Re: exact versus asymptotic p-value under normal circumstances

Posted by Bruce Weaver on
URL: http://spssx-discussion.165.s1.nabble.com/Revision-control-for-SPSS-source-code-tp4267835p4273527.html

I just discovered that Bridgette sent some output to the Hotmail account I use for posting to this list.  (Bridgette, as it says in my sig file, I only use that account for mailing to this list, and check it only occasionally.  If you want to correspond off-list, please use the everyday e-mail address given in my sig file.)  

So, here is syntax to generate her table, and the tests of interest.  Notice that the results I get do not match the ones she sent me.

* Bridgette's data.
data list list / row col kount (3f5.0).
begin data
1 1 11
1 2 16
2 1 14
2 2 35
3 1 8
3 2 21
4 1 23
4 2 27
end data.

var lab
 row "Death personality"
 col "Age group"
.
val lab
 row 1 "Cruel"
     2 "Gentle"
     3 "Impassive"
     4 "Other/variable" /
 col 1 "Under 40"
     2 "40 plus"
.
weight by kount .
crosstabs row by col / stat = chisqr.

* For those who have the Exact Tests module .

CROSSTABS
  /TABLES=row BY col / stat = chisqr
  /METHOD=EXACT TIMER(5).

* My results (using v18 with all patches under Windoze XP Professional):
* Pearson:  Asymptotic p = .213, exact p = .216 .
* Likelihood ratio:  Asymptotic p = .212, exact p = .219 .

* Output Bridgette sent me:
* Pearson:  Asymptotic p = .213, exact p = .028 .
* Likelihood ratio:  Asymptotic p = .212, exact p = .047 .


Bridgette, what version of SPSS are you using?  Have all the available patches been applied?

Also, regarding Age Group, why group?  Do you have the actual ages?

Cheers,
Bruce

Swank, Paul R wrote

Could you share a table with us. We need to compare results with other exact packages.

Dr. Paul R. Swank,
Professor and Director of Research
Children's Learning Institute
University of Texas Health Science Center-Houston


-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Bridgette Portman
Sent: Wednesday, March 30, 2011 6:44 PM
To: [hidden email]
Subject: Re: exact versus asymptotic p-value under normal circumstances

Hi Marta,

No, I'm definitely reading the exact p-value. I find this strange too --
I've spent the last couple days searching the internet for an answer to
this and I can't figure it out. My exact p-values on any table larger than
2x2 are almost invariably smaller than my asymptotic ones. I'm no longer
trusting the exact values, which leaves me unsure how to analyze the
results from tables larger than 2x2 when cell counts are low.

Has anyone used the SPSS exact tests module too? I am really stuck at this
point.

Bridgette



> El 28/03/2011 22:34, Bridgette Portman wrote:
>
>> For example:  On the crosstabs I just ran, X2=4.49, 3 degrees of
>> freedom,
>> asymptotic sig is showing up as .212, and exact sig as .028. All
>> expected
>> cell counts were greater than 5.
>
> Hi Bridgette:
>
> In my experience, exact p values are usually higher than their
> asymptotic counterparts,(and both are very similar when sample sizes are
> big). Therefore, I find strange the great discrepancy between the exact
> & asymptotic p values that you report. Is there any chance that by
> mistake you annotated the "Point probability" instead of the "Exact sig.
> (2-tailed)"?
>
> Best regards,
> Marta GG
>
> =====================
> To manage your subscription to SPSSX-L, send a message to
> [hidden email] (not to SPSSX-L), with no body text except the
> command. To leave the list, send the command
> SIGNOFF SPSSX-L
> For a list of commands to manage subscriptions, send the command
> INFO REFCARD
>
>

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD