SURVEY Enhancements in P-STAT 2.21 Release 5/6
In release 6, a change was made in the way that GROUP.STUBS with
only a summary section estimates the number of lines needed for
the next group. Instead of an arbitrary number, the groups are
counted as they are printed and the largest number seen so far
is used as an estimate of what is needed for the next group. This
will help maximize the number of groups per page without breaking
a group in the middle.
June 30, 2000 -- New significance test feature
There is an enhancement to the significance test capabilities. This
allows you to test each value of a banner variable against the combina-
tion of the other values.
The same effect can be obtained by creating N new variables where N is
the number of values for a given variable. Each new variable has two
values: the 1's are those cases with the given value and the 2's are
all the other cases. However, this requires the use of PPL and the
printout has twice as many columns as the original variable.
The subcommand takes the usual form of the TEST subcommand with the
addition of COMBINED.VALUES. The following are all legal uses:
TEST PROP COMBINED.VALUES,
TEST PROP COMBINED.VALUES .95,
TEST PROP COMBINED.VALUES .95 Var3 Var4,
TEST PROP COMBINED.VALUES Var3 Var4,
The tests can also be done using the INDEPENDENCE and MEANS tests. If
the REQUIRE subcommand is used to eliminate small cells, the the elimi-
nation is based on the combined NOT the individual values.
There is a choice of the way the printout is labeled so that the results
do not get confused with the regular between cell test or the test
against the row totals. An extra footnote is printed:
Values of each banner variable are tested against the combined other
values.
The footnote can be changed by using SURVEY.LABELS in your value labels
file with a code of 304.
(304) 'Combined values test'
If the test is blank, the footnote is not printed.
The default labeling uses the letter notation at the head of each
column. If a cell in that column is significantly different from the
combined other values, the same letter is used to mark the cell. If the
cell has the higher test values, the letter is in upper case; if not the
letter is in lower case. Since a cell cannot be different from itself,
this makes it immediately apparent that this is a combined test.
An alternative uses the CHARACTER TESTS subcommand to provide two
letters: the first to label the significantly different cells if the
difference is positive and the second if the difference is negative.
(These can be the same letter or character). If this is used, the
letters that usually mark the top of the columns are omitted.
To perform different tests within the same SURVEY command, use the 'NO
TESTS' subcommand to resets all the internal test variables to their
initial settings.
=========================================================
The SURVEY enhancements for release 5 are:
1. The JOIN subcommand which takes the three parts of a
title line and combines them into a single long title has
been enhanced so that you can specify which parts should
be joined.
JOIN T2 left joins all three pieces (LCR) into a
single title left justified for top title line 2.
JOIN T2 left LC joins the left and center pieces on
the left and leaves the right title
on the right of top title line 2.
JOIN T2 right CR leave the left title alone and joins
the center and right title on the right.
JOIN ON left LC and
JOIN ON right CR can be used to control all title lines.
The work on titles included a change so that a PostScript underline
would not extend beyond the last character when the percent character
was turned off. This sometimes caused a single character to be
omitted from the label. This is fixed in release 6.
2. For German language PostScript, a new identifier UMLAUTE.ANSI
has been added which has newer codes for the AOU umlaute characters.
A change has also been made to workaround a problem in older
PostScript translators which causes the umlautes on the first
set of surveys to be ignored.
3. FORMAT 1 in the SURVEY command underlines the labels section of
a table only if there are row totals. FORMAT 4 has been added which
is like FORMAT 1 except that the labels section is always underlined.
In release 6, FORMAT 4 can be used as an identifier as well as a
subcommand.
4. COMPUTE, used to provide an alternate base for row percents,
has been enhanced.
The normal base is either the row totals or the good.n (non-missing)
total for the values of a banner variable.
The vector provided to COMPUTE indicates which column in the table
is to be used as the base for a given calculation.
For example:
COMPUTE 0 0 1 1 1 2 2 2
specifies that the first 2 columns of the table (exclusive of the
row totals) are to have row percents based on the row totals. The
next three columns are to have row percents based on the first column
and the final three columns are to have row percents based on the
second column of the table.
The expansion of the COMPUTE subcommand provides greater control
over the row and column percents. In addition COMPUTE can now be
used in tables where additional columns such as means of a banner
variable have been created.
The numbers in the compute vector refer to the corresponding column
as defined by the ranges of the banner variables. Columns that are
SQUEEZed out are included in the COMPUTE vector. Columns that are
added by banner variable extensions ( M MED T ) are not included. Thus:
BANNER Age (1,2) Income(1,4 M) Education (1,3)
always has 9 columns as far as COMPUTE is concerned even when empty
columns are not to be printed.
The enhanced COMPUTE makes it possible to:
A. Omit percents from selected banner points.
B. Mix row and column percents on a single line.
C. Produce row percents for all banner points and omit some
column percents.
D: Produce column percents for all banner points and omit some
row percents.
--------------------------------------------------------------------------
OMITTING THE PERCENTS FROM A GIVEN COLUMN IN THE TABLE
This is accomplished by using -999 on a column reference.
COMPUTE 0 -999 -999
causes all percents to be omitted from the second and third banner points.
The percents for the row total can be omitted by using the OMIT.PERCENTS
subcommand. (OMIT.PERCENTS ROW.TOTALS)
Note: when a COMPUTE subcommand does not contain values for all banner
points, 0 (base percents on row totals) is assumed for those that
are ommitted.
--------------------------------------------------------------------------
MIXING ROW AND COLUMN PERCENTS: SINGLE ROW OUTPUT
This is accomplished by using -997 to indicate a column that should
have column percents instead of row percents. Since a separate
row for column percents is the default you must also request
ROW PERCENTS.
ROW PERCENTS, COMPUTE -997 -997 1 1 1 2 2 2
OUTPUT LINE contains C C R R R R R R (C=col percent, R=row percent)
In this example the first two banner points will have column percents
while the rest have row percents based on columns 1 and 2.
The statement ROW PERCENTS BASED ON GOOD.N in this situation causes
the two column percents to be based on the good rather than the total N
and has no effect on row percents based on a specific column.
--------------------------------------------------------------------------
MULTIPLE PERCENT OUTPUT
If both row and column percents (ROW COLUMN PERCENTS) are requested,
there are three possible arrangement of the output columns after any
columns which do not have percents (-999) are excluded.
Mixed R&C All C, Some R All R, Some C
C C R R R R R R R R R R R R R R R R R R R R
C C C C C C C C C C C C C C C C C C C C
When both row and column percents are requested, two rows are
printed. Use of -997 to cause a mixed first row is possible but
makes for a confusing printout. It is better to have one print
line with row percents and one with column percents leaving cells
blank where the percents are not desired.
COMPUTE -997 -997 0 0 1 1 2 2 (Mixed R&C)
When there are some columns of the table which should not have
row percents, the solution is the use of -998 in the COMPUTE
vector.
COMPUTE -998 -998 0 0 1 1 2 2 (All C, Some R)
When all of the columns are to have row percents while some of
the column percents are missing requires a more complex solution.
If the row percents are based on the row totals (a 0 in the COMPUTE
vector), -996 is used to indicate that the corresponding column
percent should be omitted. If the row percents are based on columns
other than the row totals, the negative column number must be used.
ROW COLUMN PERCENTS, COMPUTE 0 0 -1 -1 -2 -2 produces
R R R R R R
C C
ROW COLUMN PERCENTS, COMPUTE -996 -996 1 1 -2 2 produces
R R R R R R
C C C
SUMMARY
-999 all percents are omitted for the banner column with this value.
-998 Row percents are omitted, column percents print on a separate line.
Row and column percents must be requested.
-997 Row and column percents mixed on single line, -997 indicates a
column percent. Row percents should be requested.
-996 and -1 to -99
Column percents are omitted. If -996 is used, Row percents are
based on the row totals. If -1 to -99 is used, the
absolute value specifies the column to be used as the base
for the row percents.