note: I started this analysis to try and answer the question what might have happened if we had ranked choice voting or districting in the city. I was not able to actually run that analysis. Still I found the results interesting enough to share.

The Council Election

Mountain View City Council members are elected citywide, at-large. Voters choose N candidates, where N is equal to the number of open seats. In the 2026 election there were three open seats.

This setup makes it interesting, from a statistical standpoint. Unlike a single candidate election we can actually reach some conclusions with a bit of confidence about which candidates are viewed similarly by voters. This is because if two candidates both did well with a cohort there is a higher likelihood voters in that cohort voted for both candidates.

My analysis is below in prose. If you prefer to read the code I have published my notebook.

Geographic Breakdown

We have vote breakdown by precinct from the county elections webpage. Using this we can determine if any geographic correlation between candidates exists.

First we get the total votes per candidate by precinct, then divide by ballots cast in that precinct. Precincts do not cleanly map to neighborhood, but I’ve done my best to add a label to each precinct so it is easier to identify in the results.

Map of precincts with numbers and labels. '0003440': 'Waverly Park', '0003464': 'High School', '0003431': 'Blossom Valley', '0003419': 'Old Mountain View', '0003401': 'Shoreline West', '0003427': 'San Antonio', '0003412': 'Whisman', '0003410': 'East Whisman/Sylvan', '0003409': 'Rex Manor', '0003422': 'Rock/Bayshore', '0003402': 'Monta Loma', '0003407': 'Moffett Blvd', '0003438': 'Cuesta', '0003425': 'Gemello', '0003439': 'Martens/Cuernavaca'

And we can then generate a correlation matrix.

Table showing correlation vote percents for all candidates
Simple correlation matrices of precinct-level vote counts

But are these correlations significant? Let’s look at some scatterplots.

Paul Donahue and Robert Cox have the strongest positive correlation.

Scatterplot showing clear positive linear correlation

Paul Donahue and Eric Poicon have the strongest negative correlation.

Scatterplot showing clear negative linear correlation

Both of these look unlikely to be spurious. Very strong trend line with few outliers. But how about some less strong correlations?

IdaRose Sylvester and Robert Cox have a weak positive correlation, and so do James Kuszmaul and Erik Poicon.

Scatterplot correlation between IdaRose Sylvester and Robert Cox
Scatterplot showing correlation between James Kuszmaul and Erik Poicon

The first scatterplot shows the correlation is unlikely to be significant. A couple of data points change and the trend could invert. With the second it is not as clear cut. The correlation is weak but the data follows the trend line better.

To codify significance we compute the standard deviation and p values for each pair of candidates. Since this involves many comparisons it makes sense to control for the false discovery rate using Benjamini-Hochberg. Doing so we get a chart of which correlations are likely significant.

adjusted p value table for candidate combinations. Seven are below 0.05: Erik Poicon Paul Donahue 0.000121, Erik Poicon      Robert Cox    0.000557,  James Kuszmaul    Paul Donahue    0.018162, James Kuszmaul      Samuel Ali     0.007029, Paul Donahue      Robert Cox     0.003854, Paul Donahue      Samuel Ali    0.042596
Only seven correlations are likely significant (below 0.05)

And we can then apply a mask to the correlation matrix to get only correlations that are likely statistically significant.

Correlation table for candidate combinations which are significant: Erik Poicon Paul Donahue -0.9, Erik Poicon      Robert Cox    -0.9, Erik Poicon James Kuszmaul 0.6, James Kuszmaul    Paul Donahue    -0.7, James Kuszmaul      Samuel Ali     0.8, Paul Donahue      Robert Cox     0.8, Paul Donahue      Samuel Ali    -0.6
IdaRose, Silja and Alex do not have any strong enough signal to establish that the correlations are significant

Mapping it Out

Having established some geographic correlations between candidates, let’s look at some maps. Here is each candidates relative performance in each precinct.

Eight maps showing geographic vote trends for all candidates
Relative performance for each precinct - 1.2 means the candidate received 20% more votes in that precinct relative to their overall performance. 0.8 would indicate 20% fewer, and so on.

The strongest geographic trend in the city is the north/south divide. Erik and James did poorly in the precincts south of El Camino Real. Robert and Paul did well there. IdaRose and Silja did about the same in both parts of the city.

We do not see any geographic trends around SB 79 or California St. SB 79 allows more housing density near transit stops statewide. The areas of the city affected by the measure - Old Mountain View, Moffett, Whisman, San Antonio, etc - did not vote measurable more either for or against the candidates who opposed SB 78: Robert Cox, Paul Donahue and Silja Paymer. Likewise Samuel Ali and Alex Amoroso - the only candidates who oppose the California St changes - did not do better or worse in the precincts around California St. We cannot conclude from this that voters did not care about these issues, only that they were not significant enough to be in themselves a measurable signal.

Since IdaRose and Erik both also ran in 2024, we can compare those results. See if a similar geographic voting pattern holds

map showing similar north/south divide in 2024 and 2026 elections
map showing lack of regional trend in both 2024 and 2026

Precinct borders differ slightly, but the voting pattern holds pretty well. Eric Poicon does poorly in the south, IdaRose has no noticeable regional trend (though a consistent pocket of support in Blossom Valley).

Early/Late Votes

We can batch another way - rather than geographic, by voting time. All mail-in ballots posted on election day were mailed in at least a week before the election. We call these “early” voters. All subsequent mail-in ballots we call “late” voters. Can we see a patter of early/late voters?

Unfortunately the registrar does not provide any dataset of when the results were posted. I have however the first data download from Nov 7 (when I began this analysis). My understanding is that most of the votes included in this dataset would be ballots mailed in or dropped off before election day, with actual election day mail-in or dropoff ballots usually not appearing in the count until a week after the election. So for the mail-in votes specifically we can compare the Nov 7 data against the latest count.

Bar chart showing each candidate proportion of late votes. Lowest to highest: Robert Cox, IdaRose Sylvester, Paul Donahue, Alex Amoroso, Silja Paymer, James Kuszmaul, Erik Poicon, Samuel Ali
a

We see again many of the same correlations as before. We can further break this down by precinct.

table of all early and late vote percents for each candidate and precinct

It looks like the two southernmost precincts, 003440 and 003464, could account for most of the correlation. The geographic correlation could be bleeding here - the parts of the city that prefer certain candidates could also be full of late voters. We can mitigate this by normalizing by each precinct’s percentage of late voters. This yields a different pattern

similar table, but much lower correlation numbers than before

The early/late divide shows the negative correlation between Erik Poicon and Robert Cox. But the positive correlation with Paul is not present.

Conclusion

The North/South divide is the most significant geographic trend in the city. And displays consistent pattern of voting for or against certain candidates. This pattern is not universal. Some candidates have broad appeal, others are more targeted. The early/late divide is less clear cut. There are some interesting trends, but more data is needed for any conclusive analysis.

Two lines of research to be pursued. First, demographic analysis of the north/south divide. South of El Camino has very little multifamily housing, but single family neighborhoods also exist all over the city. Monta Loma and Rex Manor do not seem to align with this thesis - it needs more investigation. Could also consider other demographic factors such as age, income, etc.

Second, can we get enough information to actually simulate different voting systems? If we better understand the regional voting trend, get more detailed voting data, etc, we may be able to actually make a prediction as to how things might change. For example in 2024 Jose Gutierrez was positively correlated with Erik Poicon and negativly correlated with John McAlister. A ranked choice system have resulted in Poicon elected in 2024 instead of McAllister. A better model is needed to answer this question.

Anyone who knows such a model - get in touch!