The official results of Michigan's Primary have yet to be published, but the SOS's unofficial results [1] haven't changed or been updated since Monday so we're going to use them for our pollster rankings and delegate allocations.

Bernie won despite polls indicating otherwise:

Based on the results by county, we are also able to project the pledged delegate allocations within Michigan. The graphic below illustrates the projected delegate allocations in each region allocating delegates; the "AL" denotes the at-large delegates and "PL" denotes the Pledged PLEO delegates:

Bernie will end up winning 67 pledged delegates to Clinton's 63. We also published a running projection of the delegate allocations as results were released Tuesday night. For more information about delegate allocations, refer to our prior explanation.

There has been a lot of discussion about why the polling in Michigan was so far off from the final outcome. The polling was actually better in Michigan than it was in South Carolina. A total of 4 polls were accurate within 15% and the Michigan State University poll's topline deviation was single digits with a sampling error of 6.1%. Perhaps the most interesting polling event from Michigan anecdotaly validates the concept that educational institutions which conduct polls within their geographic region have above average performance; two of the three rankings published so far have been led by local universities.

The topline deviation is the absolute difference between each candidate's polled expectation and the actual outcome. The topline deviation is not related to the margin of error so they are not directly comparable. The margin of error, or sampling error, is a measure of a sample's correlation to the overall population.

In South Carolina, there was not a single poll that came within 15% of the final outcome. So despite all the media generated hoopla trying to explain why the polls were wrong, they generally weren't more wrong than previous polling. New Hampshire remains the most accurately polled state, which can generally be explained by the large number of polls conducted within the state; 106 polls were conducted in New Hampshire versus 18 in Michigan.

The main issue with certain polls in Michigan was the atrocious sample composition of the age groupings; youth were highly and illogically undersampled. We're working on a much more in-depth explanation which requires us to do quite a bit of new analysis to make the differing age groupings from each pollster comparable. As a preview, the 18-39 age group composition from the latest Mitchell Research poll was 10.95%; the actual participation of that group, based on exit polling, was 36%. That's why Mitchel Research currently holds the record for the worst performing poll by net deviation.

Here is the ranking table; like previous rankings the gender expectation methodology was used on each pollster's most recent poll with a 30 day cutoff. Polls with blank spaces did not release sufficient information to perform the deviation calculations. These polls are included for reference and are not ranked. Polls toward the top of the table have a lower deviation relative to the final result; a lower deviation implies greater accuracy.

Florida, Illinois, Missouri, North Carolina and Ohio vote on Tuesday. We're still in the process of entering all the polling data for Florida and North Carolina. Based on the current polling data, Clinton is expected to win every state with the exception of Illinois; Clinton was also expected to win Michigan.

Updated on March 13, 2016 at 8:03:07 PM CT