# Stupid question about HP50g covariance different than excel - Hewlett Packard

This is a discussion on Stupid question about HP50g covariance different than excel - Hewlett Packard ; hello, Does anyone know why the HP50g returns 6.5 as the covariance (with a linear fit) for [[3 9][2 7][4 12][5 15][6 17]] while excel returns 5.2? It looks like the 50 returns the sigma((xi-meanx)(yi-meany))/(n-1) while excel divides by n. ...

1. ## Stupid question about HP50g covariance different than excel

hello,

Does anyone know why the HP50g returns 6.5 as the covariance (with a linear
fit) for [[3 9][2 7][4 12][5 15][6 17]] while excel returns 5.2?

It looks like the 50 returns the sigma((xi-meanx)(yi-meany))/(n-1) while
excel divides by n. a little bit like the difference between the population
standard deviation and the sameple standard deviation.

regards, cyrille

2. ## Re: Stupid question about HP50g covariance different than excel

On Aug 1, 9:54*am, "cyrille de brebisson" wrote:
> hello,
>
> Does anyone know why the HP50g returns 6.5 as the covariance (with a linear
> fit) for [[3 9][2 7][4 12][5 15][6 17]] while excel returns 5.2?
>
> It looks like the 50 returns the sigma((xi-meanx)(yi-meany))/(n-1) while
> excel divides by n. a little bit like the difference between the population
> standard deviation and the sameple standard deviation.
>
> regards, cyrille

Either answer is fine as long as the user knows what definition of
sample covariance is being used. Some text books divide by n and
others divide by n-1. When you divide by n then the resulting matrix
is the maximum likelihood estimate of the covariance matrix.
Unfortunately, this estimate is biased and so some authors divide by
n-1 which results in an unbiased estimate of the covariance matrix.

--- Bertram