Wyoming held their caucus over the weekend and we were mostly successful gathering results as they occurred from social media. There were a few hiccups and some topics for discussion with regard to how primary election results are reported and how that could be improved in the next cycle.
The 14 pledged delegates ended up splitting 7-7, which our data correctly implied after some corrections. Errors were to be expected given the nature of the activity and the unreliability of the sources used; literally just random people on the internet. Despite those issues, the data was largely correct, and steps could be taken to automatically mitigate inaccuracies; some of these steps are more achievable than others.
The most obvious fix is to utilize previous results as baselines for comparison on entered data; this is non-trivial. Another notable improvement would be the knowledge of each region's delegate allotment; this happened to be known in Wyoming and was largely used to manually purge false data. In other caucus states however, the delegates in each region were not published prior.
In the case of Wyoming we ran into issues with our projection as the late reporting counties were all relatively small which ensured generally even delegate distributions. This is the reason why Bernie's delegates slide from 9-5, to 8-6 and eventually to 7-7. We also attempted to post updates on twitter, which was an exercise in revisionism as the data kept changing. There was simply too much happening and our forward projections were not automated which led to human error. We're going to stay clear of editorializing real time information in the future.
In general, this experiment proved that election results could be provided in an intuitive manner that aligns with the actual end goal of the primary election process; the collection of delegates. Lets now shift to the perspective where the raw results are provided from an official source; how should that data be reported?
Much of the polling and national reporting of primary coverage has focused on the overall result of a given state. Projections are made at the state level, despite the fact that congressional district outcomes account for most of the delegate allotments. There have been just two states that we're aware of which have reported their results at the congressional district level; Minnesota and Michigan. All primary results should be reported at the congressional district level; in the case of a caucus the associated precinct or county delegate allocations in that district should be used.
Whether the results from each congressional district are then used to extrapolate or allocate state level delegates is a more complex manner. We've defined four methods of reporting which could be used for at-large and PLEO pledged delegates at the state level:
- Report only state delegates which are definitively known; this approach would not make any forward projections. For example, if a congressional district has fully reported, only the delegates associated with that district would be reported.
- Extrapolate current results statewide and ignore unreported districts. If 3 of 7 congressional districts are reporting, the results of the unreported districts are assumed to not exist. This is what our Wyoming results page used with counties instead of congressional districts.
- Extrapolate current statewide results to unreported districts. If 3 of 7 congressional districts are reporting assign the current statewide total to the unreported districts. This process is better able to handle the intricacies of the delegate rounding, but can also give false positives as the unreported districts may not be reflective of the statewide total.
- Make forward predictions about outcomes in unreported districts based upon reported districts.
The general goal of election reporting is to provide accurate data as soon as possible. In general the first method should always be reported because it will never be wrong, it just may not be right quickly. Most media entities adhere to the first process, but they don't make it clear what districts they are reporting which is a problem.
The other 3 processes each have the potential for error, especially in primary states where it isn't always possible to know how many voters have yet to be counted. In caucuses, there is always a finite number of delegates to be assigned so the statewide projections can sometimes be made accurately before all congressional districts report. Only results which are absolutely certain should be reported as fact; there is nothing wrong with using the other 3 processes so long as the methodology and the potential for error is clearly stated.
The usage of delegates in the presidential primary is confusing to most because there are complex, multi-tiered and inconsistent rules which vary by state. This contributes to the difficulty inherent in reporting election results by delegate; each state has to be treated differently. The most obvious example of this is state parties reporting results in regions which are not applicable to delegate allocations. Washington for example reported results at the county level, but counties have no association to their congressional districts.
In the future we intend to thoroughly model the delegate allocation plans of each state to facilitate better reporting. The crowd sourced caucus results we have acquired will aide in this endeavor.