| |
 |
|
|
Science Forum Index » Statistics - Education Forum » questions about chi square and g test
Page 1 of 1
|
| Author |
Message |
| Guest |
 Posted: Tue Apr 15, 2008 8:28 am |
|
|
|
|
Hi,
I am new to statistics. I got couple questions about chi square and g
independence test. For chi square test, how do I handle all zero
columns in cross tables? Should those columns be included in the
table? all zero columns will give zero expectation for cells in those
columns and cause divided by zero errors.. For g test, what to do if
some cells have zero observations? That will cause log of zero error.
Thanks. |
|
|
| Back to top |
|
| Guest |
| Posted: Tue Apr 15, 2008 2:54 pm |
|
|
|
|
On Apr 15, 7:33 pm, Richard Ulrich <Rich.Ulr...@comcast.net> wrote:
Quote: On Tue, 15 Apr 2008 11:28:19 -0700 (PDT), ouyang....@gmail.com wrote:
Hi,
I am new to statistics. I got couple questions about chi square and g
independence test. For chi square test, how do I handle all zero
columns in cross tables? Should those columns be included in the
table? all zero columns will give zero expectation for cells in those
columns and cause divided by zero errors.. For g test, what to do if
some cells have zero observations? That will cause log of zero error.
For the ordinary contingency table, you drop
any rows or columns that total zero.
--
Rich Ulrich
http://www.pitt.edu/~wpilib/index.html
Thank you. How about zero cells in G test? |
|
|
| Back to top |
|
| Richard Ulrich |
| Posted: Tue Apr 15, 2008 6:33 pm |
|
|
|
Guest
|
On Tue, 15 Apr 2008 11:28:19 -0700 (PDT), ouyang.jie@gmail.com wrote:
Quote: Hi,
I am new to statistics. I got couple questions about chi square and g
independence test. For chi square test, how do I handle all zero
columns in cross tables? Should those columns be included in the
table? all zero columns will give zero expectation for cells in those
columns and cause divided by zero errors.. For g test, what to do if
some cells have zero observations? That will cause log of zero error.
For the ordinary contingency table, you drop
any rows or columns that total zero.
--
Rich Ulrich
http://www.pitt.edu/~wpilib/index.html |
|
|
| Back to top |
|
| Ryan |
| Posted: Wed Apr 16, 2008 1:54 am |
|
|
|
Guest
|
On Apr 15, 8:54 pm, ouyang....@gmail.com wrote:
Quote: On Apr 15, 7:33 pm, Richard Ulrich <Rich.Ulr...@comcast.net> wrote:
On Tue, 15 Apr 2008 11:28:19 -0700 (PDT), ouyang....@gmail.com wrote:
Hi,
I am new to statistics. I got couple questions about chi square and g
independence test. For chi square test, how do I handle all zero
columns in cross tables? Should those columns be included in the
table? all zero columns will give zero expectation for cells in those
columns and cause divided by zero errors.. For g test, what to do if
some cells have zero observations? That will cause log of zero error.
For the ordinary contingency table, you drop
any rows or columns that total zero.
--
Rich Ulrich
http://www.pitt.edu/~wpilib/index.html
Thank you. How about zero cells in G test?- Hide quoted text -
- Show quoted text -
Same issue b/c: G=2$B-t(B[O$B!_(Bln(O/E)]
The only option I can think of would be to drop the cells.
Ryan |
|
|
| Back to top |
|
| Guest |
| Posted: Wed Apr 16, 2008 3:48 am |
|
|
|
|
On Apr 16, 7:54 am, Ryan <Ryan.Andrew.Bl...@gmail.com> wrote:
Quote: On Apr 15, 8:54 pm, ouyang....@gmail.com wrote:
On Apr 15, 7:33 pm, Richard Ulrich <Rich.Ulr...@comcast.net> wrote:
On Tue, 15 Apr 2008 11:28:19 -0700 (PDT), ouyang....@gmail.com wrote:
Hi,
I am new to statistics. I got couple questions about chi square and g
independence test. For chi square test, how do I handle all zero
columns in cross tables? Should those columns be included in the
table? all zero columns will give zero expectation for cells in those
columns and cause divided by zero errors.. For g test, what to do if
some cells have zero observations? That will cause log of zero error.
For the ordinary contingency table, you drop
any rows or columns that total zero.
--
Rich Ulrich
http://www.pitt.edu/~wpilib/index.html
Thank you. How about zero cells in G test?- Hide quoted text -
- Show quoted text -
Same issue b/c: G=2$B-t(B[O$B!_(Bln(O/E)]
The only option I can think of would be to drop the cells.
Ryan
Or put a very small # in the cells I guess. |
|
|
| Back to top |
|
| Bruce Weaver |
| Posted: Wed Apr 16, 2008 4:50 am |
|
|
|
Guest
|
On Apr 15, 8:54 pm, ouyang....@gmail.com wrote:
Quote: On Apr 15, 7:33 pm, Richard Ulrich <Rich.Ulr...@comcast.net> wrote:
For the ordinary contingency table, you drop
any rows or columns that total zero.
--
Rich Ulrich
http://www.pitt.edu/~wpilib/index.html
Thank you. How about zero cells in G test?
The chi-square test is an approximate test. G (aka the likelihood
ratio chi-square) is just an alternative way of computing the test
statistic. So the same advice applies.
--
Bruce Weaver
bweaver@lakeheadu.ca
www.angelfire.com/wv/bwhomedir
"When all else fails, RTFM." |
|
|
| Back to top |
|
| Ray Koopman |
| Posted: Wed Apr 16, 2008 8:01 am |
|
|
|
Guest
|
On Apr 15, 5:54 pm, ouyang....@gmail.com wrote:
Quote: On Apr 15, 7:33 pm, Richard Ulrich <Rich.Ulr...@comcast.net> wrote:
On Tue, 15 Apr 2008 11:28:19 -0700 (PDT), ouyang....@gmail.com wrote:
Hi,
I am new to statistics. I got couple questions about chi square and g
independence test. For chi square test, how do I handle all zero
columns in cross tables? Should those columns be included in the
table? all zero columns will give zero expectation for cells in those
columns and cause divided by zero errors.. For g test, what to do if
some cells have zero observations? That will cause log of zero error.
For the ordinary contingency table, you drop
any rows or columns that total zero.
--
Rich Ulrich
http://www.pitt.edu/~wpilib/index.html
Thank you. How about zero cells in G test?
Individual zero cells are no problem -- they give 0 mathematically,
with no special treatment necessary, because x log x -> 0 as x -> 0.
Think of it as (log x)/(1/x). Both log x and 1/x -> infinity as x ->
0, but 1/x gets bigger faster. |
|
|
| Back to top |
|
| Guest |
| Posted: Wed Apr 16, 2008 2:26 pm |
|
|
|
|
On Apr 16, 2:01 pm, Ray Koopman <koop...@sfu.ca> wrote:
Quote: On Apr 15, 5:54 pm, ouyang....@gmail.com wrote:
On Apr 15, 7:33 pm, Richard Ulrich <Rich.Ulr...@comcast.net> wrote:
On Tue, 15 Apr 2008 11:28:19 -0700 (PDT), ouyang....@gmail.com wrote:
Hi,
I am new to statistics. I got couple questions about chi square and g
independence test. For chi square test, how do I handle all zero
columns in cross tables? Should those columns be included in the
table? all zero columns will give zero expectation for cells in those
columns and cause divided by zero errors.. For g test, what to do if
some cells have zero observations? That will cause log of zero error.
For the ordinary contingency table, you drop
any rows or columns that total zero.
--
Rich Ulrich
http://www.pitt.edu/~wpilib/index.html
Thank you. How about zero cells in G test?
Individual zero cells are no problem -- they give 0 mathematically,
with no special treatment necessary, because x log x -> 0 as x -> 0.
Think of it as (log x)/(1/x). Both log x and 1/x -> infinity as x -
0, but 1/x gets bigger faster.
Got it. Thanks. |
|
|
| Back to top |
|
| Ryan |
| Posted: Thu Apr 17, 2008 4:24 am |
|
|
|
Guest
|
On Apr 16, 2:01 pm, Ray Koopman <koop...@sfu.ca> wrote:
Quote: On Apr 15, 5:54 pm, ouyang....@gmail.com wrote:
On Apr 15, 7:33 pm, Richard Ulrich <Rich.Ulr...@comcast.net> wrote:
On Tue, 15 Apr 2008 11:28:19 -0700 (PDT), ouyang....@gmail.com wrote:
Hi,
I am new to statistics. I got couple questions about chi square and g
independence test. For chi square test, how do I handle all zero
columns in cross tables? Should those columns be included in the
table? all zero columns will give zero expectation for cells in those
columns and cause divided by zero errors.. For g test, what to do if
some cells have zero observations? That will cause log of zero error.
For the ordinary contingency table, you drop
any rows or columns that total zero.
--
Rich Ulrich
http://www.pitt.edu/~wpilib/index.html
Thank you. How about zero cells in G test?
Individual zero cells are no problem -- they give 0 mathematically,
with no special treatment necessary, because x log x -> 0 as x -> 0.
Think of it as (log x)/(1/x). Both log x and 1/x -> infinity as x -
0, but 1/x gets bigger faster.- Hide quoted text -
- Show quoted text -
Makese sense. For xlog(x), x approaches 0 faster than log(x)
approaches negative infinity. Thanks for the correction with this
limit. |
|
|
| Back to top |
|
| |
|
Page 1 of 1
All times are GMT - 5 Hours
The time now is Sun Oct 12, 2008 3:51 pm
|
|