Advertising and Environmental Stewardship: Evidence from the BP Oil Spill

Abstract

This paper explores whether private markets can incentivize environmental stewardship. We examine the consumer response to the 2010 BP oil spill and test how BP’s investment in the 2000–2008 “Beyond Petroleum” green advertising campaign affected this response. We find evidence consistent with consumer punishment: BP station margins and volumes declined by 2.9 cents per gallon and 4.2 percent, respectively, in the month after the spill. However, pre-spill advertising significantly dampened the price response, and may have reduced brand switching by BP stations. These results indicate that firms may have incentives to engage in green advertising without investments in environmental stewardship.

How does advertising shape consumer behavior and firm incentives to undertake costly and hidden investment in product quality? The answer to this question informs a debate over whether private markets can provide incentives for public goods such as environmental stewardship in production of goods and services. One view holds that advertising can provide valuable information to consumers who seek to support firms that base production decisions on environmental and sustainability concerns. Another view suggests firms may use advertising as a means of creating spurious product differentiation and brand loyalty in green markets.¹

This paper provides novel evidence on the role of advertising in green markets by studying the consumer response to the British Petroleum (BP) Deepwater Horizon oil spill in 2010, one of the largest oil-related environmental disasters to date.² Prior to the spill, BP undertook one of the most successful corporate advertising campaigns entitled “Beyond Petroleum.” Between 2000 and 2008, BP debuted a new logo—a Helios (sun) symbol—and rebranded the BP acronym (i.e., the slogan “Beyond Petroleum” replaced the name British Petroleum). These actions reflected a newly stated dedication to environmental stewardship and commitment to production methods that mitigated environmental degradation. Consumers appeared to have internalized this message as surveys fielded during the campaign consistently rated BP as the most environmentally friendly oil company during the mid-2000s (Landor Associates, Cohn & Wolfe, and Penn Schoen Berland Associates 2007, 2008).

The Beyond Petroleum campaign and subsequent BP oil spill provide a natural setting to study how advertising affects firm incentives to provide public goods. Absent third-party certification, advertising is the main mechanism through which firms make environmental quality claims to consumers. Whether consumer valuation of environmental stewardship can successfully incentivize its provision in equilibrium depends critically on whether consumers punish firms for deviating from advertised product attributes (Besley and Ghatak 2007).

Our analysis proceeds in two steps. First, we estimate the consumer response to the BP oil spill using detailed data on gasoline station prices and a select measure of sales from January 2009 to December 2010. We identify impacts by comparing outcomes for BP and a comparison group of stations in periods before and after the spill. This difference-in-differences approach indicates that there was an economically and statistically significant consumer response to the oil spill. Retail prices at BP stations declined by 2.9 cents per gallon relative to the comparison stations in the first month after the spill (May 2010). This impact represents up to an 18 percent decline in profit margin relative to industry standards. In addition, BP sales declined on average by 4.2 percent among our sample of station customers (Wright Express fleet card holders). For the entire period before BP capped the oil leak (May–September 2010), there was an average price decline of 1.1 cents per gallon.

The second step in our analysis studies how advertising modified the consumer response to the spill by combining station-level data with metropolitan-level data on BP advertising during the 2000s. Our core measure focuses on corporate advertisements (i.e., ads related to the BP Corporation, BP fuels, and environmental issues) during the Beyond Petroleum campaign (2000–2008). To address the potential endogeneity of advertising expenditures, we construct a novel instrumental variable (IV) based on the number of elections for mayors, US senators, and governors in each metropolitan statistical area (MSA) during the period of the Beyond Petroleum campaign. Electoral calendars are legally preset and should be exogenous to any unobserved area characteristics correlated with demand responses to the BP oil spill. At the same time, this instrument should be relevant because political advertising displaces other forms of advertising. This idea stems from Sinkinson and Starc (2019), who study pharmaceutical advertising using a similar IV strategy based on the primary calendar for the 2008 national election.

Our advertising results show that the negative impact of the oil spill on BP prices was significantly less severe in areas with more pre-spill advertising. Moreover, this protective effect on prices was significantly stronger in areas where consumers have “green preferences” as captured by an index measure based on the local area share of hybrid vehicles and support for environmental organizations.³ We also find suggestive evidence that advertising appeared to mitigate longer run effects of the oil spill. Utilizing an event-study framework, we find that markets with low pre-spill advertising suffered significantly greater losses in BP retail outlet share after the spill. The losses amount to a 6 percent decline relative to the mean in areas with low pre-spill advertising. Though subject to the limitations of event studies, these results are consistent with the possibility that during-spill profit losses may have been large enough to induce station owners to switch to alternative brands, and BP investments in green advertising dampened those impacts.

The evidence of a mitigating effect of the Beyond Petroleum campaign suggests that advertising provides insurance against reputational costs. This may occur because advertising and other types of corporate social responsibility shift beliefs about whether an adverse event was due to negligence or bad luck (Minor and Morgan 2011).⁴ This interpretation is consistent with the idea that advertising plays more of a persuasive rather than an informative role.⁵

In terms of contribution, this paper stands at the intersection of three literatures. First, we add to studies of the causal impact of advertising on consumer behavior. Recent work has studied the impact of advertising using price and expenditure related instruments (Dubé and Manchanda 2005; Iizuka and Jin 2005; Chou, Rashad, and Grossman 2008; Liu and Gupta 2011; Dinner, Van Heerde, and Neslin 2014) and experimental methods (Bertrand et al. 2010; Lewis and Reiley 2014; Lewis and Rao 2015). Most closely related to our analysis is Sinkinson and Starc (2019), who pioneered the use of an instrumental variable approach based on political campaigns and found positive impacts of advertising on pharmaceutical sales. Our paper complements prior studies by examining how advertising affects consumer behavior in the aftermath of a negative event. In this sense, our analysis secondly relates to a literature that shows accidents, regulatory violations, and product recalls have negative impacts on firm performance (Jarrell and Peltzman 1985; Hoffer, Pruitt, and Reilly 1988; Barber and Darrough 1996; Borenstein and Zimmerman 1988; Karpoff, Lott, and Wehrly 2005; Dranove and Jin 2010; Minor and Morgan 2011; and Freedman, Kearney, and Lederman 2012). We add to this literature by examining consumer responses to a large environmental disaster.

Third, we contribute to studies of consumer information and deception. A growing number of empirical studies find that advertising may induce consumers to make less informed purchasing decisions. This could weaken demand responses as a market disciplining mechanism to sustain firms’ commitment to product quality. For example, Jin and Kato (2006) finds that online sellers can earn price premiums for unverified product quality claims even when those are misleading. Similarly, Zinman and Zitzewitz (2016) documents deceptive advertising in the context of snowfall and ski reports. Bronnenberg et al. (2015) finds that misinformation accounts for a sizable portion of brand premiums for health products such as pain relievers. Finally, Rao and Wang (2017) finds that firms gained significant revenues by making false claims about the health attributes of consumer products. They find significant declines in demand when the Federal Trade Commission (FTC) ordered termination of misleading claims. However, this effect is heterogeneous with large responses by newcomers and more muted effects for existing customers. Our results are consistent with the idea that advertising may provide protective benefits to firms that make unsubstantiated product claims.

Overall, our findings relate to public policy debates over consumer protection and regulation. A concern is that persuasive advertising affects consumer demand and weakens incentives for firms to invest in product quality. Misleading advertising is especially policy-relevant in the environmental realm, where there are concerns about a “greenwashing” equilibrium.⁶ With this in mind, several policies could enhance efficiency in markets where consumers seek to purchase environmentally friendly goods and services. For example, both truth-in-advertising regulations and third-party evaluations of product claims have been effective in other contexts.⁷ Yet, a concern is that these policies may be difficult to apply to broad environmental campaigns or marketing that includes nature-based imagery.⁸ These considerations suggest that price-based regulations (e.g., pollution taxes or penalties) likely remain a first-best policy solution to environmental externalities. We present a detailed discussion of policy options in Section IV.

I. Background

From 2000 to 2008, BP conducted a public relations and marketing campaign that sought to align the company with environmental issues. BP introduced a new slogan, “Beyond Petroleum,” and redesigned the company logo to a green and yellow Helios sun. BP advertisements emphasized that the company was investing to make operations more efficient and reduce environmental impacts (Cherry and Sneirson 2011).⁹

This marketing appeared to impact industry and consumer stakeholders. The Beyond Petroleum campaign won two PR Week “Campaign of the Year” awards and received the Gold Effie Award from the American Marketing Association in 2007 (Solman 2008).¹⁰ Survey data suggest consumers were aware of the environmental messaging. In 2008, Landor Associates, a marketing firm, found that 33 percent of survey respondents believed BP was a “green” brand. Respondents also ranked BP as the greenest of the major petroleum companies (Landor Associates, Cohn & Wolfe, and Penn Schoen Berland Associates 2007, 2008). A poll by Marketing Week also ranked BP third in terms of companies that made the greatest commitment to environmental issues (Marketing Week 2008).¹¹

Why did BP invest in environmental branding? Prior research shows that consumers are willing to pay for environmental stewardship as a product attribute (Nimon and Beghin 1999; Forsyth, Haley, and Kozak 1999; Goett, Hudson, and Train 2000; Loureiro, McCluskey, and Mittelhammer 2001; Roe et al. 2001; Teisl, Roe, and Hicks 2002; Pelsmacker et al. 2006; Kahn 2007; Kahn and Vaughn 2009; and Kiesel and Villas-Boas 2013). In addition, the literature has found that advertising increases demand for advertised products (Ackerberg 2001; Dubé and Manchanda 2005; Bagwell 2007; Sorensen 2007; Bertrand et al. 2010; Clark, Doraszelski, and Draganska 2009; Simester et al. 2009; Friberg and Grönqvist 2012; Hastings, Hortaçsu, and Syverson 2017; Gurun, Matvos, and Seru 2016; Lewis and Reiley 2014; and Garthwaite 2014).

While there may be demand for environmental quality, consumers do not know directly whether a product has this attribute (in the absence of third-party certification). This suggests at least two motivations for green-themed advertising. First, one class of model shows that firms can use advertising as a sunk cost to signal their investment in product quality (Grossman and Shapiro 1984, Milgrom and Roberts 1986, and Cabral 2005). Second, advertising could play a persuasive role that convinces consumers that negative events are accidental and due to “bad luck” (Minor and Morgan 2011). In this sense, advertising changes customer beliefs about firm behavior and acts as insurance. Such beliefs could mitigate consumer punishment and thereby decrease the incentives for firms to invest in hard-to-observe product characteristics.¹²

The period after the Beyond Petroleum marketing campaign provides a unique opportunity to study environmental advertising due to the BP Deepwater Horizon oil spill. In April 2010, an oil well blowout sunk the Deepwater Horizon rig, and robotic monitoring devices soon discovered that oil was leaking from the damaged well. BP sought to contain the leak, but their efforts were unsuccessful until engineers installed a “containment dome” in July 2010.¹³ Officials declared that the damaged well was “effectively dead” after a relief well was completed in September 2010. Scientific experts estimated that 205.8 million gallons of oil had leaked before containment (Department of Interior 2010). Subsequent investigations found that the cause of the spill was attributable to active management decisions on behalf of BP.¹⁴

II. Data Sources and Sample Construction

We use several proprietary data sources to study the impact of the BP oil spill and advertising. This section provides a summary of our sample construction and key variables. A brief overview is as follows. We begin by creating a sample of BP and comparison group stations using data from the Oil Price Information Service (OPIS) (2009–2010). The OPIS data contain station-level information on regular grade retail gasoline prices, a select measure of sales volumes based on fleet card holders (further detailed below), and brand affiliation. Next, we match the sample of stations to wholesale gasoline prices from OPIS based on distribution terminal (“rack”) prices aggregated at the state-week level. The linked data allow us to compute net prices as a measure of retail margins. In addition, we link the sample of stations to Designated Market Area (DMA) measures of BP advertising obtained from Kantar Media.¹⁵ Finally, we link the sample of stations to local area proxies of consumer preference for environmental protection, demographics, and elections. These measures vary at the state or zip code level.

A. Station-Level Sample

We create a sample of gasoline stations using data from OPIS, which collects information from two sources. First, OPIS observes stations over time based on Wright Express fleet fuel card “swipes.”¹⁶ This information is available for stations that accept this fleet card on days when fleet card transactions happen (i.e., an individual must use their fleet card for a particular station on a particular day).¹⁷ The fleet card is widely accepted across the United States. Second, since 2009, OPIS has expanded its data collection to include reporting agreements with several gasoline refiner-marketers that provide data for additional stations that do not accept the fleet card.¹⁸ Between these two sources, OPIS has data for over 100,000 stations across the United States.

We define the sample for our analysis based on two considerations. First, most stations are available only for a portion of the years 2009–2010 or have sporadically reported prices. Given our interest in station-level variation in prices and fleet sales over time, we focus on zip codes where OPIS reporting meets minimum density criteria.¹⁹ Each zip must have at least five stations with at least three price observations per week for our entire sample period. We keep data for all stations located in this list of zip codes.

The second consideration is to ensure sufficient geographic congruity between treatment and control stations. Our empirical analysis compares prices at BP branded stations (the treatment group) to a comparison group of stations in zip codes without any BP stations. Note that the data contain information on a station’s brand over time, and we identify BP stations based on the brand observed in January 2009. Our preferred control group excludes non-BP stations in close proximity to BP stores as their prices were likely impacted by the spill as well. In the United States, the BP brand has a broad presence east of the Mississippi River, but becomes more selectively concentrated toward the West. To ensure that our empirical analysis compares treatment and comparison stations in comparable regions, we restrict the analyses to EIA areas where BP has a sufficient brand presence: the East Coast (PADD 1), the Midwest (PADD 2), and the Gulf Coast (PADD 3). Based on these criteria, we create a sample that contains 5,526 and 4,997 stations for the (separate) analyses of station-level prices and fleet sales, respectively.²⁰ Note that we test the robustness of the results for prices and fleet sales to changes in the sample inclusion criteria.²¹

B. Net Price, Fleet Card Sales, and Brand Outcomes

For stations in the sample, the OPIS data contain station-level information on retail prices for regular grade gasoline, sales to fleet card holders, and brand affiliation. The retail prices for each station are at the daily level, which we aggregate to a weekly measure because most stations do not report a price every day. We construct a net price measure (an estimate of retail price margin) after linking the sample to additional data from OPIS on wholesale prices. The wholesale prices stem from local gasoline distribution terminals (“racks”). We use state-level averages of minimum rack prices, averaged over the relevant retail gasoline formulations sold in a given state-week (e.g., 10 percent ethanol reformulated gasoline where applicable or locally relevant Reid Vapor Pressure formulations). This is a regionally appropriate measure of wholesale prices that accounts for supply shocks at a high temporal frequency that matches the retail price data.²² Formally, we construct the weekly station-level net price of station i in state j during week t as

$${\mathit{\text{netprice}}}_{i,t}\equiv {\mathit{\text{AvgRetailPrice}}}_{i,t}-{\mathit{\text{AvgWholesalePrice}}}_{j,t}.$$ (1)

As mentioned, we focus on weekly net prices because most stations do not post prices for every day during a week (data are typically available up to six days per week). In our regression specifications (detailed in Section III), we weight weekly price and quantity observations by the underlying number of daily observations within the station-week.

For stations that accept fleet cards (as opposed to stations whose parent companies only report prices to OPIS), total gasoline sales to fleet cards are recorded at the weekly level. While limited, these data represent, to our knowledge, the only station-level volume data currently available.²³ Fleet card drivers range from professional drivers to employees and members of small businesses or other entities (e.g., municipalities). Although these customers’ preferences may differ from the broader population, these data provide a glimpse into consumer behavior.

Finally, we also use the OPIS data to create a measure of BP’s share of stations within each zip code at the monthly level. The underlying data contain information on each station’s brand of gasoline at the weekly level. We aggregate this information into a monthly measure to explore longer run brand switching behavior. (Our main analysis uses the station’s initial brand in January 2009 to define treatment and comparison group indicators that do not vary over time.)

C. Advertising Measures

We link each station in the sample to measures of advertising from Kantar Media’s Ad$pender database.²⁴ Kantar uses tracking technologies and services to monitor advertising on television (cable and network), online, or in print publications such as business-to-business magazines, consumer magazines and news publications, and on internet sites. Kantar also collects outdoor and local radio advertising information from other marketing subscription services and directly from media providers (e.g., radio stations or billboard plant operators).²⁵

We use data on BP advertising for the years 2000–2010. The underlying observations are at the monthly level and provide information on the media type and market (designated market area or DMA). The data also identify the parent company (e.g., BP), distinguish between brands (e.g., BP service station versus Amoco service station), and differentiate between the specific products advertised (e.g., BP energy utilities versus BP gasoline).

Our main advertising measure is the sum of pre-spill BP advertising expenditures during the years of the Beyond Petroleum campaign (2000–2008) at the DMA level. This measure includes all advertising that focused on the BP Corporation, BP fuel products, and environmental issues. We aggregate advertising expenditures across all media as our measure of advertising exposure. This specification assumes there are stock effects of advertising on demand (Dubé and Manchanda 2005). For the robustness analysis, we also construct a measure of all advertising during the BP oil spill period (March to September 2010). We also construct a detailed measure of advertising for ancillary products and convenience stores at BP stations.²⁶

D. Elections Measure

We link each station in the sample to a measure of local area elections. This measure allows us to construct an instrumental variable to address the concern that BP advertising may be endogenous to each area’s unobserved preferences for the BP brand. Electoral calendars are plausibly exogenous to spill impacts because they are preset by law. In other words, there should be no correlation between population preferences or other unobserved characteristics that are correlated with demand responses to the BP oil spill.²⁷ The relevance of this instrument stems from the effect of political campaigns on the costs of advertising. There is cross-sectional variation in the cost of advertising due to the differences in the number of elections each city experienced during the Beyond Petroleum campaign. This logic follows Sinkinson and Starc (2019), who utilized variation in political advertising levels across DMAs in the run-up to the 2008 election to instrument for pharmaceutical advertising exposures across states. The variation in their setting is due to displacement of nonpolitical advertising due to differences in primary election dates and electoral competitiveness across states and time. In contrast, our specific instrument is the total number of elections for US senators, governors, and mayors that each MSA experienced during the Beyond Petroleum campaign years (2000–2008).²⁸

E. Local Area Characteristics and Measures of Green Preferences

Finally, we link each station to measures of local area characteristics and green preferences. Specifically, we use the 2000 US census to obtain measures of zip code level median household income. We follow prior studies to study green preferences. For example, List and Sturm (2006) uses per capita membership in environmental organizations at the state level. Kahn (2007) uses California Green Party registrations and shows that they are a significant predictor of demand for green products, such as hybrid vehicle registrations. Kahn and Vaughn (2009) creates a green index based on California referendum voting outcomes and Green Party registrations; they document that hybrid vehicles and Leadership in Energy and Environmental Design (LEED) certified buildings cluster in politically green communities. Similar to these prior studies, we collect the following measures and aggregate them into a “Green Index” score:²⁹

• Hybrids cars: share of hybrid-electric vehicle registrations in 2007 in each zip code obtained from R.L. Polk automotive data. We chose the year 2007 to exclude hybrid car purchases caused by the 2008 spike in gasoline prices.

• Sierra Club membership: per capita Sierra Club membership in 2010 at the state level created using data from the Sierra Club and the US Census Bureau.

• LEED buildings: the number of LEED-registered buildings per capita in each zip code, obtained from the US Green Building Council (accessed in June 2011).

• Green Party contributions: average per capita contributions to Green Party committees in 2003–2004 and 2007–2008 at the zip code level, computed using individual level data from the Federal Election Commission.^30,31

We first aggregate these variables by computing z-scores for each measure and taking the sum. We then create an indicator variable “Green Zip” to classify zip codes based on whether they have above or below median Green Index scores.

F. Summary Statistics

Online Appendix Table A1 provides summary statistics for all measures used for our main analysis. Station-level weekly net prices were $0.55 on average during our sample period (2009–2010). For the analysis, we rely on the station-level price difference between the pre- and post-spill period averages. The mean change in price is $0.05 in our sample, which reflects the fact that retail gas prices generally increased between 2009 and 2010. For the 645 unique zips covered by our sample of stations, the average median household income (based on the 2000 census) is approximately $48,000, which was slightly higher than the national median income of $42,000 (Denavas-Walt, Cleveland, and Roemer 2001). Online Appendix Table A1 also shows that the average DMA in our sample received about $1.64 million of BP advertising during the Beyond Petroleum campaign years (2000–2008). For this set of DMAs, there was an average of 7.63 elections during this same period.

Online Appendix Figure A2 provides aggregate summary statistics for the states covered by our sample of stations. Panel C shows that BP advertising prior to the spill was highest in several midwestern (Illinois, Indiana, and Ohio) and southeastern (Florida and Georgia) states. Panel D shows that the states with the highest number of elections during the Beyond Petroleum years were Texas, Missouri, North Carolina, and Michigan. As expected, there is no apparent pattern of geographic clustering in our measure of elections.³²

III. Empirical Analysis

A. The Consumer Response to the BP Oil Spill

We begin by examining the impact of the BP oil spill on station prices and fleet card sales using a difference-in-difference style approach. Our approach designates BP stations as the treatment group, and we use non-BP stations that are not in the same zip code as a BP station as the comparison group. As mentioned, this definition excludes non-BP stations in close proximity because the oil spill may have had an impact on these stations as well.³³

Formally, the model is a regression of station net price or fleet sales on station fixed effects, indicators for the during- and post-spill periods, and interactions of those time period dummies with an indicator of whether a station sells BP-branded gasoline:

$$y_{i,t}={\alpha }_i+{\beta }_1{\mathit{\text{during}}}_t+{\beta }_2{\mathit{\text{post}}}_t+{\theta }_1\left({\mathit{\text{during}}}_t\times B{P}_i\right)+{\theta }_2\left({\mathit{\text{post}}}_t\times B{P}_i\right)+{ϵ}_{i,t}$$ (2)

where y_i,t is an outcome for station i in period t, α_i is a station-level fixed effect, during_t is an indicator equal to one if period t is during the oil spill, post_t is an indicator equal to one if period t is after the spill (defined by the official capping of the leak in September 2010), and BP_i is an indicator of whether station i sells BP-branded gasoline. We cluster standard errors at the zip code level.³⁴

Our preferred specification aggregates net prices and quantities sold at the time-period level. This approach addresses the possibility that autocorrelation in weekly measures of net prices or fleet sales might bias standard errors (Bertrand, Duflo, and Mullainathan 2004). Specifically, we address this concern by collapsing the weekly net price and fleet sales data into averages within three time periods: a pre-spill period (January 1, 2009 through April 16, 2010), a during-spill period (April 23, 2010 through September 17, 2010), and a post-spill period (September 18, 2010 to December 31, 2010). Columns 1 and 2 of Table 1 report results from this level of aggregation. For comparison, columns 3 and 4 of Table 1 provide results using the disaggregated weekly net price and fleet sales data.³⁵

Table 1—

Oil Spill Impacts, Difference-in-Difference Estimates

	Average net price (1)	ln(average fleet sales) (2)	Weekly net price (3)	ln(weekly fleet sales) (4)
During spill	0.056 (0.003)	0.026 (0.007)	0.056 (0.002)	0.039 (0.006)
Post spill	0.006 (0.003)	−0.019 (0.009)	0.006 (0.002)	−0.011 (0.007)
BP × (during spill)	−0.011 (0.003)	−0.042 (0.011)	−0.011 (0.003)	−0.047 (0.009)
BP × (post spill)	−0.004 (0.004)	−0.033 (0.013)	−0.004 (0.003)	−0.037 (0.012)
Observations	15,807	14,400	502,094	456,244
Adjusted R2	0.946	0.970	0.608	0.855
SE cluster	Zip	Zip	Zip	Zip
Weight	Price observations	Quantity observations	Price observations	Quantity observations
Number of zips	847	836	847	836
Number of stations	5,526	4,997	5,526	4,997

Notes: The price and quantity data span January 2009 to December 2010. Columns 1 and 2 report estimates where the dependent variable is the station’s average net price and average log-quantity computed over the entire “pre-,” “during-,” and “post-” spill periods. Columns 3 and 4 report estimates when the dependent variable is the station’s weekly net price and log-quantity. Each specification regresses the dependent variable on dummies for the during-spill period, a dummy for the post-spill period, and their interactions with a dummy for BP gas station. All models control for station fixed effects.

Source: OPIS

All specifications show a negative and statistically significant effect of the oil spill on both prices and fleet sales at BP stations relative to the comparison group. BP stations experienced a relative price decrease of 1.1 cents per gallon and a 4.2 percent drop in sales from fleet customers on average in the five months after the spill.³⁶ Table 2 estimates the month-by-month change in BP prices and fleet sales relative to control stations. In the month after the spill (May 2010), BP stations experienced a peak price decline of 2.9 cents per gallon relative to comparison stations. Fleet sales impacts reached their peak in June 2010 at a 7.5 percent loss (relative to the comparison group). To put these effects in context, the National Association of Convenience Stores (NACS) estimates that the average retail markup was 16.3 cents per gallon in 2010 (NACS 2011). Using this statistic, the largest monthly point estimate represents an 18 percent decline in retail margins. However, we also see that these effects are temporary. In the post-spill period, BP retail station prices rebound, although quantities remain depressed.

Table 2—

Oil Spill Impacts by Month

	Weekly net price (1)	ln(weekly fleet sales) (2)
BP × (late April 2010)	−0.005 (0.003)	−0.010 (0.011)
BP × (May 2010)	−0.029 (0.005)	−0.042 (0.010)
BP × (June 2010)	−0.015 (0.004)	−0.075 (0.011)
BP × (July 2010)	−0.003 (0.003)	−0.053 (0.011)
BP × (Aug. 2010)	−0.014 (0.003)	−0.067 (0.012)
BP × (Sept. 2010)	0.005 (0.004)	−0.011 (0.013)
BP × (Oct. 2010)	−0.007 (0.003)	−0.025 (0.012)
BP × (Nov. 2010)	0.001 (0.004)	−0.039 (0.012)
BP × (Dec. 2010)	−0.007 (0.003)	−0.048 (0.013)
Observations	502,094	456,244
Adjusted R2	0.863	0.863
Fixed effects	Station	Station
SE cluster	Zip	Zip
Weight	Price observations	Quantity observations
Number of zips	847	836
Number of stations	5,526	4,997

Notes: The price and quantity data cover the period from January 2009 to December 2010. The dependent variables in columns 1 and 2 are weekly net price and log-quantity, respectively. Each of these dependent variables is regressed on post-spill month dummies and their interactions with a dummy for BP gas station. All models control for station fixed effects.

Source: OPIS

Figure 1 illustrates our main results by displaying the mean weekly price (level) for the BP and control stations.³⁷ The vertical lines denote the beginning and capping of the oil spill, respectively. Prior to the spill, our sample of BP stations has higher prices, on average, compared to the control group. Almost immediately following the oil spill, this BP premium collapses, consistent with the estimated relative price decline. Over the ensuing months, the BP premium begins to recover.

Figure 1.

Average Weekly Retail Price for BP and Comparison Group Stations

Notes: The figure illustrates average weekly prices for BP and non-BP competitor stations in the main analysis sample. See Section II and the online Appendix for details on the sample construction, and for a zoomed out version of the graph starting at the beginning of our sample in 2009.

Source: OPIS

We also compare these estimated impacts with a measure of public interest in the BP oil spill. Figure 2 plots estimates from Table 2 against Google search intensity for the phrase “oil spill.” In a given month, the Google search intensity is the ratio of searches relative to a baseline month of January 2004. (For example, a value of 50 indicates that searches in a month were 50 times greater than they were in January 2004.) The number of searches for the term “oil spill” intensified dramatically in early May 2010 and peaked on June 4, which was one day after a BP apology campaign began airing. The results suggest that public interest in the spill was significant around the time of the spill, and the evolution of public interest coincides with the peak price and fleet sales impacts.

Figure 2.

Google Search Intensity for BP Oil Spill Searches and Retail Impacts

Notes: The figures display the Google search intensity (blue) for the phrase “oil spill” relative to January 2004. For a given month, the Google search intensity measures the ratio of searches in that month to searches during January 2004 (the baseline). A value of 50 indicates that searches in a month were 50 times greater than in January 2004. The dots (red) plot the month-specific coefficients presented in Table 2. The dependent variables are station weekly net prices and log-quantity, respectively. Each dependent variable is regressed on post-spill month dummies and their interactions with a dummy for BP gas station. All models control for station fixed effects.

Source: OPIS and Google Insights (accessed August 16, 2011)

B. Basic Robustness Checks

We conduct three exercises to examine the robustness of our main results to alternative sample definitions. First, online Appendix Table A3 shows that the results are robust to including non-BP stations within the same zip code as BP stations in the comparison group. The estimated average price impact is attenuated slightly (from −1.1 cents to −0.7 cents per gallon), suggesting that nearby BP competitor stations partially matched the reduction in prices at BP stations.

Second, online Appendix Tables A4 and A5 repeat the main analysis on the unfiltered sample of stations in the OPIS data (located in PADD 1, 2, and 3). Recall that we define the sample to include only stations that are located in zips where there are at least five stations with three prices per week for our entire sample period. In the unrestricted OPIS sample, we obtain results that are very similar in magnitude to the results in Tables 1 and 2.

Third, we also conduct an additional analysis of stations located in areas not subject to summertime gasoline Reid Vapor Pressure (RVP) standard regulations. Such gasoline content regulations can cause local seasonal increases in gasoline prices.³⁸ This is not necessarily a threat to our analysis of net prices because state wholesale prices incorporate all relevant gasoline formulations sold in a given state-week. That is, this measure of cost captures state average price increases due to seasonal gasoline content regulations. At the same time, it is possible that, within each state, BP stations are more likely to be located in zip codes subject to RVP requirements than comparison group stations.

Panels A and B of online Appendix Figure A4 show the evolution of BP and comparison group station price levels over time for stations located in areas that are and are not subject to summertime RVP regulations, respectively. Both series show similar pre-spill trends with higher average prices at BP stations. After the spill, this premium disappears in both regulated and nonregulated areas, with an even stronger decline in the latter. Online Appendix Table A6 expands on these results by providing difference-in-difference results on the subset of zip codes that are not subject to RVP regulation.³⁹ For the net price variable, we also compute this measure after excluding restricted (RVP of 7 or 7.8) gasoline formulations in the wholesale price computation. These results are less precise, but the pattern of the estimates continues to show a significant decline in BP station margins and fleet card sales after the BP oil spill. In sum, our first set of results thus suggest that, on average, BP stations suffered losses to revenues as a result of the BP oil spill, consistent with consumer punishment of BP.⁴⁰

C. The Impact of Beyond Petroleum Advertising

To study the impact of past advertising on the consumer response to the oil spill, we estimate a modified version of the regression model from equation (2) that includes a measure of the sum of BP advertising during the Beyond Petroleum years (2000–2008) as an independent variable. On the one hand, we might expect steeper losses at BP stations in areas with heavier advertising if consumers believed this was a signal of BP’s commitment to environmental stewardship. On the other hand, such advertised claims could have swayed consumer beliefs about whether the disaster was due to bad luck or bad management, thereby mitigating price and fleet sale impacts (Minor and Morgan 2011).

To ease the interpretation of our results, we restrict the sample to the immediate pre-spill (January 1, 2009–April 16, 2010) and during-spill (April 23, 2010–September 17, 2010) periods. We use a regression model where the dependent variable is the station-level difference in average net price or total fleet sales between the pre-spill and during-spill periods. We also include main effect and interaction terms for the Green Index and zip code household income as control variables in this advertising analysis. We demean each of these interaction variables and interact them with an indicator for BP-brand affiliation.

Columns 1 and 2 of Table 3 report ordinary least squares (OLS) results. The results suggest that greater exposure to BP advertising in 2000–2008 mitigated the impact of the oil spill on BP station prices significantly (p-value < 0.05). We fail to detect a corresponding impact of pre-spill advertising on BP station fleet quantity sales. There are at least two interpretations of this pattern of results. First, a negative demand shock accompanied by an outward supply shift (due to BP lowering prices) could result in an equilibrium with lower prices but unchanged quantities. Second, sales to fleet card customers may not be representative of the population segment relevant for station price setting.

Table 3—

OLS and IV Estimates of the Impact of Pre-spill Advertising on Oil Spill Impacts

	OLS estimates		Election IV estimates
			First stage		Second stage	First stage		Second stage
	Price diff. (1)	Sales diff. (2)	Ad. spend. demeaned (3)	BP × (BP ad. spend. demeaned) (4)	Price diff. (5)	Ad. spend. demeaned (6)	BP × (BP ad. spend. demeaned) (7)	Sales diff. (8)
BP	−0.012 (0.003)	−0.035 (0.012)	−2.340 (3.041)	14.028 (1.765)	−0.015 (0.004)	−2.161 (3.025)	14.145 (1.740)	−0.032 (0.019)
Green Index	0.001 (0.001)	−0.005 (0.003)	−0.069 (0.106)	0.000 (0.000)	0.001 (0.001)	−0.079 (0.109)	0.000 (0.000)	−0.005 (0.003)
BP × (Green Index)	−0.001 (0.001)	0.010 (0.004)	0.249 (0.143)	0.180 (0.095)	−0.002 (0.001)	0.274 (0.145)	0.194 (0.096)	0.010 (0.004)
Income, demeaned	0.000 (0.000)	0.000 (0.000)	0.119 (0.025)	−0.000 (0.000)	0.001 (0.000)	0.123 (0.024)	−0.000 (0.000)	0.001 (0.001)
BP × (income, demeaned)	−0.000 (0.000)	−0.002 (0.001)	−0.020 (0.033)	0.099 (0.021)	−0.001 (0.000)	−0.022 (0.032)	0.101 (0.021)	−0.002 (0.001)
Ad spend., demeaned	0.003 (0.000)	−0.000 (0.001)			−0.001 (0.001)			−0.004 (0.003)
BP × (ad spend., demeaned)	0.001 (0.001)	0.001 (0.002)			0.004 (0.002)			0.000 (0.007)
Number of elections, 2000–2008			−1.951 (0.303)	0.000 (0.000)		−1.932 (0.302)	−0.000 (0.000)
BP × (number of elections, 2000–2008)			0.414 (0.370)	−1.538 (0.212)		0.376 (0.368)	−1.557 (0.210)
Constant	0.052 (0.002)	0.017 (0.007)	16.367 (2.476)	−0.000 (0.000)	0.056 (0.003)	16.306 (2.475)	0.000 (0.000)	0.022 (0.008)
Observations	3,748	3,424	3,748	3,748	3,748	3,424	3,424	3,424
SE cluster	Zip	Zip	Zip	Zip	Zip	Zip	Zip	Zip
Number of zips	645	637	645	645	645	637	637	637
Number of stations	3,748	3,424	3,748	3,748	3,748	3,424	3,424	3,424
Kleibergen-Paap Wald F-statistic			26.23	26.23		27.47	27.47

Notes: The sample is restricted to stations with available data on Green Index and household income. The dependent variable is the station’s price difference or log of quantity difference between the “pre-” and “during-” spill periods. The Green Index is the sum of z-scores for four variables: the hybrid share of vehicle registrations at the zip code level in 2007, Sierra Club membership, the number of LEED-registered buildings per capita, and contributions to Green Party committees. Zip code income is in year 2000 US thousand dollars.

Source: OPIS, Sierra Club, the US Green Building Council, the US Census Bureau, and Kantar Media

One concern with these OLS estimates is that advertising may be endogenous to factors that are correlated with local demand response to the BP spill. While we control for confounders such as income and environmental preferences (as proxied by the Green Index), this may not be sufficient. To address this endogeneity concern, we instrument for advertising expenditures with a novel election calendar-based instrument: a count of the number of elections scheduled in each MSA during the Beyond Petroleum campaign years (2000–2008).

Our approach stems from the idea that the number of elections is a relevant instrument because political advertising displaces other types of advertising. Using a similar IV approach, Sinkinson and Starc (2019) showed that the political advertising during the lead up to the 2008 national election resulted in significant displacement on advertising for pharmaceuticals. The exclusion restriction in our context is that cross-sectional variation in elections during the Beyond Petroleum campaign years is not correlated with unobserved factors that determine the consumer response to the BP oil spill. This assumption is plausible since the political calendars are legally preset and should be independent of retail gasoline market factors. (Online Appendix Table A2 shows that we fail to detect significant correlations between our election measure and MSA-level characteristics such as BP station share, median household income, or the Green Index.)

The remaining columns in Table 3 report the election-based IV results. The first-stage estimates in columns 3 and 6 show a highly significant negative effect of scheduled elections on BP ad spending. These results are consistent with the displacement effects detected in Sinkinson and Starc (2019). The performance of our first stage is strong in that the relevant F-statistics (i.e., 26.23 and 27.47 in the price and fleet sale first stages, respectively) compare favorably with recommended critical values.⁴¹

The second-stage results in column 5 of Table 3 show that pre-spill advertising significantly decreased the negative impact of the oil spill on net prices at BP stations (p-value < 0.05). The results imply that a one standard deviation increase in pre-spill ad spending (+$3.6 million) reduced the oil spill’s price impact by 1.44 cents per gallon. This is nearly equal to the total impact of the spill.

D. Robustness Checks for Advertising IV Results

We conduct several exercises to examine the robustness of our advertising results. First, online Appendix Table A7 presents results where the endogenous variable is BP spot TV advertising units as a measure of pre-spill campaign exposure. The results are similar, namely that a one standard deviation increase in units of spot TV advertising (+11.4 ads) is predicted to mitigate the price effect of the BP oil spill by 1.14 cents per gallon (OLS) and 2.28 cents per gallon (IV). Note that this measure counts all spot TV advertising units as equal, whereas the expenditure measure counts advertising dollars as equal.

Second, we address the concern that there is a correlation between pre-spill (2000–2008) and during-spill (March–September 2010) advertising. This is an important concern given that the Kantar data show an increase in BP advertising during the spill months. Moreover, this during-spill marketing could have affected the consumer response given that it included information about relief and mitigation efforts.⁴²

The results in online Appendix Table A8 show that controlling for during-spill advertising does not affect our main results. We obtain similar estimates for the impact of pre-spill advertising after we control for BP during-spill advertising in the IV specification. Interestingly, column 6 of online Appendix Table A8 also suggests a marginally significant positive association between during-spill advertising and fleet card sales during the spill (p-value < 0.10).⁴³

E. Impacts on Station Brand Affiliation

As a final analysis, we examine BP-brand affiliation in the period after the Deepwater Horizon oil spill. Depending on the severity of the impact on station profits, we might expect to see a change in BP station shares. Most gasoline stations are owned or leased by independent dealers who sign long-term contracts with upstream refiners to sell and market a particular brand. If expected returns to the BP brand fall sufficiently low, station owners may switch affiliations. To explore this idea, we measure changes in BP’s share of stations in a zip code before and after the oil spill.

Formally, we estimate the following event study specification:

$${\mathit{\text{MktShare}}}_{z,t}={\mu }_z+\sum _{m=-15}^{-1}{\gamma }_{m}1\left(m=t\right)+\sum _{m=1}^8{\tau }_{m}1\left(m=t\right)+{ϵ}_{z,t},$$ (3)

where the dependent variable is BP’s station share in zip code z in month t. The γ_m terms are coefficients on dummy variables for each of the pre-spill months (before April 2010) and the τ_m terms are coefficients on dummies for each month after the spill (that is, after April 2010). The term μ_z is a zip code fixed effect. The omitted month is thus April 2010. The regression coefficients measure the change in station share relative to April 2010 controlling for zip code fixed effects.

To examine the impact of advertising, we estimate equation (3) separately for zip codes in metropolitan areas with above or below median pre-spill advertising. Panels A and B of Figure 3 display the resulting coefficient estimates on the monthly time dummies (with 95 percent confidence intervals) for zip codes in above- and below-median advertising areas, respectively. Online Appendix Table A10 provides the corresponding regression results.

Figure 3.

Impacts of the Oil Spill on BP Station Market Share

Notes: This figure displays the coefficients on month indicators from a regression of the share of BP stations in each zip code-month indicators, an indicator for BP station status, the interaction of BP station status and month indicators, and zip code fixed effects. The full specification is provided in equation (3) in the text. The omitted month is April 2010, the first month of the oil spill. Results above are provided for zip codes with above and below median BP ad spending during the Beyond Petroleum campaign years of 2000–2008. All point estimates and standard errors are reported in online Appendix Table A10.

Source: OPIS and Kantar Media

While there is a statistically significant loss in market share in below-median advertising areas, there is no detectable decline in areas with higher levels of advertising. The loss in station share is economically meaningful, representing a roughly 6 percent decline (about 0.7 percent relative to the mean BP station share of 11 percent). While our estimates only pertain to a relatively short period after the spill, these results are consistent with the possibility that advertising may have dampened longer term losses to the BP brand outside of the short-run effects on prices and fleet sales. At the same time, it is important to note that this evidence is only suggestive given that we have limited ability to control for additional marketing activities that BP (and competitors) may have undertaken in the longer run period after the spill.

F. Interpretation

Table 3 provides evidence that pre-spill exposure to the Beyond Petroleum advertising campaign softened the negative impact of the BP oil spill. This suggests that firms who provide (unobserved) low levels of environmental quality in production may benefit from green advertising. This is consistent with a broader hypothesis that investments in corporate social responsibility provide reputational insurance in case of adverse events (Minor and Morgan 2011).

One concern for the interpretation of our estimates is whether the observed effects are due to environmental messaging or other advertising (e.g., local marketing by individual service stations) that also occurred during the Beyond Petroleum campaign.⁴⁴ We address this concern by relying on detailed information on the corporate entity of the advertiser and the product advertised in the Kantar data. Our core advertising measure focuses on corporate branding ads for the BP Corporation, BP fuels, and environmental issues during the Beyond Petroleum campaign years (2000–2008), which are likely to contain green messaging. To explore the underlying mechanisms further, we create a second measure that contains advertising only for local BP service stations, BP convenience stores, or ancillary products.⁴⁵

We study the direct and interaction effects of both measures of advertising. Specifically, we interact each advertising measure with the Green Zip indicator to explore whether the effects varied with environmental preferences. A caveat for this analysis is that both types of advertising may be endogenous, but we have one IV. Given this limitation, we report OLS results. One reassurance for these exploratory results is that the OLS estimates in Table 3 appear to be biased toward zero relative to the IV estimates.

Table 4 shows that the protective benefit of pre-spill BP corporate advertising (i.e., likely to contain green marketing) appears to be higher in “greener” zip codes. Column 3 shows that the interaction term between the BP indicator, the likely green advertising measure, and the Green Zip indicator is positive and significant (p-value < 0.01). Column 4 expands on these results by including the main and interaction terms for local and ancillary product advertising. In this specification, the point estimate for the three-way interaction term that includes green advertising (which reflects the protective benefits of core corporate advertising in green areas) is larger in magnitude and significant (p-value < 0.01). In contrast, the three-way interaction term that includes local and ancillary product advertising is opposite signed and marginally significant (p-value < 0.10).⁴⁶

Table 4—

Green (core) and Ancillary Advertising Effects

	Price diff. (1)	Price diff. (2)	Price diff. (3)	Price diff. (4)
BP	−0.012 (0.003)	−0.012 (0.003)	−0.009 (0.004)	−0.010 (0.004)
Green Index	0.001 (0.001)	0.001 (0.001)
BP × (Green Index)	−0.001 (0.001)	−0.001 (0.001)
Green zip dummy			−0.003 (0.004)	−0.003 (0.004)
BP × (Green zip dummy)			−0.005 (0.006)	−0.004 (0.006)
Income, demeaned	0.000 (0.000)	0.000 (0.000)	0.000 (0.000)	0.000 (0.000)
BP × (income, demeaned)	−0.000 (0.000)	−0.000 (0.000)	−0.000 (0.000)	−0.000 (0.000)
Green ad. spending	0.003 (0.000)	0.003 (0.001)	0.003 (0.000)	0.003 (0.001)
BP × (Green ad. spend.)	0.001 (0.001)	0.001 (0.001)	−0.001 (0.001)	−0.004 (0.002)
BP × (Green ad. spend.) × (Green zip)			0.003 (0.001)	0.006 (0.002)
Local/ancillary ad. spend.		−0.001 (0.003)		−0.001 (0.003)
BP × (local/ancillary ad. spend.)		0.002 (0.003)		0.008 (0.004)
BP × (local/ancillary ad. spend.) × (Green zip)				−0.007 (0.004)
Constant	0.052 (0.002)	0.052 (0.002)	0.053 (0.002)	0.053 (0.003)
Observations	3,748	3,748	3,748	3,748
SE cluster	Zip	Zip	Zip	Zip
Number of zips	645	645	645	645
Number of stations	3,748	3,748	3,748	3,748

Notes: The dependent variable is the station-level price difference (the average net price during the pre–spill period minus the average during-spill period). The advertising measures control for demeaned BP advertising expenditures during the Beyond Petroleum campaign years (2000–2008). “Green ad.” includes advertising related to the BP Corporation, BP fuels, and environmental issues. “Local/ancillary ad. spend.” includes other BP service station related ads such as for convenience stores and products and individual service stations. The indicator “Green zip” equals one for stations in zip codes whose Green Index measure is above the median.

Source: OPIS, Sierra Club, the US Green Building Council, the US Census Bureau, and Kantar Media

IV. Policy Implications

The role of private firms and markets in providing public goods has been a subject of debate going back at least to works such as Coase (1960) and Friedman (1970). Kitzmueller and Shimshack (2012) notes that the literature has broadly moved from asking whether private markets should provide corporate social responsibility (CSR) to whether the market can provide incentives for CSR.⁴⁷ One strand in this research examines how strategic market interactions between firms and activists—“private politics”—can result in CSR provision (Baron 2003, Baron and Diermeier 2007). Another set of papers analyze markets for “impure public goods,” which bundle private products with public good creation or the abatement of public “bads” (Besley and Ghatak 2001, 2007; Kotchen 2006). In these models, private provision of public goods requires that consumers value environmental stewardship and punish firms for deviating from promised (advertised) product attributes.⁴⁸

While consumers appear to be willing to punish BP in the aftermath of an environmental disaster, we find that pre-spill corporate advertising softened this response. These results provide large-scale and revealed preference evidence that consumers respond to green advertising in a way that may give firms an incentive to “greenwash” by exaggerating the environmental benefits or qualities of its products (or operations). This interpretation implies that the market’s ability to effectively reward corporate social responsibility and provide public goods may be limited if CSR is communicated through advertising.

There are at least three policies to address greenwashing. First, governments increasingly use truth-in-advertising regulations to reduce misleading environmental product claims (Rohmer 2007). For example, the US Federal Trade Commission (FTC) publishes a “Green Guide” on environmental friendliness claims. Second, third parties can evaluate environmental claims. Third, governments can use price-based mechanisms such as penalties and taxes.

A few considerations suggest that price-based policies may be most effective in encouraging firms to uphold CSR promises or provide environmental stewardship in production. Truth-in-advertising policies may have limited effectiveness because authorities often fail to punish violations (Delmas and Burbano 2011). In addition, marketing survey experiments have found that simply including nature-based images, symbols, or sounds in product advertising can promote a “green” brand image (Parguel, Benoit-Moreau, and Russell 2015). Indeed, the Beyond Petroleum campaign included elements such as changing the corporate logo to feature a helios sun and airing commercials that featured nature-based imagery and statements about environmental ideals. This type of “executional greenwashing” is likely more difficult to regulate than “claim greenwashing,” which is based on explicit claims about product attributes (Parguel, Benoit-Moreau, and Russell 2015). In addition, another concern is that consumers may have heterogeneous responses to truth-in-advertising or third-party information. For example, Rao and Wang (2017) finds that new consumers are the ones who respond most to an FTC order to terminate false ads.

Price-based mechanisms may be a first-best policy solution since penalties or taxes force firms to internalize the costs of externalities. While third-party information and truth-in-advertising regulation can mitigate greenwashing in certain contexts, they may not address the issue comprehensively. Indeed, our results indicate that green advertising can be effective in muting the kind of “punishment” response that is theoretically necessary to incentivize firm investment in practices that support environmental sustainability.

V. Conclusion

We study how environmental advertising affects consumer behavior in retail gasoline markets. After the BP Deepwater Horizon oil spill in 2010, we show that there were significant and large declines in BP station prices and gasoline sales to fleet card customers. During the decade preceding the spill, BP embarked on a large green marketing campaign, and we find that greater exposure to this advertising significantly dampened the negative impacts on retail station prices. We also find suggestive evidence that past advertising cushioned BP from longer run negative impacts on gasoline market share.

Our results are consistent with the hypothesis that advertising led consumers to believe that the spill was due to bad luck rather than to negligent practices. This supports the idea that firm expenditures on CSR function more as insurance (Minor and Morgan 2011). This suggests that advertising may fail to provide incentives for firms to undertake investments in hidden product attributes such as environmental stewardship in production. One implication of our findings is that price-based mechanisms such as penalties may be necessary to enhance efficiency in markets where consumers seek environmentally friendly products or services.