SPSS for newbies: correlation between IV and DV does not mean IVs will be significant in regression Video Lecture | SPSS: For Beginners - Data & Analytics

everyone this is Phil from statistics
mention calm in this video I want to
talk about correlation and multiple
linear regression
now it's I've seen that some students
think that if there's a correlation
between two variables of Y and X it must
mean that X is should be in the
regression model I want to show you that
this is not the case so we should be
careful in interpreting correlations or
should I say we should be careful in
using just the correlation the scatter
plots to determine what predictors
should be in the model let's look at the
scatter matrix got a plot first for this
example which is from health care
funding so I've got three variables I
have funding I've got the rate of
disease and I've got visits to clinics
so funding is healthcare funding so we
said so in this example funding is that
IV disease rate that is the DV visits
that's another IV so when I see the
effect of funding and visits to a clinic
on the rate of disease reported bring up
the scatter matrix scatter plot legacy
dialogues scatter plot matrix scatter
take the variables over to the matrix
variables box okay here is a matrix
scatter plot
it shows high positive so let's a
dependent variable is reported disease
rates this on the y-axis over reading
along here so this one shows us that
there is a positive correlation between
reported disease rates and healthcare
funding this one also shows us a
positive correlation between reported
disease and visits to health care center
is both positive and they look pretty
moderately strong at least so I suggest
that and fits a multiple regression and
though both funding and visits as IVs
now this station students think are they
should be from correlations it appears
that they should be significant let's
see if that's the case analyze got fit
multiple regression notes analyze
regression linear dependent variables
disease independent IVs funding and
visits click okay now assuming that
certain conditions of classical linear
regression more satisfied the t-test
will be valid
so there's assume that's the case let's
all over to the coefficients box
healthcare funding visits these are the
IV's healthcare funding is it
significant p value it is 0.9
it's bigger than point naught 5 so it's
not significant visits to health care
provider point naught 4 6 it's on the
boundary so we could say huh you know
use your common sense here say it's
significant would expect that so the
correlation should the scatterplot
showed us a positive correlation between
healthcare funding and reported disease
race but in a multiple regression you
can see that it is not significant so in
other words that it is not a predictor I
can to this model not predictive or
reported
disease rates how can we explain this
well ain't gonn explain it just to say
that with correlations correlation is
just looks at the linear relationship
between two things so it's very very
specific whereas when we're doing two
that multiple regression here we're
looking at a relationship between two
things but controlling for conditioned
on other things in other words we are
taking we'll look at relation between
say two specific things like in this
case we're looking at this coefficient
look at relation between health care on
disease rates but controlling for other
variables such as here visits to health
care provider and other things if I
could included it in the model okay so
that's the difference so we ought the
message here is gotta be careful about
using correlation to determine what IVs
should be in our model or what we expect
on IVs to be in our model just by using
correlation we should use determined IVs
in a model we should use the theory from
whatever field we've obtained the data
from so in this case it should be using
theory from healthcare if we're building
a model an economic model we should be
using economic theory to determine what
should be the IVs and so on right so the
main message to come out this is that
something appears to be correlated it
doesn't mean that they're going to be
significant and of