In the intervening past two weeks, Hillary Clinton testified before the House Select Committee on the events which occurred in Benghazi, Libya in 2012; her 11 hour testimony before a national TV audience was widely praised. Since my last article and the hearing, 7 new polls have been published in Iowa showing a clear increase in support for Clinton; this surge in support could be attributed to the strength of her testimony, but I'm going to present an alternative explanation.
We're going to start in 2008 and work our way into the present. Obama defeated John Edwards and Clinton in the 2008 Democratic Iowa Caucus; Obama won 37.6% to 29.7% to 29.5% respectively. Obama's victory can largely be attributed to his strong support amongst the youth demographics. The table below depicts the range of support by age within the 2008 Iowa Caucus Entrance Poll:
|2008 Iowa Caucus Age Profile||Obama||Edwards||Clinton|
|22%||17-29 years old||57%||14%||11%|
|22%||65 and older||18%||22%||45%|
Source: NY Times 
Hillary Clinton only carried the 65+ age demographic in 2008; in 2016, it depends on the poll. Lets compare two polls taken sequentially before the Benghazi hearing. I've collated the age demographic information manually for each of these polls; I may look into adding age as a tracked demographic to augment the gender and party affiliation currently collected but the different clusters make it difficult to actually compare data as we'll soon find out:
The Loras College poll has Clinton winning each age division by about 35% while the YouGov/CBS News University poll gives Bernie Sanders a 55% lead amongst 18-29 year olds. Bernie draws roughly 18% of his overall support from the 18-29 sample in the YouGov/CBS News poll while that number is just over 6% in a bigger range of 18-34 from Loras College. So which poll is correct? My intuition is to reference the results from 2008 where Clinton dominated the older demographic and got crushed in the younger divisions. Let's see if we can arrive at a more rigorous conclusion by looking at the samples and methodology in more detail:
Survey conducted with a random sample of registered voters (phone numbers drawn from official Iowa Secretary of State official voter file) who voted in either the 2012 or 2014 general election or who had registered since December 1, 2014.
Likely caucus voter was defined as those indicating they were "definitely or very likely" to vote in the 2016 Iowa Caucus. Those indicating they were "somewhat likely" were subjected to further screen question regarding their general interest in politics. Only those indicating they were "very interested" in politics were then accepted within the sample as a likely caucus voter.
The statewide sample was balanced for gender and divided evenly across Iowa’s four congressional districts. Age was balanced to match past caucus entrance polling.
Source: Loras College 
The above excerpt is from the Loras College poll release. Loras College apparently based their sample against historical registration data (which is perfectly reasonable and done by most pollsters at some level).
YouGov/CBS News do essentially the same thing but frame it in a more technical manner:
Respondents were selected for participation from available panel members to be representative of registered voters from each state in terms of age, race, and gender. A propensity score (based upon a case-control logistic regression including age, race, gender, education, born-again status, and party registration) was estimated for each respondent and responding panelists were post-stratifed upon propensity score deciles, and adjusted for differential recontact from the prior wave. A score for likelihood of voting was computed for each respondent based upon past turnout and self-reported likelihood of voting in the presidential primary.
Source: YouGov 
These two polls uses a similar sample methodology, but still portray results which are miles apart. Lets take a look at a third poll, this time from Monmouth University:
Demographic breakdowns were not released for the Monmouth University poll above, so we're only going to look at the toplines, and they clearly seem to align more closely with the Loras College poll rather than the YouGov/CBS News poll. Monmouth University did however provide this sampling statement:
The Monmouth University Poll was sponsored and conducted by the Monmouth University Polling Institute from October 22 to 25, 2015 with a statewide random sample of 400 Iowa voters drawn from a list of registered Democratic voters who voted in at least one of the last two state primary elections and indicate they are likely to attend the Democratic presidential caucuses in February 2016. This includes 300 contacted by a live interviewer on a landline telephone and 100 contacted by a live interviewer on a cell phone, in English. Monmouth is responsible for all aspects of the survey design, data weighting and analysis. Final sample is weighted for age and gender based on state registration list information on the pool of voters who participate in primary elections. Data collection support provided by Braun Research (field) and Aristotle (voter list). For results based on the total sample, one can say with 95% confidence that the error attributable to sampling has a maximum margin of plus or minus 4.9 percentage points (unadjusted for sample design). Sampling error can be larger for sub-groups (see table below). In addition to sampling error, one should bear in mind that question wording and practical difficulties in conducting surveys can introduce error or bias into the findings of opinion polls.
Source: Monmouth University 
I've added emphasis to the excerpt above to highlight the fact the Monmouth University is requiring previous participation in a primary election to qualify for inclusion into this sample. This is flawed in two manners:
(1) There have only been 2 federal level contested primary elections in Iowa since 2012; the 1st congressional district in both 2012 and 2014. Iowa has 4 congressional districts so this sample is essentially excluding 3/4 of the state's potential Democratic Caucus Electorate on the basis that they didn't participate in local primaries or uncontested federal primaries. I would generally assess that to be too restrictive of a sample as people are clearly more likely to vote for a President rather than their local school board.
(2) The filter required participation in the 2012 or 2014 primary election which excludes anybody who was between the age of 15-16 in 2012 or 17-18 in 2014. They then also excluded participants who are not yet eligible to register due to being less than 17.5 years old, but who are eligible to participate in the caucus because they will be 18 before the general.
Who can participate in the caucuses? Any person who is eligible to vote in the state of Iowa and will be at least 18 years old on Election Day, November 8, 2016, may participate in the Iowa Caucuses. These Individuals must reside in the precinct in which they wish to participate, and they must be registered as a Democrat—party registration is available on caucus night.
Source: Iowa Democrats 
The biggest issue with the Monmouth University methodology is #1 because they require primary participation as part of their threshold. Loras College is also guilty of #1, but to a lesser degree. Every pollster in existence is probably partially guilty of #2 as its difficult to get voter lists for minors who haven't proactively registered.
The biggest implication of #1 is the skew toward an older demographic; this skew also likely favors the establishment to a slightly greater degree as the party machinery is largely responsible for off year and local get-out-the-vote efforts. The skew toward age can be mitigated by weighting, but I think it may introduce some form of bias. I think this bias can be seen in the Loras College poll where Bernie Sanders loses the youngest demographic by 30; a result not seen in any other poll.
The Loras College poll also states that they intended to balance ages to "match past caucus entrance polling." It's initially tough to tell whether they did this or not because the divisions from the 2008 Entrance Poll are different than those used by Loras. I've attempted to extrapolate the 2008 groupings for comparison purposes; I just assumed an even linear distribution of each age within each group:
|Age||2008||Loras||YouGov/CBS News||Monmouth Univ.|
Loras did not weight their sample to match the 2008 outcome and as a result their sample skews to the older age groups. The older age groups predominately favor Clinton, so Clinton's lead grew. The inclusion of an older demographic shouldn't be construed as a slight against the Loras College poll because nobody knows what the turnout will be. By the same token, the result isn't particularly relevant because the sample selection strongly implies the topline result. This isn't an earth shattering conclusion; if you include more of a demographic that favors a particular candidate, that candidate will likely do better. I'm going to call this the expected demographic outcome and further define this concept in a future article using gender preference data.
There are four additional polls which we haven't analyzed with respect to their age breakdowns. The ORC International/CNN News poll only provided partial information for the age sub-samples; I can tell you that Hillary won the under-50 group 50-47%, the 50-64 group by 40% and the 65+ group by 49%, all of which seems congruent with the concept that Bernie does better with younger voters. The Gravis Marketing poll did not release any age based demographic crosstabs so they are being excluded entirely from this analysis.
This leaves a Public Policy Polling result and a poll from Douglas Fulmer/Monmouth College/KBUR which both use funky age groupings, but skew older:
The data seems to be divided at this point as to whether Hillary Clinton's new lead will persist, but I think we have two clear variables; Bernie's support amongst younger voters and the youth participation rate. The first can easily be measured by polling while the second is more of a statistical guess. The most recent batch of 7 polls seems to fall within the low youth support for Bernie and low youth turnout, which is Bernie's worst case scenario.
I'll leave you with the age demographics for the last 3 competitive Iowa Caucuses; they are trending younger:
|Age||2000 ||2004 ||2008|
Source: Archive.org/CNN News , CNN News 
Updated on November 10, 2015 at 7:27:34 PM CT