Applications to research: Field: "Determinants
of Abortion Policy in Developed Nations"?
 What accounts for variations across nations in
abortion policies?
 Proposed ten explicit hypotheses, Examples:
 The greater the proportion of Catholics in a
country, the more conservative the abortion
policy.
 The greater the strength of leftist, noncommunist
parties, the more liberal the abortion policies.
 Your research papers should contain hypotheses which
are equally explicit in FORM as Field's.
 Field's tests of her hypotheses:
 Used data from 29 nations.
 Created abortion policy scale according to legal
grounds for abortion:
 1 = No acceptable reasons
 2 = lifesaving only
 3 = lifesaving and healthpreserving
 4 = medical/social circumstances allowable
 5 = Social reasons other than age and number of
births
 6 = Simply on request of the woman
 Treated ordinal data as interval
 Scale scores varied from Ireland (1) to East
Germany (6)
 Findings for all 29 nations and for 22 nonCommunist
nations:
 ALL 29 Nations: correlation between % Catholic and
abortion scale: r = .47
 NONCOMMUNIST: correlation between % Catholic and
abortion scale: r = .80
 NONCOMMUNIST: correlation between % Socialist in
govt and scale: r = .51
Earlier versions of SPSS
contained a simple PLOT procedure that also produced
regression statistics. SPSS 10 has dropped that simple
procedure, but one can get a plot with regression
statistics. As described by Ms. Wuyi Wang at SPSS, Inc.,
this is the procedure, with illustrations from
me:
 Go to
Graphs>Interactive>Scatterplot
 drag and drop
the variables to the vertical axis and horizontal
axis.
 It should look like
this:
 Rightclick on each
variable to make sure that Scale is
selected
 [SPSS
interprets data without decimals as
"Ordinal"]
 (if the variable is
tagged as Ordinal no regression line will be shown).
 Click on the Fit
tab. Select Regression from the dropdown list
(the default is None).
 Leave all other
settings as default.
Click OK. A
scatterplot made this way has the regression line and
Rsquared.
Assumptions about the distributions of variables involved
in correlational analysis

 For maximum utility in analysis, correlation and
regression assumes that both variables have unimodal,
symmetrical distributions  at least that one or the
other variable is not highly skewed in either
direction

 In a technical sense  and using a term to be
defined explicitly later  both variables are assumed to
approximate a normal distribution, which looks
like this:

 Problems arise if either variable departs from a
normal distribution
 If one variable is skewed away from a normal
distribution, and the other is not, the correlation can
never equal 1.
 If both variables are skewed away from normal, the
relationship is likely to be artificially high.
How to convert skewed distributions to one that are more
"normal"
 Use the COMPUTE command in SPSS to transform
the variable by pulling in the outliers
 See the transformation
of a variable on "CIVIL DISORDER" computed for
nations across the world
