|HP RPN's : deficiencies in stat summation|
Message #2 Posted by Karl Schneider on 25 June 2004, 1:07 a.m.,
in response to message #1 by bill platt
Hi, Bill --
Wow! That's dredging the archives. Yep, I wrote that, so here's the "short essay" I had in mind:
The usability of statistical-summation functions on RPN HP calc's (not including the RPL HP-28/48/49 and the algebraic-based 27S/17B/17BII) falls a bit short, in my estimation. This basic functionality was implemented more sensibly and thoughtfully even in older Casio and low-end TI calc's, as I see it.
On many HP RPN calculators:
1. Entering 2-variable data pairs puts the values in reverse order.
This can be overcome on the RPN models by entering the values in reverse order, or with x<->y before SIGMA_+, but this is annoying. On algebraic models by HP and others, x INPUT y SIGMA_+ maintains the order of the data pair.
2. 2-variable summation and calculation of standard deviation (SD) are automatically performed, even when unnecessary.
This sometimes causes problems, if "leftover" value(s) are used for y-summation in 1-variable analysis, then SD is calculated. The "debris data" might cause a sqrt(neg) error when the SD function tries to calculate both x_SD and y_SD. Erroneous y-summation registers must be cleared before the x_SD answers can be obtained.
The 32S/32SII get around this by providing separate commands for x_SD and y_SD. The Casio and TI models typically have separate user-selected modes for 1-variable and 2-variable analyses.
3. The ability to specify repeated data or data-pairs is not provided.
By contrast, TI offers "FREQ" for entering repeated data. Casio calc's offer a similar capability.
4. Sample SD's, rather than population SD's, are calculated.
I never quite understood HP's emphasis on sample SD's. I would think that the most common applications for basic stat on a calculator would be a teacher computing mean and SD for class scores, or a student doing the same for laboratory observations. For either of these cases, the population SD would certainly be appropriate. It is no more difficult to obtain a sample SD by first subtracting the mean as a datum, as to obtain a population SD by first adding the mean.
Again, the 32S/32SII/33S offer both population and sample SD's. Menus made this choice easier to provide.
There are other gripes, many of which pertain only to certain models. For instance:
- The HP-41 didn't even have 2-variable stat built in (it could be added with a Stat Pac).
- The 34C/11C/15C omit certain useful functions (e.g., x-estimate) and kluge others together as one function (y-estimate and coefficient of regression). This situation probably stemmed from lack of sufficient available keyboard space.
Much of this is "water under the bridge", but I view this as one instance of something that HP didn't implement as well as its competitors.
-- Karl S.
Edited: 25 June 2004, 1:14 a.m.