You have already stated that the answers that I get in comparing the 37-marker haplotypes of two people are correct. I presume that what you mean by being correct is that you get the same answers. The only way to get those answers is to pretend that there are 37 people, not 2, and that there are 2 characteristics being compared, and not 37. Whether you do this from first principles, or using Correl in Excel, or using Mathematica or some other package is irrelevant. You can do this for all possible pairings of two people out of 600 odd people and it still doesn't alter the fact that you are using a method of calculation that is appropriate for calculating each pairwise correlation coefficient as if there were 37 people instead of 2, each with 2 characteristics instead of 37. Nothing can alter that. Sandy -----Original Message----- From: dna-r1b1c7-bounces@rootsweb.com [mailto:dna-r1b1c7-bounces@rootsweb.com] On Behalf Of Bill Howard Sent: 17 July 2011 12:54 To: dna-r1b1c7@rootsweb.com Subject: [R-M222] Calculation of a correlation coeficient Sandy, You are wrong. Please don't mislead the post readers without understanding what I did and how I did it. You don't use Excel's process on a matrix; you used it on only a string of values. You need to use the approach on a matrix of ROWS of numbers, not just a couple of columns.

You still did it wrong, Sandy. Read my reply again. And please think more carefully about what I wrote. I think we are wasting the readers' time where there are more important issues to which we should be paying attention. (Oops, I see that my full reply did not get on the list. Too bad, because it gave more reasons why you are wrong. I did not state in my full reply that the answers you got were correct, but I did say that you had approached it wrong and that I did not use the CORREL function in Excel, as you contended.) - Bye from Bill Howard On Jul 17, 2011, at 10:05 AM, Alexander Paterson wrote: > You have already stated that the answers that I get in comparing the > 37-marker haplotypes of two people are correct. I presume that what you mean > by being correct is that you get the same answers. > > The only way to get those answers is to pretend that there are 37 people, > not 2, and that there are 2 characteristics being compared, and not 37. > Whether you do this from first principles, or using Correl in Excel, or > using Mathematica or some other package is irrelevant. > > You can do this for all possible pairings of two people out of 600 odd > people and it still doesn't alter the fact that you are using a method of > calculation that is appropriate for calculating each pairwise correlation > coefficient as if there were 37 people instead of 2, each with 2 > characteristics instead of 37. > > Nothing can alter that. > > > Sandy > > > > -----Original Message----- > From: dna-r1b1c7-bounces@rootsweb.com > [mailto:dna-r1b1c7-bounces@rootsweb.com] On Behalf Of Bill Howard > Sent: 17 July 2011 12:54 > To: dna-r1b1c7@rootsweb.com > Subject: [R-M222] Calculation of a correlation coeficient > > Sandy, > > You are wrong. Please don't mislead the post readers without understanding > what I did and how I did it. You don't use Excel's process on a matrix; you > used it on only a string of values. You need to use the approach on a > matrix of ROWS of numbers, not just a couple of columns. > > > R1b1c7 Research and Links: > > http://clanmaclochlainn.com/R1b1c7/ > ------------------------------- > To unsubscribe from the list, please send an email to DNA-R1B1C7-request@rootsweb.com with the word 'unsubscribe' without the quotes in the subject and the body of the message

The fact of the matter is that there is no such thing as a multivariate correlation coefficient. It simply doesn't exist. My guess is that you assumed that such a concept existed and tried to calculate it. Now you're surprised and offended that someone doubts the validity of your calculations. I repeat: there is no such thing as a multivariate correlation coefficient. The only correlation coefficient known to man is that for a bivariate normal distribution. As soon as you move to 3 or more variables, the concept of a correlation coefficient is replaced by something called R-squared (also known as the coefficient of determination), which, in mathematical modelling, is defined as that proportion of the variance of the variable that you are trying to model, that is explained by the model. -----Original Message----- From: dna-r1b1c7-bounces@rootsweb.com [mailto:dna-r1b1c7-bounces@rootsweb.com] On Behalf Of Bill Howard Sent: 17 July 2011 17:00 To: dna-r1b1c7@rootsweb.com Subject: Re: [R-M222] Calculation of a correlation coeficient You still did it wrong, Sandy. Read my reply again. And please think more carefully about what I wrote. I think we are wasting the readers' time where there are more important issues to which we should be paying attention. (Oops, I see that my full reply did not get on the list. Too bad, because it gave more reasons why you are wrong. I did not state in my full reply that the answers you got were correct, but I did say that you had approached it wrong and that I did not use the CORREL function in Excel, as you contended.) - Bye from Bill Howard On Jul 17, 2011, at 10:05 AM, Alexander Paterson wrote: