Date: Thu, 27 Jun 2002 10:58:20 -0400
Sender: "SPSSX(r) Discussion" <SPSSX-L@LISTSERV.UGA.EDU>
From: "Arthur J. Kendall" <Arthur.Kendall@VERIZON.NET>
Subject: Re: Rounding in output--accounting for only 100% of cases
Content-Type: text/plain; charset=iso-8859-1
I concur. ASA's Quantitative Literacy project has made efforts to get math
teachers to include data analysis thinking in their curricula.
The disclaimer should be something like
" All percentages are within plus or minus 5 percentage points except as noted.
Of course, the percentages as presented may not sum to exactly 100% due to
rounding." ( Put in 4 or 6 or whatever your 95% confidence level margin of error
is, i.e., 2* standard error)
If you do not anticipate feigned naiveti to "score points", drop the "of
Except when dealing with rare events or close comparisons with very large
denominators, percentages from samples should be from 0% to 100%. Think of the
sample sizes necessary to make .1% meaningful! Aristotelians would consider
the use of the extra decimal positions to be committing the "fallacy of
precision". Cognitive psychologists would consider it obscuring your message by
inducing unnecessary cognitive effort.
"Petersen, Hauke" wrote:
> Hello Patricia,
> regardless of rounding-method, as soon as you have more than 2
> categories, they will not always sum to 100%, unless you tailor
> rounding to that specific table.
> Besides, especially people teaching math should feel perfectly
> comfortable with that kind of rounding error, IMHO.
> -----Urspr|ngliche Nachricht-----
> Von: Patricia Cleland [mailto:Patricia.Cleland@eqao.com]
> Gesendet: Donnerstag, 27. Juni 2002 15:10
> An: SPSSX-L@vm.marist.edu
> Betreff: Rounding in output--accounting for only 100% of cases
> As near as I can see, SPSS rounds up starting at .5 in frequency (pivot)
> tables. That is 95.5=96 and 4.5=5. This has the effect of accounting for
> 101% of your cases. (This is particularly embarrassing when you're
> reporting the results of math assessments. People who teach math notice
> things like that.) What do other people and organizations do, besides
> issuing the 'Totals may not come to 100% due to rounding error" disclaimer?
> Is there a way to get SPSS to use "banker's rounding" in the output? That
> is, .5 rounds to the nearest even number, so that 1.5=2 and 2.5=2. That way
> you round up half the time and down half the time and only account for 100%
> of your cases, i.e., 95.5=96 and 4.5=4 for a total 100.