Skip to content

"chi-square" | "chi-squared" | "chi squared" | "chi square" ?

10 messages · Martin Maechler, Dénes Tóth, Rolf Turner +6 more

Martin Maechler
# 
As it's Friday ..

and I also really want to clean up help files and similar R documents,
both in R's own sources and in my new 'DPQ' CRAN package :

As a trained mathematician, I'm uneasy if a thing has
several easily confusable names, .. but as somewhat
humanistically educated person, I know that natural languages,
English in this case, are much more flexible than computer
languages or math... 

Anyway, back to the question(s) .. which I had asked myself a
couple of months ago, and already remained slightly undecided:

The 0-th (meta-)question of course is

  0. Is it worth using only one written form for the
     ?? - distribution, e.g. "everywhere" in R?

The answer is not obvious, as already the first few words of the
(English) Wikipedia clearly convey:

The URL is  https://en.wikipedia.org/wiki/Chi-squared_distribution
and the main title therefore also
    "Chi-squared distribution"

Then it reads
So, in title and 1st paragraph its "chi-squared", but then
everywhere(?) the text used "chi-square".

Undoubtedly, Wilson & Hilferty (1931) has been an important
paper and they use "Chi-square" in the title;
also  Johnson, Kotz & Balakrishnan (1995)
see R's help page ?pchisq use  "Chi-square" in the title of
chapter 18 and then, diplomatically for chapter 29,
 "Noncentral ??-Distributions" as title.

So it seems, that historically and using prestigious sources,
"chi-square" to dominate (notably if we do not count "??" as an
alternative).

Things look a bit different when I study R's sources; on one
hand, I find all 4 forms (s.Subject); then in the "R source
history", I see

  $ svn log -c11342
  ------------------------------------------------------------------------
  r11342 | <....> | 2000-11-14 ...

  Use `chi-squared'.
  ------------------------------------------------------------------------

which changed 16 (if I counted correctly) cases of 'chi-square' to 'chi-squared'.

I have not found any R-core internal (or public) reasoning about
that change, but had kept it in mind and often worked along that "goal".

As a consequence, "statistically" speaking, much of R's own use has been
standardized to use "chi-squared"; but as I mentioned, I still
find all  4  variants even in "R base" package help files
(which of course I now could quite quickly change  (using Emacs M-x grep, plus a script);
but

... "as it is Friday" ... I'm interested to hear what others
think, notably if you are native English (or "American" ;-)
speaking and/or have some extra good knowledge on such
matters...

Martin Maechler
ETH Zurich
Dénes Tóth
# 
Dear Martin,

Others struggle with this inconsistency as well; I found this discussion 
useful: 
https://math.stackexchange.com/questions/1098138/chi-square-or-chi-squared

Denes
On 10/18/19 2:51 PM, Martin Maechler wrote:
Rolf Turner
# 
I have the vague impression that "chi-squared" is more common in British 
usage and "chi-square" more common in American usage.  I'm pretty sure 
that either is acceptable, although "chi-squared" sounds much better to 
my ear.

Of course within a given document (or collection of related documents) 
consistency is mandatory.

cheers,

Rolf
On 19/10/19 1:51 AM, Martin Maechler wrote:

            

      
      
      
    

  
  


  


      

    

    

      
      

        


  

        

      Jim Lemon
    

    

      
      #
    

  


  

      
I have thought about this one myself, and just reading the posts and
links has afforded me a more informed viewpoint. My guess is that it
boils down to a contest between mathematics and prosody. To speakers
of English, "square"  in the mathematical sense implies the active
form such as "I square this number". When our focus is on the number
itself, it is usually expressed as "This number has been squared". My
suggestion is that while "chi-square" may be more correct in the
derivation of the statistic, "chi-squared" is more consistent with
colloquial usage in using the passive form. It may also avoid
confusion with the use of "square" as a noun, in which the preceding
word is often an adjective (e.g. "a red square").

Jim

        
On Sat, Oct 19, 2019 at 12:19 AM D?nes T?th <toth.denes at kogentum.hu> wrote:

            

      
      
      
    

        

      
      
    

            

      
      
      
    

        

      
      
    

  
  


  


      

    

    

      
      

        


  

        

      Rich Shepard
    

    

      
      #
    

  


  

      
On Sat, 19 Oct 2019, Jim Lemon wrote:

        

            

      
      
      
    

        
Jim,

This is a cogent suggestion that's pragmatic and defensible.

Thank you,

Rich

  
  


  


      

    

    

      
      

        


  

        

      Richard M. Heiberger
    

    

      
      #
    

  


  

      
What a delightful question.  Bill Cochran discussed this in class
one day about 50 years ago.  He said the British usage (which I think
he said was chi-squared,
as is consistent with the other memories in this thread)
is what he learned and previously used.  But he had been in the US for
so long that he was now using
the American preference (chi-square).

Rich

On Fri, Oct 18, 2019 at 8:51 AM Martin Maechler

        
<maechler at stat.math.ethz.ch> wrote:

            

      
      
      
    

        

      
      
    

  
  


  


      

    

    

      
      

        


  

        

      Dalthorp, Daniel
    

    

      
      #
    

  


  

      
oh my...

I'd like to see the statistics on it before jumping to a conclusion that
the American preference is "chi-square" and the British preference is
"chi-squared". I don't see that at all.

------
In keeping with the pronunciation of x^2 and 3^2, maybe "chi-squared" makes
the most sense,.

The "chi-square"? Because the iterated dentals in "chi-squared
distribution" and "chi-squared test" are a little cumbersome to pronounce,
an even slightly lazy pronunciation would sound like "chi-square
distribution" and "chi-square test". There's no need to write it that way
though.

-Dan

        
On Fri, Oct 18, 2019 at 2:28 PM Richard M. Heiberger <rmh at temple.edu> wrote:

        

            

      
      
      
    

        

      
      
    

        

      
      
    

        

  
    

      
      
    

  


  


      

    

    
2 days later

    

      
      

        


  

        

      Richard O'Keefe
    

    

      
      #
    

  


  

      
Pearson's original paper uses both \chi and \chi^2 frequently but
never spells out how to pronounce the latter.
Try another question: when talking about \sigma^2 do you say
"sigma-square" (which sounds rather odd) or "sigma-squared" (which
sounds more natural)?  If you say sigma-square, say chi-square.  If
you say sigma-squared, say chi-squared.
For what it's worth, the multu=i-volume "Encyclopedia of Statistical
Sciences", 2nd edition, uses both variants.
And so does Kendal & Stuart, volume 2, 1961, although "-squared" seems
to predominate.

On Sat, 19 Oct 2019 at 11:27, Dalthorp, Daniel via R-help

        
<r-help at r-project.org> wrote:

            

      
      
      
    

        

      
      
    

        

      
      
    

            

      
      
      
    

        

      
      
    

  
  


  


      

    

    

      
      

        


  

        

      Ivan Krylov
    

    

      
      #
    

  


  

      
On Fri, 18 Oct 2019 15:25:59 -0700

        
"Dalthorp, Daniel via R-help" <r-help at r-project.org> wrote:

        

            

      
      
      
    

        
One way to get some data on this would be to count Scopus hits for
various usages in articles with different affiliations, with a query
like this:

{<usage #1>}
AND PUBYEAR > 1980
AND AFFILCOUNTRY(<country #1>)
AND NOT (
 AFFILCOUNTRY(<country #2>)
 OR {<usage #2>} OR {<usage #3>} OR {<usage #4>}
)

The year cutoff is here to show only the "modern usage" (the trends
look the same whether I leave it in or not). Intersections (papers with
authors from both countries and/or using more than one form) are a
minority and don't seem to reverse any trends, either. Here are the
results:

              UK    US
chi-square  4666 30159
chi-squared 1374  4798
chi square   769  3844
chi squared  142   197

"chi-square" seems to be the most popular form.

  
    

      
      
    

  


  


      

    

    

      
      

        


  

        

      Dalthorp, Daniel
    

    

      
      #
    

  


  

      
Interesting!

Both UK and US show quite a bit more "square" than "squared" in your
sample, with maybe an even stronger tendency for "square" in US than in UK.

I like "squared" better because it fits better with standard English
pronunciation of, say, \sigma^2 or x^2 or r^2.

        
On Mon, Oct 21, 2019 at 9:46 AM Ivan Krylov <krylov.r00t at gmail.com> wrote: