TI-89 stats oddity (corr)

[Dave Britten]

I was messing around with some curve fitting on my 89 Titanium, since, while it would probably score dead last if we ranked calculators by usability * documentation quality / feature set, it's got a pretty decent stat package on board. (And it's actually not too shabby for programming, if you don't mind the horrible keyboard layout.)

If you run a linear regression, you get a value for corr. Nothing weird there. If you run a polynomial regression, you get R^2. But if you run one of the intrinsically linear regressions - LnReg, ExpEeg, and PowerReg, you get nothing indicating the quality of the fit.

Is there a mathematical reason for that, or is it just some really specific manifestation of laziness? My TI-86 and HP 48 both give correlation for those regression types, but nothing for polynomial (and the 48 just doesn't do polynomial period). The TI-36X Pro gives r and r^2 for the main four models, and R^2 for polynomial. My Casio fx-9860g Slim behaves similarly to the 36X Pro, and also reports MSe for a number of the models, which I'm not familiar with.

[Gerald H]

MSe is the Mean Squared Error, the average of the squared differences between predicted & actual values.

The 38G has a similar statistic, RELERR.

[Dave Britten]

(05-30-2015 01:55 PM)Gerald H Wrote:  MSe is the Mean Squared Error, the average of the squared differences between predicted & actual values.

The 38G has a similar statistic, RELERR.

Yeah, I dug through the Casio manual just a little while ago and found that. Along with tons of other formulas implemented in various functions within the machine. For as awful as it is to program a Casio (lists are essentially unworkable; doing ANYTHING to an empty list throws a dim error), it's got a great feature set, and rather good manual.

Still not sure what's up with the inconsistent reporting of fit quality across TI machines.