Diplomacy Lessons

John Brady Kiesling, former U.S. Foreign Service Officer

For OSCE/ODIHR's official account of the Armenian elections.

Charting Electoral Fraud:

Turnout Distribution Analysis as a Tool for Election Assessment

Summary: 

An above-average voter turnout in a given precinct is often a signal of electoral fraud. Local oligarchs have a selfish interest in delivering the maximum number of votes for the winning candidate or party in electoral districts they control. Vote-buying, intimidation, and ballot box stuffing raise reported voter turnout as well as the number of votes for the winning candidate, but only in the precincts in which they occur. Most precincts are honest, even in a fraudulent election. It is worth the trouble to compile the turnout and vote count from every precinct into a single spreadsheet. By aggregating precinct-level election results by voter turnout, one can generate a graph of how far voter behavior diverges from the bell-curve distribution of an honest election. This graph gives policy-makers a useful tool for judging whether a given election was free and fair.

Methodology

The 1998 Armenian presidential elections were the starting point for this research.  As an OSCE election observer, I witnessed substantial election fraud (ballot-box stuffing, carousel voting) in some precincts and impeccably honest and transparent procedures in others.  There was an obvious correlation between observed election misconduct and precincts that reported extremely high voter turnout and massive support for the winning candidate. What was less obvious was the extent to which such misconduct distorted the overall election results. This paper is product of my efforts to find a reasonably quick and vivid way to permit policy-makers to evaluate the quality of elections. 

Voter turnout in a free election tends to be relatively consistent from precinct to precinct, with the discrepancies explainable by urban-rural or other obvious differences. Given detailed election data (results from each of several hundred precincts and several hundred thousand voters), in an honest, uncoerced election, one would expect a fairly symmetrical bell curve of vote distribution. Most votes would be cast in voting precincts  where voter turnout clustered around the overall turnout percentage reported for the election as a whole.  

Using precinct voting data from the 2000 U.S. presidential elections in Minnesota and Florida, I graphed the number of votes cast for each candidate against the turnout, grouping precincts in two-percent bands (e.g., summing the votes of all precincts reporting voter turnout between 49 and 51 %).[For details of the method see Note 1]   Figure 1 gives the outcome of my turnout distribution analysis for the  4000 precincts in Minnesota. Total turnout was 2.45 million voters, 75.2% of registered (including late-registering) voters.

                                                            Figure 1

As one would expect in Minnesota, a state with a strong rule-of-law culture, the results were a clean distribution curve with a maximum at 76%, consistent with the official 75.2 % overall voter turnout.  Results for Bush and Gore tracked closely, albeit with a shift to favor Bush as turnout increased.  As an additional test, I graphed on a secondary y-axis (right side scale) the difference in vote percent between Gore and Bush at each level of voter turnout.  

In Florida (Figure 2) the results were similar.  With some 5900 precincts included, and 5.8 million voters, the bell curve was relatively smooth.  The voting peak, at 70% turnout, was slightly higher than the calculated overall turnout of 66.7%.  Deducing the cause of the sharp uptick in the distribution curve for Bush voters is beyond the scope of this paper. 

                                                                        Figure 2

The Armenian Presidential Elections

If Minnesota and presumably Florida show the predictable pattern for an honest election, Figure 3 shows a very different curve, the results of the highly controversial 1996 Armenian presidential elections in which incumbent Levon Ter-Petrosian defeated opposition candidate Vazgen Manukian by a reported 51.9% to 41.1%. 

                                                            Figure 3

The result of graphing 1218 precincts or multiple-precinct communities and 1.25 million valid votes by turnout (grouped in 2.5% bands) was a curve that diverged dramatically from a normal distribution.  The peak of the distribution curve was at 52.5 %, well below the official turnout of 58.8%.  There were numerous spikes.  Voting behavior at the peak of the distribution curve showed a consistent, clear advantage to opposition challenger Manukian.  Only in the areas of elevated turnout did the pattern shift, with a strong correlation between increased turnout and increased margin of victory for the incumbent. 

There are possible explanations for substantial differences in voting behavior and turnout.  Segregating Armenian results by urban/rural and small/large precinct, however, showed no significant difference in voter behavior.  Nor could the urban/rural distinction explain the massive oversupply of votes at unusually high turnouts.  Given the huge percentage of the Armenian population that had migrated to Russia or elsewhere for economic reasons, and the primitive state of most voter registries, turnouts much beyond 70% would have been physically impossible in most areas of the country.

Election night 1996 was marked by a mysterious interruption in the counting process, and thousands of citizens later took to the streets to protest suspected fraud.  The official curiosity of OSCE observers focused on a missing 22,000 ballot coupons in Yerevan.  These coupons were an additional security measure, and their number was of interest because the discrepancy they signaled was just enough to put Ter-Petrosian over the 50% threshold for a first-round victory.  It was assumed, however, that Ter-Petrosian had in any case a wide lead.  The question of the coupons was dropped.

After Ter-Petrosian's ouster in 1998, various Armenian personalities felt freer to disclose their discomfort in having played unethical roles in his reelection.  It became clear, in retrospect, that there were two phases to the election fraud.  The first had been widespread fraud in many precincts on election day.  This had been enough to give Ter-Petrosian a sizeable lead over Manukian, but not enough to cross the 50% threshold.  Late at night, as the shortfall became apparent, senior officials in Yerevan scrambled to produce extra votes.  In their haste to correct the results, they failed to cope with the accounting problem of 22,000 more ballots than coupons.

The 1998 Armenian elections, however, were in significantly less clear.  Figure 4 shows the distribution of votes in the first round, in which Robert Kocharian, the de facto incumbent, won a plurality against multiple rivals, though his main opposition rival, Karen Demirchian, enjoyed a small but consistent lead in lower turnout bands.

                                                            Figure 4         

                                                Table One

That the election was problematic is underscored by Table One, which shows a major increase in voter turnout in 1998 compared to the 1996 presidential elections.  This was followed by an even more striking increase in turnout –109,000 additional voters -- from the first round to the second round run-off with Karen Demirchian.  Election observers detected no increase or if anything a slight decrease in voter numbers from two weeks before.  The turnout distribution chart shows a large discrepancy between officially reported turnout of 68% and the peak of the turnout distribution curve at 57.5%.  Study of Figure 5 (note shift of x-axis direction from previous graphs) shows that the credibility of Kocharian's crushing victory over Demirchian in the run-off was undermined by the near-draw at the peak of the distribution curve.

                                                            Figure 5

The final chart, Figure 6, shows the utility of consolidating first and second round precinct voting data in a single spreadsheet.  Doing so gives the analyst a powerful tool for statistical analysis, by comparing precinct-by-precinct changes in voter turnout.  Figure 6 aggregates precincts based on the number of additional voters they reported in the second round.  Statistically speaking, every new voter (a vote cast in the second-round but not the first) was an additional vote for Kocharian.  In the polling stations in which turnout officially decreased (in effect the precincts that conducted the vote honestly), Kocharian and Demirchian ran a statistical dead heat.  On a secondary axis, with the green dotted line, is plotted the absolute number of precincts reporting a given change in voter turnout, with the maximum at zero change in number of voters. 

                                                            Figure 6

The Fraud Problem

The Kocharian government would have preferred a narrow but plausible victory to a landslide.[3]  Most Armenian electoral precincts had at least one committed opposition member in the electoral commission.  Thus, a plurality of precincts had reasonably fair balloting and accurate counting and reporting, and a further large group of precincts had only minor irregularities -- typically the use of state and parastatal mechanisms to assure that a maximum number of voters arrived at the polls fortified with the promise of cash or other benefits. 

Local officials and clan/tribal/economic leaders, however, have a tradition of justifying their privileged local status by delivering as close as possible to 100 % of the possible votes for the expected winner.  Such a high turnout cannot legitimately be achieved in Armenia, where electoral registers still list hundreds of thousands of Armenians who emigrated and did not vote.  Aggressive fraud, such as large-scale ballot box stuffing, miscounting or invalidation of opposition ballots, or falsified protocols, requires active complicity from election commissions.  In the 1998 Armenian elections, observers found enough ballot boxes containing stacks of identically marked ballots to document the common practice of spending the final minutes of election day forging the signatures of absent voters and casting ballots on their behalf.[4] 

Conclusion

Turnout distribution analysis offers a useful snapshot of the extent to which outside forces have intervened, legally or not, to distort the normal distribution curve of voter behavior.  In the Armenian case, the graphical picture coincides with a reasonably sober assessment of what took place on the ground.  Tested on additional elections and improved by more sophisticated statistical methods, it is a useful complement to the tools already available to election observers to shape the political response of the international community to a fraudulent election.

Access to precinct-level election data is usable electronic form should be a basic minimum demand of any international organization or citizen watchdog group anywhere in the world.  The file should include, at a minimum, region, municipality, and polling precinct identifiers, number of registered voters, number of valid ballots, number of spoiled/invalid ballots, and number of votes for each candidate.  Since election-to-election comparison is a powerful tool for detecting vote fraud, the same precinct names and boundaries should apply from one election to the next, to the extent possible.