HP 20b statistical distributions
02-26-2015, 06:13 PM (This post was last modified: 02-26-2015 06:17 PM by Dieter.)
Post: #1
 Dieter Senior Member Posts: 2,397 Joined: Dec 2013
HP 20b statistical distributions
I wanted to find out if the 20b could be an option for flashing it into a WP34s. It arrived today and I tried this and that with the original 20b firmware. One of the first things were the statistical distributions. Here I got some unexpected results: while the Normal CDF ("LTND") seems to work fine, the inverse is way off. For instance LTND–1(0,9) returns 1,34479283373 instead of the expected 1,28155156554. Likewise LTND–1(0,05) yields –1,68555786757 instead of –1,64485362695. Other distributions are more or less off either. For instance the Student quantile for 0,9 with 20 dof is returned as 1,32534081283 instead of 1,32534070699. Again, the CDF is fine while the quantile is off.

What's going on there? I cannot remember the 30b had the same problem.

Dieter
02-26-2015, 06:53 PM
Post: #2
 rprosperi Senior Member Posts: 5,022 Joined: Dec 2013
RE: HP 20b statistical distributions
(02-26-2015 06:13 PM)Dieter Wrote:  I wanted to find out if the 20b could be an option for flashing it into a WP34s. It arrived today and I tried this and that with the original 20b firmware. One of the first things were the statistical distributions. Here I got some unexpected results: while the Normal CDF ("LTND") seems to work fine, the inverse is way off. For instance LTND–1(0,9) returns 1,34479283373 instead of the expected 1,28155156554. Likewise LTND–1(0,05) yields –1,68555786757 instead of –1,64485362695. Other distributions are more or less off either. For instance the Student quantile for 0,9 with 20 dof is returned as 1,32534081283 instead of 1,32534070699. Again, the CDF is fine while the quantile is off.

What's going on there? I cannot remember the 30b had the same problem.

Dieter

When introduced, the 30b was described as having some new features (e.g. programming) and also some bug fixes (either Jake's or Gene's article). I think you simply found one .

--Bob Prosperi
02-26-2015, 07:58 PM (This post was last modified: 02-26-2015 08:00 PM by Gerald H.)
Post: #3
 Gerald H Senior Member Posts: 1,458 Joined: May 2014
RE: HP 20b statistical distributions
You might find this interesting:

02-26-2015, 09:10 PM (This post was last modified: 02-27-2015 07:03 AM by Dieter.)
Post: #4
 Dieter Senior Member Posts: 2,397 Joined: Dec 2013
RE: HP 20b statistical distributions
(02-26-2015 07:58 PM)Gerald H Wrote:  You might find this interesting:

Hmmm... unlike your 30b examples, I consistently get wrong results. ;-)

In the meantime I found out that the Normal quantile seems to work fine for probabilities ≤ 1E–14 and in the center between 0,1354 and 0,8646. For other arguments (i.e. most cases with significance for real life) the results are off. I also tried some Student quantiles far out in the tails (with 20 dof), but it did not get better than 5 decimals, or a few more closer to the center. #-)

Dieter

Edit: updated the range where the Normal quantile seems to be correct.
02-26-2015, 10:57 PM
Post: #5
 Dieter Senior Member Posts: 2,397 Joined: Dec 2013
RE: HP 20b statistical distributions
(02-26-2015 06:13 PM)Dieter Wrote:  ...while the Normal CDF ("LTND") seems to work fine, the inverse is way off. For instance LTND–1(0,9) returns 1,34479283373 instead of the expected 1,28155156554.

I just read HP's PDF #27 on "HP 20b probability distributions". Et voilà: page 5 shows an example where exactly this value – LTND–1(0,9) – is calculated. Guess what, the result shown there is the same wrong 1,34 as mentioned above. I now remember I came across this PDF some time ago and wondered why this result was far off. So the error is even recorded in an "official" HP document.

Incredible... #-\

BTW, my 20b's firmware version is 11 25 2008. I wonder if there is a newer bugfixed version.

Dieter
02-26-2015, 11:03 PM
Post: #6
 Marcus von Cube Senior Member Posts: 760 Joined: Dec 2013
RE: HP 20b statistical distributions
(02-26-2015 10:57 PM)Dieter Wrote:  BTW, my 20b's firmware version is 11 25 2008. I wonder if there is a newer bugfixed version.

You can load the 30b firmware into the 20b. It still behaves as a 20b then but should have the bugfixes included. I can mail you a copy.

Marcus von Cube
Wehrheim, Germany
http://www.mvcsys.de
http://wp34s.sf.net
http://mvcsys.de/doc/basic-compare.html
02-26-2015, 11:10 PM
Post: #7
 Dieter Senior Member Posts: 2,397 Joined: Dec 2013
RE: HP 20b statistical distributions
(02-26-2015 11:03 PM)Marcus von Cube Wrote:  You can load the 30b firmware into the 20b. It still behaves as a 20b then but should have the bugfixes included. I can mail you a copy.

That's fine – as soon as I know that my programing cable is not defective. Some time ago (December 2012) you kindly flashed a 30b for me. I also sent you my cable. Do you still remember if it was OK, or did you have to use a different cable?

Dieter
02-27-2015, 10:04 PM
Post: #8
 Marcus von Cube Senior Member Posts: 760 Joined: Dec 2013
RE: HP 20b statistical distributions
(02-26-2015 11:10 PM)Dieter Wrote:  I also sent you my cable. Do you still remember if it was OK, or did you have to use a different cable?

I don't remember. From the recent discussions I conclude that it may have a broken ERASE button or cable connection. ERASE can be easily performed by a simple wire jumper from the top right to the bottom left pin of the connector on the back off the calculator. Short these pins and either issue a reset through the little hole in the back or press ON. This should ready the device for flashing.

Marcus von Cube
Wehrheim, Germany
http://www.mvcsys.de
http://wp34s.sf.net
http://mvcsys.de/doc/basic-compare.html
 « Next Oldest | Next Newest »

User(s) browsing this thread: 1 Guest(s)