SPSS multinomial logistic regression (2 of 2) Video Lecture - SPSS

okay are there going to be two other
things I'm going to do here related
things first thing you might ask me but
look I don't want to compare the
breakfast
how likely someone is going to go for
the breakfast bar relative to the cereal
bar and I kind of compare a breakfast
bar so oatmeal how likely is someone
going to go for the breakfast MA
compared to the oatmeal well mmm the
simplest way to do this well involving
no calculation is to rerun it but set
the reference category for oatmeal say I
want compare now oatmeal I mean
breakfast bar to oatmeal so I don't want
a thing left out a cereal we can tell
SPSS that by going by rerunning it so
analyze regression multinomial logistic
okay look notice here look what
reference category we can choose we can
choose which one it is so that moment is
set by default to last category we want
custom and what we want to do is want to
set oatmeal to the reference how do we
do that well we got to get code for it
so let's go back to the data and just
toggle right so oatmeal this oatmeal on
the third line alright let's just see
what the number that comes up when I
press this button - okay oatmeal is
coded - so we go doesn't matter which
window we do it in we go analyze
regression multinomial logistic
let's say I'm going to click on the
reference category and we're going to
click custom and we want to because we
want oatmeal to set this reference now
before I press ok you're just going to
get a guess what's going to happen to
the boxes here well we're going to
retain breakfast bar yes we're going to
retain oatmeal no instead that oatmeal
should be replaced by cereal let's just
see that happen yep there you go
breakfast bar and cereal so if I want to
compare breakfast bar now to oatmeal
look for a gender that's a negative
number so that tells us that a male
compared to female is less likely to go
for the breakfast bar and the oatmeal
okay
I emphasize again in this video I'm just
looking just at the co-efficiency I am
not focusing on any of the test
statistics I'm not looking at
significance I'm not looking to see if
the model is any good I'm just assuming
that this model is is the best fit okay
now I just want to say that this odds
here not 0.81 not 0.819 I could have got
this same result
from the first regression so I didn't
have to run a second regression I could
have just got it from the first
regression and but then you have to do a
bit of math you have to write some
numbers down and I think that's not
where we need to be so you know we just
want to kind of get the values out but
just know that if say you're in an exam
you've given this you want to compare
cereal bar over better Sparta cereal bar
you ought to calculate the odds you can
actually do that you don't need to rerun
this thing alright okay
that's the first thing that's the first
extension the second one is let me put
this question to you this result that
I've got to get the odds ratio did I
have to kind of run a multinomial
logistic regression
could I have got the same result just
running crosstabs because if you think
about it I've got two categorical
variables no choice of breakfast or
breakfast which has got you know three
things three groups cereal breakfast bar
oatmeal and then I've got gender so
isn't this the same as doing a test of
Association because once you've got the
test of Association you got the table
you can calculate the odds ratio that
way as well can't you the answer is yes
so why don't we do that now just to show
you that I can get the same result what
I want to show you is what do I want to
show you I want to show you that I can
say get the same odds ratio that we
calculated here for males or females for
the going for the breakfast balanced
cereal using crosstabs okay let's do
that
I'm going to run cross steps now so
that's analyse the script tips cross
tabs well by the way before I run it I
want to say to you it doesn't mean that
you cannot just if you want to do
multinomial that you can just do cross
tabs and I'll tell you why
once I've done this so you see why that
multinomial isn't is needed you can't
just do cross tabs instead of
multinomial whenever you need to do
multinomial okay okay I just want to get
these numbers right because I probably
you can't see it that well because one
thing I don't let by SPSS this table
here you can't really enlarge it easily
and still make it look nice let me just
write down the figures here I want to
keep bear in mind this not 0.864
so let me just well I copied the numbers
out you just kind of think about what
we're doing here so I've got the
crosstab table gives me male and female
about a choice of breakfast bar type of
bar got oats and I got cereal and then
I've got the numbers coming in so let's
have a look for then one time seven one
five one five five six five months and
four and then let's calculate the totals
okay total number of males add those
three figures up four to four add those
numbers up four five six right this is
the crosstab so this is the cross steps
table so it's saying 104 104 men who
took breakfast bar okay and say here 155
ladies took oatmeal and calculate the
odds for breakfast bar over cereal bar
okay for men remember the formula
so how it works is like this let's see
mmm notation wise let's make let's
change color let's go
odds M for men all right that is equal
to the probability that they go for the
breakfast bar say breakfast bar / Pulte
here that they go for cereal so we've
got to calculate property that goes for
the breakfast bar 104 out of 4 - 4 go
for that so 104 divided by 4 - 4 the top
line for cereal or proportion of them
well 165 out of 2 and 424 so 165 divided
by 4 to 4 and call that using fractions
rule fractions here that divide the same
is turning upside down so you got four
two four four one six five that cancels
with that so you've got 104 divided by
165 and that's four odds for men and for
breakfast pie over cereals do the same
for females so P breakfast bar over PC
ok ladies better one two seven over four
five six and cereal one seven four over
four five six
I can just
both fractions turn the second one to
divide then being divided upside down so
we've got one two seven one seven four
say em F now let's go over here the odds
ratio here for main competitor ladies
it's going to be that M divided by that
F so it's 104 divided by 165 divided by
this thing here which same as turning
upside down two times one two seven okay
now put that into your calculator okay
I just don't up my calculator I get
their number no point eight six okay
let's just go here that's that's enough
okay that's the odds ratio not point
eight six three six let's go back up
here oh look
not point eight six my Forte's three
decimal places so the crosstabs is
giving you the same result as using them
running the multinomial regression Wow
so does that mean that we whenever we
need multinomial logistic that we can
just run across steps instead answer is
no no here it's fine and the reason is
because in this model in this example
you see my explanatory variables a
categorical
remember the crosstabs is a tool that
can be used when my variables are
categorical so here I've got two
categorical as the Y's categorical
choice breakfast breakfast and so is my
explanatory variable but the multinomial
can also we can have indexes we can have
continuous things for instance here they
group the data age category but if they
had the broad if they're the actual age
of the person so this is continuous then
we could have put that in our model as
well as an explanatory variable as a
predictor whereas you know that this
continues you can put that you know
crosstabs couldn't handle them so to
recap a multinomial logit model justic
model is when your response variable
contains more than two categories in
fact if it's two categories it just
collapse down will be just the same as
the binary logit model so it's only
differs from the logic model if it you
have if the response variable has more
than two outcome outcomes okay and and
if we think of the most choices and if
you'd you're going for one of those
choices those choice could depend on a
set of explanatory variables and those
expander variables could be categorical
it could be continuous or it could be a
mix of them and with this model we can
talk about you through the coefficients
we can talk about how the x's are
related to the choices could actually
also use a mole to obtain probabilities
of here an individual going for a
certain choice so when are you looking
are likely to kind of encounter such
possibility of using such a model is
like when you've got the survey data
from a questionnaire and something that
you're interested in is categorical with
more than two outcomes and those
outcomes there that variable is nominal
it's Ana nominal scale so order doesn't
matter if they're all it does matter
then you would run an ordinal it's just
ordinal
regression religiously regression okay
so you can have go of your own data set
or this data set which is an SPSS it's
called a serial dot SAV okay good luck