Bud Light and the price of a political brand

In the average ZIP code in this sample, Bud Light made up 13.7% of beer units sold in the week before April 3, 2023. A year later it made up 10.5%, and in the last week of 2024, 9.6%. The share fell by 1.0 point in the first week and did not recover. The fall was larger where the 2020 vote was more Republican: compared with the 52 weeks before the event, Bud Light's share in the 52 weeks after was 17% lower in ZIPs under 40% Republican and 29% lower in ZIPs at least 70% Republican.

This case builds that measurement in steps: a two-by-two comparison, a regression across all 1,156 ZIPs, a weekly panel with fixed effects, and an event study that estimates the gap week by week. It ends with a design that looks reasonable and overstates the loss.

The event: a sponsored post on Saturday, April 1, 2023

On April 1, 2023, the transgender influencer Dylan Mulvaney posted a sponsored Instagram video promoting a Bud Light March Madness contest. It showed a can with her face on it that the brand had made to mark the first anniversary of her transition. A conservative backlash followed within days. Bud Light, America's best-selling beer for more than two decades, became the target of one of the most prominent consumer boycotts in recent memory.¹

Sources: contemporaneous coverage by CNN, NBC News, NPR, Fortune and others.¹

The post went up inside the week of March 27, which the data code as before the controversy. The drop shows up in the week of April 3, called week 0 below. Four questions go to the data: how large was the drop, how fast did it come, did it last, and where was it concentrated? The last bears most on how to read the event. If the loss scaled with local politics, it tracks a political reaction, whereas a brand having a bad spring would show up everywhere.

The data: 1,156 ZIP codes by 157 weeks

The panel has one row per ZIP code and week, from the week of January 3, 2022 to the week of December 30, 2024, a total of 181,492 ZIP-weeks. Every ZIP pools at least two convenience stores in every week, about three on average.Data. Point-of-sale data from PDI Technologies convenience stores, obtained through Dewey. The stores are mostly in the South and Midwest; the Northeast is thin. Demographics are ACS-based ZIP tables and CDC PLACES estimates. The 2020 presidential vote by ZIP code area comes from Fekrazad (2025).² The outcome is Bud Light's share of beer units in the ZIP that week, in percent. A unit is one scanned item, so a six-pack and a single can each count once, and shares compare brands within beer.

Work with shares, because sales swing with the calendar

Beer volume rises and falls with seasons, holidays and weather. Across the 65 weeks before the event, units per store in the average ZIP ranged from 492 to 722 a week. A brand's share of the beer a store sells in the same week compares brands under the same conditions. A break in the share is a change in what customers choose.

The picture: the share fell in the first week and has not come back

In the week before the controversy, Bud Light had 13.7% of beer units in the average ZIP. It had 12.7% in week 0, 11.5% in week 4 and 10.5% in week 52. By week 91, the last week of 2024, it had 9.6%. The brand was already losing share before the event: the average was 15.0% in the first week of 2022, so about 1.3 points of the decline to 13.7% came before April 2023.

Before the event the more Republican a ZIP, the more Bud Light it bought: over the 52 weeks before April 3, 2023 the share averaged 16.0% in ZIPs at least 70% Republican and 13.1% in ZIPs under 40% Republican. Within a month the gap between those two groups narrowed from 2.7 points in week −1 to 0.6 points in week 4. The share kept falling after the company's April 14 statement and the April 21–23 suspension of two marketing executives, from 12.0% in week 1 to 11.5% in week 4.

A two-by-two comparison: 2.4 points of extra loss in the most Republican ZIPs

The simplest way to measure a difference in the drop compares two groups before and after. Difference-in-differences subtracts the comparison group's change from the treated group's change. Take as treated the 293 ZIPs at least 70% Republican, as comparison the 226 ZIPs under 40% Republican, and as before and after the 52 weeks on each side of April 3, 2023:

Changes are computed from unrounded means, so a change can differ from the difference of the rounded columns by 0.01.

Both groups went through the same seasons, prices and national news, so subtracting the comparison group's change removes what the two share. What remains, 2.44 points, is the extra loss in the Republican ZIPs. The assumption is called parallel trends: without the boycott the two groups' shares would have changed by the same number of points.³ Figure 2 lets you change the groups and the window.

The estimate depends on the window. For the 70%+ and under-40% groups it is −1.65 points with 4 weeks on each side, −1.97 with 12, −2.15 with 26 and −2.44 with 52. It grows with the window because the gap between the groups was already narrowing before the event (2.9 points on average in the 52 weeks before, 2.7 in the 4 weeks before) and kept narrowing for months after it.

Part of a larger loss in points is arithmetic. The Republican ZIPs started higher, at 15.99 against 13.06, so a proportional loss costs them more points. If the 70%+ ZIPs had lost the same 17.3% as the under-40% ZIPs, they would have lost 2.77 points. The observed 4.70 points leaves 1.93 points that proportion does not explain, and the percent changes (−29.4% against −17.3%) differ by 12.1 percentage points.

Using every ZIP: each 10 points of Republican share goes with 0.51 points more loss

The two-by-two leaves out the 637 ZIPs in the middle two groups. To use all of them, give each ZIP one number, the change in its Bud Light share (the average of the 52 weeks after minus the average of the 52 weeks before), and regress that change on the ZIP's Republican vote share. A regression fits the straight line closest to the dots; its slope is the average difference in the outcome between two ZIPs that differ by one unit of the explanatory variable.

The fitted slope is b = −0.051 points per point of Republican share, or −0.51 per 10 points (standard error 0.028 per 10 points, robust to residuals whose spread differs across ZIPs). From a 40% to a 70% Republican ZIP that adds up to 30 × −0.051 = −1.5 points, on a base of about 14. Of the 1,156 ZIPs, 1,130 lost share and 26 did not.

Other variables also line up with the loss, but Republican share has the most explanatory power. Its R² is 0.22, against 0.14 for population density, 0.06 for Hispanic share, 0.03 for Black share, 0.01 for college and 0.00 for income. ZIPs with more Hispanic residents or more density lost less, which is the opposite sign to Republican share, and both slopes shrink once the other variables, Republican share among them, are held fixed: density from +1.13 to +0.31 points per tenfold increase, Hispanic share from +0.33 to +0.12 per 10 points. The Republican slope falls only from −0.51 to −0.46 per 10 points (standard error 0.05). The map shows where the losses fall.

The weekly panel: ZIP and week fixed effects

The same comparison can use all 181,492 ZIP-weeks instead of two averages per ZIP. The regression gives every ZIP its own intercept and every week its own intercept. These intercepts are called fixed effects. The ZIP effect α_i absorbs permanent differences between places, such as a ZIP that always buys more Bud Light. The week effect γ_t absorbs whatever hit every ZIP in the same week: seasons, holidays, national news and the brand's slow national slide.

Post_t is 1 from the week of April 3, 2023 and 0 before. The coefficient β is a difference-in-differences with a continuous dose: how many points more the share fell for each extra 10 points of Republican vote share. It is −0.550 with a standard error of 0.030. The standard error is clustered by ZIP, which allows the 157 weekly observations of one ZIP to be correlated with each other; ignoring that correlation makes standard errors too small.⁴ A 70% Republican ZIP loses 3 × 0.550 = 1.65 points more than a 40% Republican ZIP, 11% of the average pre-boycott share of 14.5%. In percent of a ZIP's own earlier share, each 10 points of Republican share adds 2.6 percentage points to the loss.

First three rows: standard errors clustered by ZIP. Last row: robust (HC1) standard errors. The county-by-week row keeps ZIPs in counties with more than one ZIP.

The gradient barely moves when the loss is allowed to vary with education, income, race, ethnicity and density. Of the five added interactions, three (% Black, % Hispanic and urban) have coefficients more than two standard errors from zero. Comparing ZIPs only with others in the same county and week gives −0.491, so local weather, events and distributor promotions do not produce the gradient. The one-row-per-ZIP slope of −0.512 differs from the panel because it uses 52 weeks on each side, while the panel compares all 92 weeks after with all 65 weeks before. With every week, the one-row-per-ZIP regression gives −0.550 exactly.Units. Per point of Republican share the panel slope is −0.055 and the 52-week slope is −0.051. This case uses per-10-point units throughout so that the two compare directly.

The event study: the gap opens in the first four weeks and stays

A difference-in-differences averages over the whole after-period. An event study estimates the gap separately for each four-week period, relative to the four weeks just before the event.

Here β_k is the extra change in share per 10 points of Republican share in period k, relative to period −1 (weeks −4 to −1), which is the reference and set to zero. Period 0 is weeks 0 to 3, period 1 is weeks 4 to 7, and so on.

The gap opens at once: −0.35 points per 10 points of Republican share in weeks 0 to 3 and −0.45 in weeks 4 to 7. It does not close. In every four-week period from week 32 to the end of 2024 the estimate is between −0.50 and −0.54. In the weekly version the estimate is −0.23 in week 0, −0.34 in week 1 and −0.43 in week 4, so the gap builds over about a month.

The pre-period is not flat. The estimates start at +0.21 (standard error 0.06) in early 2022 and fall toward zero, reaching −0.01 to +0.01 in periods −4 to −2; the mean of the 16 pre-period estimates is +0.07. Bud Light was losing share a little faster in Republican ZIPs throughout 2022, where it started higher. The drift is small next to the break of −0.35 within four weeks, but it makes the long-run level of the gap less certain than the break itself. A sudden break at a known date, concentrated where the cue predicts, is harder to explain by a slow trend than a gradual divergence would be. A pre-period that looks flat does not prove parallel trends, because tests on the pre-period have limited power and can distort the estimates that pass them, so the pre-period belongs on the same chart as the break.⁵ ZIP-specific trends and percent changes are natural follow-ups.

Where the share went: rival light lagers and other beer, with Modelo up everywhere

Shares in a ZIP add to 100, so Bud Light's loss of 3.47 points across all ZIPs is a gain somewhere else. Most of the gain went to all other beer (+2.40 points), Modelo (+0.97), Coors Light (+0.61) and Miller Lite (+0.47). The gains of the two closest substitutes rise with Republican share: Coors Light from +0.19 points in ZIPs under 40% Republican to +1.05 in ZIPs at least 70% Republican, Miller Lite from +0.05 to +0.89, and all other beer from +1.54 to +3.27.

Modelo gained +1.05 points in the under-40% ZIPs and +0.94 in the 70%+ ZIPs, so where Bud Light lost the most, Modelo gained no more. These data do not link Modelo's rise to the boycott. Anheuser-Busch's other brands lost share too, more in Republican ZIPs: Budweiser from −0.47 to −0.66 points, Michelob Ultra from −0.07 to −0.39 and Natural Light from −0.01 to −0.32. The parent company's other brands show losses too, in smaller amounts.

The control-group trap: rivals that received the switchers

A tempting design takes Coors Light and Miller Lite as the control group for Bud Light: same category, same stores, same weeks. Bud Light's own before-and-after change, over all 157 weeks (92 after, 65 before), is −3.92 points (standard error 0.063).Windows. This −3.92 is larger than the −3.47 in the 52-week comparison because it averages all 92 post-event weeks, including the months when the share was lowest. Against Coors Light and Miller Lite, with ZIP-by-brand and ZIP-by-week fixed effects, the estimate is −4.34 (standard error 0.073). The extra 0.41 points is the controls' average gain (the two headline numbers are rounded).

The controls were treated too, because they received the switchers. A difference-in-differences assumes the control group is unaffected by the treatment. When the customers who leave the treated product land in the control group, subtracting the control's gain counts the same switchers twice, once as a loss and once as a gain. With shares that add to 100, no brand in the data is untouched: whatever Bud Light lost went to the others.

Who reacted how: six kinds of places

Case 2 grouped ZIPs by their 2022 convenience-store baskets into six segments, and 1,143 of the 1,156 ZIPs here are also in that case. Bud Light's share fell 27% in the rural red South and Texas segment (from 16.9% by 4.5 points) and 17% in the Deep South segment with large Black populations (from 17.8% by 3.0 points). Those were the two segments with the highest Bud Light shares in 2022 and they reacted differently. The Deep South segment is also the least Republican of the six, at 44% on average, against 70% in the rural red South.

The West shows why points and percent can tell different stories. Its share before the event was 8.5%, the lowest of the six, and it fell 2.5 points, the smallest change in points and the largest in percent.

What these data cannot show

ZIP vote shares describe places. The data show the share fell more in more Republican places, which is an ecological pattern. A ZIP that voted 70% Republican has 30% of voters who did not, and nothing here says whether the people who stopped buying Bud Light were Republicans. Answering that needs purchase records for individual households.

The panel covers convenience stores, mostly in the South and Midwest, and ZIPs with at least two stores reporting every week. The national ranking in which Modelo overtook Bud Light came from all-channel retail data, so this panel cannot test it. It can say only that Modelo's gain in these stores did not vary with politics. Shares can also fall while a brand's own volume holds, if rivals grow, and a unit is one scanned item whatever the pack size. The data show a change in what customers choose within beer, not a count of cases lost.

The event study dates the break to the week of April 3 and places it in Republican ZIPs. It cannot say which part of the episode caused it: the sponsored post, the first boycott calls, or the company's response, all of which fall within three weeks. The pre-period drift means the level of the gap after the break, and not the break itself, depends on how the earlier trend would have continued. The data end in December 2024, so "no recovery" means none through that date.

Questions for discussion

1. By the end of week 1 the average share was 1.7 points below its level in the week before the event. The CEO's statement came on April 14 and the executives' leave on April 21–23. Using Figure 1, what did the company know at the end of week 1, and what would you have done with it?
2. Bud Light's average share drifted down by about 1.3 points between January 2022 and the week before the event. Which weekly series would you monitor to detect a boycott early, at what level of aggregation (region, ZIP politics, store), and what threshold would have separated April 2023 from ordinary drift?
3. In Figure 2 keep the 70%+ ZIPs as treated and use 55–70%, then 40–55%, as the comparison. How does the estimate change, and which assumption does each choice make about what would have happened without the boycott?
4. The event-study estimates fall from +0.21 to zero before April 2023. If that drift had continued, would the post-event gap be larger or smaller than −0.5? What would you do to check?
5. In Figure 7 add Modelo, then all other beer, as controls. Which choice gives the largest estimate, and why? Is any brand in a share variable a clean control?
6. A brand manager asks whether the customers Bud Light lost were Republicans. What can the ZIP-level results say, what can they not say, and what data would you need?

library(fixest)
feols(bud_light_share ~ post_controversy:rep_share_2020 | zip + week,
      data = df, cluster = ~zip)

import pyfixest as pf
pf.feols("bud_light_share ~ post_controversy:rep_share_2020 | zip + week",
         data=df, vcov={"CRV1": "zip"})

@misc{singh2026bud,
  author = {Singh, Vishal},
  title  = {Bud Light and the price of a political brand},
  year   = {2026},
  note   = {Teaching case, NYU Stern School of Business},
  url    = {https://vishalsingh.org}
}