home
***
CD-ROM
|
disk
|
FTP
|
other
***
search
/
Monster Media 1993 #2
/
Image.iso
/
math
/
ksprob21.zip
/
KSDOCS.EXE
/
KSPDNS.EXE
/
NOT.DOC
< prev
next >
Wrap
Text File
|
1990-05-28
|
7KB
|
127 lines
Creating a series of graphics screens. Joseph C. Hudson
The screens in not1.com - not6.com illustrate an idea in
elementary statistics. I will briefly explain the idea, how
the screens were created, and how they can be used.
1. The idea
Tables of Student's t percentage points often stop at 30 or so
df, with a last line giving standard normal percentiles. It is
implied that when df is above 30, normal percentiles and p values
are adequate approximations to those of the Student's t. The
graphs in not1.scn through not6.scn compare the normal and t
distributions, with not4.scn, not5.scn and not6.scn showing
details of the 32 df t and normal relevant to the impliction
mentioned above.
The series of graphs in not1.com through not6.com explore the
relationship between the t and normal distributions, with a
detailed look at the right tail areas of the normal and 32 df t.
Since programs are available to compute Student's t percentage
points and p values for any df, Why use an approximation that,
for "close calls", can lead to the wrong conclusion? The normal
approximation to the t is not conservative. Better rational
function approximations can be had - see ksprbas.bas for an
example of one.
2. how these files were created.
Once the idea is clear the screen contents can be decided on:
screen contents
not1 std normal pdf and Student's t pdf with 2 df
not2 std normal pdf and Student's t pdf with 8 df
not3 std normal pdf and Student's t pdf with 32 df
not4 std normal and 32 df t reliab fcns (1-cdf) -5 < x < 5
not5 std normal and 32 df t reliab fcns (1-cdf) 1.5 < x < 3
not6 % diff between the fcns shown in not5, 100 * (t - n) / t
The functions plotted on these screens cannot be represented by
simple formulas that plotting programs can take as input. It was
necessary, then, to use kspdat to generate data sets containing
values of the functions and then plot the data.
I first laid out the relationships between the graph files and
data files:
graph file data files needed The *.c## are produced
not1 not.d01, not.d02 by kspdat - they give
not2 not.d01, not.d03 column names and missing
not3 not.d01, not.d04 value info for the cor-
not4 not.d05, not.d06 esponding .d## files.
not5 not.d07, not.d08 The .c## files are not
not6 not.d09 needed here.
You might wonder why not.d01 - not.d04 are separate files. The
kspdat program can easily create a 5 column data file with the
contents of these files as columns. I would have created just one
file insetad of four if the graphing program I used could handle
it that way. It can't. Hence the separate files. (The program -
gnuplot - is really great. occasional quirks like this are not
enough to stop me from using it.)
There are three .spc files that can be read in by kspdat to
produce these output files. notp.spc has the specs for not.d01 -
not.do4. notr1.spc has the specs for not.d05 and not.d06.
notr2.spc has the specs for not.d07 and not.d08.
not.d09 has the common column 1 from not.d07 and not.d08 and a
second column created by combining the second columns in these
two files. Since most spreadsheets and general purpose statistics
programs can do this kind of thing and output not.d09, I won't
describe the (hopefully) obvious here, except to say that if you
output one table instead of two using notr2.spc in kspdat, you
will have a three column data file with the contents of not.d07
and not.d08 that can be read by most stat programs.
With the data files in hand, it only remains to produce the
graphs and the executable files that contain them. For this, I
use gnuplot, version 2.0, and grabber. grabber produces the .com
files from the graphics screens. It is shareware and the current
version costs about $25. gnuplot is copyrighted but free. I don't
see it distributed in the usual shareware channels, but it seems
to be available on academic bbs's. If you can't find it anywhere,
send me a postage-paid return mailer and a blank formatted disk
and I'll be glad to send you the current version I have. (Please
don't send money - I don't want to get involved in any copyright
infringements, no matter how innocent.)
The file not.gnu contains gnuplot commands to produce the 6
graphs. To produce the .com files, I
a. set my vga into cga emulation.
b. loaded grabber.
c. ran gnuplot, used the command load"b:not.gnu to execute the
commands in not.gnu (carriage returning past a meaningless grerr
message at first).
d. as each screen appeared, used grabber's hot key to capture the
screen in a .com file.
e. hit a carriage return to go on to the next screen.
The current version of grabber can do vga screen captures (The
copy I have (and registered) does not) and gnuplot 2.0 does a
gorgeous job in vga mode, so If you have vga, use it for this.
The quality is much much better.
3. How to use the results
You can, of course, simply type the name of the .com files to
display the screens. I find that using dirmagic as a controller
program makes a nice presentation manager. Dirmagic presents a
menu of the files and allows them to be executed by pressing
ctrl-enter.
You can use a cga screen capture program to extract the screens
from the .com files, and then use the screens in a slide show
presentation program, like present or flicker.
It might be nice to automate this process so that a program could
take the commands of kspdat and quickly produce the graphs
directly. Maybe some day.