Date: Thu, 21 Jun 2001 16:23:33 +0100
Reply-To: "Petersen, Hauke" <hauke.petersen@INTERACTIVEMEDIA.DE>
Sender: "SPSSX(r) Discussion" <SPSSX-L@LISTSERV.UGA.EDU>
From: "Petersen, Hauke" <hauke.petersen@INTERACTIVEMEDIA.DE>
Subject: Re: Analyses of Weighted Data in SPSS - Explanation
I am not sure if i remember correctly how crosstabs works,
but the topic of fractional weights came up on this list
some time ago.
As far as i remember, crosstabs uses the nonfractional part
and rounds at random to the next integer with a probability
given by the fractional part.
Von: Gonzalo Kmaid [mailto:gkmaid@INTERNET.COM.UY]
Gesendet am: Donnerstag, 21. Juni 2001 00:24
Betreff: Re: Analyses of Weighted Data in SPSS - Explanation
Shail and Hector,
Sometime ago I posted a question about Spss and weighting.
My concern then, was that old stuff weighted in an old DOS SPSS/PC yielded
slightly different results than the same crosstab but run in SPSS for
I remember some posts talking about some restriccions (no cell with less
than one integer case or something similar) imposed by the "exact tests"
new options. Also, some posts talked about the tables module being able to
reproduce the same results (because there were no restricciones impposed by
the exact test routine).
At 18:40 20/06/01 -0300, you wrote:
>I knew that SPSS rounds (up or down) any weighted frequencies that have
>a fractional part, which only happens if the weights themselves are not
>integers (and this is a good reason to avoid fractional weights). This
>could cause rounding differences or errors for the same total obtained
>in different statistical procedures (e.g. in DIFFERENT tables involving
>different variables). But I cannot fathom why the rounding is different
>for two procedures that should theoretically yield the same result (as
>in the SAME table produced by CROSSTABS or TABLES).
>Universidad del Salvador
>Buenos Aires, Argentina
>Shail Dobson wrote:
>> Sorry I did not give enough information about my observation. Here are
>> details to my earlier e-mail:
>> Version: SPSS 10.1 for windows
>> Data: weighed data
>> Here are syntax:
>> BY weight variable.
>> For Crosstabs:
>> /TABLES= childbirth BY county
>> /FORMAT= AVALUE TABLES
>> /CELLS= COUNT ROW .
>> For Custom Table:
>> * Basic Tables.
>> /FORMAT BLANK MISSING('.')
>> /TABLES childbirth
>> BY country > (STATISTICS)
>> count( ( F5.0 ))
>> cpct( ( PCT5.1 ) 'Row %':country ).
>> My purpose was to do a crosstab between variable childbirth and country.
>> found that the Crosstabs (in Descriptive Statistics) output numbers are
>> slightly different than those produced through in Custom Tables.
>> There is a great likelihood that I am missing something.... I really
>> like to know the reason why numbers are different in these setting if
>> weighted data is used.
>> >-----Original Message-----
>> >From: Shail Dobson [mailto:dobson_shail@HOTMAIL.COM]
>> >Sent: Wednesday, June 20, 2001 12:35 PM
>> >To: SPSSX-L@LISTSERV.UGA.EDU
>> >Subject: Analyses of Weighted Data in SPSS
>> >While analysing weighted data, I made the following observation, hoping
>> >may find it useful.
>> >If you have weighted data, Crosstabs under Descriptive Statistics DO NOT
>> >give the accurate numbers because the Crosstabs option does not take
>> >account the decimal places (it simply round up the number to zero
>> >however, the Customer Tables option produced the more accurate number
>> >because it do includes decimals places.
>> >I wonder why SPSS do that?!
>> Get Your Private, Free E-mail from MSN Hotmail at http://www.hotmail.com.
Carlos Ma Morales 944/801