Escobari and Hoover’s ‘grade difference’ is due to geography, not ‘fraud’

This is the eighth article in a series of articles on Diego Escobari and Gary Hoover’s report on the 2019 Bolivian presidential election. Their conclusions do not stand up to scrutiny, as we see in our report. Nickels to dimes. Here we expand on the various claims and conclusions that Escobari and Hoover make in their article. Links to previous posts: Part one, part two, Part Three, part four, part five, part six as well as part seven.

In a previous post, we explored the confounding effects of rural location and socioeconomic status on naive fraud estimates. We noted that if we could divide polling stations into groups so that confounding effects did not differ within each group, then we could begin to unravel the influence of, say, Internet connectivity on whether a precinct was included in a TSE announcement.

The most obvious way to manage such factors is to assume that voters within small geographical areas are virtually indistinguishable. Of course, the smaller the geographical area, the more truth there will be in this assumption. As a first approximation, we expect that voters in the same precinct will have a more similar socioeconomic status or rural area than voters in different precincts, even within the same municipality. At least we can hope.

Consider column 4 Escobari and Hoover difference scores. There, the polling stations are grouped by municipality. In practice, this meant conducting exactly the same analysis as before, but only after eliminating the average differences in the majority of votes between municipalities. It is important to note that the adjustment takes place at the municipal level in order to maintain differences between polling stations in the same municipality. Since the averages depend on the weighting scheme, the adjustments will be different if we take into account the number of valid votes in each polling station.

In Figure 1, we see the unadjusted and (weighted) adjustments for polling places in two municipalities: New York (USA) and Acasio (Potosi). On the left, we see that New York actively supported Mesa, while Acasio supported Morales. On the right, we have adjusted the margins to only account for differences between municipalities.

figure 1

Official and municipality-adjusted results in New York and Acasio

Sources: TSE and author’s calculations.

In Figure 2, we see how applying the adjustment to all municipalities affects the overall trend. We see the municipality explaining most but not all of the ARRIVAL support trend.

figure 2

Official and municipality-adjusted dynamics of the voting gap

Sources: TSE and author’s calculations.

Our iteration and adjustment of the results adjusted by the municipality are presented along with their reported results in Table 1.

Table 1

Reproducing and reanalyzing Escobari and Hoover’s difference model adjusted for municipality

As published replication Right voters suspended
(one) (2) (3) (four)
variable
MALFUNCTION 7.243 (0.437) 7.243 (0.437) 7.214 (0.436) 7.766 (0.447)
Constant 11.28 (0.162) 11.28 (0.162) 11.31 (0.162) 9.32 (0.166)
Observations 34 529 34 529 34 551 34 551
R2 0.640 0.640 0.640 0.627

Sources: TSE and author’s calculations.

Because municipal adjustments (“fixed effects”) are defined in this particular model, the coefficients are more difficult to interpret, even if polling stations are weighted by size. However, when weighted, the average difference is still in line with the data: 9.32 + 0.16 x 7.7766 = 10.56, where 16 percent of the votes were excluded from the TSE announcement. More importantly, the OFF effect is greatly reduced. More than half of the measured effect came from differences in municipalities, and was not confirmed in the later stages of the count. This provides strong evidence that the problem of nickels versus dimes is a serious issue that needs to be fully considered when evaluating fraud.

As we decrease the size of the group, the differences between polling stations within each group also decrease. Accordingly, there is less variation within each group that can be explained by the exclusion from the TSE declaration. The more we take into account distorting factors, the smaller the effect of OFF. The smallest practical unit we can use is the lot.

In figure 3, we see that the plots almost completely determine the overall trend. Very little remains in the adjusted results.

Figure 3

Official and constituency-adjusted trends in vote margins

Sources: TSE and author’s calculations.

table 2

Reproducing and reanalyzing Escobari and Hoover’s plot-adjusted difference model

As published replication Right voters suspended
(one) (2) (3) (four)
variable
MALFUNCTION 0.365 (0.194) 0.377 (0.194) 0.360 (0.193) 0.287 (0.192)
Constant 12.39 (0.0631) 12.39 (0.0631) 12.41 (0.0630) 10.52 (0.0632)
Observations 34 529 34 529 34 551 34 551
R2 0.958 0.958 0.958 0.958

Sources: TSE and author’s calculations.

Instead of a 16 percentage point increase, we see that, on average, in this divided precinct, the excluded polling stations favored Morales by only an additional 0.3 percentage points—or about 3,000 net votes. This difference is far from politically significant, as the model explains everything but Morales’ final lead of 0.046 percentage points.

Again, this does not mean that 3,000 votes was a scam; this means that we have yet to offer any alternative explanation for the unexpected difference. For example, if the last name is associated with both Morales’ support and reporting delays, the difference could be explained by that. According to Escobari and Hoover, voters with last names beginning with “Z” voted more strongly for Mesa than others. Perhaps such surnames are associated with a certain socio-economic status. In any case, at the end of the alphabet, “Z” voters are assigned to the precincts with the most numbers in that precinct, and therefore tend to show up earlier (taking into account the “small precinct” effect discussed in post #). 3). So we are seeing polling place bias where opposition “Z” voters report disproportionately early and are therefore more likely to be included in the TSE announcement. In turn, this would make us underestimate the support for Morales in the polling stations excluded from the announcement. Worse, Escobari and Hoover do not weight precincts by the number of voters, so smaller precincts with more “Z” names have an undue influence on the analysis, exaggerating the difference.[1]

We do not investigate whether 3,000 votes can be counted in this way or, alternatively, whether other explanations (perhaps including fraud) would be required. However, even if we assume that the entire 3,000-vote difference was due to fraud, that alone would bring Morales’ lead down to 10.52 percentage points—not enough to change the outcome of the election.

This has two important implications. First, geography can do a lot to explain the rise in support for Morales. Even assuming a worst-case interpretation of the results, the officially recorded rise in support for Morales among late polling places in split precincts is very small. Second, because 84 percent of the late polls were in these divided tracts, there is little room for politically sensitive fraud. Escobari and Hoover’s 16.26 percentage points do not provide a reliable estimate of fraud in the late polls.

As an illustration, note that Escobari and Hoover estimate fraud at just under 160,000 votes. If we generously quadruple the number of surprise votes received from late polling stations in divided tracts (12,000 instead of 3,000), there will be 148,000 “rigged” votes among 153,890 valid votes cast in late polling stations. In other words, to be consistent with Escobari and Hoover’s estimate of 16.26 percentage points (see Fig. previous post), late polling places officially reported to have supported Morales by more than 50 percentage points, actually backed Mesa by about 45 percentage points.

Not only would this be a shockingly obvious level of manipulation, it would have to be manipulation of votes in the polls overwhelmingly in favor of Mesa. These will be the precincts where the majority of the jurors will be Mesa voters, where the Mesa voters will witness the ballot counting, and where a representative of the CC (Mesa party) is likely to be present and offer a copy of the act as a security measure. Fraud in favor of Morales does not make sense in these polling stations. And if fake acts were later offered as replacements, then where are the copies of the originals? The OAS audit report indicates that its audit team compared the official records with the copies received, but the report did not mention any discrepancies in the numbers.

In the absence of any credible way to manipulate votes of this magnitude, we fall back on a much more plausible explanation: Escobari and Hoover’s interpretation of their results as a measure of fraud is wrong.

In the next post, we will begin to explore approaches to separating the influence of geography from the influence of possible manipulations at the parcel level.

[1] On the other hand, they report that “Y” voters favored Morales, so the effect could be reversed in many areas.