In WAR, There Are No Winners But All Are Lou-sers

Is Nancy Pelosi really one of the weakest Democratic candidates for Congress - or was her opponent really one of the strongest Republican candidates?

Some methods suggest Nancy Pelosi faced an unusually strong challenger in 2024. Analysis of precinct-level vote records and Census data reveal that Pelosi underperformed modestly in the most Asian American neighborhoods. However, this trend was also observed in 2022 and at the presidential level, providing little evidence that the identity of her opponent drove this result.

I’ve made the replication code for this blog post publically available, here as a GitHub repository.

Speaker emerita Nancy Pelosi recently made headlines when she was spotted visiting fellow Italian-American Stefanie Germanotta at the Mayhem Ball in San Francisco earlier this month. “It was a fabulous show in San Francisco,” Pelosi wrote from her official account in a quote tweet of Pop Crave. “The most fun I’ve had in a long time.”

And, to be sure, it seems from the outside that for Pelosi there’s been a shortage of fun to be had recently after November’s elections swept Donald Trump back into the White House and Republicans into majorities in both chambers of Congress. Pelosi herself had an unremarkable November on paper, winning 81 percent of the vote and cruising to victory over her Republican challenger Bruce Lou.

But the headline commanding margin belies some interesting quirks to this particular victory. First, Lou’s seemingly paltry 19 percent still marked the highest vote share of any Republican challenger Pelosi has faced since 1990.¹ This suggests that despite losing in a blowout, perhaps Lou managed to find appeal that previous challengers to Pelosi failed to muster.

¹ This claim, circulated by both Lou’s profile on the San Francisco Republican Party website and by the high-profile journalistic outlet The Saratoga Falcon is somewhat of a misleading technicality: Democrat Shahid Buttar won just over 22 percent in 2020 and unaffiliated challenger Preston Picus won 19.1 percent in 2016. But given that crossover voting for Republicans is much less common in San Francisco, the claim is not without its merits.

² WAR, or “wins above replacement,” refers to the recent attempts by analysts to quantify candidate strength. The term borrows from baseball parlance.

Second, a lot of ink has been recently spilled in the WAR² wars. Split Ticket’s 2024 House WAR metrics suggested that the WAR in CA-11 was a whopping R+10.6, suggesting that Lou ran 10.6 points above a “replacement” Republican candidate. The recently-released metric from G. Elliot Morris and Mark Rieke claims instead that Pelosi’s WAR was +4, suggesting that actually her performance was slightly better than what we would expect from a replacement Democrat. Say what you will about WAR metrics (for example, that they are highly sensitive to researcher specification and struggle to attribute results correctly to one candidate’s overperformance versus their opponent’s underperformance) but clearly these approaches still are inconclusive. Considering the district directly and in context provides a way to adjudicate between these competing findings.

Third, others have noted³ that San Francisco saw strong shifts towards Republicans in Asian neighborhoods but have failed to put the trend into context. Considering the historical context, down-ballot performance, trends, and neighborhood-level diversity adds to this analysis.

³ Astute readers may notice I commented in this thread. The OP kindly ignored my request for help finding data, so any complaints about delays in this piece should be directed to the other Zachary.

⁴ Bruce could be considered the Nancy Pelosi of 2015 quiz bowl in Northern California. For the uninitiated, quiz bowl teams are traditionally composed of four players. Bruce essentially played solo, beating four-person teams so handily that while winning our last tournament together (the 2015 NAQT Northern California State Championship) he went undefeated, 13-0, and had more than double the PPG of the second-highest individual scorer (which, while nobody asked, was also about ten times my PPG at the same tournament).

Fourth, and perhaps most importantly in motivating me to actually write this piece up, I actually know Lou from our time playing quiz bowl in high school.⁴ So, with these points in mind, I wrote up a short post to answer a few questions. Was Lou’s (relatively) strong performance buoyed by a particular ability to leverage appeals to the Asian community, or did his loss simply mark a continuation of declining support for an aging Pelosi?

The Substance

• Lou’s “historic” performance was driven by broader trends, not individual strength. While Lou won 19% (the highest for a Republican challenger to Pelosi since 1990), this reflects ongoing erosion in Pelosi’s support rather than exceptional candidate appeal—her vote share fell 3 percentage points from 2022.

• Asian American neighborhoods showed mixed results for Lou, not systematic support. Although Lou performed relatively well in some Asian-heavy precincts, he actually underperformed predictions in core Asian American areas like Chinatown and parts of the Richmond and Sunset districts.

• Historical partisanship matters more than demographics. Precinct-level analysis reveals that 2022 voting patterns and baseline Republican support better predicted Lou’s performance than Asian American population percentages.

• Lou ran behind Trump in diverse neighborhoods. In racially diverse areas like Bayview-Hunters Point, Lou underperformed Trump’s presidential vote share by 3-4 percentage points, suggesting limited crossover appeal beyond the top of the ticket.

• In this post, I assume some basic familiarity with the neighborhoods of San Francisco. Neighborhood names are included in the interactive maps, and also you can scroll to the end of the gallery to find a labelled map of San Francisco neighborhoods.

Bruce Almighty

Before getting into the analysis, for the unfamiliar, who is Bruce Lou? As mentioned earlier, I first met Lou when we were both high school seniors: he at Saratoga High School, I at Davis Senior High School.⁵ He went on to attend UC Berkeley⁶ where he graduated with a degree in computer science in 2019.

⁵ I did not consult nor contact Lou ahead on the contents of this article before publishing this piece, though he did briefly DM me asking about it because I accidentally made a draft visible earlier than intended.

⁶ Go Bears?

⁷ A quick and dirty search on the FEC website suggests that he did make some small donations to the California Republican Party prior to entering the race.

After graduating, Lou worked as a software engineer and at True Search. He also started a consulting firm, Stingray, and helped fundraise/distribute pepper spray to Chinatown residents in 2021. During this time period, he had no notable public appearances in politics.⁷

Lou announced his run against Nancy Pelosi on November 1, 2023, almost a full year ahead of the election. His campaign’s press release identifies Lou as “‘Jeopardy!’ Winner Bruce Lou” alongside a photo of Lou with the late Alex Trebek.

From the start Lou faced a clear uphill battle. San Francisco has been a Democratic enclave for decades, and Pelosi started the cycle with $3.5 million cash on hand at the end of the previous cycle. Lou raised slightly more than $208,000 over the course of the campaign.⁸ After placing second in the primary with 8.6 percent of the vote, Lou was officially set to challenge Pelosi in the general.

⁸ Credit where it’s due, Lou raised more than similarly long-shot candidates. Anna Cheng Kramer in CA-15 to the south raised $160,000 and in CA-02 to the north, Chris Coulombe raised nearly $124,000. All three Republican challengers failed to crack 30 percent of the vote.

Lou on the campaign trail. Source: @RealBruceLou on X.

Pelosi’s status as a nationally unpopular figure⁹ means that her challengers have frequently garnered national attention, and Lou was no exception. Lou parlayed his primary victory into a social media audience of more than 15,000 followers on X¹⁰ where he frequently posted videos¹¹ filmed in front of the San Francisco skyline directly addressing Pelosi. But there was one person Lou failed to engage with these pillories: Pelosi herself. Neither her official X account nor her personal account mention Lou once. He also struggled to get coverage in local mainstream news outlets;¹² nods to Lou’s campaign in The San Francisco Chronicle are limited to a closing paragraph noting his primary victory and a quick quote in a story on Chinese-American Trump voters.

⁹ Forgive the RealClearPolitics link.

¹⁰ The “everything app,” fka. Twitter.

¹¹ His X.com account has 151 photos and videos. As a comparison, his campaign website has five total press releases, one of which predates the campaign and one of which is announcing his initial filing for Congress.

¹² With the exception of The Saratoga Falcon, of course.

It should be no surprise, then, that when November rolled around Pelosi romped Lou by a sixty-point margin, 81.0 - 19.0 percent. But, as previously noted, there were some interesting findings under the hood of what seemed like another standard Democratic blowout. In this piece, I use the precinct-level results and Census data to investigate.

Anatomy of a Fall

Did Lou post a historically high performance for Republicans in CA-11 because of individual factors, or was he just the latest standard bearer in a district slowly trending against its 85-year-old incumbent? First, we should take a look at the political geography at play. To start, I created this precinct-level map of the November 2024 race between Nancy Pelosi and Bruce Lou in California’s eleventh congressional district.

To do this, I first downloaded the 2024 statement of vote data at the precinct level from the California Statewide Database,¹³ grabbed the shapefiles from the Statewide Database geographic data page, then used the city of San Francisco Department of Elections map to determine which precincts in San Francisco county are not in the CA-11 boundaries and thus should be deleted. Because Pelosi won every single precinct with an outright majority, I’ve instead colored the map to reflect whether Pelosi won a precinct by more or less than the citywide margin of victory.

¹³ I personally would not have named my statewide elections database “Statewide Database,” given the nonspecificity on both geography and data contents, but I have to admit it is a bit of a power move.

Precinct-level map of the November general election, normalized to citywide results. Blue indicates precincts where Pelosi won a higher share of the vote than her citywide average; red indicates precincts where Lous won a higher share of the vote than his citywide average. Click to expand.

Glancing at the map, there are a few clear patterns. Pelosi does the best in the center of the city, particularly in the Noe Valley and Buena Vista Heights neighborhoods. Meanwhile, relative to the citywide results Lou does well in the Marina District, Chinatown, the Richmond District, the Sunset District and Parkside. Lou performed especially well in Bayview-Hunters Point and Silver Terrace; his closest precinct was in the latter, where he closed the margin to six percentage points and lost 53.3-46.7 percent.

Readers familiar with San Francisco will note that most of these neighborhoods have considerable Asian American populations. Rather than eyeballing it, though, we can use the 2020 Census data to visualize the Asian American population in San Francisco. I started by grabbing Census data at the block level from IPUMS,¹⁴ then using the crosswalk provided by the Statewide Database to determine racial demographics at the electoral precinct level. I then mapped the results with ggplot:

¹⁴ Citation: Steven Manson, Jonathan Schroeder, David Van Riper, Katherine Knowles, Tracy Kugler, Finn Roberts, and Steven Ruggles. IPUMS National Historical Geographic Information System: Version 19.0 [dataset]. Minneapolis, MN: IPUMS. 2024. http://doi.org/10.18128/D050.V19.0

Precinct-level map of Asian American population in the district. Click to expand.

We can see that indeed, with the primary exception of the Marina District, the map of the Asian American population in San Francisco seems suspiciously similar to the map of neighborhoods where Lou improved on the citywide average. But, as I said, eyeballing these comparisons can only get us so far. First, a simple scatterplot should convince you that this relationship is more than suggestive.

Scatterplot of Lou performance relative to citywide results, by Asian American population. Click to expand.

Of course, here we can still clearly see that there are several precincts that are less than 25 percent Asian American where Lou outpaced his citywide margin, and several precincts that are 25-50 percent Asian American where Lou underperformed his citywide margin. One obvious explanation is that these approaches do not account for historical partisanship; perhaps the areas where Lou did best are historically (relative) Republican strongholds, that also coincidentally happen to include several Asian American neighborhoods. To address this, we can look at the shift in results between 2022 and 2024 in this district.

Precinct-level map of 2022-2024 shifts in the district. Red indicates a precinct where Lou won a higher percent of the vote than 2022 Republican candidate John Dennis. Blue indicates a precinct where Pelosi won a higher percent of the vote in 2024 than she had in the same precinct in 2022. Click to expand.

Between 2022 and 2024, Pelosi’s support fell by about 3 percent. This map shows that the turn against Pelosi was distributed relatively evenly throughout the city, though there were some pronounced shifts around Market Street and in the Bayview-Hunters Point area. To draw out how neighborhoods moved relative to expectations, we can also visualize whether precincts shifted more between 2022 and 2024 than the city as a whole.

Precinct-level map of 2022-2024 shifts in the district normalized to citywide results. Blue indicates a precinct that had less of a shift to Republicans between 2022 and 2024 than the city as a whole. Red indicates a precinct that shifted more towards Republicans between 2022 and 2024 than the city overall. Click to expand.

This draws out the previous conclusion: by and large, what seemed originally like a tidy story of geographic shifts becomes somewhat of an oversimplification. We can see that the 2022-2024 shift was indeed stronger in the Market and Bayview-Hunters Point regions, but baring a slightly stronger than expected move in the Tenderloin and Chinatown there is a far less certain link between neighborhoods’ Asian American populations and the 2024 performance. The scatterplot tells a similarly opaque story.

Scatterplot of 2022-2024 shifts in the district normalized to citywide results, by precinct Asian American population. Click to expand.

Everything Everywhere All At Once

So far, the picture on this election remains a bit confusing. We know that Lou did somewhat better in the west and the south of the city, and that this performance seems somewhat tied to the precinct’s partisanship and number of Asian American voters. But how do we adjudicate between these two factors? And, dear reader, why force you to trust me when I can let you look directly at the data?

Here we get to the main point of the piece. I use a number of bivariate regressions to predict Lou’s vote share, map the predictions against the actual results, and discuss the relative explanatory power of each.

First, I regress Lou’s vote share on the percent of a precinct that is Asian American. This uses the precinct’s Asian American population to predict Lou’s expected vote share. The residuals from this model are thus the difference between how much of the vote Lou actually received, and how much of the vote we would expect Lou to get in a precinct based only on its racial demographics. The results are visible in this interactive map, which you can scroll and zoom around at your pleasure.

From this map, a few interesting things pop out. Intuitively, Lou’s previously-noted performance in the Marina greatly outpaces what a naive prediction using only Asian American population as a predictor would forecast. There are precincts on the West and South side of the city that have relatively large Asian American populations where Lou still gets more of the vote than the race-based predictions would suggest. But there are also Asian American neighborhoods where Lou underperforms the race-based predictions, primarily in the Inner and Outer Sunset, the Richmond District, and (perhaps most notably) Chinatown.

As previously discussed, we might worry that this analysis omits the political factors. And indeed, if we plot the residuals against the precinct’s 2022 vote history we can see that the prediction error of a model using only Asian American population to predict Lou’s vote share is correlated with how strongly the precinct supported Nancy Pelosi in 2022.

Scatterplot of residuals from the above map, by precinct’s 2022 vote history. Click to expand.

Thus, I next regress Lou’s vote share on the 2022 results in a given precinct. This uses the precinct’s 2022 electoral results to predict Lou’s vote share. To restate, what this does is give us a prediction of how well Lou should do in a given precinct based on its 2022 voting record. We can then use this prediction to see whether Lou did better or worse than the historical trends would suggest. Again, the results from this model are plotted in this interactive map, where you can zoom in, scroll, and select precincts to view the results.

You might notice that the results resemble (but are not identical to!) the earlier map of the normalized 2022-2024 shift. This is because the regression is mechanically doing something very similar. The conclusions, therefore, are essentially the same, with Lou posting overperformances of between five to seven points in Bayview-Hunters Point, the Tenderloin, and South of Market and modestly underperforming the historical prediction in much of the rest of the city.

Also notable is that if we plot the residuals from this model against the precinct’s share of Asian Americans, there does not appear to be an immediate and clear pattern. There is, of course, still a positive association, but this suggests that the percent of a precinct that is Asian American has a much smaller effect on Lou’s performance than the precinct’s historical partisanship.

Scatterplot of residuals from the above map, by precinct’s Asian American population. Click to expand.

Finally, I regress Lou’s vote share on Donald Trump’s vote share in a given precinct. This uses the precinct’s support for Donald Trump to predict Lou’s vote share. Essentially, the goal of this is to use a measure for baseline 2024 partisanship in a precinct to predict how well we would expect Lou to do, based purely on that measure. One advantage to this approach is that because these elections were held simultaneously, any differences can be attributed to actual changes in vote choice rather than compositional changes in the electorate. It should also be noted that, for context, Trump won 14.7 percent of the vote in CA-11 and thus underran Lou by about 4.3 percent.

Here, a few things pop out. The Marina remains a strong spot for Lou; in most precincts he overperforms predictions from presidential vote by about 5 percentage points on average, and generally outruns Trump by around 10 percentage points. Meanwhile, in the Western Addition and some of the Tenderloin Lou generally ran even with Trump’s vote share, which is an underperformance from the prediction of about 3 percentage points.

But the results in the Bayview-Hunters Point area are most interesting to me. Here, Lou generally ran even with Trump. In a few precincts, he even underperformed Trump’s vote share by about 4 percentage points, meaning that in these precincts he won about 9 percent less of the vote than we would have expected based on these neighborhoods’ votes for Trump. This is all with the exception of the single Silver Terrace precinct where Lou did notably well; his 46 percent of the vote marks about a 3 percentage point overperformance of what we might have expected based on that precinct’s 39 percent vote for Trump.

Again, if we plot the residuals from this model against the precinct’s share of Asian Americans, there does not appear to be an immediate and clear pattern. The association between the precinct’s Asian American population and Lou’s performance relative to Trump’s seems even weaker than the previous association between precinct’s Asian American population and Lou’s performance relative to previous election results.

Scatterplot of residuals from the above map, by Asian American population. Click to expand.

But, to those who know San Francisco geography, the previously-mentioned pattern in Bayview-Hunters Point is notable for another reason. Bayview-Hunters Point is a historically Black neighborhood, and today is one of San Francisco’s most diverse neighborhoods. Is it possible that Lou’s apparent underperformance of Trump in this neighborhood is being driven by non-Asian, non-white voters who voted Trump but did not vote for Lou, either due to downballot lag or other reasons?

To investigate, I use the Census data to calculate a measure of diversity for each precinct. Specifically, I calculate a Blau index. For a geography with \(G\) groups, we can take the proportion \(p_g\) of individuals belonging to any given group \(g\). The index is constructed by taking the sum of each \(p_g\) squared, and subtracting the total from \(1\). Thus by construction we get a measure of diversity, \(D_{blau}\) that increases as the geography gets more diverse. If the entire population is only from a single group, \(G=1\) and \(\sum_{g=1}^G p_g^2 = 1\), so \(D_{blau}=1-1=0\). As the number of groups increases or as the relative dispersion across groups increases, thus \(D_{blau}\) approaches 1. Specifically, I calculate the diversity index using the following equation:

\[D_{blau} = 1-\sum_{g=1}^G p_g^2\]

Plotting the results produces a result that resembles (but is not identical to) the map of how Lou performed relative to expectations generated from the top of the ticket. The most diverse precincts are concentrated mainly in Bayview-Hunters Point, the Mission, south of Market, the Tenderloin and Ingleside. Meanwhile, Chinatown, the Marina District, Noe Valley, Twin Peaks, and Parkside are among the least diverse neighborhoods.

Precinct-level map of diversity (Blau index) in CA-11. Click to expand.

Returning to the original question, we can plot the residuals from the regression of Lou’s vote share on Donald Trump’s vote share against the diversity levels in a given precinct. Interestingly, there appears to be a bit of a u-shape, where Lou underperformed relative to the top of the ticket in both the most and least diverse precincts.

Scatterplot of residuals from regression of Lou’s vote share on Trump vote share, by Blau index. Click to expand.

TL;DR: The Usual Suspects

To recap, what can we take away from all of this? I set out to answer a few questions. First, it is an undisputable fact that Lou won a higher share of the vote than any Republican challenger to Nancy Pelosi in the past 35 years. Does that mean that he did extraordinarily well? Overall, the evidence suggests that Lou did about as well as we might expect from the current trends. Pelosi’s overall support fell 3 percent despite a relatively large shift in the overall environment between 2022 and 2024.

Was Lou’s performance boosted by overperformances in Asian American neighborhoods? Lou did get more of the vote in neighborhoods that have considerable Asian American populations than he won citywide, but those neighborhoods also historically have given Republicans more of the vote than the rest of the city has. After accounting for the 2022-2024 shift, there appears to be only a minimal shift in the vote for Lou that can be attributed to the precinct’s Asian American population.

One thing missing from this piece is a consideration of turnout effects. Frankly, this is mostly because I’ve spent more time on this piece than I would have liked to. If I have time in the future, I may come back to that, but for now I leave it for any curious readers to use the Github repository to extend the analysis.

Finally, considering results from the top of the ticket does suggest that, if the urban swing against Harris proves to be a trend rather than a one-off swing, future Republican challengers in CA-11 may benefit from broader electoral shifts rather than candidate-specific factors. The Marina District’s strong support for Lou despite its limited Asian American population, combined with his mixed results in diverse neighborhoods like Bayview-Hunters Point, reinforces that electoral success in San Francisco’s changing political landscape could be complicated by the broader partisan realignment.

For now, though, my take is that Bruce Lou’s 2024 performance appears to be less about breaking new ground and more about riding existing currents against an aging incumbent whose coasted on her national profile.

Extra Maps

Neighborhoods of San Francisco, as defined by the city planning department. Click to expand.

As a treat, I’ve left this interactive map of the general results for you to look through.

Data Sources

• Election Data
• Shapefiles
• CD boundaries
• Steven Manson, Jonathan Schroeder, David Van Riper, Katherine Knowles, Tracy Kugler, Finn Roberts, and Steven Ruggles. IPUMS National Historical Geographic Information System: Version 19.0 [dataset]. Minneapolis, MN: IPUMS. 2024. http://doi.org/10.18128/D050.V19.0