question
Stephen J Bensman
notsjb at LSU.EDU
Thu Feb 1 09:36:25 EST 2007
Beautiful, but I was not "pleading" for examination for an exogenous
subject variable but "suggesting" it. I myself never do a correlation or
regression analysis without first scatter plotting the data and then
examining the outliers. It is in the outliers and residuals that the
essence of the data is to be found. I think that Ali adopted the right
approach with the chi-squared test of independence, given the highly skewed
data within a restricted range. It is now just a matter of classifying the
articles into subject subsets and examining the composition of his cells in
these terms to see whether there is in fact a subject variable in action.
A dollar to your donut that he will find that there is one, given the
strength of the relationship that he found. The relationship of number of
references to number of citations is too strong to be an random event, and
there either has to be a functional explanation like review vs. research
articles or some subject variable in operation. I can see no logical
reason for number of references generating number of citations. If the
latter is the case, then we can all become famous by gang-footnoting. Soon
editors would not only be resticting the number of pages but also the
number of references.
SB
"Quentin L. Burrell" <quentinburrell at MANX.NET>@LISTSERV.UTK.EDU> on
01/31/2007 02:40:02 PM
Please respond to ASIS&T Special Interest Group on Metrics
<SIGMETRICS at LISTSERV.UTK.EDU>
Sent by: ASIS&T Special Interest Group on Metrics
<SIGMETRICS at LISTSERV.UTK.EDU>
To: SIGMETRICS at LISTSERV.UTK.EDU
cc: (bcc: Stephen J Bensman/notsjb/LSU)
Subject: Re: [SIGMETRICS] question
I apologise for the fact that my last response to Steven Morris (SM) was
sent out before proper completion - my intention was to include endorsement
also of the points raised by Stephen Bensman (SJB).
(I also apologise to any list members who find the following somewhat
patronising, but over the (many) years of teaching and reading regression
analyses I have found that a major problem is that of performing the
appropriate analysis without looking at the data!)
To illustrate at a very simple (and artificial) level, consider a set of 6
papers in e.g. scientometrics for which the (references, citations) counts
are (0,2), (1,1), (2,0), (18,20), (19,19) and (20,18) respectively. A
staightforward regression analysis yields :
citations = 0.16 + 0.98references
with Rsquared = 0.97 (correlation coefficient = 0.984).
Impressive, but following SM if we plot the scatter diagram we see two
"clusters" made up of the first 3 points and the last 3 which reflect very
different "patterns". If we now find from the context of the data that the
first set correspond (say) to mathematical presentations and the latter to
non mathematical ones, then maybe we should be analysing them separately.
If
we do, then we find the correllation coefficient for each category is -1,
perfect negative correlation in each case!
This simplistic presentation is also hinting at SJBs plea for a more subtle
analysis including an "exogenous subject variable".
Thanks to SM and SJB for highlighting some of the possible pitfalls in
regression analysis, but thanks also to Ali for his interesting analysis.
I think that my main point is that when performing any sort of
mathematical/statistical analysis one has to take full account of the data
context, not just the data.
Best wishes
Quentin
*****************************************
Dr Quentin L Burrell
Isle of Man International Business School
The Nunnery
Old Castletown Road
Douglas
Isle of Man IM2 1QB
via United Kingdom
q.burrell at ibs.ac.im
www.ibs.ac.im
----- Original Message -----
From: "Stephen J Bensman" <notsjb at LSU.EDU>
To: <SIGMETRICS at LISTSERV.UTK.EDU>
Sent: Tuesday, January 30, 2007 5:17 PM
Subject: Re: [SIGMETRICS] question
> Adminstrative info for SIGMETRICS (for example unsubscribe):
> http://web.utk.edu/~gwhitney/sigmetrics.html
>
>
>
>
>
> OK, Ali, I have read your paper, and a nice piece of work it is.
However,
> I want to make one criticism. You failed to classify your 467
> scientometric papers into subject subsets. It may well be that certain
> scientometric topics may both have more references per paper and be more
> prone to be cited. Therefore, your finding of the high postive
> relationship between the number of reference and the number of citations
> may be an artifact of an exogenous subject variable. I hope that you do
> not take this as a criticism but as an opportunity to squeeze another
> paper
> out of the same set of data.
>
> SB
>
>
>
>
> ali uzun <azun at METU.EDU.TR>@LISTSERV.UTK.EDU> on 01/30/2007 01:37:26 AM
>
> Please respond to ASIS&T Special Interest Group on Metrics
> <SIGMETRICS at LISTSERV.UTK.EDU>
>
> Sent by: ASIS&T Special Interest Group on Metrics
> <SIGMETRICS at LISTSERV.UTK.EDU>
>
>
> To: SIGMETRICS at LISTSERV.UTK.EDU
> cc: (bcc: Stephen J Bensman/notsjb/LSU)
>
> Subject: Re: [SIGMETRICS] question
>
> Adminstrative info for SIGMETRICS (for example unsubscribe):
> http://web.utk.edu/~gwhitney/sigmetrics.html
>
> -----Dear Stephen,
> I am sending an electronic version of the paper. The statistical
> ralationship between the two categories (citations received and
> referances listed) is of predictive type. There is no cause and effect
> relation.
>
> Prof. Dr. Ali Uzun
> Depr. Stat. Middle East Technical Univ. Ankara-Turkey.
> -------------
>> Adminstrative info for SIGMETRICS (for example unsubscribe):
>> http://web.utk.edu/~gwhitney/sigmetrics.html
>>
>> Ali,
>> That is very interesting. Something I would not expect. It does
> support
>> Narin. Can you send me an electronic version of your article to
>> notsjb at lsu.edu? Do you have any idea why there is such a
> relationship?
>>
>> SB
>>
>>
>>
>>
>> ali uzun <azun at METU.EDU.TR>@LISTSERV.UTK.EDU> on 01/29/2007 01:44:05
> AM
>>
>> Please respond to ASIS&T Special Interest Group on Metrics
>> <SIGMETRICS at LISTSERV.UTK.EDU>
>>
>> Sent by: ASIS&T Special Interest Group on Metrics
>> <SIGMETRICS at LISTSERV.UTK.EDU>
>>
>>
>> To: SIGMETRICS at LISTSERV.UTK.EDU
>> cc: (bcc: Stephen J Bensman/notsjb/LSU)
>>
>> Subject: Re: [SIGMETRICS] question
>>
>> Adminstrative info for SIGMETRICS (for example unsubscribe):
>> http://web.utk.edu/~gwhitney/sigmetrics.html
>>
>> ------Dear Ronald,
>> A sample of 467 artiles (not including reviews) published from 1999
> to
>> 2003 in the journal Scientometrics has shown that there is a linear
>> correlation (correlation coefficient of 0.799) between the number of
>> times an article is cited and the number of references it contains.
>> This was supported by a Chi-Square test of independence between the
>> two indicators at 0.01 level of significance (Uzun, A. (2006).
>> Proceedings of the International Workshop on Webometrics,
> Informetrics
>> and Scientometrics, 87-91,10-12 May 2006, Nancy-France).
>> > Adminstrative info for SIGMETRICS (for example unsubscribe):
>> > http://web.utk.edu/~gwhitney/sigmetrics.html
>> >
>> > Dear colleagues,
>> >
>> > Is there a positive correlation between the length of a reference
>> list of a
>> > publication and the number of citations received? Is this true (or
>> not) in
>> > general, i.e. considering all types of publication? And what if
> one
>> only
>> > considers 'normal articles', this is when reviews and letters (and
>> other short
>> > communications) are not taken into account?
>> >
>> > Can someone point me to a reference?
>> >
>> > Thanks!
>> >
>> > Ronald
>> >
>> >
>> > --
>> > Ronald Rousseau
>> > KHBO (Association K.U.Leuven)- Industrial Sciences and Technology
>> > Zeedijk 101 B-8400 Oostende Belgium
>> > Guest Professor at the Antwerp University School for Library and
>> Information
>> > Science (UA - IBW)
>> > E-mail: ronald.rousseau at khbo.be
>> > web page: http://users.telenet.be/ronald.rousseau
>> >
>> >
>> >
>> > ----------------------------------------------------------------
>> > This message was sent using IMP 3.2.8, the Internet Messaging
>> Program.
>> >
>> (See attached file: France1.pdf)
>
More information about the SIGMETRICS
mailing list