2
the
role
of
administrative
changes
upon
turnout
rates.
Among
the
techniques
used,
the
most
common
by
far
are
the
use
of
aggregate
turnout
data
and
surveys
of
voters.
Aggregate
data
analyses
typically
consist
of
measures
of
the
overall
number
of
votes
counted
during
an
election,
within
a
given
electoral
area,
and
controls
based
upon
aggregate
demographic
and
political
variables
available
for
that
region
(Vercelloti
and
Andersen
2006).
Such
studies
have
strengths
in
demonstrating
actual
change
in
voting
behavior
in
real
world
populations,
but
suffer
from
a
host
of
ecological
inference
problems.
Even
if
turnout
reductions
are
shown,
the
data
cannot
be
forced
to
reveal
what
drives
those
reductions.
Data
may
demonstrate,
for
example,
that
allowing
citizens
to
register
to
vote
in
motor
vehicle
offices
increases
registration
rates
in
areas,
but
it
cannot
say
that
it
is
those
motor
vehicle
centers
that
drive
those
increases.
Lacking
data
on
precisely
how
many
registrations
are
coming
specifically
from
those
offices,
their
impact
on
registration
rates
is
logical,
but
speculative
1
.
Another
way
to
examine
the
effect
of
changes
in
administrative
procedure
is
to
survey
the
population,
asking
them
about
their
voting
behavior
and
personal
characteristics.
Using
logistic
regressions,
researchers
generate
individual
level
predicted
probabilities
that
demonstrate
how
different
regulations
alter
the
probability
of
any
given
person
voting.
This
method,
while
more
precise
on
the
individual
level,
also
suffers
from
drawbacks.
Survey
data
on
voting
relies
upon
individuals
reporting
their
own
prior
voting
behavior
(Rosenstone
and
Wolfinger
1978,
Timpone
1998,
Mycoff
et
al
2007).
These
reports
are
biased
both
by
fallible
memories
and
by
social
desirability.
Some
respondents
may
genuinely
not
remember
whether
they
voted
or
not,
or
whether
they
are
correctly
registered
to
vote
(Berinsky
1999,
Berinsky
2002).
Surveys
generally
have
a
lag
between
the
election
and
when
questions
about
voting
are
asked,
allowing
time
for
even
honest,
motivated
respondents
to
forget
their
own
actions
(for
an
interesting
exception,
see
Barreto
et
al
2007).
Additionally,
and
perhaps
related,
is
a
recurrent
finding
that
survey
data
on
voting
is
almost
always
inflated
by
the
social
desirability
of
having
voted.
Even
those
people
who
may
have
had
good
reasons
for
not
having
voted
often
succumb
to
the
pressure
of
wishing
to
declare
having
voted.
The
act
of
voting
is
often
considered
an
admirable,
socially
desirable
action,
leading
people
to
answer
in
the
affirmative,
even
if
they
were
unable
or
unwilling
to
vote
when
previously
given
the
opportunity.
This
biases
data
by
inflating
the
predicted
turnout
rates
of
voters.
This
would
be
especially
damaging
to
researchers
trying
to
illustrate
reductions
in
turnout
by
people
caused
by
administrative
changes,
because
socially
desirable
responses
mask
the
deterrent
effect
of
the
laws
being
investigated.
Regrettably,
few
alternative
measures
of
voting
behavior
have
been
available,
forcing
researchers
to
become
both
more
creative
and
more
sophisticated
with
their
1
Such data is now available in the 2006 Election Day Survey and through individual
states.
Convention