Extended Analysis of Election 2008

This page is a working document and is not to be quoted or referenced

Alternate Measures

The formal analysis shown in the main Election 2008 page is based on network variance (Netvar) but our extended analysis program includes other measures. The independent Covar statistic addresses the variance of inter-reg correlations, and has a different spectrum of response to some categories of events. For events identified as celebrations, the Covar tends to yield a larger effect than the Netvar measure. This analysis is applied to the formal dataset for the election with the result shown in the following figure. The blue tic-marks show the first results at 7:00 pm ET, and the declaration that Barack Obama had won at 11:00 pm. Obama spoke at about midnight. The "celebratory" trend continues until about 04:00 ET. As we have seen for other celebration events, the Covar does show a larger response than the Netvar.

If we take the Covar result for the whole day of the elections, which begins at 06:00 Eastern Standard Time, we see that the cumulative deviation shows a very steep slope. By the time the outcome is clear, Obama has given his speech and several hours of ongoing celebration have been underway the departure reaches approximately 3 Sigma. Then at about 04:00 the data return to the normal character of a random walk. The time of the declaration for Obama at 11:00 ET is marked with a vertical line.

Another independent measure is the momentary absolute variance among the REG devices (Devvar). The following figure shows a substantial increase in the Devvar beginning at 17:00 ET, 6 hours prior to the announcement that Obama had won the election, peaking about 2 hours prior to the announcement. The variance then shows a decreased level for a few hours before resuming normal fluctuation around expectation. (This analysis is preliminary, and will be replaced as time allows.)

Here is the same analysis extended over a week beginning with Nov 4. This is the device variance plotted as the accumulated deviation from expectation. In the following figure, the mean of this dataset is used to estimate the variance of the REGs. Beginning roughly a day after the election, the deviation of the variance is persistenly large for about two days, after which it resumes the expected level random walk. For a comparison, please note the plot (in green) of this same measure using data from Sept. 11 2001, which up to that time showed the most extreme device variance excursion in the database. We will later have more precise assessments, but as of Nov 16, this is a reasonable picture of what the device variance did over this period of time. There is little question that it was not random, but interpretation must remain difficult. The world situation is hugely complex, with wars and misery in the middle east and Africa, and the economy in dire straits in most countries. The timing relative to the election of Barack Obama is striking, but not definitive.