SPSS for newbies: Measuring strength of association in chi square test of independence Video Lecture | SPSS: For Beginners - Data & Analytics

hello everyone Phil from Statistics
mentor.com
and we're continuing the video on
crosstabs and the first part where I
give you a gentle introduction to
crosstabs we introduced the idea of what
crosstabs is useful just to recap it's
used to test whether there is an
association or relationship between two
qualitative variables and then we should
look at an example of smoking and lung
cancer in that video I said that one of
the common things I see students doing a
mistake that students make is to report
that because the p-value is really small
that there is a strong relationship
between say smoking and lung cancer and
I pointed out that is not correct
because the lower the p-value just tells
us that the is a strong evidence of
relationship against the null then I
said that in spss there are options to
measure the strength of relationship if
that's what we want to do and indeed
it's a good idea and that's what I'm
going to do to do it that's the theme of
this here video so the strength of
relationship now the strength of
relationship depends on whether the
measure we use depends on whether our
variables are nominal or ordinal or a
mix of both and here I just got a
lecture slide boring or lecture slide if
the nominal measures of the Association
if yeah if we use a nominal measure of
Association those that measure takes the
value between zero and two one with zero
meaning is no Association closer is to
one that means a stronger the
Association for ordinal measures the
Association is on a scale to minus two
plus one with minus one meaning very
very perfect negative association the
closer is the plus point that means
strong at the positive Association now
if it's it between the two extreme
values zero to one the nominal case or
minus one to plus one in the
case well how do you know how strong it
is or not strong it well it really
depends on the field of application
whether it's like psychology or
engineering or biological sciences so
this is where you look at the journal
papers and see what your peers have kind
of reported in those papers and compare
your values with what they've got and
then on base of that say whether it's
moderate mid strong weak but generally
converse a is a rule of thumb well
sometimes you can say if they lessen
point to that is there's no association
between point two and point six days
weak point six and point two point eight
moderate point eight one is strong but
this is only like a guideline again I
said this differs from field to field of
application right so in SPSS gives us a
whole load of choices for the measure so
if we go back to the data here recall
that to do crosstab score analyze
descriptive statistics and then we put
in the two attributes then we go to
statistics now under statistics we click
on chi-square to get the chi-square test
but then look underneath there is a box
nominal and this a box ordinal each one
of those their whole load of check boxes
we could check each of those basically
given those options on calculating the
strength of relationship between our two
quality variables so here is a table
here is a table to guide you on what to
use now some lecturers prefer to use one
rather than the other one because
there's so many choices is there a right
or wrong so sometimes no but let's have
a look so if we do look nominal you can
use this P of Phi is a Greek word Greek
letter if we're dealing with both
nominal or one is normal one is ordinal
and this only works if you've got a
two-by-two table meaning that one
attribute has two categories like
smoking has two categories because it's
smoking or non-smoking or gender male or
female
Cramer's V ice of Phi but it's for
asymmetric tables meaning that if your
tables are not that square ie 2x2 3x3
4x4 I those two numbers do not match ok
so an example for a symmetric one is
maybe 2 by 3 lambda this is phenomenal
only symmetric or asymmetric tables fine
contingency coefficients for nominal is
minimum maximum values depending on size
at the table carrying ordinal right if
we've got ordinal gamma can apply it
when both are ordinal one is all non
nominal so long as the nominal has more
than two levels they say so in that case
I'll make some I'll just note that if
it's a mix of nominal and ordinal for
using the third case where we've got one
is nominal one is ordinal then the sign
is not meaningful so the sign here not
meaningful so just look at the case we
just look at the magnitude all right if
we go back to our example smoking and
was it smoking and lung cancer
let's do continue to the coefficient
because that's nominal we could also do
Phi because they're both phenomenal okay
so could even use Phi we could use
lambda as well we'd use contingency
coefficient you see so this one saying
there is no right wrong could sometimes
depend on preference so we could here
here use Phi we could use lambda could
use contingency coefficient we don't use
creme is V here because it's for
asymmetric tables we've got a square
table we've got 2 by 2 let's just do
contingency coefficient now analyze
analyze descriptive statistics boxes I
did before and under here we go for
contingency coefficient and say okay
under the contingency coefficient is on
a scale from zero to around one but not
necessarily equal to one so the
chi-square test tells us we reject the
null so there is a relationship how
strong is that relationship let's look
down here symmetric to where it says
symmetric measures continue to the
coefficient that's about your 0.61 five
right so on a scale from zero to one
that's over half isn't it so that's
going to quite moderate to stronger and
is it significant yes okay so that is a
figure we would report so they come to
this this chi-square tells us there's
evidence rivaling some relationship and
this measure here conditions the
coefficient tells us that this say this
moderate moderate least moderately
strong association between lung cancer
and smoking which is significant this
value significantly different from zero
all right so that's a show you some
measures of strength of relationship you
using cross tabs so hope this has been
useful never go okay all for now