Saturday, March 18, 2017

Chuck Hagel and Nebraska Election Fraud

One of the earliest suspect elections due to electronic voting was the 1996 Nebraska Senate race. Chuck Hagel, a political newcomer who had only recently moved to Nebraska, managed to upset a popular Democratic governor in a 14% landslide, despite the very best polls for Hagel showing a tie. What Hagel failed to disclose was that he was chairman and CEO of AIS (precursor to ES&S), the Omaha company that made the state's voting machines, prior to running for Senate.

Hagel's story is a clear illustration of the disturbing conflicts of interest in the elections industry. And the lengths he went to to hide his connections to AIS suggest something nefarious was going on. So naturally, many election integrity analysts believe Hagel's election was stolen. But on their own, Hagel's upset and conflicts of interest aren't very strong evidence of election fraud. All they offer is reasonable suspicion.

Lulu Friesdat's report, An Electoral System in Crisis, suggests this:
It would be instructive to do a statistical analysis on one of the races that [Victoria Collier] cites as an “up-set” like Chuck Hagel’s 1996 Nebraska Senate victory.
Aside from polling discrepancies, election integrity analysts have two techniques of directly analyzing the vote counts (within a county, city, or town) for irregularities:
  • Cumulative vote share: Sort the precincts by total number of votes. Add them, in that order, to a running vote total, calculating candidates' vote percentages as precincts accumulate. Graph their vote shares by accumulated total votes.
  • Precinct share by size: Sort the precincts by total number of votes. Calculate the candidates' vote percentages in each precinct, and simply graph them all by precinct size.
These analyses are based on the idea that a candidate's vote share shouldn't be significantly associated with precinct size. In a clean election, these graphs should be relatively flat. On the other hand, if an election is rigged, the fraud is more likely to be targeted towards larger precincts (where tampering is less likely to be caught and has more effect), which would make these graphs show an irregular increase for a certain candidate.

However, it's not as simple as that. There are potential demographic reasons for the larger precincts to more strongly support a candidate. But that doesn't make these analyses (CVS and precinct share) useless. It just means they need to be more carefully controlled.

If a demographic effect is present, it should affect all elections, both suspect and not. We can run these analyses and compare suspect/competitive elections to innocuous/noncompetitive elections. If an irregular trend appears in the former but not the latter, it's harder to dismiss it as a demographic artifact.

For Hagel's 1996 election, we'll look specifically at Douglas County NE, the state's largest county which contains Omaha. In the CVS graphs, irregularities are indicated by upslopes/downslopes after the graph flattens out. All these graphs can be verified from the official Douglas County results. I also have my spreadsheets here.

1987 Omaha mayoral recall

This was a recall election against Omaha mayor Mike Boyle, after firing police chief Robert Wadman for insubordination. With an upslope of ~10% for recalling Boyle (and corresponding downslope for opposing it), the CVS graph seems fairly regular. The support for recalling Boyle only has a minor linear increase with precinct size.

1996 Republican presidential primary

An upslope of ~10% for Dole (and corresponding downslope for Buchanan) can be observed in the CVS graph. For both candidates, vote share is slightly linearly associated with precinct size, though it's close to flat. Neither graph is irregular, as would be expected in a noncompetitive election like this.

1996 Republican Senate primary

Similar to the GOP presidential primary, the GOP Senate primary exhibits a ~10% upslope for Hagel, along with a slight linear increase for Hagel with precinct size (and corresponding decreases for Stenberg). Hagel's victory in this primary over popular attorney general Don Stenberg was one of his upsets considered suspect, but it shows little signs of irregularity.

1996 Senate election

The pattern in Hagel's general election against Ben Nelson is more irregular than any of the previous ones. In the CVS graph, his upslope is ~15%, and it starts much more abruptly than any of the previous ones. Uniquely, it actually reverses the initial trend of Nelson winning. The precinct share graph is also different: rather than a slight linear pattern, there's a very well-defined upward curve for Hagel with precinct size. This fits the suspicion that Hagel's Senate victory was due to vote tampering.

2002 Democratic Senate primary

Both the CVS graph and precinct share graphs are almost perfectly flat. This lack of irregularities makes sense given the uncompetitiveness of the election, in which Matulka won by 65%.

2002 Senate election

Hagel's 2002 reelection is the most irregular of any of these. There's a massive ~20% upslope for Hagel (and corresponding downslope for Matulka) in the CVS graph, once again beginning quite abruptly, along with the curved pattern in the precinct vote share graph seen in 1996. However, this election still seemed to be a shoo-in for Hagel: it wasn't considered competitive, and the CVS initially flattened out at 60%, a landslide in its own right. So it might seem weird that this race would be a target for fraud. But given Hagel's presidential aspirations, and that this election gave him the largest margin of victory in Nebraska's history, it would have served a beneficial political purpose for Hagel.

2004 Democratic presidential primary

Like the 2002 Dem Senate primary, both graphs are almost perfectly flat, showing no irregularities. This makes sense, since Kerry won by 75%, indicating the election wasn't competitive.

2004 presidential election

This election has some signs of irregularities, though they're weaker. Bush's upslope is ~12%, slightly larger than normal, and the curve seen in precinct share graphs for Hagel's 1996 and 2002 elections appears here as well. It might seem strange that a solid-red state like Nebraska would be rigged in a presidential election. However, the 2004 presidential election had swing states like Ohio and Florida being rigged, as well as noncompetitive states to pad Bush's popular vote margin, so this strategy would make sense. According to the unadjusted exit polls, Nebraska was red-shifted by 3%.

2006 Senate election

In the noncompetitive 2006 Senate election, where Ben Nelson led substantially, there was only a ~10% upslope for his Republican opponent. The precinct share graph, meanwhile, showed a slight linear decrease for Nelson and a slight increase for his opponent.

2008 Senate election

This noncompetitive 2008 Senate race, where Scott Kleeb lost by about 18% to his Republican opponent, showed very little signs of irregularities. The Republican upslope was even less than 10%, and the precinct share graphs show amorphous blobs, giving vote share no meaningful association with precinct size.


As we can see, Hagel's 1996 and 2002 elections, as well as possibly Bush's 2004 election, are the only examined Douglas County elections showing statistical irregularities. And those are some of the only elections where fraud would be likely/expected.

Democratic primaries (2002 Senate and 2004 presidential) show completely flat trends. Republican primaries and all general elections (aside from the aforementioned three) have minor (10% or less) CVS upslopes for the Republican, with linear or unassociated trends in the precinct share graphs. Almost all of these elections were noncompetitive and weren't suspect, so fraud wouldn't be expected. The 10% upslope is likely a demographic effect.

Both of Hagel's general elections, along with Bush's 2004 election, had markedly different patterns. They all exhibited CVS upslopes greater than 10% for the Republican, and rather than a linear or flat precinct vote share pattern, it was a noticeable upward curve for the Republican. All of these elections have reasons to be considered suspect: 1996 because of Hagel's shocking upset and AIS conflict of interest, 2002 because of Hagel's continued AIS ties and political ambitions, and 2004 because of the exit poll red shift and the fraudulent padding of the popular vote for Bush.

Hagel's 1996 election has been considered one of the first electronically stolen races. However, this suspicion has mainly been based on his polling upset, conflict of interest with AIS, and suspicious ethical violations (failing to disclose his AIS ties). The fact that Hagel's elections have an irregular statistical pattern that shows up only in suspect Douglas County elections offers much stronger proof that Hagel's Senate race was rigged.

It is worth noting that Hagel still would have won Douglas County in 1996 had the normal 10% upslope appeared. But it would have been significantly closer, about 49% to 47% for Hagel. And given the reasonable assumption that other Nebraska counties were similarly rigged in favor of Hagel, it's quite likely that his 1996 Senate victory was indeed stolen.


One last interesting thing to note, although it may be nothing, is that the 1996 precinct-level results on the Douglas County elections website were initially corrupt. When I tried downloading the PDF, it wouldn't open. I had to email the elections department, after which they gave me a working copy and fixed it on their website. Perhaps it was simply coincidence, but I found it quite odd that the results of Hagel's suspect Senate election was the one file on their site that happened to be corrupted.