Date:         Thu, 31 Oct 2002 12:38:40 +0000
Reply-To:     "R. Allan Reese" <R.A.Reese@gri.hull.ac.uk>
Sender:       "SPSSX(r) Discussion" <SPSSX-L@LISTSERV.UGA.EDU>
From:         "R. Allan Reese" <R.A.Reese@gri.hull.ac.uk>
Subject:      Mult Response - "Misleading cases"
Content-Type: TEXT/PLAIN; charset=US-ASCII

The subject line is based on the title of a book by AP Herbert, and the
message is to draw attention to a feature of Mult Response that is
documented and logical, but so potentially dangerous as to be flagged as
needing attention. It's a feature rather than a bug, but a feature that is
as helpful as Macbeth's witches.

The problem emerged when I showed a client how to use MR with multiple
dichotomies and compared the marginal figures with totals she had
tabulated for individual variables.  MR showed lower counts and a lower
overall total.  MR reported missing cases, though there were none in her
data.  We tried "/MISSING=INCLUDE"  but nothing changed.  MR frequencies
showed the counts she already had; MR crosstabs had different margins.

Cutting straight to the answer, "For a multiple-response group, a case is
considered missing if none of the components has valid values falling
within the tabulated range for that case."  The keywords in subcommands
affect the treatment of missing values in the component variables.
Logical, but not what the client wanted or expected.  In the CROSSTABS
command, "/MISSING=INCLUDE" creates additional rows or columns, so there
ought to be an equivalent in MR.

Here's a work-round example:

* Define additional category for "no response selected".
count  norow = rowvar1 to rowvarN (1).
recode norow (0=1)(else=0).
count  nocol = colvar1 to colvarM (1).
recode nocol (0=1)(else=0).
mult response  groups = $row (rowvar1 to rowvarN norow (1))
                        $col (colvar1 to colvarM nocol (1))
   / tables = $row by $col.

Request to SPSS:
It is probably a bad idea to redefine the action of existing syntax,
but treating non-responses as MISSING is inconsistent with the rest of
SPSS.  The output might be adapted to read, instead of "missing cases",
"cases excluded because of no responses in set <name> = <number>", which
would be repeated for each group variable in the table.  The action
achieved above could be provided through  a new option in the groups
subcommand, eg, /groups = $row (rowvar1 to rowvarN (1)(noresponse)).


R. Allan Reese                       Email: r.a.reese@gri.hull.ac.uk
Associate Manager GRI                Direct voice:   +44 1482 466845
Graduate School                      Voice messages: +44 1482 466844
Hull University, Hull HU6 7RX, UK.   Fax:            +44 1482 466436
====================================================================