SPSS for newbies: wrong use of t-statistic/test in multiple linear regression Video Lecture | SPSS: For Beginners - Data & Analytics

hi everyone its Phil from statistics
mentor.com
today we're going to be looking at the
incorrect use of the t-test T statistic
in multiple regression I'm going to
demonstrate this for an example
education data to my dependent variable
earnings which is a scale variable it's
continuous so that means I can apply the
standard regression model and we're
going to regress that the earnings on
years of experience of each person
number of years of schooling the number
of siblings of the person has in their
family and the weight of the person now
what we're going to do is fit the
multiple regression regressing earnings
on these four things so here obviously
we couldn't imagine that weight has no
impact on earnings but I put in here
because I want to show you demonstrate
to you this incorrect use of the T
statistic
all right so analyze regression linear
then we'll move into the dependent
variable box of earnings and then we'll
move this lot into the dependents group
and reselect as a block so it's much
quicker to move over rather than one by
one I do that by pressing the top one
then putting my finger on the the up
arrow and then press weight
okay and then transfer over and then
we'll go okay right the the standard
three tables come up three four table
sorry I'm looking at the coefficients
box here okay here that he stats for
each of the coefficients
that's the intercept and these are the
four independent variables so what we've
got here is multiple regression now
experience is that significant for the
t-test yes cause the p-value is given by
this column of stuff p-value is 0.01 7
so it's less than 0.05 so we reject the
null as insignificant conclude is
significant from zero school
yes significant siblings 0.5 that's
bigger than naught point naught 5 so it
is in significant weight bigger than its
point 2 1 - bigger than not 0.5 so it is
insignificant now this step you might
think why see students doing is they'll
say AHA insignificant insignificant from
0 that means we can drop them so what
they will do then is drop both of them
in one go so we run the regression of
earnings on experience experience
schooling and that's it now that is
wrong because because if you drop these
two I can't look by the T you're
actually treating it as if you are doing
an F test for the joint significance of
a subset of IV of IVs with the T you can
only drop things one at a time okay
that's the correct use of the T I don't
explain why in this video just know as a
rule
so we drop the least significant one
so 0.52 is bigger than 0.2 one they're
both insignificant but this one is less
significant than this one here so we
would drop siblings only so we would
refit it but dropping siblings to
analyze Russian linear will drop
siblings so to take that back to your
side and go okay look at the new
coefficients box simple siblings has
been dropped
check the significance also significant
apart from weight 0.24 so now we would
drop weight linear
okay and now everything is significant
so that that's the brings us to the end
of this video just know that with T if
you're using the t-test you drop things
one at a time drop that and if more than
one IV one of these slope coefficients
is insignificant you drop the least
significant I had the one of the highest
p-value drop that and then refit okay
hope that's helped