Because 2014 saw record low turnout for Iowa Democrats, especially among younger voters, this led to a scenario all across Iowa where there more voters showing up at a precinct than the number of delegates allotted for them.
For example, let's say we have two precincts, A and B. Each is expecting 400 people to show up and based on that projection, they have been each been allotted 10 delegates. On caucus night, 600 people show up to A and 400 people show up to B, but it doesn't matter: Each precinct will still only get to send ten delegates.
Overall, most of the precincts that had delegate allotments that were disproportional to the number of voters that showed up tended to be be precincts with higher populations of younger voters and those precincts with delegates that were proportional to the number of voters that showed up tended to not have populations of younger voters.
Considering that Sanders won over 90% of the 17-29 demographic, this is no small point.
This happened in Iowa City's 21st precinct. Iowa City is home to the University of Iowa and this precinct has a lot of student voters. The precinct was to allot nine delegates and were expecting a crowd of no more than 400 people. However on caucus night, 759 people showed up. Of those people, 582 went for Bernie and 172 for Hillary.
Even though, almost twice as many people showed-up as expected, the delegates allocated were still sent forward as if approximately 400 people showed up-- effectively leaving hundreds of mostly Bernie Sanders not having their votes counted.
As I reported earlier, a similar scenario happened in 2008, where then-Senator Barack Obama's vote was undercounted when the final results came in because of similar higher- than-expected turnout.
In 2008, it didn't matter, because Obama had a big enough margin to overcome the undercount. However, in 2016, with the razor-thin margin, it has apparently tilted the outcome, making Bernie settle for a tie instead of a narrow win.