Virginia Primary accurately predicted by Bullitics Threshold; Observes late shift in all primaries

By: bullitics
Published On: 2/13/2008 1:46:37 PM

By: Ricardo Rossello, PhD
Yosem Companys, PhD

Note:  Thanks to all RK readers, for using the Bullitics polling mechanism (in the right  hand side column of the RK main page).  

As you may remember, political pundits were stunned when pollsters were unable to predict the New Hampshire primary outcome.  Yet Bullitics successfully predicted not only the outcome but also the trend of this primary in real time (http://www.raisingkaine.com/showDiary.do?diaryId=12231).  Similarly, while many polls grossly mischaracterized the California and South Carolina primaries, Bullitics was successfully able to predict both the trends and outcomes of these contests.  

How could Bullitics achieve these feats?  The secret lies in a novel method called threshold analysis.  
In scientific terms, the threshold analysis is the percentage of unique voters who have voted on a question who rated a particular option by an amount equal to or greater than the threshold amount.  In other words, if a person has a threshold of 50% for Hillary Clinton and 50% for Barack Obama, it means that he or she views Obama and Clinton equally.  If one of the two candidates has a higher than 50% allocation, that means the user is inclined to vote for that candidate.  If a voter has voted multiple times on the question during the period of analysis, the most recent vote is used in this calculation (you can also see the trends and changes as in figure 1, http://www.supertuesdayafterma... ).  Threshold analysis can be performed by registered users of Bullitics.com.

By using threshold analysis, Bullitics was able to predict the Virginia primary results with outstanding accuracy.  The sample size for VA users was the largest of any state thus far (n=1,421).  Figure 1 depicts the daily trends of likely voters that had a threshold for either Clinton or Obama greater than 50% from Feb 8th - Feb 12th.  At 2pm EST (Tuesday), the % of voters that had Clinton at more than 50% allocation was 33%, while Obama had 62%.   Comparing these results to the actual turnout (Clinton 35%, Obama 65%, with 99% of the precincts reporting), suggests that the 50% Threshold analysis, when two options are compared, is an effective way to evaluate an electoral turnout.  In other words, when a voter allocated more than 50% of their vote to Obama or Clinton, they tended to vote for that candidate!

The numbers, however, reveal some intriguing insights.  First, notice that before the February 9th primary results were revealed, Clinton and Obama were in a virtual dead heat.  But once Obama swept through these states, the momentum shifted his way.  The "bump" Obama generated from the Feb 9th primaries was significant; a momentum shift was quantified at 10% points per day!  This suggests that early outcomes may have a significant influence on later ones.  

What is especially surprising, however, is that Bullitics observed a small trend towards Clinton and away from Obama on the day of the primary.  This is consistent with other "late shifts" in favor of Clinton that Bullitics has identified in past primaries.  On average, these shifts are within the range of 7-10%.  In Virgina, however, the shift was too little too late, as the primary results demonstrate.  

But why does this "primary day" effect occur, and why does it seem to benefit Clinton so consistently?  One explanation may be that voters know Clinton better and may feel more comfortable reverting to someone they know than someone they don't.  We will continue to monitor this shift in future primaries, as it could have a decisive effect wherever these are close primaries as was the case in New Hampshire.

feedback@bullitics.com


Comments