An attentive reader noticed my reference
yesterday to David Yepsen's report
that "only 2 percent of likely Democratic caucusgoers" surveyed in the recent Des Moines Register poll
"are under age 25." My reader asked simply, "is 2% under 25 a problem?" Put
another way, my reader is asking, what is the expected age composition of
"likely caucus goers" and how does this poll compare? Unfortunately, for
reasons I'll explain below, that is a tough question to answer, although the 2%
estimate does appear low in comparison to 2004.
Let's set aside for a moment the nearly impossible task of
guessing the demographics of the voters that will turn out this coming January.
Consider instead what ought to be a much easier question: What were the
demographics of the Democratic caucus goers in 2004? Even that question, it turns
out, leads to series of riddles.
The table below shows the age composition of the Democratic
caucus participants measured in two ways. On the left, courtesy of the Iowa
Democratic Party, we have the age composition of the actual participants, based
on matching those who signed in at their local precincts to the list of
registered voters provided by the Iowa Secretary of State (which includes the
age of each voter). On the right is the age distribution of voters surveyed in
the network "entrance poll."
The caucus goers interviewed in the entrance poll are
significantly younger than those on the list of past caucus goers. But wait! The
reported ages are (roughly) four years apart. The entrance poll shows the age
of the respondent on Caucus day 2004, but the data from the voter list reflects
the ages of 2004 participants as of today.
In other words, those who were 17 to 24 years old in 2004 are now mostly age
21-25. So the 17-24 category in the age breakout on the left is missing at
least half of those who were 17-24 four years ago (I don't have access to the
file, and so cannot attempt to recalculate ages to reflect their age as of
However, the four year shift in age does not appear to
explain everything, and the roughly ten percentage point difference carries
though to the over 50 category.
In thinking about the differences between these two age
estimates, we should probably consider some of the potential shortcomings of
both data sources.
Let's start with the entrance
poll, conducted by the National
Election Pool. In many ways, their procedures for the Iowa caucuses are similar to those used for
election exit polls. They start with a random sample of precincts (50 in this
case) and send interviewers out to each location with paper questionnaires to
be filled out by randomly selected voters.
In Iowa, however, the
unique nature of the Iowa
caucuses require different procedures. Since the caucuses are meetings that
begin at 6:30 p.m., all voters arrive shortly before that hour rather than
streaming in throughout the day. The pollsters send two interviewers to each
caucus location with the task of gathering as many completed interviews as
possible as the participants arrive. We call it an entrance poll since they interview
voters on their way in.
Given the time crunch, they do not attempt to record the observed
gender, race and approximate age of those who refuse to be interviewed. They
also have no official headcount to compare to the precinct level result
(results are based on estimated state delegates chosen and reflect two rounds
of voting where the supporters of those with less than fifteen percent of the
vote are forced to "realign"
with a different candidate). So the pollsters cannot use their standard
procedures to attempt to "correct" either the demographics at the precinct
level (against their observations of all randomly selected voters) or the vote
preferences of sampled voters (at sampled precincts or in the state as a
Moreover, many of us learned in the aftermath of the 2004
exit pool controversies about a potential source of error that might introduce
error into their age estimates. In 2004, at least, exit pollsters depended
mostly on younger interviewers that had a much
harder time completing interviews than their older colleagues. So the
potential for a skew to younger respondents is real, especially without any
record of the approximate age of voters that refuse to participate in the
We should also consider that the adults streaming into each
caucus location included some out of state organizers (most famously the
thousands of young Howard Dean Perfect
Storm volunteers) and others who could not participate in the caucus vote
but may have been willing to fill out an exit poll questionnaire when
Now consider the actual
vote history data. The Iowa
Democratic Party reports that 124,000 Iowans participated in the 2004
Democatic caucus, but at least three campaigns have confirmed for me that on
the vaunted Voter Activation Network (VAN) list maintained by the Party, only
about 95,000 voters are identified as 2004 caucus participants. What happened
to the roughly one participant in four that seems to be missing?
I put that question to Carrie Giddins, the communications
director for the Iowa Democratic Party. Here is her response:
1. There were some caucus goers
from 2004 who did not sign in. In 2004, we had one of the largest caucuses in
history and the sheer volume of people was more than some precincts had
experience dealing with. Because of this, there were caucus goers who did not
get fully signed in before they caucused and therefore we did not have enough
information to identify them after the fact.
2. The New Democratic Caucus
Attendee forms required individuals to write down information about themselves
so that we could identify those caucus goers after the caucuses were over. Not
all of that information was completed in its entirety and some of the completed
forms were simply illegible. This left us no way to identify those caucus goers
in our records.
3. Some caucus goers may not
have been entered into our records due to data entry errors.
4. The VAN is a dynamic environment.
It is not intended to show historically what the attendance was at the 2004
Caucus. Therefore, anyone who has been removed from the SOS rolls, including
people who are deceased or who have moved, is no longer included in the VAN. At
this time, if we turn off the suppressions that we have internally in the VAN
the number of caucus goers goes up by about 7,000 people, so the number of 2004
caucus goers whose status may have changed in 4 years to the extent that they
have been removed from the SOS is likely not an insignificant amount of this
Chris Sullentrop's 2004 dispatch
for Slate confirms that in one Des Moines precinct, there
were "so many" new and first-time caucus participants "that the organizers ran
out of forms to register them."
I should add here that least one 2008 campaign tells me they consider the reported 124,000 turnout too optimistic. They believe the real total was closer to 105,000.
But either way, we can easily hypothesize a number of
reasons why the past caucus goers identified on the VAN list may be older than
those who actually participated. Based on the accounts above, it appears as if
the voters that never signed in were more likely to be first time caucus goers
(who tend to be younger). Other missing voters may have registered elsewhere
since (and more mobile voters tend to be younger). On the other hand, those
2004 participants purged because they are no longer living probably skewed
So what is the precise age composition in 2004? We have a
general sense, but precise percentages are unknowable. What is the "right"
percentage of 17-24 year olds? Your guess is as good as mine, although it was
probably somewhere in the range of 6% to 14%. As such, the 2% on the Register
poll does look a bit low by comparison.
But even if we knew the precise age composition for the 2004
caucuses, we would have only a general sense of the potential demographic
composition this time around. Only 60,000 Iowa Democrats participated in the 2000 caucuses, so using
their demographics as a model would have been misleading four years later.
We also know that several campaigns are spending heavily to identify
and motivate potential supporters among the hundreds of thousands of registered
voters (and potential registrants) that have not participated before. Those
they persuade to caucus will not make up their minds about participating until
the final weeks of the campaign. Thus, when it comes to younger Iowa voters, as
reader "FlyOnTheWall" put
it earlier today, "we don't have the first f---ng clue what voters younger
than 25 are likely to do."
If you remember nothing else about this post, remember this:
Since late July, we have seen 13 different Democratic polls in Iowa taken by eleven
different pollsters. Each pollster does things differently, so we have eleven different conceptions of Iowa's "likely caucus goers."
Take a look at our Iowa chart (above), take into account the up-or-down, overlapping
spread in the results for each candidate, and the only sensible conclusion is
that Iowa is currently a competitive three-way race. The varying conceptions of
the likely electorate create a margin of potential real world error far more
important here than mere sampling error. And of course the results may look
very different in early January.