03-17-2015, 08:09 AM

As the 38G is no longer produced it doesn't have any USPs but may have "Unique Collecting Points". If so in the FAQs here

http://www.hphomeview.com/

you find #56:

56. What is 'RelError'?

The value of RelError is available on all models (38G, 40G, 39G, 40gs, 39gs) but it only appears in the STAT list on the later models. On the HP38G you have to type RelErr into the HOME view to get its value.

Suppose that you have a set of bivariate (2VAR) data which is not linear and you have used the SYMB SETUP view to change to an exponential fit. You may have already noticed that the value of the correlation coefficient does not change and so, on the 38G, there seems to be no way to tell numerically whether your change was justified. The reason for this is that the correlation coefficient is a linear measure and, mathematically, can't be used with other fits. The value of RelErr is there to compare other curves. RelErr is defined as the measure of the relative error in predicted values when compared to data values and its formula is

RelErr = Σ(j = 1, n, (yj - ŷ)^2/Σ(j = 1, n, (yj)^2).

Unlike the correlation coefficient, RelErr is not constrained to a range from zero to 1, with 1 being best. RelErr has no upper limit and the lower the value of RelErr, the better the fit is considered to be.

Do any other calculators have this statistic?

http://www.hphomeview.com/

you find #56:

56. What is 'RelError'?

The value of RelError is available on all models (38G, 40G, 39G, 40gs, 39gs) but it only appears in the STAT list on the later models. On the HP38G you have to type RelErr into the HOME view to get its value.

Suppose that you have a set of bivariate (2VAR) data which is not linear and you have used the SYMB SETUP view to change to an exponential fit. You may have already noticed that the value of the correlation coefficient does not change and so, on the 38G, there seems to be no way to tell numerically whether your change was justified. The reason for this is that the correlation coefficient is a linear measure and, mathematically, can't be used with other fits. The value of RelErr is there to compare other curves. RelErr is defined as the measure of the relative error in predicted values when compared to data values and its formula is

RelErr = Σ(j = 1, n, (yj - ŷ)^2/Σ(j = 1, n, (yj)^2).

Unlike the correlation coefficient, RelErr is not constrained to a range from zero to 1, with 1 being best. RelErr has no upper limit and the lower the value of RelErr, the better the fit is considered to be.

Do any other calculators have this statistic?