SPSS for newbies: misuse of the Durbin Watson test in regression Video Lecture | SPSS: For Beginners - Data & Analytics

hi this video is only going to take a
few minutes and it's going to save you a
lot of hassle because it's about the
common mistake students make in projects
the incorrect use of Durbin Watson
testing regression and the data comes
from a questionnaire first we'll explain
what Durbin Watson tests does and then
we'll explain why it's not relevant when
you've got data coming from a
questionnaire when we're studying
regression models your lecturer your
professor would have told you a number
of conditions that must be satisfied by
your regression model for your
estimators to be nice namely you might
have heard the acronym blue best linear
unbiased estimate right number of these
assumptions concern the error term
otherwise known as the disturbance term
and one of these assumptions that error
term is that the errors should be
uncorrelated or your professor might
rephrase it another way that if then
correlated and if another assumption of
normality holds then that's the same as
saying that the errors are independent
okay so in other words error should be
uncorrelated or your professor might
have said they should be independent for
our purposes doesn't matter too much now
the Durbin Watson tests whether the
error terms are correlated or not now
it's very loosely speaking but since
we're just doing applied SPSS here
that's all we need to say okay all right
so how do we run it well we first have
to run a regression model to be able to
perform the test so say I've collected
us run out and done a survey I've asked
for to people about their earnings and
their use of work experience and the
years of schooling and I build a very
simple regression model that could help
to explain earnings from experience and
schooling so how to run the regression
well here earnings is a continuous scale
so we're just going to use a standard
regression so go analyze regression
linear then I transfer to the deep end
box
the earnings and the other two as
predictors of independent variables now
to get the Durbin Watson test we go to
statistics and then we run down to where
it says residuals and we check Durbin
Watson and then continue and now okay
where we'll see the Durbin Watson is in
the summary model summary and it's right
here Durbin Watson here the value is two
point six one now don't Watson takes the
values between zero to four inclusive if
it's around two that means we do not
reject the null of no correlation
between the error terms and again I'm
speaking loosely here all right so
that's what we want we want the Durbin
Watson to be very close to two how close
to two is enough because here it's two
point four six is that close enough to
two well just as we had tables for the T
tests we got table for the Durbin Watson
test as well
there's no p-value SPSS doesn't give you
a p-value for the Durbin Watson for you
to conveniently interpret okay but in
any case it's not useful it's not
relevant because we've got questionnaire
data so that moves on to the next point
so why is it not relevant because we've
got questionnaire data well Durbin
Watson requires that the data to have a
natural ordering okay and if your
undergrad a kind of example of that is
time series in time series data comes is
measured through time so for example
interest rates in quarter one quarter
two quarter three quarter four of 2014
say so that's a natural ordering there
you cannot say that quarter one comes
after quarter three okay whereas if we
look at our data here each row
represents a respondent one person who's
filled out the questionnaire this like
the person one this would be like person
three who
to see that this person one comes before
person to and person three this choice
is just arbitrary how I've done it
alright so if I jiggle these jiggle
these keep trying jiggling jiggling it
and might be able to get that the Durbin
Watson test is rejected okay so in other
words it's sensitive power jiggle but
then there's no kind of rationale what
why I should place one person above
another in other words the ordering here
there is no ordering and that comes from
it being a questionnaire okay and that's
it so in a nutshell questionnaire you
don't question it in a survey don't do
the Durbin Watson test okay it's
irrelevant all right that's it