the right.
>
> I usually try other transformations (percent change, logarithmic base 2, box-cox,
anything) rather that a transformation to ranks.
>
> --- On Sat, 2/28/09, lexie75@... <lexie75@... > wrote:
> From: lexie75@... <lexie75@... >
> Subject: [Statisticians_ group] Non parametric ANCOVA
> To: Statisticians_ group@yahoogroup s.co.in
> Date: Saturday, February 28, 2009, 2:49 AM
>
>
>
>
>
>
>
>
>
>
>
>
> Hi everyone
>
> I'm tearing my hair out trying to find out how to run a Non parametric ANCOVA in SPSS.
can
>
> I get people's advice? Can I rank the data and then run the ANCOVA as done in SPSS? I
>
> think ti might be more complicated than that but if anyone knows could you direct me
to
>
> how I can get the SPSS syntax or how to calculate this test? I can find information that
non
>
> para ANCOVAs exist but none of the papers say how to actually do it beyond using
ranks.
>
> Any ideas?
>
> Claire
>

#3301

From: "lexie75@..." <lexie75@...>
Date:: Sat Feb 28, 2009 6:06 pm
Subject:: Re: Non parametric ANCOVA

lexie75...

Offline

Hello
Thank you for the reply, it's appreciated.  I have already transformed the
variables using
square root, log10 and inverse but each of these transformed variables violate
the levene
and box tests.  I want to test for differences in gender between 6 different
delinquency
variables.  Because of the nature of the variables, it was not surprising they
would be
positively skewed, but I can't get them acceptable for parametric tests.  So
ideally I would
have run a MANCOVA if my variables had transformed the way I wanted them to.  So
I feel
bit stuck as to what to do.  All I could think of was to run a set of non
parametric
ANCOVAs and adjust them using a bonferroni correction.  I have tried work out if
it is possible to add a covariate to the Kruskal-Wallis Test but I can;t work
out how.
Claire

--- In Statisticians_group@..., Andrew Hartley <khahstats@...>
wrote:
>
> Clair,
> yes that's the basic approach. However, why would you want to do it?
Nonparametric
ANCOVA takes away your ability to estimate mean differences in the outcome
(dependent
variable) between levels of class variables (& a whole lot more). If you have a
customer/client & they want to test whether 2 levels of a class variable are
associated with
different mean responses, well, you already know they are associated with
different
means, so what's the point?
>
> Even if such testing was inherently meaningful, the ranks don't come even
close to being
symmetrically distributed, as long as there's much overlap between the responses
associated with the levels of the class variable. If the responses for Level A
of a class
variable are shifted to the right, say, of those of Level B, then the responses
for Level A will
have a long tail to the left, & those for Level B will have such a tail to the
right.
>
> I usually try other transformations (percent change, logarithmic base 2,
box-cox,
anything) rather that a transformation to ranks.
>
> --- On Sat, 2/28/09, lexie75@... <lexie75@...> wrote:
> From: lexie75@... <lexie75@...>
> Subject: [Statisticians_group] Non parametric ANCOVA
> To: Statisticians_group@...
> Date: Saturday, February 28, 2009, 2:49 AM
>
>
>
>
>
>
>
>
>
>
>
>
>             Hi everyone
>
> I'm tearing my hair out trying to find out how to run a Non parametric ANCOVA
in SPSS.
can
>
> I get people's advice?  Can I rank the data and then run the ANCOVA as done in
SPSS?  I
>
> think ti might be more complicated than that but if anyone knows could you
direct me
to
>
> how I can get the SPSS syntax or how to calculate this test?  I can find
information that
non
>
> para ANCOVAs exist but none of the papers say how to actually do it beyond
using
ranks.
>
> Any ideas?
>
> Claire
>

#3300

From: Nishant Jain <nishantjain18@...>
Date:: Sat Feb 28, 2009 5:34 pm
Subject:: Re: Customer Satisfaction Index

nishantjain18

Offline

Nupur -

There could be different formulae for calculating CSI, depending upon the type of study.

However, if you want to study the basic concept, I am sure this article can do - http://en.wikipedia.org/wiki/American_Customer_Satisfaction_Index#Methodology

�

I am sure the questions you asked would be different from the ones indicated here, also you need to decide upon the weights to be assigned.

�

I remember a study in which all the questions were given equal weightage, in that case, the number of respondents answering that question (valid responses, excluding DK/CS) was taken into consideration.

I can give you the entire methodology used.... but I would not prefer to... for obvious resons :-)

�

Nishant

�

On Sat, Feb 28, 2009 at 2:22 PM, nupur choudhury <me_nupur79@...> wrote:

Hello,

Can anyone help me to calculate Customer Satisfaction Index(CSI)

Add more friends to your messenger and enjoy! Invite them now.

--
Nishant J

#3299

From: Andrew Hartley <khahstats@...>
Date:: Sat Feb 28, 2009 12:06 pm
Subject:: Re: Non parametric ANCOVA

khahstats

Offline

Clair,
yes that's the basic approach. However, why would you want to do it? Nonparametric ANCOVA takes away your ability to estimate mean differences in the outcome (dependent variable) between levels of class variables (& a whole lot more). If you have a customer/client & they want to test whether 2 levels of a class variable are associated with different mean responses, well, you already know they are associated with different means, so what's the point?

Even if such testing was inherently meaningful, the ranks don't come even close to being symmetrically distributed, as long as there's much overlap between the responses associated with the levels of the class variable. If the responses for Level A of a class variable are shifted to the right, say, of those of Level B, then the responses for Level A will have a long tail to the left, & those for Level B will have such a tail to the right.

I usually try other transformations (percent change, logarithmic base 2, box-cox, anything) rather that a transformation to ranks.

--- On Sat, 2/28/09, lexie75@... <lexie75@...> wrote:

From: lexie75@... <lexie75@...>
Subject: [Statisticians_group] Non parametric ANCOVA
To: Statisticians_group@...
Date: Saturday, February 28, 2009, 2:49 AM

Hi everyone
I'm tearing my hair out trying to find out how to run a Non parametric ANCOVA in SPSS. can
I get people's advice? Can I rank the data and then run the ANCOVA as done in SPSS? I
think ti might be more complicated than that but if anyone knows could you direct me to
how I can get the SPSS syntax or how to calculate this test? I can find information that non
para ANCOVAs exist but none of the papers say how to actually do it beyond using ranks.
Any ideas?
Claire

#3298

From: nupur choudhury <me_nupur79@...>
Date:: Sat Feb 28, 2009 8:52 am
Subject:: Customer Satisfaction Index

me_nupur79

Offline

Hello,

Can anyone help me to calculate Customer Satisfaction Index(CSI)

Add more friends to your messenger and enjoy! Invite them now.

#3297

From: "lexie75@..." <lexie75@...>
Date:: Sat Feb 28, 2009 7:49 am
Subject:: Non parametric ANCOVA

lexie75...

Offline

Hi everyone
I'm tearing my hair out trying to find out how to run a Non parametric ANCOVA in
SPSS.  can
I get people's advice?  Can I rank the data and then run the ANCOVA as done in
SPSS?  I
think ti might be more complicated than that but if anyone knows could you
direct me to
how I can get the SPSS syntax or how to calculate this test?  I can find
information that non
para ANCOVAs exist but none of the papers say how to actually do it beyond using
ranks.
Any ideas?
Claire

#3296

From: jitadeepa nayak <im_jitu@...>
Date:: Fri Feb 27, 2009 5:20 am
Subject:: Help to install Beamer

im_jitu

Offline

Can anyone help me to download and install Beamer-Latex?
If anyone can send me the installation setup file, then please provide me that.

Thanks in advance......

#3295

From: "Neethu Thomas" <gtg634t@...>
Date:: Thu Feb 26, 2009 6:25 pm
Subject:: Re: time series analysis

thomasneethu

Offline

1. Brockwell, P.J. and Davis, R.A. (1991), Introduction to Time Series
and Forecasting, 2nd
edition, Springer-Verlag, New York.
2. Tsay, R.S. (2005), Analysis of Financial Time Series, 2nd Edition,
Wiley Series in Probability
and Statistics.
Other recommended books:
1. . Fan, J., Yao, Q. (2003). Nonlinear Time Series. Springer-Verlag,
New York.
2. Box, G.E.P., Jenkins, G.M. and Reinsel, G.C. (1994). Time Series
Analysis: Forecasting
and Control, 3rd Edition, Prentice Hall, New Jersey.
3. Chat�eld, C. (1996). The Analysis of Time Series, 5th edition,
Chapman and Hall, New
York.
4. Shumway, R.H., Sto�er, D.S. (2006), Time Series Analysis and Its
Applications (with R
examples), Springer-Verlag, New York.

#3294

From: KAUSHIK BHATTACHARJEE <kabonline07@...>
Date:: Thu Feb 26, 2009 6:23 pm
Subject:: Re: time series analysis

kabonline07

Offline

so many good books are there.

All econometrics book devote 2-3 chapters on it.

to start with...Chatfield is good. Cochran -available in net is also good.Hamilton is always there for reference.

Hopethis helps.

Kaushik

From: ashwini_mahapatra <ashwini_mahapatra@...>
To: Statisticians_group@...
Sent: Thursday, February 26, 2009 11:22:31 PM
Subject: [Statisticians_group] time series analysis

can some one help me a book on time series

#3293

From: "ashwini_mahapatra" <ashwini_mahapatra@...>
Date:: Thu Feb 26, 2009 5:52 pm
Subject:: time series analysis

ashwini_maha...

Offline

can some one help me a book on time series

#3292

From: dhssresearcher@...
Date:: Thu Feb 26, 2009 4:14 pm
Subject:: Re:DON'T DRINK MENTOS & COCA-COLA TOGETHR-SOS

geodssphq

Offline

Sorry this is spam. Or rather, an urban legend.

I'd check out Snopes.com for this.

Part of science, and statistics is bursting balloons. We all know of great ideas and theories that just don't work. Personally, I like the one about if you cut off tails on lab rats, they'll learn not to grow tails in later generations.

I'd argue the worst users of statistics are police departments. For a host of reasons.

A close second is medical literature, for a lot of reasons. Having published medical lit and read more that a lot of people, including going through the footnotes to check articles, some of the claims are just amazing. Like I was taught, the devil is in the details.

In this case, the story doesn't hold water... or Coke.

A Good Credit Score is 700 or Above. See yours in just 2 easy steps!

#3291

From: Siva Satish <siva_satish@...>
Date:: Wed Feb 25, 2009 12:27 pm
Subject:: DON'T DRINK MENTOS & COCA-COLA TOGETHR-SOS

siva_satish

Offline

Last week a little boy died in Brazil after eating MENTOS and drinking Coca-Cola / PEPSI together. One year before the same accident happened with another boy in Brazil. Please check the experiment that has been done by mixing Coka-Cola (or Coka-Cola Light) with MENTOS .

So be careful with your self eating MENTOS (POLO's) and drinking COCA-COLA or PEPSI together.
CHECK THIS OUT...

PLZ PASS THIS INFORMATION TO AS MANY PEOPLE AS POSSIBLE SPECIALLY TO THE CHILDREN'S. BECAUSE IN OUR COUNTRY MENTOS AND COCA-COLA BOTH ARE VERY POPULAR
TO THE CHILDREN.

Get your preferred Email name!
Now you can @ymail.com and @rocketmail.com.

#3290

From: Jay Warner at AT_T <jay.warner3@...>
Date:: Tue Feb 24, 2009 5:37 pm
Subject:: Re: How best to select optimal class intervals ?

jaywarner2008

Offline

embedded.

On Feb 23, 2009, at 11:46 PM, KAUSHIK BHATTACHARJEE wrote:

> Dear Jay Warner,
> #1 is well taken. I will do that.

Cool!

> #2 is also understood&known. Do you have a soft copy of the reference?

The text is heavy on the graphics, and so far as I know is NOT
available on line.  Check with any local universities for: Tufte,
Edward R., The Visual Display of Quantitative Information, Graphics
Press, 1983.

IMHO, if they don't have it, they are not teaching how to present
statistics to statisticians. And others.

Cheers,
Jay

> #3 is donealready. It is not lognoraml. May be mixture of 2/3
> normals or extreme value for the tails  or noraml inverse Gaussian
> will yield better result

The procedure remains the same.  Find a transformation that makes it
'more or less' Normal, if you can.  In this 'space' set out equal
intervals.  Back transform the interval boundaries to the 'real'
space, and histogram away!

Good luck,
Jay
>
> Statisticians: anyone has done that in SAS/MATLAB?want to share ?
>
> Kaushik
>
>
> From: Jay Warner at AT_T <jay.warner3@...>
> To: Statisticians_group@...
> Sent: Tuesday, February 24, 2009 9:17:28 AM
> Subject: Re: [Statisticians_group] How best to select optimal class
> intervals ?
>
> 1) Why do you want unequal class intervals?  Why not have a lot of
> intervals (many of which would be empty, out on that tail).?
>
>
> 2) If you do select different interval widths, be sure to keep the
> _area_ of the displayed histogram appropriate.  That is, for an
> interval twice the width of 'most' of them, you would not have a
> column that was twice the width and at the numerical height as
> given by the frequency in that column.  It would be half as high.
> Now the total area is appropriate to the data.  REf:  See Tufte,
> Edward R., The Visual Display of Quantitative Information, Graphics
> Press, 1983.  And your histogram will be _much_ more informative.
>
> 3) Is your 'fat tail' indicative of a lognormal type distribution?
> Then select your class intervals so that they are equally spaced on
> a log scale.  But make your plot on a regular, linear scale.  And
> heed item 2) above.
>
> Would this work?
>
> Jay
> On Feb 23, 2009, at 9:09 AM, KAUSHIK BHATTACHARJEE wrote:
>
>>
>> Dear Staticians,
>> How best to select optimal class intervals in case of continious
>> variables with large deviatuions( 'fat tails' in finance
>> parlance)? e.g. I have values for X1& X2 and percentilesbelow.
>> Quantile                 X1               X2
>>
>>                                      100% Max
>> 189.300561             115.45807
>>                                      99%
>> 129.160831             95.63896
>>                                      95%
>> 100.927531              72.43824
>>                                      90%
>> 86.680550             61.84146
>>
>>                                      75% Q3
>> 64.212005             42.0278
>>                                      50% Median
>> 34.488379             31.15345
>>                                      25% Q1
>> 14.039510             7.12425
>>                                      10%
>> 0.539843             -2.39189
>>                                      5%
>> -6.694915             -7.27948
>>                                      1%
>> -31.509901              -21.17243
>>                                      0% Min
>> -106.203049             -53.77416
>>
>> Number of observations are unequl (1646 ad1780).  Since cental
>> tendncies are same ,  I want to do a test of homogenity.For that
>> how best (optimally) select the (unequal) class intervals? Any
>> help/idea/reference ?
>>
>>
>> Also has anyone done modelling(and testing) of mixture normal
>> distributions in MATLAB or SAS?
>>
>>
>>
>>
>> Kaushik Bhattacharjee
>> Assistant Professor
>> IIMT,ICFAI University
>> ICFAI Business School,Hyderabad
>> RR District.501203
>> Contact No.9396677684
>>
>>
>>
>>
>
> Jay Warner
> Principal Scientist
> Warner Consulting, Inc.
> 4444 North Green Bay Road
> Racine, WI 53404-1216
> USA
>
> Ph:       262.634.9100
> FAX:     262.681.1133
> email:  quality@a2q. com
> Temporary:  jay.warner3@ sbcglobal. net
>
> web:    www.a2q.com
>
> The A2Q Method (tm) -- What do you want to improve today?
>
>
>
>
>
>
>
>

Jay Warner
Principal Scientist
Warner Consulting, Inc.
4444 North Green Bay Road
Racine, WI 53404-1216
USA

Ph:       262.634.9100
FAX:     262.681.1133
email:  quality@...
Temporary:  jay.warner3@...

web:    www.a2q.com

The A2Q Method (tm) -- What do you want to improve today?

#3289

From: Andrew Hartley <khahstats@...>
Date:: Tue Feb 24, 2009 12:12 pm
Subject:: Re: How best to select optimal class intervals ?

khahstats

Offline

Kaushik,

good thoughts. Yes many people do test means & variances, using p-values. Whenever I do so, though, I feel increasingly guilty about taking somebody's money for performing a procedure that I feel helps so little. But I console myself by pressing my co-workers & our clients to migrate to what I think are better ways.

You seem to admit the arbitrariness of testing, by mentioning "subjective" & "a roundabout manner." I am fairly certain that we do our clients a great disservice by settling for this arbitrariness, without seeking methods & results we can be more certain will truly add value to their research.

I agree that nobody knows the truth with utter certainty. That type of knowing is not the goal. Rather we can & should seek to shift & enhance our incomplete degrees of certainty about populations & scientific hypotheses.

--- On Tue, 2/24/09, KAUSHIK BHATTACHARJEE <kabonline07@...> wrote:

From: KAUSHIK BHATTACHARJEE <kabonline07@...>
Subject: Re: [Statisticians_group] How best to select optimal class intervals ?
To: Statisticians_group@...
Date: Tuesday, February 24, 2009, 12:07 AM

Dear Andrew Hartley,

yes you are correct at a phylosophical level... yet we do test equality of means& variances based on samples (being fully aware that the first two momentsare simply inadequate to charcterise any distribution- both may come from distributions with no moments ) and if the p-values are not verylow(again a subjective value) conclude in a roundabout manner : 'three is no reason to believe that the two distributions are significantly different'.

It is implied that by test of homogenity I meant something of that nature(equality of proportions in diff classes-perhaps)

It is immaterial what I want to hear---if they are different(who knows the truthby the way?),I should conclude they are different and I have to do statistical tests for that(I can not quote Nestor for not doing it)-which again is not fault free.

Regds.

Kaushik

From: Andrew Hartley <khahstats@yahoo. com>
To: Statisticians_ group@yahoogroup s.co.in
Sent: Tuesday, February 24, 2009 7:54:43 AM
Subject: Re: [Statisticians_ group] How best to select optimal class intervals ?

Mr/Ms Bhattacharjee,
the populations from which x1 & x2 were sampled are non-homogeneous. There. That was easy. As Nester has said (see his "applied statistician' s creed"), no 2 populations are the same in any respect. To conclude that, you don't need any statistical test.

That's probably not what you wanted to hear. Hopefully, though, I can console you somewhat by suggesting also that the more useful question is not "are these 2 populations the same" but "how certain can we be that they differ by a given amount?"

--- On Mon, 2/23/09, KAUSHIK BHATTACHARJEE <kabonline07@ yahoo.com> wrote:

From: KAUSHIK BHATTACHARJEE <kabonline07@ yahoo.com>
Subject: [Statisticians_ group] How best to select optimal class intervals ?
To: "statistics gropu" <statisticians_ group@yahoogroup s.co.in>
Date: Monday, February 23, 2009, 10:09 AM

Dear Staticians,

How best to select optimal class intervals in case of continious variables with large deviatuions( 'fat tails' in finance parlance)? e.g. I have values for X1& X2 and percentilesbelow.

Quantile                 X1               X2

                                     100% Max       189.300561             115.45807
                                     99%            129.160831             95.63896
                                     95%            100.927531              72.43824
                                     90%             86.680550             61.84146
                                     75% Q3          64.212005             42.0278
                                     50% Median      34.488379             31.15345
                                     25% Q1          14.039510             7.12425
                                     10%              0.539843             -2.39189
                                     5%              -6.694915             -7.27948
                                     1%             -31.509901              -21.17243
                                     0% Min        -106.203049             -53.77416
Number of observations are unequl (1646 ad1780). Since cental tendncies are same , I want to do a test of homogenity.For that how best (optimally) select the (unequal) class intervals? Any help/idea/reference ?

Also has anyone done modelling(and testing) of mixture normal distributions in MATLAB or SAS?

Kaushik Bhattacharjee
Assistant Professor
IIMT,ICFAI University
ICFAI Business School,Hyderabad
RR District.501203
Contact No.9396677684

#3288

From: ashwini mahapatra <ashwini_mahapatra@...>
Date:: Tue Feb 24, 2009 11:38 am
Subject:: Re: my book

ashwini_maha...

Offline

I Felt this was a nice observation and startled by the fact that how true this is --

In life i have always encountered a series of good events and series of bad events and never a singe random occurrence of good or bad

Thanks

Ashwini

--- On Sat, 2/21/09, S�ren Nielsen <srentospace@...> wrote:

From: S�ren Nielsen <srentospace@...>
Subject: [Statisticians_group] my book
To: Statisticians_group@...
Date: Saturday, February 21, 2009, 7:25 AM

I have added a file "lumpy statistic.pdf" in the file section.

I would like somebody to read it.

Maybe it is silly, but then tell me!

Best regards S�ren Nielsen

#3287

From: KAUSHIK BHATTACHARJEE <kabonline07@...>
Date:: Tue Feb 24, 2009 5:46 am
Subject:: Re: How best to select optimal class intervals ?

kabonline07

Offline

Dear Jay Warner,

#1 is well taken. I will do that.

#2 is also understood&known.Do you have asoft copy of the refernce?

#3 is donealready. It is not lognoraml. May be mixture of 2/3 normals or extreme value for the tails or noraml inverse Gaussian will yield better result

Statisticians: anyone has done that in SAS/MATLAB?want to share ?

Kaushik

From: Jay Warner at AT_T <jay.warner3@...>
To: Statisticians_group@...
Sent: Tuesday, February 24, 2009 9:17:28 AM
Subject: Re: [Statisticians_group] How best to select optimal class intervals ?

1) Why do you want unequal class intervals? Why not have a lot of intervals (many of which would be empty, out on that tail).?

2) If you do select different interval widths, be sure to keep the _area_ of the displayed histogram appropriate. That is, for an interval twice the width of 'most' of them, you would not have a column that was twice the width and at the numerical height as given by the frequency in that column. It would be half as high. Now the total area is appropriate to the data. REf: See Tufte, Edward R., The Visual Display of Quantitative Information, Graphics Press, 1983. And your histogram will be _much_ more informative.

3) Is your 'fat tail' indicative of a lognormal type distribution? Then select your class intervals so that they are equally spaced on a log scale. But make your plot on a regular, linear scale. And heed item 2) above.

Would this work?

Jay

On Feb 23, 2009, at 9:09 AM, KAUSHIK BHATTACHARJEE wrote:

Dear Staticians,

How best to select optimal class intervals in case of continious variables with large deviatuions( 'fat tails' in finance parlance)? e.g. I have values for X1& X2 and percentilesbelow.

Quantile                 X1               X2

                                     100% Max       189.300561             115.45807

                                     99%            129.160831             95.63896

                                     95%            100.927531              72.43824

                                     90%             86.680550             61.84146

                                     75% Q3          64.212005             42.0278

                                     50% Median      34.488379             31.15345

                                     25% Q1          14.039510             7.12425

                                     10%              0.539843             -2.39189

                                     5%              -6.694915             -7.27948

                                     1%             -31.509901              -21.17243

                                     0% Min        -106.203049             -53.77416

Number of observations are unequl (1646 ad1780). Since cental tendncies are same , I want to do a test of homogenity.For that how best (optimally) select the (unequal) class intervals? Any help/idea/reference ?

Also has anyone done modelling(and testing) of mixture normal distributions in MATLAB or SAS?

Kaushik Bhattacharjee
Assistant Professor
IIMT,ICFAI University
ICFAI Business School,Hyderabad
RR District.501203
Contact No.9396677684

Jay Warner

Principal Scientist

Warner Consulting, Inc.

4444 North Green Bay Road

Racine, WI 53404-1216

USA

Ph: 262.634.9100

FAX: 262.681.1133

email: quality@a2q. com

Temporary: jay.warner3@ sbcglobal. net

web: www.a2q.com

The A2Q Method (tm) -- What do you want to improve today?

#3286

From: KAUSHIK BHATTACHARJEE <kabonline07@...>
Date:: Tue Feb 24, 2009 5:07 am
Subject:: Re: How best to select optimal class intervals ?

kabonline07

Offline

Dear Andrew Hartley,

yes you are correct at a phylosophical level... yet we do test equality of means& variances based on samples (being fully aware that the first two momentsare simply inadequate to charcterise any distribution-both may come from distributions with no moments ) and if the p-values are not verylow(again a subjective value) conclude in a roundabout manner : 'three is no reason to believe that the two distributions are significantly different'.

It is implied that by test of homogenity I meant something of that nature(equality of proportions in diff classes-perhaps)

It is immaterial what I want to hear---if they are different(who knows the truthby the way?),I should conclude they are different and I have to do statistical tests for that(I can not quote Nestor for not doing it)-which again is not fault free.

Regds.

Kaushik

From: Andrew Hartley <khahstats@...>
To: Statisticians_group@...
Sent: Tuesday, February 24, 2009 7:54:43 AM
Subject: Re: [Statisticians_group] How best to select optimal class intervals ?

Mr/Ms Bhattacharjee,
the populations from which x1 & x2 were sampled are non-homogeneous. There. That was easy. As Nester has said (see his "applied statistician' s creed"), no 2 populations are the same in any respect. To conclude that, you don't need any statistical test.

That's probably not what you wanted to hear. Hopefully, though, I can console you somewhat by suggesting also that the more useful question is not "are these 2 populations the same" but "how certain can we be that they differ by a given amount?"

--- On Mon, 2/23/09, KAUSHIK BHATTACHARJEE <kabonline07@ yahoo.com> wrote:

From: KAUSHIK BHATTACHARJEE <kabonline07@ yahoo.com>
Subject: [Statisticians_ group] How best to select optimal class intervals ?
To: "statistics gropu" <statisticians_ group@yahoogroup s.co.in>
Date: Monday, February 23, 2009, 10:09 AM

Dear Staticians,

How best to select optimal class intervals in case of continious variables with large deviatuions( 'fat tails' in finance parlance)? e.g. I have values for X1& X2 and percentilesbelow.

Quantile                 X1               X2

                                     100% Max       189.300561             115.45807

                                     99%            129.160831             95.63896

                                     95%            100.927531              72.43824

                                     90%             86.680550             61.84146

                                     75% Q3          64.212005             42.0278

                                     50% Median      34.488379             31.15345

                                     25% Q1          14.039510             7.12425

                                     10%              0.539843             -2.39189

                                     5%              -6.694915             -7.27948

                                     1%             -31.509901              -21.17243

                                     0% Min        -106.203049             -53.77416

Number of observations are unequl (1646 ad1780). Since cental tendncies are same , I want to do a test of homogenity.For that how best (optimally) select the (unequal) class intervals? Any help/idea/reference ?

Also has anyone done modelling(and testing) of mixture normal distributions in MATLAB or SAS?

Kaushik Bhattacharjee
Assistant Professor
IIMT,ICFAI University
ICFAI Business School,Hyderabad
RR District.501203
Contact No.9396677684

#3285

From: Jay Warner at AT_T <jay.warner3@...>
Date:: Tue Feb 24, 2009 3:47 am
Subject:: Re: How best to select optimal class intervals ?

jaywarner2008

Offline

1) Why do you want unequal class intervals?� Why not have a lot of intervals (many of which would be empty, out on that tail).?

2) If you do select different interval widths, be sure to keep the _area_ of the displayed histogram appropriate.� That is, for an interval twice the width of 'most' of them, you would not have a column that was twice the width and at the numerical height as given by the frequency in that column.� It would be half as high.� Now the total area is appropriate to the data.� REf:� See�Tufte, Edward R., The Visual Display of Quantitative Information, Graphics Press, 1983.� And your�histogram will be _much_ more informative.

3) Is your 'fat tail' indicative of a lognormal type distribution?� Then select your class intervals so that they are equally spaced on a log scale.� But make your plot on a regular, linear scale.� And heed item 2) above.

Would this work?

Jay

On Feb 23, 2009, at 9:09 AM, KAUSHIK BHATTACHARJEE wrote:

�
Dear Staticians,
How best to select optimal class intervals in case of continious variables with large deviatuions('fat tails' in finance parlance)? e.g. I have values for X1& X2 and percentilesbelow.
Quantile�� X1�� X2
�
�� 100% Max�� 189.300561�� 115.45807
�� 99%�� 129.160831�� 95.63896
�� 95%�� 100.927531�� 72.43824� ��
�� 90%�� 86.680550�� 61.84146
��75% Q3�� 64.212005�� 42.0278
�� 50% Median�� 34.488379�� 31.15345
�� 25% Q1�� 14.039510�� 7.12425
�� 10%�� 0.539843�� -2.39189
�� 5%�� -6.694915�� -7.27948
�� 1%�� -31.509901�� -21.17243��
�� 0% Min�� -106.203049�� -53.77416
Number of observations are unequl (1646 ad1780).� Since cental tendncies are same ,� I want to do a test of homogenity.For that how best (optimally) select the (unequal) class intervals? Any help/idea/reference?
�
�
Also has anyone done modelling(and testing)�of mixture normal distributions in MATLAB or SAS?
�
�

�
Kaushik Bhattacharjee
Assistant Professor
IIMT,ICFAI University
ICFAI Business School,Hyderabad
RR District.501203
Contact No.9396677684

Jay Warner

Principal Scientist

Warner Consulting, Inc.

4444 North Green Bay Road

Racine, WI 53404-1216

USA

Ph: � �� 262.634.9100

FAX:� � �262.681.1133

email:��quality@...

Temporary:� jay.warner3@...

web: � �www.a2q.com

The A2Q Method (tm) -- What do you want to improve today?

#3284

From: Andrew Hartley <khahstats@...>
Date:: Tue Feb 24, 2009 2:24 am
Subject:: Re: How best to select optimal class intervals ?

khahstats

Offline

Mr/Ms Bhattacharjee,
the populations from which x1 & x2 were sampled are non-homogeneous. There. That was easy. As Nester has said (see his "applied statistician's creed"), no 2 populations are the same in any respect. To conclude that, you don't need any statistical test.

That's probably not what you wanted to hear. Hopefully, though, I can console you somewhat by suggesting also that the more useful question is not "are these 2 populations the same" but "how certain can we be that they differ by a given amount?"

--- On Mon, 2/23/09, KAUSHIK BHATTACHARJEE <kabonline07@...> wrote:

From: KAUSHIK BHATTACHARJEE <kabonline07@...>
Subject: [Statisticians_group] How best to select optimal class intervals ?
To: "statistics gropu" <statisticians_group@...>
Date: Monday, February 23, 2009, 10:09 AM

Dear Staticians,

How best to select optimal class intervals in case of continious variables with large deviatuions( 'fat tails' in finance parlance)? e.g. I have values for X1& X2 and percentilesbelow.

Quantile                 X1               X2

                                     100% Max       189.300561             115.45807

                                     99%            129.160831             95.63896

                                     95%            100.927531              72.43824

                                     90%             86.680550             61.84146

                                     75% Q3          64.212005             42.0278

                                     50% Median      34.488379             31.15345

                                     25% Q1          14.039510             7.12425

                                     10%              0.539843             -2.39189

                                     5%              -6.694915             -7.27948

                                     1%             -31.509901              -21.17243

                                     0% Min        -106.203049             -53.77416

Number of observations are unequl (1646 ad1780). Since cental tendncies are same , I want to do a test of homogenity.For that how best (optimally) select the (unequal) class intervals? Any help/idea/reference ?

Also has anyone done modelling(and testing) of mixture normal distributions in MATLAB or SAS?

Kaushik Bhattacharjee
Assistant Professor
IIMT,ICFAI University
ICFAI Business School,Hyderabad
RR District.501203
Contact No.9396677684

#3283

From: KAUSHIK BHATTACHARJEE <kabonline07@...>
Date:: Mon Feb 23, 2009 3:09 pm
Subject:: How best to select optimal class intervals ?

kabonline07

Offline

Dear Staticians,

How best to select optimal class intervals in case of continious variables with large deviatuions('fat tails' in finance parlance)? e.g. I have values for X1& X2 and percentilesbelow.

Quantile X1 X2

100% Max 189.300561 115.45807

99% 129.160831 95.63896

95% 100.927531 72.43824

90% 86.680550 61.84146

75% Q3 64.212005 42.0278

50% Median 34.488379 31.15345

25% Q1 14.039510 7.12425

10% 0.539843 -2.39189

5% -6.694915 -7.27948

1% -31.509901 -21.17243

0% Min -106.203049 -53.77416

Number of observations are unequl (1646 ad1780). Since cental tendncies are same , I want to do a test of homogenity.For that how best (optimally) select the (unequal) class intervals? Any help/idea/reference?

Also has anyone done modelling(and testing) of mixture normal distributions in MATLAB or SAS?

Kaushik Bhattacharjee
Assistant Professor
IIMT,ICFAI University
ICFAI Business School,Hyderabad
RR District.501203
Contact No.9396677684

#3282

From: Muhammad Ibrahim Shamsi <abrahamshamsi@...>
Date:: Mon Feb 23, 2009 6:05 am
Subject:: Fw: Fwd: Gul Ahmed Textile Mills Ltd. is hiring Supply Chain Officer

abrahamshamsi

Offline

----- Forwarded Message ----
From: Ibrahim Shamsi <ibrahim.shamsi@...>
To: Muhammad Ibrahim Shamsi <abrahamshamsi@...>
Sent: Monday, February 23, 2009 11:00:58
Subject: Fwd: Gul Ahmed Textile Mills Ltd. is hiring Supply Chain Officer

---------- Forwarded message ----------
From: ali hussain <alihussain.bme@...>
Date: Mon, Feb 23, 2009 at 10:30 AM
Subject: Fwd: Gul Ahmed Textile Mills Ltd. is hiring Supply Chain Officer
To: ibrahim.shamsi@...

Dear Ali Hussain Ronaque

A new job has been posted in one of the categories matching your resume.

Details
Company Name	Gul Ahmed Textile Mills Ltd.
Job Code	22084
Posting Date	19 Feb 2009
Closing Date	31 Mar 2009
Industry	Textile/Garment/Accessories/Fashion
Job Tenure	Permanent
Gender Requirement	None
Summary	? Knowledge of main functions of Supply Chain ? Good computer literacy ? Strong communication skills
City	Karachi
Country	Pakistan
Experience Required	1-5 Year(s)
Management Level	Mid Career (2+ years of experience)
Availability	Full Time
Work Location	Onsite
Location	Landhi- Karachi
Age Requirement	25 - 30 Year(s)
Minimum Education	Graduate (Preferably Statistics)
Description	Experience Required: Textile Buying House/FMCG/Courier Companies (TCS/DHL/Fed Ex)/Textile Mills with large customers. Statistical Software's experience will be preferred. Responsibilities; • Develop and manage Material Resource Planning & ensure availability of product. • Inspection of documents on incoming orders to ensure quality and quantity compliance. • Prompt resolution to any and all shipment challenges • Identify systems, tools needed to implement planning/forecasting systems.

Skill Set :

Management

Customer Services Management

General Management

Information Systems Management

Investor Relations Management

Operations Management

Pricing Policies & Plans

Strategic Planning

Supply Chain Management

Planning & Development

Business/System Analysts

Enterprise Resource Planning

Sales & Marketing

Business Plans

Competitive Analysis

Direct Marketing

Marketing Communications

Marketing Plans

Public Relations

Retailing

Sales Management

Merchandising

To apply for this job, please click : http://www.naukrijunction.com/UserServices/ViewJob.aspx?JobId=52450

If you have any queries, please contact support@....

You are receiving this email because you signed up for an account at http://www.naukrijunction.com. To unsubscribe, click here

We want to thank you for using NaukriJunction and always value your comments and feedback.

NaukriJunction Team
http://www.naukrijunction.com

You are receiving this email because you signed up for an account at http://www.naukrijunction.com with login ID: alihussainbme.

In case you have forgotten your password, please go here to recover it:http://www.naukrijunction.com/RecoverPswd.aspx?Email=alihussain.bme@...

--
Ibrahim Shamsi

New Email names for you!
Get the Email name you've always wanted on the new @ymail and @rocketmail.
Hurry before someone else does!

Reply | Forward | Messages in this Topic (1)

#3281

From: vishal merchant <vishal_merchant@...>
Date:: Sat Feb 21, 2009 5:37 pm
Subject:: Re: can anybody help me to find the books...

vishal_merchant

Offline

Hi,
This is the download link to Multivariate ebooks you requested. See if its useful to you. Ebooks are in pdf format.
http://www.ziddu.com/download/3593923/MultivariateE-book.rar.html

vishal

--- On Fri, 20/2/09, subhabrata bhattacharya <bhutta_subha@...> wrote:

From: subhabrata bhattacharya <bhutta_subha@...>
Subject: [Statisticians_group] can anybody help me to find the books...
To: "stat" <statisticians_group@...>
Date: Friday, 20 February, 2009, 5:44 PM

Hi All,

I am MSc in Statistics & working.

Can anyone supply a e-book discussed with practical examples on

1.Logistic regression

2.Multivariate Analysis(stress on Discriminant Analysis)

Thanks in advance.

Thanks & Regards,

Subho.

Subhabrata Bhattacharya

E-mail : bhutta_subha@ yahoo.co. in

bhuttasubha@ gmail.com

Download prohibited? No problem. CHAT from any browser, without download.

Add more friends to your messenger and enjoy! Invite them now.

#3280

From: ehsan sabaghiayan <e_sabaghian82@...>
Date:: Sat Feb 21, 2009 8:08 am
Subject:: Re: can anybody help me to find the books...

e_sabaghian82

Offline

I recommend this book: Applied Logistic Regression, W.Hosmer and Stanley Lemeshow, 2nd Edition. That is a great book with point of view practical. You can find it here www.gigapedia.com

Regards

Ehsan

--- On Fri, 2/20/09, subhabrata bhattacharya <bhutta_subha@...> wrote:

From: subhabrata bhattacharya <bhutta_subha@...>
Subject: [Statisticians_group] can anybody help me to find the books...
To: "stat" <statisticians_group@...>
Date: Friday, February 20, 2009, 4:14 AM

Hi All,

I am MSc in Statistics & working.

Can anyone supply a e-book discussed with practical examples on

1.Logistic regression

2.Multivariate Analysis(stress on Discriminant Analysis)

Thanks in advance.

Thanks & Regards,

Subho.

Subhabrata Bhattacharya

E-mail : bhutta_subha@ yahoo.co. in

bhuttasubha@ gmail.com

Download prohibited? No problem. CHAT from any browser, without download.

#3279

From: S�ren Nielsen <srentospace@...>
Date:: Sat Feb 21, 2009 1:55 am
Subject:: my book

srentospace

Offline

I have added a file "lumpy statistic.pdf" in the file section.

I would like somebody to read it.

Maybe it is silly, but then tell me!

Best regards S�ren Nielsen

#3278

From: Sandra Romany <sandy009@...>
Date:: Fri Feb 20, 2009 1:51 pm
Subject:: Re: can anybody help me to find the books...

sandy009

Offline

http://www.psychstat.missouristate.edu/multibook/mlt00.htm

The above link is hypertext on Multivariate Analysis by David W. Stockburger. You may wish to enter "Statistics hypertext" on google.

Sandra

Trinidad and Tobago

--- On Fri, 20/2/09, subhabrata bhattacharya <bhutta_subha@...> wrote:

From: subhabrata bhattacharya <bhutta_subha@...>
Subject: [Statisticians_group] can anybody help me to find the books...
To: "stat" <statisticians_group@...>
Date: Friday, 20 February, 2009, 8:14 AM

Hi All,

I am MSc in Statistics & working.

Can anyone supply a e-book discussed with practical examples on

1.Logistic regression

2.Multivariate Analysis(stress on Discriminant Analysis)

Thanks in advance.

Thanks & Regards,

Subho.

Subhabrata Bhattacharya

E-mail : bhutta_subha@ yahoo.co. in

bhuttasubha@ gmail.com

Download prohibited? No problem. CHAT from any browser, without download.

#3277

From: subhabrata bhattacharya <bhutta_subha@...>
Date:: Fri Feb 20, 2009 12:14 pm
Subject:: can anybody help me to find the books...

bhutta_subha

Offline

Hi All,

I am MSc in Statistics & working.

Can anyone supply a e-book discussed with practical examples on

1.Logistic regression

2.Multivariate Analysis(stress on Discriminant Analysis)

Thanks in advance.

Thanks & Regards,

Subho.

Subhabrata Bhattacharya

E-mail : bhutta_subha@...

bhuttasubha@...

Download prohibited? No problem. CHAT from any browser, without download.

#3276

From: Statisticians_group@...
Date:: Wed Feb 18, 2009 11:44 am
Subject:: New file uploaded to Statisticians_group

Statisticians_group@...

Hello,

This email message is a notification to let you know that
a file has been uploaded to the Files area of the Statisticians_group
group.

   File        : /lumpy statistic..pdf
   Uploaded by : srentospace <srentospace@...>
   Description :

You can access this file at the URL:
http://in.groups.yahoo.com/group/Statisticians_group/files/lumpy%20statistic..pd\
f

To learn more about file sharing for your group, please visit:
http://help.yahoo.com/help/in/groups/files

Regards,

srentospace <srentospace@...>

Reply | Forward | Messages in this Topic (26)

#3275

From: "Nina Rankine" <nrankine@...>
Date:: Wed Feb 18, 2009 1:03 am
Subject:: Applied Math Role For Fortune 500 Client In New Jersey

nina.rankine593

Offline

Hello Everyone,

The qualified candidate for this role must reside in New Jersey. Please circulate with your network.

Interested parties should email resumes to Nina Rankine: nrankine@....

Please see below for the job description for the role.

Applied Math Specialist

General Summary:

The general nature of the job is to apply professional statistical methods that develop the subscriber insight used to differentiate interactions with Client’s customers according to their preferences and profitability in order to increase ARPU, reduce churn, and target the acquisition of high value prospects. The responsibility level of the job is strategic in that the position’s work product is foundational not only within the context of CRM but across a number of business functions. The purpose of the job is to support the Manager in arming Client’s marketing function with subscriber level insight that informs Client’s efforts against the aforementioned objectives.

Responsibilities:

% Of Time	Duties and Responsibilities
40%	1. Manage development of plans to attain the analytic goals requested by the Manager-Applied Math, implement them, and maintain ownership of directing each initiative to completion. Will marshal, lead and direct the cross-functional teams necessary to complete the Manager’s requests.
20%	2. Per the Manager’s directive, develop and empirically evaluate hypothesis generation / testing and subsequent target identification, proactive uncovering and utilization of internal and external data, manage data processing resources both internal and external, and the generation and automation of customer insight.
20%	3. Within the context of the team’s overarching goals, manages predictive modeling, data mining and decision support capabilities related to new customer acquisition, product / service purchase propensity, and proactive account maintenance. Recommends and drives improvements within each of the aforementioned subject domains.
20%	4. Identification of opportunities to acquire new customers by leveraging knowledge of marketing objectives and requirements in database queries, analysis and automation of insight generated. Directs the varied functional teams to the opportunity’s output as represented in the data.

Scope:

§ This headcount addition was approved by the COO to support the $1.05B incremental revenue commitment to which CRM has agreed. This position requires both creative and critical thinking skills in developing leverageable subscriber insight in pursuit of incremental revenue.

§ The data managed by the Applied Mathematics function is a significant contributor, and will become a more frequent contributor, to internally developed subscriber analyses, and provides critical input to Client’s segmentation process

§ The data managed by the Applied Math function is at the core of the Client’s CRM efforts and foundational to its purpose

§ The analyses and insight produced by Applied Math impacts revenue, gross and net adds

§ Applied Math is responsible for developing, automating CRMs insight generation function

Knowledge, Skills and Abilities

· Experience: 3+ years data warehouse and quantitative reporting / analysis experience; 3+ years in telecommunications or wireless industry; experience with development, testing, and management of predictive modeling, data mining and decision support tools

· Other: Strong technical and business background. Competence with analysis, reporting and presentation tools (ex., SAS, Excel, Powerpoint). Strong competence in database and data management language (ex., Perl, SQL, MS Access); understanding of, and proven ability to apply, CRM precepts; superior planning skills; results oriented with strong sense of urgency, proactive and flexible; organized and process oriented; leverages technology and forward thinking mindset

· Education: Bachelor’s of Statistics / Applied Mathematics, Information Systems; graduate degree preferred

· Certifications: Project management; programming; Teradata Table Maintenance and Data Warehousing

Regards,

J. Anderson & Associates
An Executive Search Firm
Nina Rankine

Executive Recruiter
973-794-5312
www.jandersoninc.com

No virus found in this outgoing message.
Checked by AVG.
Version: 7.5.552 / Virus Database: 270.10.25/1955 - Release Date: 2/16/2009 6:55 AM

Reply | Forward | Messages in this Topic (1)

#3274

From: Pablo <les_yeux_clos@...>
Date:: Tue Feb 17, 2009 11:52 pm
Subject:: RE: Barttlett's test of sphercity--why such name?

dehn_nielsen

Offline