Jump to content

  • Quick Navigation
Photo
- - - - -

Correlation coefficients etc


  • You cannot start a new topic
  • Please log in to reply
6 replies to this topic

poppysnoss

    Grade - MIFSQN

  • IFSQN Member
  • 146 posts
  • 10 thanks
1
Neutral

  • United Kingdom
    United Kingdom

Posted 24 March 2010 - 10:52 PM

Hello.

I am currently writing up a project I have been working on for a chemistry test.

As part of the testing, I have drawn up a calibration graph on microsoft excel, whereby I enter results from my standards and this produces a calibration curve.

My problem is I need to state when I would accept the graph and when I would not, ie introduce some sort of regression limit or correlation coefficient and state what would be acceptable or not. :dunno:

I am not statistical minded, but have figured out how to add a correlation coefficient using the formula function on excel based on Pearsons (whoever that is!) However, the result comes out very high ie 0.997, even if I throw a spanner in the works and put a dodgy result in for one of my standards. :doh:

If anyone is still reading this and has even a remote idea what I'm talking about, would you have any idea what acceptable linear criteria should be used for acceptance of a graph? So far, they all look good by eye, but I don't have anything set in stone for when it's not.

Many thanks in advance

Poppy


Charles.C

    Grade - FIFSQN

  • IFSQN Moderator
  • 18,984 posts
  • 5284 thanks
1,256
Excellent

  • Earth
    Earth
  • Gender:Male
  • Interests:SF
    TV
    Movies

Posted 26 March 2010 - 03:44 AM

Dear poppy,

:off_topic:

Sadly I hv absolutely no idea regarding yr technical query but as to yr lack of knowledge on Pearson, Oh Dear. ( some pros and cons - http://en.wikipedia....ki/Karl_Pearson )

How about Fisher ?, famous for experimental designs et al, eg

Fisher is said to have devised the test following a comment from Muriel Bristol, who claimed to be able to detect whether the tea or the milk was added first to her cup;

. Although Fisher was undoubtedly a genius, I fell asleep at the end of the 3rd page/6 of the procedure so I will refrain from posting the originallink.

I do hope somebody can help with yr question.

Rgds / Charles.C

Kind Regards,

 

Charles.C


FSSM

    Grade - MIFSQN

  • IFSQN Member
  • 207 posts
  • 34 thanks
0
Neutral

  • Mexico
    Mexico

Posted 26 March 2010 - 04:23 PM

Maybe it depends on the error that is acceptable for the detection method. Like not more or less than 1 unit, if you find that a calibration x,y pair is not within that range, then, at least that point of the calibration curve might not be acceptable.

Regards,

FSSM



AS NUR

    Grade - PIFSQN

  • IFSQN Principal
  • 582 posts
  • 58 thanks
9
Neutral

  • Indonesia
    Indonesia
  • Gender:Male
  • Location:east java, indonesia

Posted 29 March 2010 - 01:25 AM

Hello.

I am currently writing up a project I have been working on for a chemistry test.

As part of the testing, I have drawn up a calibration graph on microsoft excel, whereby I enter results from my standards and this produces a calibration curve.

My problem is I need to state when I would accept the graph and when I would not, ie introduce some sort of regression limit or correlation coefficient and state what would be acceptable or not. Posted Image

I am not statistical minded, but have figured out how to add a correlation coefficient using the formula function on excel based on Pearsons (whoever that is!) However, the result comes out very high ie 0.997, even if I throw a spanner in the works and put a dodgy result in for one of my standards. Posted Image

If anyone is still reading this and has even a remote idea what I'm talking about, would you have any idea what acceptable linear criteria should be used for acceptance of a graph? So far, they all look good by eye, but I don't have anything set in stone for when it's not.

Many thanks in advance

Poppy



Dear poppy..

As I know if you make graph (calibration graph) between standard solution list (independent parameter (X)) and the result (dependent parameter (Y).. you have to known the correlation between X and Y as linear equation Y = aX + c .. and to accept the graph you have to calculate coefficient of Relation (R square). and R square ~ ±1 thats mean strong relationship between X and Y.. IMEX i decide limit of R square is ±0.8.. so if you mean 0.997 is R square, you can state there is strong corellation between x and Y.. for Me i accept that graph. so you can use the equation to predict unknown sample..

hope can help you


AS Nur


Charles.C

    Grade - FIFSQN

  • IFSQN Moderator
  • 18,984 posts
  • 5284 thanks
1,256
Excellent

  • Earth
    Earth
  • Gender:Male
  • Interests:SF
    TV
    Movies

Posted 31 March 2010 - 05:52 AM

Dear Poppy,

I hv done a little searching for you. I am no statistician nor hv any experience this subject so anyone is welcome to correct or comment on the items below.

I deduce you are trying to interpret how good / usable the fit of yr calculated straight line is to the data.

This extract seemed to be close to what you were seeking –

21. What is the minimum acceptable value of the coefficient of determination (R2)?
It depends on the accuracy required. As a rough rule of thumb, if you need an accuracy of about 0.5%, you need an R2 of 0.9998; if a 1% error is good enough, an R2 of 0.997 will do; and if a 5% error is acceptable, an R2 of 0.97 will do. The bottom line is that the R2 must be pretty darned close to 1.0 for quantitative results in analytical chemistry.

http://terpconnect.umd.edu/~toh/models/CalibrationCurve.html

Below is some further background on this subject.

Since probabilities are involved, there is no black/white answer just like the choice of 90, 95 percents in hypothesis testing. The links typically state that it depends on parameters like yr specific topic and what you are trying to do with the data. The last reference below warns that these various coefficients are often not too “robust” and one must be careful about outliers etc

I noticed 2 popular parameters, the correlation coefficient you already mentioned and the “coefficient of determination” (R squared). (there seems to be some confusion in terminology between authors regarding r, R, corr.coeff., coeff.determ. etc so you need to make sure what each reference is numerically referring to).

Anyway here are the links and you can see if they are useful –

http://mathbits.com/...correlation.htm

Attached File  coefficient of determination.png   119.02KB   0 downloads

http://www.tc3.edu/i...i83/regress.htm

http://cnx.org/content/m17077/latest/
http://cnx.org/content/m17098/latest/
(presumably also possible to generate 99% data if need more “confidence”)

http://en.wikipedia....ion_coefficient
(esp see later section on robustness)

added - and one more :smile: -

http://path.upmc.edu...ers/linear.html

Rgds / Charles.C

Kind Regards,

 

Charles.C


Simon

    IFSQN...it's My Life

  • IFSQN Admin
  • 12,541 posts
  • 1321 thanks
731
Excellent

  • United Kingdom
    United Kingdom
  • Gender:Male
  • Location:Manchester
  • Interests:Married to Michelle, Father of three boys (Oliver, Jacob and Louis). I enjoy cycling, walking and travelling, watching sport, especially football and Manchester United. Oh and I love food and beer and wine.

Posted 06 April 2010 - 07:00 PM

How you doing with this poppy - did you get the answer? I hate stats. :thumbdown:


hand-pointing-down.gif
 
Get FREE bitesize education with IFSQN webinar recordings.
 
Download this handy excel for desktop access to over 140 Food Safety Friday's webinar recordings.
https://www.ifsqn.com/fsf/Free%20Food%20Safety%20Videos.xlsx

 
Check out IFSQN’s extensive library of FREE food safety videos
https://www.ifsqn.com/food_safety_videos.html

 

recommend-us-on-facebook.png


poppysnoss

    Grade - MIFSQN

  • IFSQN Member
  • 146 posts
  • 10 thanks
1
Neutral

  • United Kingdom
    United Kingdom

Posted 06 April 2010 - 09:32 PM

How you doing with this poppy - did you get the answer? I hate stats. :thumbdown:


I'm still working on it, Simon. Have been away for a few days and am just about to throw myself back into it. Deadline is looming... :unsure:

Thanks for the replies above, very much appreciated. I am working my way through them but hate stats sooo much!





0 user(s) are reading this topic

0 members, 0 guests, 0 anonymous users