|
|
Re: [Fathom & TP] Creating a collection from a frequency table?
Posted:
Nov 19, 2007 11:10 AM
|
|
The original question (if I understand it, correct me, Kevin, if I'm
wrong) was about doing analyses on data when you have the frequency
table but not the cases.
Suppose you know that in your town there are 76,312 Republicans,
55,400 Democrats and 24,800 Greens, Independents, etc. You want to do
some kind of analysis as if you had all 150,000-ish cases, but you
don't want to enter them all.
Wouldn't that be great?? Alas, Your Basic Fathom doesn't do what you
want, however:
* Once upon a time there was a concept called the "Virtual
Collection" that Bill might weigh in on as to whether it might
eventually be resurrected
* You COULD make a collection with 150,000+ cases with an attribute
"affiliation" and a formula involving caseIndex, and although this
would work for modest-sized situations, the large size here would
bring Fathom to its knees, depending on your hardware.
* You can do a lot of what you might want to do in Fathom's Test and
Estimate objects without ever having case-by-case data. For example,
suppose you want to do a Test for Independence to see if party
affiliation is associated with sex (and you have the sex breakdown of
the party numbers above).
> Create a test object
> Choose "Test of Independence"
> Edit the numbers in the box to reflect numbers and names of
categories
> Enter the frequencies
and shazam! You have your test! (You can collect measures from this
test as usual if you want to change things and graph results)
And finally, not actually a response to THIS issue,
Corey mentioned some trick with bar charts. I don't remember what I
was alleged to have said! But it might have to do with the little-
used (so maybe unknown :) feature whereby you double-click the bar-
chart formula -- which is count( ) by default, so you get frequencies
-- and alter it.
So suppose you have a bar chart showing frequencies of males and
females in a Census data set. Here, you have the case-by-case data.
Now you want to show the difference in mean income.
> Double-click count( ) at the bottom of the graph
> edit it to show mean(income)
and now the bar chart compares mean income instead of count. This is
great for many things such as showing the actual pattern detected by
a chisquare procedure and displaying unusual exploratory data: for
example, suppose we didn't want to display the MEAN income but rather
the 67th percentile. Just change the formula!
Best to all, and happy eating,
Tim
------------------------------------
Tim Erickson * Epistemological Engineering
5269 Miles Avenue, Oakland CA 94618 * 510.653.3377
http://www.eeps.com
To unsubscribe, send mail to
majordomo@mathforum.org
In the body of the message, use the following command:
unsubscribe fathomtp
To reach the (actual, human) owner of the list, send your message to:
owner-fathomtp@mathforum.org
|
|
