ADVERTISING AS INSURANCE OR COMMITMENT? EVIDENCE …€¦ · (Solman, 2008).5 Anecdotally,...

1

ADVERTISING AS INSURANCE OR COMMITMENT? EVIDENCE FROM THE BP OIL SPILL*

by

Lint Barrage† Eric Chyn‡ Justine Hastings♦

Brown University University of Michigan Brown University NBER NBER

This Draft: May 2016

ABSTRACT This paper explores how advertising impacts the consumer response to news about unobserved product quality. Specifically, we estimate how British Petroleum’s (BP) 2000-2008 “Beyond Petroleum” advertising campaign affected the impact of the 2010 BP oil spill. We find that BP station margins declined by 4.2 cents per gallon, and volumes declined by 3.6 percent after the spill. However, pre-spill advertising significantly dampened the price response in the short-run, and reduced the fraction of BP stations switching brand affiliation in the long-run. Our results suggest that advertising provides insurance against adverse events. We discuss implications for private provision of environmental stewardship. *Previous versions of this manuscript were circulated with the title: “Advertising, Reputation, and Environmental Stewardship: Evidence from the BP Oil Spill.” We thank Ryan Kellogg, Matthew Kahn and Richard Schmalensee and Jesse Shapiro for helpful comments. Phillip Ross provided outstanding research assistance. Hastings gratefully acknowledges funding through Brown University, Department of Economics and Population Studies and Training Center. Chyn gratefully acknowledges support from an NICHD training grant to the Population Studies Center at the University of Michigan (T32 HD007339). The Online Appendix is available at the following URL: www.justinehastings.com/images/downloads/BCH_OnlineAppendix. †[email protected], Dept. of Economics, Robinson Hall, 64 Waterman Str., Providence, RI 02912 ‡[email protected] ♦[email protected]

2

1 Introduction

How does advertising shape consumer behavior and firm incentives to

undertake costly, hidden investments in product quality? Theoretical models

generate ambiguous predictions as to whether advertising serves as an informative

commitment to provide product quality (Shapiro, 1983; Cabral, 2005), or plays a

persuasive role that protects firms even in the event of negative product news

(Minor and Morgan, 2011). Hence, the relationship between advertising and

product quality is an open empirical question.

This paper provides novel evidence on this question by studying the impact

of advertising on consumers’ response to news about product quality. Specifically,

we study the consumer response to the British Petroleum (BP) Deepwater Horizon

Oil Spill in 2010, one of the largest oil-related environmental disasters to date.1

Prior to the spill, BP undertook one of the largest and most successful corporate

advertising campaigns entitled “Beyond Petroleum.” Between 2000 and 2008, BP

rebranded its gasoline stations with a new logo – a Helios (sun) symbol – and a new

name behind the BP acronym (Beyond Petroleum replaced British Petroleum).

Both moves were designed to reflect the company’s newly stated dedication to

environmental stewardship – a commitment to take more expensive production

decisions to mitigate environmental degradation. The campaign launched with a

$200 million budget and won a prestigious advertising award from the American

Marketing Association in 2007. Anecdotally, these marketing efforts appeared to

have an effect as U.S. consumer surveys and press reviews consistently rated BP as

1 In April 2010, an oil well blowout caused multiple explosions and led to the eventual sinking of the Deepwater Horizon oil drilling rig. An estimated 205.8 million gallons of oil flowed from the well in the ensuing weeks (National Commission, 2011). Despite containment efforts, the spill led to the world’s largest accidental release of oil into marine waters. On November 5, 2012, BP formally pled guilty to charges of environmental crimes, and agreed to pay $4 billion to settle its criminal case with the United States government (United States of America v. BP Exploration & Production, Inc. CDN: 2:12-cr-00292-SSV-DEK).

3

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 oil spill are a natural

setting for measuring the impact of news about unobserved quality on consumer

demand, and testing whether pre-period advertising investments dampened or

amplified the demand response. We combine detailed data on gasoline station

prices and sales from January 2009 to March 2011 with supplemental data on both

metropolitan-level BP advertising data during the 2000s and measures of local area

environmental preferences. This allows us to estimate the impact of the spill on

retail demand for BP gasoline and examine how effects varied over time and across

areas with different levels of pre-spill advertising exposure and green preferences.

We find the following. First, there was a significant consumer response to

the BP oil spill. BP retail prices declined 4.2 cents per gallon relative to non-BP

stations in neighboring markets. This represents a 25 percent decrease in margins

relative to industry standards. In addition, BP volumes declined by 3.6 percent

among our sample of station customers (fleet card holders). Further, over the course

of the spill, BP prices and volumes fell with increasing intensity: the negative

impact of the spill peaked at a 6.1 cents per gallon price decrease and a 6.7 percent

volume loss in August 2010.

Second, the estimated impact is significantly stronger in areas where

consumers exhibit greener preferences. Following List and Sturm (2006), Kahn

(2007), and Kahn and Vaughn (2009), we create a Green Index based on local

demand for green products, as well as memberships in and contributions to

environmental organizations. We find the impact was more intense in areas with

stronger green preferences and less intense in higher-income areas, all else equal.

The positive correlation between green preferences and income mitigated the

impact on BP retail performance in “green” markets.

4

Third, we find that the consumer response to the spill was significantly

reduced by pre-spill exposure to BP advertising. We measure advertising using data

from Kantar Media (formerly known as CMR, TNS Media Intelligence, and KMR

Group).2 The data include BP’s monthly advertising units and expenditures across

newspaper, billboard, radio, television, and internet by metropolitan area. Our core

ad spending 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 use market-level TV spot prices as an instrument for variation in

BP advertising across cities. We find that the impact of the oil spill on BP prices

was significantly less severe in areas with more BP pre-spill advertising. These

results are robust to a variety of specification checks such as controlling for BP’s

corporate advertising during the spill and for other types of advertising that may

have affected demand for BP-branded retail gasoline stations.

Finally, we also find long-term effects of the oil spill. The impact on BP

prices and quantities changed sharply after the leak was sealed in September 2010.

BP prices increased to slightly higher than pre-spill averages relative to stations in

comparison markets; however, fleet card volume sales remained significantly

lower.3 In addition, we find that markets with low pre-spill advertising suffered

greater losses in BP retail outlet share. We find significant losses in BP’s share of

stations beginning around the time of the largest price impacts. The losses amount

to a 5 percent decline relative to the mean and occur only in areas with low pre-

spill advertising, suggesting that in these areas, during-spill profit losses may have

been large enough to cause station owners to switch to alternative brands.

2 TNS Media Intelligence acquired Competitive Media Research (CMR) in 2003. Kantar acquired both TNS and KMR group in 2008 (Chou et al. 2008, Clark et al. 2009). 3 We provide a discussion of the interpretation of price versus quantity effects in Section 4.

5

Overall, our results suggest that BP’s investment in the Beyond Petroleum

advertising campaign cushioned the impact of the spill on demand. There are

several possible explanations for this result. Minor and Morgan (2011) argue that

expenditures on corporate social responsibility can provide insurance against

reputational costs after product recalls by shifting beliefs about whether the event

was due to negligence or bad luck. In this sense, advertising plays more of a

persuasive role (Dixit and Norman, 1978; Schmalensee, 1976; Becker and Murphy,

1993) than an informative role (Butters, 1977; Grossman and Shapiro, 1984),

shifting valuations for a good rather than providing information and commitment

to quality. Alternatively, this effect could also be generated by positive brand

recognition or non-environmental brand value (such as habit formation) that

buoyed demand despite revelations of lower-than-advertised environmental quality

(Clark et al., 2009). While we only observe one history of BP advertising, we

provide suggestive evidence on the protective effect of reputation-building through

the environmentally-themed Beyond Petroleum campaign versus local and

ancillary product ads that are more likely to affect demand through the latter

channel. While both seem to have a positive effect, our results are consistent with

a larger protective effect of environmentally-themed corporate advertisements in

greener areas.

Our short- and long-run findings have potential implications for public

policy. Specifically, governments (or other organizations) may be able to enhance

market efficiency by monitoring environmental stewardship claims. Such efforts

may provide additional incentives for firms to internalize externalities.

2 Background

In July 2000, BP launched a $200 million public relations campaign focused

on aligning the BP brand with environmental issues (PR Watch, 2010). The

company introduced a new slogan, “Beyond Petroleum,” and redesigned its logo to

6

a green and yellow Helios sun. New advertising focused on environmental

stewardship4 and emphasized that BP was making its operations more efficient and

working to reduce environmental impacts (Cherry and Sneirson, 2011). The

campaign won two PR Week “Campaign of the Year” awards and received the

prestigious Gold Effie Award from the American Marketing Association in 2007

(Solman, 2008).5

Anecdotally, consumers appeared to retain this environmental messaging.

In 2008, the marketing firm Landor Associates surveyed consumers, asking “How

green do you consider [BP] to be?” Survey results showed 33 percent believed BP

was a “green” brand, and respondents ranked BP as the greenest of the major

petroleum companies (Landor Associates, Cohn & Wolfe, and Penn, Schoen, &

Berland Associates, 2007, 2008). A 2008 poll of 1,000 U.K. marketers ranked BP

as third when asked which company made the greatest commitment to

environmental issues (Marketing Week, 2008).6

Why did BP undertake this costly investment in environmental branding?

Broadly speaking, empirical work has found that advertising generally increases

demand for advertised products (e.g., Ackerberg, 2001; Bagwell, 2007; Dube and

Manchanda, 2005; Bertrand et al., 2010; Clark et al., 2009; Simester et al., 2009;

Lewis and Reiley, 2008; Hastings et al., 2013; Gurun et al. 2013). Previous research

has also shown that consumers are willing to pay for environmental stewardship as

a product attribute (e.g., Kiesel and Villas-Boas, 2013; Kahn and Vaughn, 2009;

Kahn, 2007; Teisl et al., 2002; Roe et al., 2001; Nimon and Beghin, 1999; Goett et

4 For example, one TV ad featured a narrator asking “Is it possible to drive a car and still have a clean environment?” and “Can business go further and be a force for good?” Speaking on the behalf of BP, the narrator affirms: “We think so” (BBC News, 2000). 5 PR Week, Brand Development Campaign of the Year (winner), International Campaign of the Year (honorable mention), Internal Communications Campaign of the Year (winner) for “Taking BP Beyond” (PR Week, 2010) 6 At the same time, several environmental and advocacy groups, such as Greenpeace and Corpwatch, criticized BP’s re-branding as “greenwashing” (Corpwatch, 2000).

7

al., 2000; Forsyth et al., 1999; De Pelsmacker et al., 2006; Loureiro et al., 2001).

Yet, while there may be demand for environmental quality, consumers do not know

whether a product has this attribute in the absence of third party certification.

Since environmental quality is unobserved at the time of purchase, this

suggests that there are at least two different motivations for firms to invest in

advertising. On the one hand, some theoretical models have shown that firms

investing in hard-to-observe product attributes (such as environmental stewardship)

can use advertising as a sunk cost to credibly signal their investment in product

quality (Shapiro, 1983; Milgrom and Roberts, 1983; Cabral, 2005). Alternatively,

advertising could play a persuasive role that convinces consumers that negative

events are accidental and occur due to “bad luck.” This model was proposed by

Morgan and Minor (2011) in the context of corporate social responsibility claims,

and shares a persuasive flavor with Dixit and Norman (1978), Schmalensee (1976)

and Becker and Murphy (1993). In this context, advertising can change customers’

beliefs about underlying firm actions and acts as insurance to reduce the chance

that customers interpret bad outcomes as due to shirking. This mitigates consumer

punishment, decreasing firm incentives to follow through with product quality

promises.7,8 In this sense advertising is a substitute for - instead of complement to

- investments in unobserved product quality.

7 More broadly, models of ex-ante unobservable product quality provision have found that firms must face financial sanctions for false product quality claims (such as advertising) as incentives for equilibrium quality provision (see Cabral (2005) for a survey of this literature). Models of private provision of public goods have similarly formalized this point (Besley and Gathak, 2007). In addition, punishment may be more difficult if deviation is hard to detect. In our setting, negative news about environmental stewardship may only occur probabilistically. Consumers must infer events are the result of shirking on quality promises, and decrease demand accordingly. 8 Several studies have analyzed the impacts of negative product news on demand, such as recalls of consumer products (e.g., Crafton et al., 1981; Reilly and Hoffer, 1983; Minor and Morgan, 2011; Freedman et al., 2012), airplane crashes (e.g., Borenstein and Zimmerman, 1988) and lawsuits involving medical services (Dranove et al., 2012). They do not examine advertising and baseline claims of product quality.

8

With this in mind, the BP Deepwater Horizon oil spill provides a unique

setting to test whether advertising plays more of an informative or persuasive role.

Shortly after the conclusion of the Beyond Petroleum advertising campaign, an oil

well blowout caused multiple explosions and the eventual sinking of the Deepwater

Horizon rig in April 2010. Afterward, robotic monitoring devices discovered that

oil was leaking from the damaged well. Over the next few months BP engineers

sought to contain the oil leak, but were unsuccessful until a “containment dome”

was placed over the leaking well in July 2010. 10 With the capping of the well,

government-appointed scientists estimated that nearly 205.8 million gallons of oil

had leaked from the well (Department of Interior, 2010). On September 19, 2010,

BP completed the relief well, and officials declared that the damaged well was

“effectively dead.” Subsequent investigations confirmed that the cause of the spill

was attributable to active management decisions on behalf of BP.12

Our analysis begins by estimating the impact of the BP oil spill on station-

level retail gasoline prices and volumes (as measured in our customer sample of

fleet card holders). We then examine how the consumer response varied across

markets that varied in two key dimensions: their willingness to pay for

environmental products (measured using a variety of proxies) and their exposure to

BP’s corporate advertising preceding the spill. The latter constitutes our test for

whether advertising had a persuasive effect. Specifically, we examine whether BP

stations suffered greater losses in markets which received high levels of pre-spill

advertising. In addition, we explore whether advertising had an impact in the long-

by examining changes in the share of stations affiliated with the BP brand.

10 Aigner et al. (2010). 12 A non-partisan commission found that “the immediate cause of the blowout could be traced to a series of identifiable mistakes made by BP” and its contractors, further concluding that “(w)hether purposeful or not, many of the decisions that BP, Halliburton and Transocean made that increased the risk of the Macondo blowout clearly saved those companies significant time (and money)” (National Commission, 2011). The Department of Justice concluded that “the explosion of the rig was a disaster that resulted from BP’s culture of privileging profit over prudence” (DOJ, 2011).

9

3 Data

3.1 Gasoline data

We use data on retail gasoline prices, sales to fleet-card customers, and station

brand affiliations to estimate the impact of the BP oil spill on gasoline prices, sales,

and long-run branding decisions. The data come from the Oil Price Information

Service (OPIS), which collects information on gasoline station prices and sales

from two sources. First, OPIS records information on prices and volumes from

Wright Express fleet fuel card “swipes”. Wright Express reports the last transaction

of the day at each station to OPIS and calculates a price based on that transaction’s

total sales amount and gallons sold.13 This information is available only for stations

that accept this fleet card and available only on days when fleet card transactions

happen (i.e., an individual must use their fleet card for a price to be recorded for a

particular station on a particular day).14 The fleet card is widely accepted across the

U.S. Second, since 2009, OPIS has expanded its data collection to include reporting

agreements with several gasoline refiner-marketers that provide retail prices for

some stations that do not accept the fleet card.15

Between these two sources, the OPIS data have a price observation for over

100,000 stations in the United States. However, most stations are available only for

a portion of the years 2009-2011 or have sporadically reported prices. Given our

interest in station-level variation in prices and sales over time, we focus on zip

13 As with all scanner data, this can result in errors in prices. Because only the last purchase of the day is reported, it is more difficult to clean out errors than in scanner data for which many purchases are recorded for the same product each day. Prices are more accurate in recent years as more purchases are recorded for more stations each week and the data become easier for Wright Express and OPIS to clean. We drop only one percent of price observations based on large one-day changes in prices indicative of an error in data. Note that for gasoline stations that offer personalized discounts (e.g. grocery store chains), variation in OPIS retail prices may reflect both changes in street price as well as differences in per-gallon discounts available to the customer who post the last purchase of the day. 14See also Busse et al. (2013) for another description of these data. 15 For a list of stations that accept the fleet card see www.wrightexpress.com.

10

codes in which OPIS reporting meets minimum density criteria.16 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.

In our empirical results, we compare prices at BP stations to a control group

of stores in zip codes without any BP stations present. To be clear, this control

group excludes non-BP stations in close proximity to BP stores as their prices were

likely impacted by the spill as well. This leaves us with a sample of 7,503 stations.

As a robustness check, we reproduce our main analysis using all of the OPIS data,

regardless of whether stations are missing large portions of data or whether most

competitors in the station’s area are not in the OPIS data. The results for this

unfiltered sample are very similar and can be found in Online Appendix Section II.

For stations in our sample that accept fleet cards (as opposed to stations

whose parent companies only report prices to OPIS), we observe weekly total

gasoline sold through fleet cards. Although fleet card customer preferences may

be different than the population average, these data provide a glimpse into the

consumer response to the events of interest. While limited, these data represent, to

our knowledge, the only station-level volume data currently available.17 We follow

an analogous procedure to select zip codes with sufficient fleet sales coverage (see

Online Appendix). For the volumes data, we are left with 6,735 stations of which

6,709 are also in our price sample. Again robustness checks using the entire sample

of treatment and control stations produce very similar results and are reported in

the Online Appendix.

In addition to prices and fleet sales, each observation includes a station's

location, brand of gasoline, and brand of convenience store in each week. Our main

16 Further details on how we clean the data and define our sample are in the Online Appendix. 17 The alternative panel data on gasoline sales volumes of which we are aware are state-aggregated (over all brands and suppliers) sales volumes reported to the Energy Information Administration (EIA) by oil companies through survey responses (Hastings and Shapiro, 2013).

11

analysis uses each station’s initial brand in our sample (from January 2009) to

categorize it as a BP or non-BP station in order to avoid potential brand endogeneity

due to stations switching away from the BP brand after the spill. We analyze such

switching behavior in a separate analysis in Section 4.3.

Finally, we use weekly gasoline spot prices from the Energy Information

Administration (EIA) to compute a measure of retail margins (EIA, 2011).

Specifically, we define a weekly station-level net price as the average price for

station i in week t less the average New York spot price in week t:18

it it tnetprice AveRetailPrice EIANewYorkSpot (1) We focus on weekly net prices to abstract from daily variation and 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, we weight weekly price

and quantity observations by the underlying number of daily observations within

the week.

3.2 Advertising data

We measure advertising using Kantar Media Ad$pender data which report

expenditures by date and marketplace for more than three million brands across 18

media formats.19 Kantar uses tracking technologies and services to monitor

television advertising on both cable and network stations, print media expenditures

from over a thousand business-to-business and consumer magazine and news

publications, and internet sites. They collect outdoor and local radio advertising

18 We use the NY spot price instead of the Gulf spot price because several hurricanes hit this area during our sample period, causing a few instances of spot price spikes that were not reflected in our NY spot or retail price series. 19 The 18 media types provided by Kantar Media include network television, spot television, cable television, Spanish language network television, syndication, magazines, business-to-business magazines, Sunday magazines, Hispanic magazines, local magazines, national newspapers, local newspapers, Hispanic newspapers, network radio, national spot radio, local radio, U.S. Internet and outdoor activities.

12

information from other marketing subscription services and directly from media

providers (e.g., radio stations or billboard plant operators).20 Given a fixed

combination of time period, market, and media type, advertising expenditure data

are hierarchically categorized through product levels that identify the parent

company (e.g., BP vs. Shell), distinguish between brands (e.g., BP service station

vs. Amoco service station) and differentiate between products to which a brand is

attached (e.g., BP energy utilities vs. BP gasoline).

Our data set tracks BP advertising from 2000 through 2011 and all other

advertising from 2007 through the 2011.21 In our main specification we use

advertisements during the years of the Beyond Petroleum campaign (2000-2008)

that focused on the BP Corporation, BP fuel products, and environmental issues.

Our main analysis aggregates all advertising expenditures across all media as our

measure of advertising exposure. This specification assumes there are stock effects

of advertising on demand (Dube and Manchanda, 2005).

Since BP advertising may be endogenous to each area’s unobserved

preference for the BP brand, we instrument for BP’s advertising using television

advertising spot prices across all industries and product categories. We focus

specifically on the quantity-weighted average spot television advertising price from

2007-2008. This price provides a measure of advertising cost differences across

metropolitan areas.22 Our identifying assumption is that cross-sectional differences

in demand and supply for general spot television advertising do not lead to

differences in the consumer response to the BP oil spill other than through their

20 For more details, see Ad$pender manual (Kantar Media, 2011). See also other papers that have used these data, including Saffer and Dave (2006), Reuter and Zitzewitz, (2006), Chou et al., (2008), Clark et al. (2009) and Gurun et al. (2013). 21 Ad$pender data licenses cover a rolling five year period; historic data must be purchased separately and at a significant premium. 22 We match the Kantar data, which are at the Designated Market Area (DMA) level, to zip codes using the county-DMA correspondence provided by Gentzkow and Shapiro (2008), in conjunction with a county-zip correspondence from the U.S. Department of Housing and Urban Development.

13

impact on BP advertising levels. Note that previous studies in advertising use this

type of instrument (Dube and Manchanda 2005; Izuka and Jin 2005; Choi, Shin-

Yi, and Grossman, 2008; Liu and Gupta 2011; Dinner, Van Heerde, and Neslin

2014). We discuss the plausibility of the identifying assumption in section 4.2.1.

3.3 Measures of Green Preferences

The literature characterizes green preferences in a variety of ways. For example,

List and Sturm (2006) use 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) create a green index based on

California referendum voting outcomes and Green Party registrations; they

document that hybrid vehicles and LEED-certified (“green”) buildings cluster in

politically green communities. Building on this literature, we compile and combine

the following measures to create a green index:23

1) Hybrids: 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.

2) Sierra: Per capita Sierra Club membership in 2010 at the state level created

using data from the Sierra Club and the U.S. Census Bureau.

3) LEED: The number of LEED-registered buildings per capita in each zip,

obtained from the U.S. Green Building Council (accessed in June 2011).

23 We also experimented with including measures of Democratic Party committee contributions and Barack Obama’s vote share from the 2008 presidential election. However, these measures appeared to decrease the explanatory power of the green index.

14

4) 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.24,25

We aggregate these variables into a single “Green Index” by computing Z-

scores for each of the measures and summing them. We also consider each zip

code’s hybrid vehicle share as an alternative measure of green preferences.

4 Empirical Analysis

4.1 Pooled results

We begin by examining the impact of the BP oil spill on station prices and

fleet card sales. We regress station net price or fleet sales on station fixed effects,

indictors for during- and post- spill periods, and interactions of those time period

dummies with an indicator of whether a station sells BP-branded gasoline:

1 2 1 2it i t t t i t i ity during post during BP post BP

(2)

Here, ity is either average net price or the log average fleet sales for station i in

period t, i is a station-level fixed effect, duringt is an indicator if period t is during

the oil spill, postt is an indicator if period t is after the spill, and BPi is an indicator

of whether station i sells BP-branded gasoline.

We aggregate daily prices and quantities at two levels. First, a concern is

that autocorrelation in net prices or fleet sales data might bias the standard errors

(Bertrand et al., 2004). To address this, we collapse all weekly net price and fleet

sales data into averages within three time periods: a pre-spill period (January 01,

2009 through April 16, 2010), a during-spill period (April 23, 2010 through

24 The Federal Election Commission data cover all individual contributions over $200. 25 To maintain comparability with income data, contributions are converted to 1999 dollars using the Bureau of Labor Statistics’ CPI inflation calculator.

15

September 17, 2010), and a post-spill period (through March 2011). Results from

this aggregation are presented in Table 1, columns 1 and 2. Second, we use weekly

net price and fleet sales data for comparison in Table 1, columns 3 and 4.26

Across specifications we find that there is a negative, economically and

statistically significant effect of the oil spill on both prices and sales at BP stations

relative to the control group. BP stations experienced a relative price decrease of

4.2 cents per gallon and a 3.6 percent drop in sales from fleet customers.27 This

decrease in net price is substantial, given that the National Association of

Convenience Stores estimates that the average retail mark-up was 16.3 cents per

gallon in 2010 (NACS, 2011). Using this statistic, the point estimate represents a

26 percent decline in retail margins. These effects are, however, temporary: in the

post-spill period, retail station prices at BP stations rebound although quantities

remain depressed.

Figure 1 displays the mean weekly price (level) for the BP and control

stations in our sample. The vertical lines denote the beginning and sealing of the

oil spill, respectively. For much of the period 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, the mean price for BP falls below the control

price until the spill is capped. Several months following the spill, BP’s prices rise

above control station prices. This pattern is consistent with the following

interpretation: advertising increased demand from marginal consumers pre-spill,

those consumers decreased demand during and after the spill. BP re-optimized post-

spill to their new demand curve to sell to the most loyal, but smaller subset of

26 In both specifications, the aggregate observations for each station in each time period are weighted by the number of underlying observations from the disaggregated (daily) data. 27 Because our measure of volume comes from fleet sales, we prefer reduced-form regressions for price and quantity. Using our data to estimate structural parameters of the change in preferences resulting from the spill would require an assumption that fleet sale demand is the same as non-fleet sale demand (which we do not observe). In addition, as prices and sales are not available at all stations, estimating a demand system based on a random utility model is problematic.

16

consumers. If these consumers were less price elastic, BP’s new equilibrium price

should increase and quantity sold should fall.

Table 2 estimates the month-by-month change in BP prices and fleet sales

relative to control stations. After the spill, BP stations experienced a small,

immediate drop in net price (1 cent per gallon) with no discernible impact on fleet

sales. Net prices continued to fall, bottoming out in August at -6.1 cents per gallon.

During the same month, BP stations experienced a 6.7 percent reduction in fleet

sales compared to control stations. At this point, nearly 205.8 million gallons of oil

had spilled into the Gulf and only 17 percent had been captured by BP’s

containment efforts (New York Times, 2010).28 By October, the price impact had

declined to 0.5 cents per gallon, with quantities remaining lowered by 2.4 percent.

Figure 2 plots the point estimates from Table 2 against Google search

intensity relative to January 2004 for the phrase “oil spill.” For a given month, the

Google search intensity is measured as the ratio of searches in that month to

searches during a baseline month. Here, the baseline month is January 2004, so a

value of 50 indicates that searches in a baseline 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 4th, one day after a

BP apology campaign began airing. The results suggest that public interest in the

spill was significant and that the relative magnitude of the price response appears

to lag the spike in online searches.

Our identifying assumption is that, aside from the oil spill, there was no

shock to gasoline prices (and quantity sold to fleet vehicles) that affected BP and

competitor stations differentially from non-BP/non-BP competitor stations in the

aftermath of the oil spill. Although plausible, this assumption could be violated if,

for example, BP stations are more likely to be in zip codes that are less (more) likely

28 Among the rest, eight percent had been burned or skimmed, 25 percent evaporated or dissolved, 24 percent dispersed either naturally or chemically and 26 percent still at sea or on shore.

17

to be subject to summertime gasoline Reid Vapor Pressure (RVP) standard

regulations than zip codes in which our control group stations lie.29 This could

disproportionately drive down (up) the relative price of gasoline in markets with

BP stations in the summer, as content regulations can cause local seasonal increases

in gasoline prices through increased production costs. Because the BP spill

occurred during the spring and summer of 2010, differential regulations could be a

confounding factor.

Table 3 restricts the sample to zip codes with no seasonal gasoline content

regulation (uniform RVP of 9.0). The results show a stronger overall BP price

decrease of 7.5 cents per gallon. Fleet sales impacts cease to be significant, although

the point estimate remains negative. It should be noted that the Table 3 specification

reduces our sample size by over 70 percent. Indeed, when considering a larger

sample of standard RVP zip codes from the unfiltered OPIS data (i.e., not restricted

to our list of “good” sample zip codes), the quantity impacts are stronger and remain

highly significant in this specification as well (see Online Appendix Section II).

Overall, seasonal changes in RVP gasoline content requirements do not appear to

be driving our results.

These findings suggest that, on average, BP stations suffered losses to

revenues as a result of the BP oil spill. Our results are consistent with both short-

run punishment and a more permanent loss of some customers post-spill. They are

consistent with models of trust, where a consumer expects a firm to behave a certain

way and punishes it for deviating from that behavior for a period of time, and with

reputation models, where consumers expect firms to be a particular type (e.g., high

quality) and update their beliefs permanently in response to an experience

sufficiently different from their expectation. Trust models primarily address moral

hazard (e.g., shirking on promised quality effort), whereas reputation models

29 See Brown et al. (2008) and Auffhammer and Kellogg (2011) for detailed descriptions of gasoline content regulations.

18

primarily deal with adverse selection (e.g., low quality types pretending to be high

quality types). Both may have happened for different consumers, generating the

observed changes in prices and sales during and after the spill.

Note that trust models that involve many consumers suffer from a similar

problem to voting; punishment is not individually rational as each individual

consumer’s demand is not sufficiently large enough to affect aggregate outcomes

or incentives.30 This may explain why consumers organize boycotts as coordinated

responses to firm behavior, as many did during the BP spill.31,32 Alternatively, Fehr

and Gaechter (2000) find in laboratory experiments that subjects are willing to

expend resources to punish deviating players even in a single-shot trust game,

where such punishment cannot incentivize better future behavior, suggesting that

punishment of bad behavior may have intrinsic value.

30 See the literature on the paradox of not voting (e.g., Downs, 1957; Olson, 1965; Palfrey and Rosenthal, 1985; Feddersen, 2004). 31 Calls for boycotting BP stations were issued by voices including Public Citizen, Jesse Jackson, and the Backstreet Boys, who reportedly completed their 2010 tour without stopping at BP stations to refuel their tour bus (Backstreet Boys, 2010). 32 Models of civic duty, peer pressure and group voting have been put forward as social mechanisms to overcome the paradox of not voting. See for example Gerber and Green (2000), Green and Gerber (2004) and Coate and Conlin (2004).

19

4.2 Interaction and advertising effects

Table 4 examines how the price and sales impacts vary with measures of

local green preferences and income. We merge onto our base data zip code level

income data from the 2000 U.S. Census, the share of all registered cars in a zip code

that are hybrid vehicles, and our Green Index as described in Section 3. We focus

on the pre-spill versus during-spill periods to facilitate interpretation of interaction

terms. Our regression reduces to a pure difference-in-difference estimation, with

the difference in net price or total sales during the spill versus the pre-spill period

at each station i as the dependent variable. We demean each of our interaction

variables (income in 2000 U.S. thousands of dollars, hybrid share of registered

vehicles) and interact them with an indicator for BP brand affiliation.

The first two columns repeat the results in Table 1 on the subsample of

stations for which the Green Index, hybrid car shares, and income data are all

available. The results are essentially unchanged. Columns 3 and 4 add controls and

interactions for income and hybrid shares. Income has a positive and significant

association with the price changes at BP stations, indicating that the negative impact

of the spill was abated in high-income areas. A one standard deviation increase in

income (of $15,563) implies a 1.55 cents per gallon (0.001*$15.563) smaller price

decrease than the average. This difference represents an approximately 39 percent

reduction in the price decrease relative to the overall impact of -4 cents per gallon.

The smaller price effects seen in high income areas may be driven by gasoline

station selection and by higher valuation of convenience. We find a negative and

significant association between income and quantity sold through fleet cards. A one

standard deviation increase in income at the zip code level reduces BP volumes

during the spill period by an additional 3 percentage points (-0.002*$15.563)

relative to our sample mean of -3.6 percent. Thus, while BP prices drop less in high-

20

income areas, BP fleet card customer sales drop more, though we note that fleet

card sales may not be reflective of overall demand relevant for price setting.

Price effects were larger in areas with larger shares of hybrid vehicles. The

results imply that a one-standard deviation increase in hybrid vehicle share is

associated with an additional 0.6 cent per gallon (-0.012*0.5%) drop in BP retail

gasoline prices in the aftermath of the spill. However, the hybrid vehicle share

interaction term is not a significant predictor of changes in BP sales after the spill.

Columns 5 and 6 of Table 4 substitute our Green Index for percentage of hybrid

vehicles, as described in Section 3, compiling measures of green preferences used

by List and Sturm (2000), Kahn (2007) and Kahn and Vaughn (2009). Using this

measure, we again find that greener areas responded more strongly to the BP oil

spill. The coefficient on Green Index implies that a one standard deviation increase

in the Index intensifies price decreases by 0.94 cents per gallon (-0.006*1.56), or a

23.4 percent further decrease relative to a mean decrease of 4 cents per gallon. We

do not find a significant interaction effect between the Green Index and changes in

fleet-card volume sold at BP stations, however fleet card sales may not be reflective

of overall demand relevant for price setting.

Finally, Table 5 adds interactions with demeaned BP advertising

expenditures to test if advertising during the Beyond Petroleum campaign is

associated with higher or lower price and sales impacts. Our main specification

measures advertising as total expenditures aggregated over all forms of advertising

in our Kantar data, which includes television, newspapers, magazines, radio,

billboards and Internet spending (Clark et al., 2009) for ads that focused on the BP

Corporation, BP fuel products, and environmental issues during the Beyond

Petroleum campaign years (2000-2008). If this advertising convinced consumers

of BP’s commitment to the environment through investments in production

processes that provide an environmental public good (or reduce negative

externalities), one might expect to see steeper losses at BP stations in areas with

21

heavier Beyond Petroleum advertising. On the other hand, in the early days of the

spill, such advertised claims could have swayed consumers’ beliefs about whether

the disaster was due to bad luck or bad management, leading to softer price and

sales impacts (Minor and Morgan, 2011).

The first two columns of Table 5 replicate the benchmark results from Table

1 for the sample of stations that have income, green preference, and advertising

data available. The average impact of the spill is slightly smaller in this sample, but

remains economically and statistically significant. Columns 3 and 4 add demeaned

advertising and its interactions with an indicator if the station was a BP station and

an indicator for the post-spill period. The results suggest that pre-spill exposure to

BP advertising significantly dampened the impact of the oil spill. The point estimate

on the interaction term BP*Advertising suggests that a one standard deviation

increase in advertising expenditure softened the price impact of the spill by about

1 cent per gallon (0.003*3.4), resulting in a 24 percent decline in the price impact

of the spill. The effects of the spill on BP station prices in high income and high

Green Index areas remain unchanged; the coefficients on these interaction terms

are similar to those in Table 4. We find no significantly different effect of the spill

on quantities sold in areas exposed to more versus less advertising. On the one hand,

a negative demand shock accompanied by an outward supply shift (i.e., BP

lowering prices sufficiently) may result in an equilibrium with lower prices but

unchanged quantities. On the other hand, sales to fleet card customers may not be

representative of the population segment relevant for station price-setting, as

discussed previously.

4.2.1 Instrumental Variables and Identification of Advertising Effects

Advertising may be endogenous to other factors that are correlated with

local demand response to the BP spill. For example, advertising may be correlated

with BP station market share. Market share may also be correlated with customer

perceptions of BP brand quality or with the set of alternative non-BP brand stations

22

they could substitute towards. Suppose that advertising prices were correlated with

BP’s share of gasoline stations in a metropolitan area or with the number of gasoline

station options. In this case, advertising would be correlated with consumer

response to the oil spill as BP customers would have fewer non-BP gasoline options

nearby, and would therefore be less responsive to the spill in their choice of station.

To address this endogeneity concern, we instrument for advertising

expenditures using spot television advertising prices. Several papers in the

literature develop similar instruments for advertising (Dube and Manchanda 2005;

Izuka and Jin 2005; Choi, Shin-Yi, and Grossman, 2008; Liu and Gupta 2011;

Dinner, Van Heerde, and Neslin 2014).33 We use the quantity-weighted average

spot price in the late Beyond Petroleum campaign years (2007-2008), when we

have advertising data for all brands and all products in all product categories and

industries (e.g., automobiles, clothing, etc.). First stage results are reported in full

in the Online Appendix Table A0. To summarize, spot TV advertising prices are a

highly significant predictor of advertising expenditures. The Shea’s partial R-

squared value is 0.69 in the first stage. Formal tests of instrument relevance strongly

reject the null that the first stage coefficients on the excluded instruments are equal

to zero (e.g., the Angrist-Pischke F-statistic leads to a rejection of the null with a p-

value<0.0000).

The instrumental variables results in columns 5 and 6 of Table 5 are very

similar in magnitude to the OLS results in columns 3 and 4. That is, our IV results

confirm that the price effects of the spill were softer in areas where BP advertised

33 Most similarly, Dube and Manchanda (2005) use the list price of gross rating points (an advertising measure), Choi, Shin-Yi, and Grossman (2008) use the price of advertising computed as dollars per seconds of messages aired (as well as the number of households in a DMA with a television set), and Izuka and Jin (2005) compute average wages in advertising-related occupations to capture advertising costs. Also relying on broad advertising market measures are Liu and Gupta (2011), who instrument for statin drug advertising with average advertising expenditures across all pharmaceutical firms and other drugs, and Dinner, Van Heerde, and Neslin (2014), who use non-direct competitor firm’s advertising expenditures as instrument for firm’s advertising expenditures.

23

more heavily during the Beyond Petroleum Campaign years. Indeed, the coefficient

on price is stronger in the IV specification (0.4 cents per gallon spill impact

protection per $1 million additional advertising expenditure), suggesting, if

anything, that BP advertising was potentially higher in areas where it would have

been punished more.

The instrument is valid under the assumption that spot prices are determined

by the broad advertising market. This assumption would be violated if spot prices

were instead determined by factors endogenous to the demand elasticity at BP

gasoline stations per se; these factors would dampen the demand response to the

oil spill in the absence of increased advertising. In the Online Appendix we

investigate correlations between our instrument and other local area characteristics

that could affect the demand response to the spill. Table A5 shows that spot TV

prices vary positively and significantly with population density, but there is no

detectable relationship with retail gasoline market concentration (HHI), BP station

share, or gasoline station density. This suggests that spot advertising prices are

orthogonal to key factors that might impact demand response at BP stations to the

BP spill, such as BP market share and retail gasoline brand market concentration.

We also conduct several specification checks which directly control for the

characteristics of local markets and which could affect the demand elasticity of BP

gasoline stations. Columns 3 and 6 in Table A6 report results from specifications

which add our measure of BP’s market share in the metropolitan area to our main

advertising IV specification. The results further confirm that BP station share is

uncorrelated with our instrument since the point estimates on advertising’s

interaction with BP are very similar to our results in columns 5 and 6 of Table 5.

Similarly, Table A7 also tests the robustness of our IV results by adding interactions

with measures of the number of gas stations per square mile at the zip code level to

our IV specifications. Adding these measures to our IV estimation has no impact

on our advertising results, further confirming that our IV findings are not driven by

24

station density or concentration through more or fewer stations to substitute towards

in response to the spill.

Since our instrument is specific to TV expenditures, we conduct another

robustness check for our analysis by focusing on BP’s spot TV advertising only

(the excluded media are billboards, newspaper, radio and online spending). Online

Appendix Tables A8-A9 show that focusing only on BP’s spot TV advertising

yields very similar results to our main analysis based on all media expenditures: a

one standard deviation increase in BP’s spot TV expenditures (+$2.2 mil) reduces

the oil spill’s impact on BP prices by 0.9 (OLS) and 1.3 (IV) cents per gallon. The

instrument yields slightly higher Shea’s partial R-squared values in the first stage

regression as spot TV market prices are stronger determinants of spot TV

advertising for BP than they are for all-media advertising. As before, we find no

statistically significant advertising effect on quantities. Lastly, when we measure

TV advertising in units of advertising we get similar results to using expenditures,

namely that a one standard deviation increase in units of spot TV advertising

(+1,080 ads) is predicted to mitigate the price effect of the BP oil spill by 1.1 (OLS)

and 3.2 (IV) cents per gallon. Note that this measure counts all spot TV advertising

units as equal whereas the expenditure measure counts advertising dollars as equal.

4.2.2 Interpretation

In summary, the positive and significant impact of advertising suggests that,

rather than responding more strongly to the spill, consumers in high-advertising

metropolitan areas were less likely to shift away from BP, lowering the impact of

the spill on BP station prices. This result suggests that firms that provide low

environmental quality in production may benefit from environmentally themed

corporate advertising. Our results provide empirical support for the notion that

investments in corporate branding may provide reputational insurance in case of

adverse events, as suggested by Minor and Morgan (2011) for firm branding

through investments in corporate social responsibility

25

Two main issues arise in interpreting these results. First, it may be the case

that during-spill advertising is correlated with pre-spill advertising. Our data show

an increase in BP advertising during the spill. These marketing efforts included

informational advertising about relief and mitigation efforts (Tracy, 2010), which

could have stemmed the impact of the spill on demand. We thus control for BP

advertising during the oil spill in an augmented version of the main specification in

Table 5. Table A10 shows that our estimates are robust to including during-spill

advertising. Interestingly, column 2 shows that the price impact of during-spill

advertising is also precisely estimated and has a slightly larger positive effect (per

dollar of advertising) on reducing the consumer response to the oil spill.

A second issue for our interpretation is controlling for other forms of

advertising that may have affected demand at BP stations and been positively

correlated with Beyond Petroleum advertising (e.g., local ads by individual service

stations and convenience stores). To address this concern, we exploit the fact that

the Kantar data contain information on the corporate entity of the advertiser and the

product advertised. Our main advertising measure focuses on corporate branding

ads for the BP Corporation, BP fuels, and environmental issues, which were also

likely to have contained Beyond Petroleum messaging. For our supplementary

analysis, we create a second measure of advertising specific to local BP service

stations, BP convenience stores, and ancillary products. (See the Online Appendix

for further details.)

Using these data, we compare the effect of both categories of advertising.

One caveat for this analysis is that both types of advertising may be endogenous,

but we have only one instrument. Given this limitation, we report OLS results only.

One reassurance for these results is that the similarity between the OLS and IV

estimates in our main specification suggests that the endogeneity bias in these

advertising estimates is minimal.

26

Table A11 shows that the estimated effect of our core corporate advertising

measure from the Beyond Petroleum campaign is robust to controlling for other

types of advertising that may have affected demand for BP retail gasoline stations.

Column 2 shows that the point estimate for the impact of our core advertising

measure is only slightly smaller than our main specification estimate re-produced

in Column 1. Specifically, the point estimate shrinks from 0.3 to 0.2 cents per

gallon per $1 million of corporate advertising during the campaign. Although

imprecise, the point estimate for local and ancillary products advertising is positive,

which suggests that these ads also cushioned the consumer response to the oil spill

at BP stations. This may have occurred through channels such as habit formation

or consumer loyalty (e.g.,to a local station owners).34

4.2.3 Long-run impact on station brand affiliation

Depending on the severity of the impact on station owners’ profits, we

might expect to see a long-run impact on BP through loss of station share as retailers

switch affiliations to other brands. 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 of gasoline.35 If expected returns to the BP brand fall low

enough, station owners may switch brand affiliations. This is a second, longer-term

measure of the spill’s impact on demand and long-run supply. We measure changes

in BP’s share of stations across zip codes before and after the oil spill, as well as

how these patterns differ with BP advertising.

Specifically, we estimate the following specification:

34 Prior literature suggests that advertising may operate through these additional channels. For example, Clark et al. (2009) also use Kantar advertising data linked to survey data on quality and brand awareness for firms across many sectors. They find that advertising has a larger impact on brand awareness than on quality perception (they do not, however, distinguish between advertising campaigns targeted at communicating quality versus brand awareness). 35 Although many stations are not convenience stores, the National Association of Convenience Stores describes contracting and pricing generally among its members (NACS, 2012)

27

, ∑ 1 ∑ 1 , (3) where the dependent variable is BP’s station share in zip code z in month t, are

coefficients on dummy variables for each of the pre-spill months (before April

2010), are coefficients on dummies for each month after the spill (that is, after

April 2010) and are zip code fixed effects. 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. We estimate (3) separately for zip codes

in metropolitan areas with above or below median BP ad spending during the

Beyond Petroleum campaign years of 2000-2008. Figures 3A and 3B 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.

Table A1 in the Online Appendix provides the corresponding regression tables.

The figures show no significant decline in station share in zip codes in high-

Beyond Petroleum advertising areas, but a significant loss in below-median areas.

The losses appear about six months after the oil spill, coinciding with the largest

monthly drop in prices and sales volumes according to Figure 2. The loss in station

share is sizeable, representing a five percent decline (-0.5% relative to a sample

mean station share of 9.67%). The comparison of outlet share changes between

areas with high and low pre-spill advertising suggests that advertising dampened

longer term losses to BP in addition to softening the short-run negative impact of

the spill on prices and sales.

4.3 Implications for Corporate Social Responsibility

An emerging applied theory literature has set out to explain the economic

forces behind the private provision of public goods, motivated in part by the

increasing popularity of corporate social responsibility (CSR) and environmental

28

branding (the Beyond Petroleum campaign being one example).36 One strand in

this research examines how strategic market interactions between firms and

activists – “private politics” – can result in CSR provision (e.g., 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).37 In these

models, private provision of public goods requires (i) consumers to value

environmental stewardship, and (ii) consumers to punish firms for deviating from

promised (advertised) product attributes.38

While we find that consumers value environmental stewardship, we also

find that pre-spill corporate advertising during the Beyond Petroleum campaign

softened the negative demand shift away from BP-branded gasoline. This finding

is consistent with the idea that advertising provided reputational insurance, thus

playing a persuasive role rather than serving as a commitment for BP to invest in

environmental quality.

Ideally, we would differentiate the effects of advertised environmental

stewardship from the effects of generic corporate branding that may also cushion

against a negative demand shock. This would be done by observing the impacts of

two separate advertising campaigns pre-spill, one with green messaging and one

36 The majority of Americans now expect companies to engage in socially responsible practices such as environmental stewardship in production (Fleishman-Hillard and National Consumers League, 2007). Companies appear to be responding: A 2011 KPMG study found that 95 percent of Global Fortune 250 companies publicly report their social and environmental efforts (KPMG, 2011). In 2008, more than 3,000 companies provided reports dedicated solely to highlighting corporate social and environmental activities (Lydenberg and Wood, 2010). 37 Kitzmueller and Shimsack (2012) discuss these papers in a review on the CSR literature. 38 Other empirical evidence linking CSR investments and social bads include Kotchen and Moon (2011), who provide backward-looking evidence that firms with past “social irresponsibility” subsequently invest in CSR. They regress combinations of companies’ current Kinder, Lydenberg, Domini Research & Analytics social responsibility indices on lagged values to test if past poor ratings (as measures of corporate social “irresponsibility”) predict future good ratings (as measures of corporate social “responsibility”). Relatedly, Eichholtz et al. (2009) find that firms in certain ‘dirty’ industries, such as oil and mining, are more likely to lease green office space.

29

without. We can provide suggestive evidence by comparing the effects of our core

corporate advertising measure with the effects of local and ancillary product ads.

To do this we augment our advertising specification by adding interactions between

the indicator for BP stations, each measure of advertising, and an indicator for

whether a station is located in a zip code that has an above median green index

score. While the estimates for these additional interaction terms are noisy, the

results in Column 4 in Table A11 suggest that in high-green-preference markets,

the Beyond Petroleum advertising had a larger dampening impact on demand

response to the spill. In low-green-preference markets, however, the local station

advertising had the larger dampening impact on demand response. This suggests

that green advertising had larger protective effects where customers value the

green-ness of their gasoline, while in markets where gas station loyalty is more

likely driven by ancillary product services, advertising those products may have

been more effective at preserving demand (perhaps habit formation is a potential

mechanism here). This suggests that firm have incentives to build an advertising

cushion on the dimension that local customers value most.

Overall, our results are consistent with the notion that consumers value

environmental stewardship, but that their response to green advertising may give

firms an incentive to “greenwash”.39 Though suggestive, 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.

These findings support the need for public or private environmental certification to

monitor green product claims and suggest that regulation may be necessary to

provide the incentives for firms to internalize the environmental repercussions of

their production decisions.

39 Greenwashing describes when firms mislead consumers about the environmental benefits and qualities associated with its products.

30

5 Conclusion

This paper studies how advertising affects the consumer response to new

information about product quality. We explore this topic in the context of BP’s

2000-2008 Beyond Petroleum advertising campaign and the subsequent BP

Deepwater Horizon oil spill. Specifically, we estimate the effect of the oil spill on

BP gasoline prices and sales, and examine how the spill’s impact varied over time

and across areas with different levels of green preferences, demographics, and

exposure to BP corporate advertising. We find a statistically and economically

significant (relative) decline in BP stations’ prices and gasoline fleet card customer

sales. This is consistent with a demand shift away from BP-branded gasoline in

response to the spill. We also find that station margins suffered significantly larger

losses in areas that exhibit green preferences as measured by proxies such as hybrid

vehicle ownership or Green Party donations. This finding relates to a literature

linking political green preferences with consumers’ retail purchasing behavior (e.g.,

Kahn, 2007; Kahn and Vaughn, 2009) and provides evidence that consumers may

be voting with their wallets to incentivize environmental protection.

Our analysis also shows that pre-spill exposure to BP advertising

significantly dampened the spill’s impacts on BP stations’ prices. During the

decade preceding the oil spill, BP embarked on a large and celebrated marketing

campaign to brand itself as an environmentally friendly company. In the absence

of formal certification schemes, advertising is a way for firms to signal and commit

to product quality, including for environmental stewardship. However, our results

suggest that corporate advertising may have led consumers to attribute the oil spill

to bad luck rather than to negligent practices, potentially playing a persuasive rather

than an informative role about environmental practices. This is consistent with the

notion that expenditures on CSR may function more as insurance (Minor and

Morgan, 2011). Finally, we also find that advertising cushioned BP from long-run,

31

negative impacts on sales as it decreased the fraction of gasoline stations who re-

branded to other brands in the aftermath of the spill.

We conclude that our results suggest that advertising may fail to provide

incentives for firms to undertake investments in hidden product quality attributes

such as environmental stewardship in production. With regards to green advertising

in particular, one implication of this finding is that there may be a need for public

or private environmental certification to monitor green product claims, and that

regulation may be necessary to provide the incentives for firms to internalize the

environmental repercussions of their production decisions.

32

References

Ackerberg, Daniel, “Empirically Distinguishing Informative and Prestige Effects of Advertising,” RAND Journal of Economics, 32, no. 2 (2001), 316-333.

Aigner, Erin, Joe Burgess, Shan Carter, Joanne Nurse, Haeyoun Park, Amy Schoenfeld, and Archie Tse, “Tracking the Oil Spill in the Gulf,” New York Times, Updated August 2, 2010. URL: http://www.nytimes.com/interactive/2010/05/01/us/20100501-oil-spill-tracker.html [Accessed: 03/10/2013]

Auffhammer, Maximilian and Ryan Kellogg, “Clearing the Air? The Effects of Gasoline Content Regulation on Air Quality,” American Economic Review, 101, no. 6 (2011), 2687-2722.

Backstreet Boys Website, 2010, URL (accessed June 2011): http://backstreetboys.com/news/11968

Bagwell, Kyle, “The Economic Analysis of Advertising,” in Handbook of Industrial Organization, vol. 3, 1st ed., Mark Armstrong and Robert Porter, eds. (New York: Elsevier, 2007.)

Baron, David. P., “Private Politics,” Journal of Economics & Management Strategy, 12, (2003), 31–66.

Baron, David and Daniel Diermeier, “Strategic Activism and Nonmarket Strategy,” Journal of Economics and Management Strategy, 16, (2007), 599-634.

BBC News, “BP Goes Green,” available at http://news.bbc.co.uk/2/hi/business/849475.stm, 2000, accessed 01/07/2013.

Becker, Gary and Kevin Murphy, “A Simple Theory of Advertising as a Good or Bad,” Quarterly Journal of Economics, 108, no. 4 (1993), 941-964. Bertrand, Marianne, Esther Duflo, and Sendhil Mullainathan, “How Much Should We Trust Differences-In-Differences Estimates?” Quarterly Journal of Economics, 119, no. 1 (2004), 249-275. Bertrand, Marianne, Dean Karlan, Sendhil Mullainathan, Eldar Shafir, and Jonathan Zinman, “What’s Advertising Content Worth? Evidence from a

33

Consumer Credit Marketing Field Experiment,” Quarterly Journal of Economics, 125, no. 1 (2010), 263-306.

Besley, Timothy and Maitreesh Ghatak, “Government versus Private Ownership of Public Goods,” Quarterly Journal of Economics, 116, no. 4 (2001), 1343-1372.

–––––––––––, “Retailing Public Goods: The Economics of Corporate Social Responsibility,” Journal of Public Economics, 91, no. 9 (2007), 1645-1663.

Borenstein, Severin and Martin Zimmerman, “Market Incentives for Safe Commercial Airline Operation,” American Economic Review, 78, no 5 (1988), 913-935.

Brown, Jennifer, Justine Hastings, Erin T. Mansur, and Sofia B. Villas-Boas, “Reformulating Competition? Gasoline Content Regulation and Wholesale Gasoline Prices,” Journal of Environmental Economics and Management, 55, no. 1 (2008), 1-19.

Busse, Meghan, Christopher Knittel, and Florian Zettelmeyer, “Are Consumers Myopic? Evidence from New and Used Car Purchases,” American Economic Review, 103, no. 1 (2013), 220-56.

Butters, Gerard R., “Equilibrium Distributions of Sales and Advertising Prices,” Review of Economic Studies, 44, no. 3 (1977), 465-91.

Cabral, Luís M. B., “The Economics of Trust and Reputation: A Primer,” New York University and CEPR Working Paper, 2005.

Chou, Shin-Yi, Inas Rashad, and Michael Grossman, “Fast Food Restaurant Advertising on Television and Its Influence on Childhood Obesity,” Journal of Law and Economics, 51, no. 4 (2008), 599-618.

Cherry, Miriam A. and Judd F. Sneirson, “Beyond Profit: Rethinking Corporate Social Responsibility and Greenwashing After the BP Oil Disaster,” Tulane Law Review, 85, no. 4 (2011), 983.

Clark, C. Robert, Ulrich Doraszelski, and Michaela Draganska, “The Effect of Advertising on Brand Awareness and Perceived Quality: An Empirical Investigation Using Panel Data,” Quantitative Marketing Economics, 7, no. 2 (2009), 207-236.

34

CNN, “BP Apology Campaign Begins Airing,” June 3, 2010, available at http://www.cnn.com/2010/US/06/02/oil.spill.bp.apology/index.html, accessed on January 7, 2013.

Coate, Stephen and Michael Conlin, “A Group Rule-Utilitarian Approach to Voter Turnout: Theory and Evidence,” American Economic Review, 94, no. 5 (2004), 1476-1504.

Corpwatch, “BP: Beyond Petroleum or Beyond Preposterous,” December 14, 2000, available at http://www.corpwatch.org/article.php?id=219, accessed on January 7, 2013.

Crafton, Steven, George E. Hoffer, and Robert Reilly, “Testing the Impact of Recalls on the Demand for Automobiles,” Economic Inquiry, 19, no 4 (1981), 694-703.

Department of Interior. “U.S. Scientific Teams Refine Estimates of Oil Flow from BP’s Well Prior to Capping,” August 2, 2010, available at http://www.doi.gov/news/pressreleases/US-Scientific-Teams-Refine-Estimates-of-Oil-Flow-from-BP-Well-Prior-to-Capping.cfm, accessed on January 7, 2013.

Department of Justice, “BP Exploration and Production Inc. Agrees to Plead Guilty to Felony Manslaughter, Environmental Crimes and Obstruction of Congress Surrounding Deepwater Horizon Incident,” November 15, 2011, available at http://www.justice.gov/opa/pr/2012/November/12-ag-1369.html, accessed on January 7, 2013.

De Pelsmacker, Patrick, Wim Janssens, Ellen Sterckx, and Caroline Mielants, “Fair-trade Beliefs, Attitudes and Buying Behavior of Belgian Consumers,” International Journal of Nonprofit and Voluntary Sector Marketing, 11, no. 2 (2006), 125-138.

Dinner, Isaac M., Harald J. van Heerde, and Scott A. Neslin. "Driving Online and Offline Sales: The Cross-channel Effects of Traditional, Online Display, and Paid Search Advertising." Journal of Marketing Research (2014).

Dixit, Avinash, and Victor Norman. "Advertising and welfare." Bell Journal of Economics (1978): 1-17Dranove, David, Subramaniam Ramanarayanan, and

35

Yasutora Watanabe, “Delivering Bad News: Market Responses to Negligence,” Journal of Law and Economics, 55, no. 1 (2012), 1-25.

Dube, Jean-Pierre and Puneet Manchanda, “Differences in Dynamic Brand Competition Across Markets: An Empirical Analysis,” Marketing Science, 24, no. 1 (2005), 81-95.

Downs, Anthony, An Economic Theory of Democracy, (New York: Harper and Row, 1957).

Eichholtz, Piet, Nils Kok, and John M. Quigley, “Why Do Companies Rent Green? Real Property and Corporate Social Responsibility,” Berkeley Program on Housing and Urban Policy Working Paper W09-004, University of California, Berkeley: Institute of Business and Economic Research, 2009.

Energy Information Administration, Spot Prices, URL (accessed 2011): http://www.eia.gov/dnav/pet/pet_pri_spt_s1_d.htm

Environmental Leader News, January 18, 2008. URL (accessed May 2015): http://www.environmentalleader.com/2008/01/15/beyond-petroleum-pays-off-for-bp/Feddersen, Timothy J., “Rational Choice Theory and Paradox of Not Voting,” Journal of Economic Perspectives, 18, no. 1 (2004), 99-112.

Fehr, Ernst, and Simon Gaechter, "Cooperation and Punishment in Public Goods Experiments," American Economic Review, 90, no. 4 (2000), 980-994.Fisk, Margaret Cronin, and Laurel Brubaker Calkins, “BP Gulf of Mexico Spill, From Disaster to Trial: Timeline,” Bloomberg, February 24, 2013, URL: http://www.bloomberg.com/news/2013-02-25/bp-gulf-of-mexico-spill-from-disaster-to-trial-timeline.htm [Accessed: 3/15/2013]. Fleishman-Hillard and National Consumers League, “CCSR: Collaborative Corporate Social Responsibility,” 2007, http://innovation.fleishmanhillard.com/index.php/2010/02/09/ccsr-collaborative-corporate-social-responsibility/, accessed on January 7, 2013.

Forsyth, Keith, David Harley, and Robert Kozak, “Will Consumers Pay More for Certified Wood Products?” Journal of Forestry, 97, no. 2 (1999), 18-22.

Freedman, Seth, Melissa Kearney, and Mara Lederman, “Product Recalls, Imperfect Information, and Spillover Effects: Lessons from the Consumer

36

Response to the 2007 Toy Recalls,” Review of Economics and Statistics, 94, no. 2 (2012), 499-516.

Gentzkow, Matthew, and Jesse M. Shapiro, “Introduction of Television to the United States Media Market, 1946-1960,” Inter-University Consortium for Political and Social Research, ICPSR Study No. 22720, 2008.

Gerber, Alan S. and Donald Green, “The Effects of Personal Canvassing, Telephone Calls, and Direct Mail on Voter Turnout: A Field Experiment,” American Political Science Review, 94, no. 3 (2000), 653-664.

Goett, Andrew A., Kathleen Hudson, and Kenneth E. Train, “Customer’s Choice Among Retail Energy Suppliers: The Willingness-to-Pay for Service Attributes,” Energy Journal, 21, no. 4 (2000), 1-28.

Green, Donald and Alan S. Gerber, Get Out the Vote! (Washington D.C.; Brookings Institution Press,2004).

Grossman, Gene M. and Carl Shapiro, “Informative Advertising with Differentiated Products,” Review of Economic Studies, 51, no. 1 (1984), 63-81.

Gurun, Umit G., Gregor Matvos and Amit Seru, “Advertising Expensive Mortgages,” NBER Working Paper No. 18910, 2013.

Hastings, Justine, Ali Hortascu, and Chad Syverson, “Advertising and Competition in Privatized Social Security: The Case of Mexico,” NBER Working Paper No. 18881, 2013.

Hastings, Justine and Jesse Shapiro, “Fungibility and Consumer Choice: Evidence from Commodity Price Shocks,” The Quarterly Journal of Economics, 128, no. 4 (2013), 1449-1498.

Iizuka, T., & Jin, G. Z. (2005) “Drug advertising and health habit” NBER Working Paper w11770.Kahn, Mathew, “Do Greens Drive Hummers or Hybrids? Environmental Ideology as a Determinant of Consumer Choice,” Journal of Environmental Economics and Management, 54, no. 2 (2007), 129-145.

Kahn, Mathew and Ryan Vaughn, “Green Market Geography: The Spatial Clustering of Hybrid Vehicles and LEED Registered Buildings,” Bell Journal of Economic Analysis and Policy, 9, no. 2 (2009), 1-22.

37

Kantar Media Ad$pender Manual, 2011, URL (accessed May 2015): http://products.kantarmediana.com/documents/AdSpenderManual.pdf

Kiesel, Kristin and Sofia Villas-Boas, “Can Information Costs Affect Consumer Choice? Nutritional Labels in a Supermarket Experiment, ” International Journal of Industrial Organization, 31, no. 2 (2013), 153–163.

Kitzmueller, Markus and Jay Shimsack, “Economic Perspective on Corporate Social Responsibility,” Journal of Economic Literature, 50, no. 1 (2012), 51-84.

Kotchen, Matthew J., “Green Markets and Private Provision of Public Goods,” Journal of Political Economy, 114, no. 4 (2006), 816-834.

Kotchen, Matthew. J., and Jon Jungbien Moon, “Corporate social responsibility for irresponsibility,” NBER Working Paper No. w17254, 2011.

KPMG, “KPMG International Survey of Corporate Responsibility Reporting 2011,” 2011, available at http://www.kpmg.com/Global/en/IssuesAndInsights/ArticlesPublications/corporate-responsibility/Pages/2011-survey.aspx, accessed on April 3, 2013.

Landor Associates, Cohn & Wolfe, and Penn, Schoen, & Berland Associates, “2008 ImagePower® Green Brands Survey,” 2008.

Landor Associates, Cohn & Wolfe, and Penn, Schoen, & Berland Associates, “Green Brands 2.0; An ImagePower® Survey,” May 2007.

Lewis, Randall and David H. Reiley, “Does Retail Advertising Work? Measuring the Effects of Advertising on Sales via a Controlled Experiment on Yahoo!” Yahoo Research Manuscript, 2008.

List, John A. and Daniel Sturm, “How Elections Matter: Theory and Evidence from Environmental Policy,” Quarterly Journal of Economics, 121, no. 4(2006), 1249-1281.

Liu, Qiang, and Sachin Gupta. "The impact of direct-to-consumer advertising of prescription drugs on physician visits and drug requests: empirical findings and public policy implications." International Journal of Research in Marketing 28.3 (2011): 205-217.

38

Loureiro, Maria L., Jill J. McCluskey, and Ronald C. Mittlehammer, “Assessing Consumer Preferences for Organic, Eco-Labeled and Regular Apples,” Journal of Agricultural and Resource Economics, 26, no. 2 (2001), 404-416.

Lydenberg, Steve and David Wood, How to Read a Corporate Social Responsibility Report: A User’s Guide,. (Boston: Boston College Institute for Responsible Investment, 2010.

Marketing Week, “M&S Tops WM/YouGov Green Issues Marketers Poll,” March 26, 2008, available at http://www.marketingweek.co.uk/ms-tops-mw/yougov-green-issues-marketers-poll/2060134.article, accessed on January 7, 2013.

Milgrom, Paul, and John Roberts. "Price and advertising signals of product quality." The Journal of Political Economy (1986): 796-821.

Minor, Dyland, and John Morgan, “CSR as Reputation Insurance: Primum Non Nocere,” California Management Review, 53, no. 3 (2011), 40-59.

National Association for Convenience & Fuel Retailing (NACS), 2011, URL (accessed November 27, 2012): http://www.nacsonline.com/NACS/Resources/campaigns/GasPrices_2011/Documents/GasPriceKit2011.pdf

National Association for Convenience & Fuel Retailing (NACS), 2012, URL (accessed November 27, 2011): http://www.nacsonline.com/NACS/Resources/campaigns/GasPrices_2012/Documents/NACSFuelsReport2012__HowBrandedStationsOperate_Pink.pdf.

National Commission on the BP Deepwater Horizon Oil Spill and Offshore Drilling, Chief Counsel’s Report, 2011, available at http://www.oilspillcommission.gov/chief-counsels-report, accessed 01/07/2013.

New York Times, “Tracking the Oil Spill in the Gulf,” August 2, 2010, available at http://www.nytimes.com/interactive/2010/05/01/us/20100501-oil-spill-tracker.html, accessed on January 7, 2013.

Nimon, Wesley and John Beghin, “Are Eco-Labels Valuable? Evidence from the Apparel Industry,” American Journal of Agricultural Economics, 81, no. 4 (1999), 801-811.

39

Olson, Mancur, The Logic of Collective Action. (Cambridge: Harvard University Press,1965.)

Palfrey, Thomas R., and Howard Rosenthal, “Voter Participation and Strategic Uncertainty,” American Political Science Review, 79, no. 1 (1985), 62-78.

PR Watch. “BP’s Beyond Petroleum Campaign Losing its Sheen,” May 3, 2010, available at http://www.prwatch.org/node/9038, accessed on January 7, 2013.

PR Week, Brand Development Campaign of the Year (winner), International Campaign of the Year (honorable mention), Internal Communications Campaign of the Year (winner) for ‘Taking BP Beyond’”, 2001, available at http://www.ogilvypr.com/about/awards, accessed on January 7, 2013.

Reilly, Robert J. and George E. Hoffer, “Will Retarding the Information Flow On Automobile Recalls Affect Consumer Demand?” Economic Inquiry, 21, no. 3 (1983), 444-447.

Reuter, Jonathan and Eric Zitzewitz, “Do Ads Influence Editors? Advertising and Bias in the Financial Media,” Quarterly Journal of Economics, 121, no. 1 (2006), 197-227.

Robertson, Campbell, and Clifford Krauss, “Gulf Spill Is the Largest of Its Kind, Scientists Say” New York Times, August 2, 2010.

Roe, Brian, Mario F. Teisl, Alan Levy, and Matthew Russell, “U.S. Consumers’ Willingness to Pay for Green Electricity,” Energy Policy, 29 (2001), 917-925.

Saffer, Henry, and Dhaval Dave. "Alcohol advertising and alcohol consumption by adolescents." Health Economics 15.6 (2006): 617-637. Schmalensee, Richard, “A Model of Promotional Competition in Oligopoly,” Review of Economic Studies, 43, no. 3 (1976), 493-507.

Shapiro, Carl, “Optimal Pricing of Experience Goods,” Bell Journal of Economics, 14, autumn (1983), 497-507.

Simester, Duncan, Jeffrey Hu, Erik Brynjolfsson, and Eric Anderson, “Dynamics of Retail Advertising: Evidence from a Field Experiment,” Economic Inquiry, 47, no. 3 (2009), 482-499.

40

Solman, Greg, “BP: Coloring Public Opinion?” Adweek, January 14, 2008, available at http://www.adweek.com/news/advertising/bp-coloring-public-opinion-91662, accessed on January 7, 2013.

Teisl, Mario F., Brian Roe, and Robert Hicks, “Can Eco-Labels Tune a Market? Evidence from Dolphin-Safe Labeling,” Journal of Environmental Economics and Management, 43, no. 3 (2002), 339-359.

Tracy, Tennile, “BP Tripled Its Ad Budget After Oil Spill,” The Wall Street Journal, September 1, 2010, available at http://www.wsj.com/articles/SB10001424052748703882304575465683723697708, accessed on May 10, 2015.

United States of America v. BP Exploration & Production, Inc. CDN: 2:12-cr-00292-SSV-DEK, available at http://www.justice.gov/criminal/vns/caseup/bpexploration.html, accessed on January 7, 2013.

29

TABLE 1: OIL SPILL IMPACT: BASIC DIFFERENCE ESTIMATES (1) (2) (3) (4) VARIABLES Average Net Price Ln (Ave. Fleet Sales) Weekly Net Price Ln(Weekly Fleet Sales) During-spill 0.072** 0.019** 0.071** 0.032** (0.001) (0.004) (0.001) (0.003) Post-spill -0.062** -0.025** -0.062** -0.021** (0.001) (0.005) (0.001) (0.004) BP*During-spill -0.042** -0.036** -0.042** -0.040** (0.002) (0.009) (0.002) (0.008) BP*Post-spill 0.025** -0.027* 0.025** -0.027** (0.002) (0.011) (0.001) (0.009) Observations 21,421 19,430 763,985 695,166 Adjusted R-squared 0.933 0.965 0.741 0.852 S.E.cluster station Station station station Weight price observation quantity observation price observation quantity observation # stations 7,503 6,735 7,503 6,735 Notes: Source: OPIS. The price and quantity data cover the period from January 2009 to March 2011. 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. Standard errors are clustered by station. Significance at 1%**, 5%*.

30

TABLE 2: OIL SPILL IMPACT BY MONTH VARIABLE Weekly Net Price Ln(Weekly Fleet Sales) (1) (2) BP*late_Apr'10 -0.011** 0.003 (0.002) (0.010) BP*May'10 -0.041** -0.030** (0.002) (0.009) BP*Jun'10 -0.049** -0.063** (0.002) (0.010) BP*Jul'10 -0.044** -0.049** (0.002) (0.009) BP*Aug'10 -0.061** -0.067** (0.002) (0.010) BP*Sep'10 -0.029** -0.010 (0.002) (0.010) BP*Oct'10 -0.005** -0.024* (0.002) (0.010) BP*Nov'10 0.021** -0.040** (0.002) (0.010) BP*Dec'10 0.052** -0.044** (0.002) (0.011) BP*Jan'11 0.049** -0.031** (0.002) (0.011) BP*Feb'11 0.022** 0.012 (0.002) (0.011) BP*Mar'11 0.028** -0.033** (0.002) (0.011) Observations 763,985 695,166 Adjusted R-squared 0.839 0.860 Fixed Effects station Station S.E.cluster station Station Weight price observation quantity observation # stations 7,503 6,735 Notes: Source: OPIS. The price and quantity data cover the period from January 2009 to March 2011. 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. Standard errors are clustered by station. Significance at 1%**, 5%*.

31

TABLE 3: OIL SPILL IMPACT AND REID VAPOR PRESSURE REGULATION VARIABLE Average Net Price Ln(Ave. Fleet Sales) Weekly Net Price Ln(Weekly Fleet Sales) (1) (2) (3) (4) During-spill 0.075** 0.011 0.075** 0.024** (0.003) (0.009) (0.002) (0.007) Post-spill -0.076** -0.040** -0.076** -0.038** (0.001) (0.011) (0.001) (0.009) BP*During-spill -0.075** -0.023 -0.075** -0.027 (0.004) (0.020) (0.003) (0.017) BP*Post-spill 0.020** -0.039 0.021** -0.038 (0.003) (0.024) (0.002) (0.020) Observations 6,010 5,350 211,285 190,283 Adjusted R-squared 0.886 0.958 0.645 0.849 Fixed Effects Station Station Station Station S.E.cluster Station Station Station Station Weight price observation quantity observation price observation quantity observation # stations 2,122 1,871 2,122 1,871 Notes: Source: OPIS. The sample covers the period from January 2009 to March 2011. Sample restricted to states meeting the standard summertime Reid Vapor Pressure (RVP) 9.0 psi limit. The coefficients reported are from regressions of BP retail price and log-quantity on the during-spill dummy, the dummy for post-spill period, and their interactions with a dummy for BP gas station. 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 where the dependent variable is the individual station's weekly net price and log-quantity. All models control for station fixed effects. Standard errors are clustered by station. Significance at 1%**, 5%*.

32

TABLE 4: IMPACT OF OIL SPILL AS A FUNCTION OF GREEN PREFERENCES

(1) (2) (3) (4) (5) (6)

DEP. VARIABLE: Price Diff Sales Diff Price Diff Sales Diff Price Diff Sales Diff

BP -0.043** -0.036** -0.041** -0.036** -0.041** -0.033**

(0.002) (0.009) (0.002) (0.010) (0.003) (0.010)

Pct hybrid, Demeaned 0.008** -0.003

(0.002) (0.009)

BP*(Pct hybrid, Demeaned) -0.012* 0.039

(0.005) (0.021)

Income, Demeaned -0.000 0.000 -0.000 0.000

(0.000) (0.000) (0.000) (0.000)

BP*(Income, Demeaned) 0.001** -0.002* 0.001** -0.001

(0.000) (0.001) (0.000) (0.001)

Green Index 0.006** -0.002

(0.001) (0.002)

BP*(Green Index) -0.006** 0.013

(0.002) (0.008)

Constant 0.073** 0.016** 0.073** 0.017** 0.074** 0.016**

(0.001) (0.004) (0.001) (0.004) (0.001) (0.004)

Observations 6,388 5,868 6,388 5,868 6,388 5,868

Adjusted R-squared 0.050 0.002 0.057 0.003 0.070 0.002

# stations 6,388 5,868 6,388 5,868 6,388 5,868 Notes: Sources: OPIS, Sierra Club, the U.S. Green Building Council, the U.S. Census and Kantar Media. The sample is restricted to stations with available data on Green Index and household income. Columns (1) and (2) report the benchmark estimates from Table 1 for the sample of stations that has income, green index, and hybrid car share data available. The dependent variable is the station's price difference or log-quantity difference between the “pre” and “during” spill periods. Columns (3) and (4) add median household income and hybrid vehicle shares as control variables. Columns (5) and (6) add income and the Green Index. 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 2000 U.S. $thousands. Significance at 1%**, 5%*.

TABLE 5: OLS AND IV ESTIMATES OF OIL SPILL IMPACT INCLUDING INTERACTIONS WITH GREEN PREFERENCES AND PRE-SPILL ADVERTISING

OLS ESTIMATES OLS ESTIMATES 2SLS ESTIMATES

DEPENDENT VARIABLE: Price Diff Sales Diff Price Diff Sales Diff Price Diff Sales Diff

(1) (2) (3) (4) (5) (6)

BP -0.035** -0.031** -0.042** -0.029* -0.044** -0.025*

(0.002) (0.010) (0.003) (0.011) (0.003) (0.012)

Green Index 0.006** -0.001 0.005** -0.002

(0.001) (0.003) (0.001) (0.003)

BP*(Green Index) -0.007** 0.010 -0.007** 0.010

(0.002) (0.008) (0.002) (0.009)

Income, Demeaned 0.000 0.000 0.000 0.000

(0.000) (0.000) (0.000) (0.000)

BP*(Income, Demeaned) 0.001** -0.002* 0.000* -0.002*

(0.000) (0.001) (0.000) (0.001)

Ad spending, Demeaned -0.000 0.000 -0.001** -0.000

(0.000) (0.001) (0.000) (0.001)

BP*(Ad spending, Demeaned) 0.003** 0.000 0.004** -0.001

(0.000) (0.002) (0.001) (0.002)

Constant 0.067** 0.013** 0.067** 0.013** 0.062** 0.014**

(0.001) (0.004) (0.001) (0.004) (0.001) (0.005)

# observations 5,088 4,662 5,088 4,662 5,002 4,582

# stations 5088 4662 5088 4662 5,002 4,582

R-squared 0.039 0.002 0.074 0.002 0.075 0.003

Notes: Source: OPIS, Sierra Club, R.L. Polk, the U.S. Green Building Council, and U.S. Census. The sample is restricted to stations with available data on Green Index, household income, and BP advertising expenditures. Columns (1) and (2) report the benchmark estimates from Table 1 for the stations that have income, Green Index, and advertising data available. The dependent variable is the station's price difference or log-quantity difference. Columns (3) and (4) report results with added controls for Green Index, demeaned median household income, and demeaned cumulative BP advertising expenditures during the 'Beyond Petroleum' campaign years for the BP Corporation, BP fuels, and environmental issues. Expenditures are in $millions, with mean $1.5 and std. $3.4 mil. The regressors of interests are the interactions of these variables with the BP gas station dummy. The price difference is the average net price in the during-spill period minus the pre-spill period. The log-quantity difference is the log average quantity in the during-spill period minus the pre-spill period. Columns (5) and (6) report 2SLS estimates instrumenting BP advertising expenditures with the DMA average spot TV ad price across all industries and products in 2007-2008. First stage results are in the Online Appendix. The Green Index is 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 2000 U.S. $thousands. Significance at 1%**, 5%*.

2

FIGURE 1 AVERAGE WEEKLY PRICE (LEVEL) FOR BP AND CONTROL STATIONS

JANUARY 2010 TO MARCH 2011

Notes: Source: OPIS. The figure displays average weekly prices for BP and non-BP competitor stations in our sample of 7,503 stores. See text and Online Appendix for details on our sample construction, and for a zoomed out version of the graph starting at the beginning of our sample in 2009.

Oil Spill Oil Leak Capped

2.6

2.8

33.

23.

43.

6

Avg

. Ret

ail P

rice,

$/g

al

Jan 10 Mar 10 Jun 10 Oct 10 Jan 11 Mar 11

BP Non−BP, Non−BP Competitor

3

FIGURE 2: GOOGLE SEARCH INTENSITY OF BP OIL SPILL RELATED SEARCHES

Panel A. Google Intensity and Price Coefficients

Panel B. Panel A. Google Intensity and Quantity Coefficients

Notes: Source: OPIS and Google Insights (accessed 8/16/2011). The figures display in blue the Google search intensity 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 the baseline month. A value of 50 thus indicates that searches in a month were 50 times greater than in January 2004. The red lines with markers 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.

−.0

5−

.025

0.0

25.0

5

Pric

e C

oeffi

cien

ts (

$)

−50

−25

025

50

Goo

gle

Sea

rch

Inte

nsity

01 Jan 2010 01 Apr 2010 01 Jul 2010 01 Oct 2010

’Oil spill’ SearchesBP Monthly Price Coefficients

−.1

−.0

50

.05

.1.1

5

Vol

ume

Coe

ffici

ents

−50

−25

025

50

Goo

gle

Sea

rch

Inte

nsity

01 Jan 2010 01 Apr 2010 01 Jul 2010 01 Oct 2010

’Oil spill’ SearchesBP monthly Volume Coefficients

4

FIGURE 3A: BP MARKET SHARE TIME-DUMMY COEFFICIENTS,

ABOVE MEDIAN ADVERTISING SPENDING

FIGURE 3B: BP MARKET SHARE TIME-DUMMY COEFFICIENTS,

BELOW MEDIAN ADVERTISING SPENDING

Notes: Sources: OPIS and Kantar Ad$pender. This figure displays the coefficients on monthly time dummies –relative to the omitted April 2010 oil spill month – from a regression of the share of BP stations in each zip code-month on these time dummies as well as zip code fixed-effects (see specification (3) from the text). The regression was estimated separately for zip codes in metro areas with above and below median BP ad spending during the Beyond Petroleum campaign years of 2000-2008. The corresponding regression results can be found in the Online Appendix.

−.0

1−

.005

0.0

05.0

1

Jan 09 Jul 09 Jan 10 Jul 10 Jan 11

Coefficients 95% CI

−.0

1−

.005

0.0

05.0

1

Jan 09 Jul 09 Jan 10 Jul 10 Jan 11

Coefficients 95% CI

5

ONLINE APPENDIX

(For Online Publication )

ADVERTISING AS INSURANCE OR COMMITMENT? EVIDENCE FROM THE BP OIL SPILL*

By

Lint Barrage, Brown University and NBER

[email protected]

Eric Chyn, University of Michigan [email protected]

Justine Hastings, Brown University and NBER

[email protected]

This draft May 2016

6

*Previous versions of this manuscript were circulated with the title: “Advertising, Reputation, and Environmental Stewardship: Evidence from the BP Oil Spill.” We thank Ryan Kellogg, Matthew Kahn and Richard Schmalensee for helpful comments. Phillip Ross provided outstanding research assistance. Hastings gratefully acknowledges funding through Brown University, Department of Economics and Population Studies and Training Center. Chyn gratefully acknowledges support in part from an NICHD training grant to the Population Studies Center at the University of Michigan (T32 HD007339).

7

Table of Contents

Section 1: Additional Results from Main Analysis

Figure A1: Average Weekly Price for BP and Control Stations 2009-2011 ............................. 3

Table A0: First Stage Results for Table 5 Advertising Spending IV Regression ..................... 4

Table A1: Market Share Impacts Above and Blow Median Ad Spending................................ 5

Section 2: Specification Checks

Table A2: Unfiltered Data Basic Oil Spill Impacts ................................................................... 7

Table A3: Unfiltered Data Oil Spill Impacts by Month ............................................................ 8

Table A4: Unfiltered Data Basic Oil Spill Impacts and RVP Regulation ................................. 9

Table A5: Determinants of Spot Prices ................................................................................... 11

Table A6: Robustness to Controls for BP Market share ......................................................... 13

Table A7: Robustness to Controls for Gas Station Density .................................................... 14

Table A8: Robustness Check: Spot TV Advertising Expenditures ......................................... 16

Table A9: Robustness Check: Spot TV Advertising Units ..................................................... 17

Table A10: Robustness to Controls for During-Spill Spending .............................................. 18

Table A11: Core Corporate vs. Other Advertising and Green Zip Triple Interactions ........... 19

Section 3: Details and Supporting Materials

OPIS Data Description Details and Sample Construction ....................................................... 25

Table A12: Number of Stations across Sample Cuts .............................................................. 26

8

Section 1: Additional Results from Main Analysis

Figure A1: AVERAGE WEEKLY PRICE (LEVEL) FOR BP AND CONTROL STATIONS JANUARY 2009 TO MARCH 2011

Notes: Source: OPIS. The figure displays average weekly prices for BP and non-BP competitor stations in our sample of 7,503 stores. See text and appendix for details on our sample construction.

Oil Spill Oil Leak Capped

1.5

22.

53

3.5

Avg

. Ret

ail P

rice,

$/g

al

Jan 09 Oct 09 Jun 10 Mar 11

BP Non−BP, Non−BP Competitor

9

TABLE A0: FIRST STAGE RESULTS FOR TABLE 5, BP AD SPENDING 2SLS RESULTS

Price Difference Sales Difference BP Adspend BP*(BP Adspend BP Adspend BP*(BP Adspend VARIABLES Demeaned Demeaned) Demeaned Demeaned) BP 1.310*** 0.824*** 1.324*** 0.849*** (0.110) (0.045) (0.114) (0.048) Green Index -0.263*** 0.000 -0.272*** 0.000 (0.024) (0.010) (0.025) (0.011) BP*(Green Index) 0.118 -0.145*** 0.160* -0.112*** (0.083) (0.034) (0.086) (0.036) Income, Demeaned 0.003 -0.000 0.002 0.000 (0.030) (0.012) (0.003) (0.001) BP*(Income, Demeaned) 0.189** 0.192*** 0.018** 0.020*** (0.077) (0.032) (0.008) (0.003) Spot TV Ad Price, Demeaned 0.012*** -0.000 0.012*** 0.000 (0.000) (0.000) (0.000) (0.000) BP*(Spot TV Ad Price, Dm.) 0.003*** 0.014*** 0.002*** 0.014*** (0.000) (0.000) (0.000) (0.000) Constant -0.486*** -0.000 -0.475*** -0.000 (0.045) (0.018) (0.047) (0.020) # Observations 5,002 5,002 4,582 4,582 Shea’s Partial R-squared 0.687 0.817 0.690 0.817 Angrist-Pischke F-Stat 6348 11579 5831 10443 AP F-stat p-value 0.000 0.000 0.000 0. 000

Notes: Source: OPIS, Sierra Club, R.L. Polk, the U.S. Green Building Council, and U.S. Census. The sample is restricted to stations with available data on Green Index, household income, and BP advertising expenditures. Columns (1) and (2) report the first stage estimation results for the ‘price effects’ regression; Columns (3)-(4) do so for the ‘sales effect’ regressions of Table 5. The specification controls for Green Index, demeaned median household income, and instruments for demeaned cumulative BP advertising expenditures during the 'Beyond Petroleum' campaign years for the BP Corporation, BP fuels, and environmental issues. Expenditures are in $millions, with mean $1.5 and std. $3.4 mil. The instruments are the metropolitan-area average TV spot advertising price (across industries) over period 2007-2008, and the spot price interacted with a BP dummy. The Green Index is 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 2000 U.S. $thousands. Significance at 1%***, 5%** and 10%*.

10

TABLE A1: BP STATION MARKET SHARE IMPACTS BY AD SPENDING

Above Median Ad Spend Below Median Ad Spend VARIABLES BP Station Share BP Station Share

Jan '09 -0.001 0.003 (0.002) (0.002) Feb '09 -0.001 0.003 (0.002) (0.002) Mar '09 -0.001 0.003 (0.002) (0.002) Apr '09 -0.001 0.002 (0.002) (0.002) May '09 -0.001 0.004* (0.002) (0.002) June '09 -0.003* 0.002 (0.002) (0.002) July '09 -0.002 0.003 (0.002) (0.002) Aug '09 -0.001 0.001 (0.001) (0.002) Sep '09 -0.001 0.001 (0.001) (0.002) Oct '09 -0.000 0.002 (0.001) (0.002) Nov '09 -0.000 0.002 (0.001) (0.001) Dec '09 0.001 0.001 (0.001) (0.001) Jan '10 0.001 0.000 (0.001) (0.001) Feb '10 0.000 -0.000 (0.001) (0.001) Mar '10 0.001 -0.000 (0.001) (0.001) May'10 -0.000 -0.000 (0.001) (0.001) Jun'10 -0.000 -0.000 (0.001) (0.001) Jul'10 0.000 -0.000 (0.001) (0.001)

11

Aug'10 -0.000 -0.002 (0.001) (0.002) Sep'10 -0.002 -0.003 (0.001) (0.002) Oct'10 -0.001 -0.005** (0.001) (0.002) Nov'10 -0.001 -0.006** (0.001) (0.002) Dec'10 -0.002 -0.005** (0.001) (0.002) Jan'11 -0.002 -0.005** (0.001) (0.002) Feb'11 -0.002 -0.004* (0.001) (0.002) Mar'11 -0.003** -0.005* (0.001) (0.002) Observations 15,687 6,912 Adj. R-squared 0.964 0.960 Fixed effects Zip Zip S.E. cluster Zip Zip

Notes: Sources: OPIS and Kantar Ad$pender. Dependent variable is the share of stations in a zip-month selling BP-branded gasoline. The regressions are estimated separately for zip codes in metro areas with above and below median BP ad spending during the Beyond Petroleum campaign years of 2000-2008. We include zip code fixed effects in the specification. Standard errors are clustered by zip. Significance at 1%**, 5%*.

12

Section 2: Specification Checks

TABLE A2: UNFILTERED DATA BASIC OIL SPILL IMPACTS (1) (2) (3) (4) VARIABLES Average Net Price Ln(Ave. Fleet Sales) Weekly Net Price Ln(Weekly Fleet Sales) During 0.059** 0.029** 0.059** 0.047** (0.000) (0.001) (0.000) (0.001)

Post -0.049** -0.019** -0.049** -0.012** (0.000) (0.002) (0.000) (0.001)

BP*during -0.025** -0.036** -0.024** -0.043** (0.001) (0.004) (0.001) (0.003)

BP*post 0.017** -0.017** 0.017** -0.021** (0.001) (0.004) (0.000) (0.004)

Observations 228,455 208,659 7,707,300 7,215,198

Adjusted R-squared 0.947 0.965 0.773 0.853

Fixed Effects Station Station station Station

S.E. cluster Station Station station Station

Weight price observation quantity observation price observation quantity observation # stations 81,402 72,875 81,402 72,875 Notes: Source: OPIS. The sample covers the period from January 2009 to March 2011. Columns (1) and (2) report estimates from specifications in which the dependent variable is set to the individual station’s average net price and average log-quantity computed over the “pre-,” “during-,” and “post-” spill periods. Columns (3) and (4) report estimates when the dependent variable is set to the individual station's weekly net price and log-quantity. Each specification regresses the dependent variable on an indicator variable for the during-spill period, a dummy for post-spill period, and their interactions with a dummy for BP gas station. All models control for station fixed effects. Standard errors are clustered by station. Significance at 1%**, 5%*.

13

TABLE A3:UNFILTERED DATA OIL SPIL IMPACTS BY MONTH

Variable Weekly Net Price Weekly Fleet Sales (1) (2) BP*late_Apr'10 0.000 -0.003 (0.001) (0.004) BP*May'10 -0.027** -0.032** -0.001 (0.003) BP*Jun'10 -0.030** -0.064** (0.001) (0.004) BP*Jul'10 -0.028** -0.054** (0.001) (0.004) BP*Aug'10 -0.039** -0.062** (0.001) (0.004) BP*Sep'10 -0.007** -0.019** (0.001) (0.004) BP*Oct'10 0.001* -0.028** (0.001) (0.004) BP*Nov'10 0.014** -0.046** (0.001) (0.004) BP*Dec'10 0.031** -0.029** (0.001) (0.004) BP*Jan'11 0.031** -0.020** (0.001) (0.004) BP*Feb'11 0.017** 0.024** (0.001) (0.004) BP*Mar'11 0.018** -0.021** (0.001) (0.004) Observations 7,707,300 7,215,198 Adjusted R-squared 0.859 0.858 Fixed Effects Station Station S.E. cluster Station Station Weight price observation quantity observation # stations 81,402 72,875 Notes: Source: OPIS. The sample for price and quantity data covers the period from January 2009 to March 2011. 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 effects. Standard errors are clustered by station. Significance at 1%**, 5%*.

14

TABLE A4: UNFILTERED DATA BASIC OIL SPILL IMPACTS AND RVP REGULATION

VARIABLE Average Net Price Ln(Ave. Fleet Sales) Weekly Net Price Ln(Weekly Fleet Sales) (1) (2) (3) (4)

during 0.075** 0.033** 0.075** 0.051** (0.001) (0.003) (0.001) (0.002) Post -0.065** -0.027** -0.065** -0.020** (0.001) (0.004) (0.000) (0.003) BP*during -0.065** -0.060** -0.064** -0.065** (0.001) (0.009) (0.001) (0.007) BP*post 0.018** -0.043** 0.019** -0.045** (0.001) (0.01) (0.001) (0.008) Observations 56,296 50,510 1,984,578 1,743,183 Adjusted R-squared 0.899 0.962 0.645 0.850 Fixed Effects Station Station Station Station S.E. cluster Station Station Station Station Weight price observation quantity observation price observation quantity observation # stations 21,149 18,679 21,699 19,159 Notes: Source: OPIS. The sample for price and quantity data covers the period from January 2009 to March 2011. Sample restricted to states meeting the standard summertime Reid Vapor Pressure (RVP) 9.0 psi limit. The coefficients reported are from regressions of BP retail price and log-quantity on the during-spill dummy, the dummy for post-spill period, and the interactions of these indicator variables with a dummy for the BP gas station. Columns (1) and (2) report estimates from specifications in which the dependent variable is set to the individual station's average net price and average log-quantity computed over the “pre-,” "during-," and "post-" spill periods. Columns (3) and (4) report estimates from specifications in which the dependent variable is set to the individual station's weekly net price and log-quantity. All models control for station effects. Standard errors are clustered by station. Significance at 1%**, 5%*.

15

Notes on Table A5: Determinants of Advertising Spot Prices

To help provide context for our instrumental variable strategy in Section 4.2, we examine the

determinants of industry-wide TV advertising spot prices. Specifically, we focus on the quantity-weighted

average spot television price from 2007-2008 across metropolitan areas.40 We compute these spot prices

from Kantar Media Ad$pender data as described in Section 3.2. Table A5 provides the results from our

cross-sectional analysis of (logged) spot prices.

Column 1 focuses on the impact of (logged) population density on spot prices. Our estimates

suggest that a one percent increase in metropolitan population density increases spot prices by 0.61 percent.

Columns 2 through 4 present results after adding additional measures of metropolitan area characteristics.

Notably, this analysis does not detect any evidence that spot prices depend on BP’s market share, the

gasoline market HHI or the density of gas stations.41 We do find that metropolitan area average household

income has a positive association with spot prices: a one percent increase in average household increases

spot prices by 0.73 percent. Notice that the estimated impact of population density remains positive in each

specification, although this elasticity attenuates as additional controls are added into the regression.

40 We match the Kantar data, which are at the Designated Market Area (DMA) level, to zip codes using the county-DMA correspondence provided by Gentzkow and Shapiro (2008), in conjunction with a county-zip correspondence from the U.S. Department of Housing and Urban Development 41 Column 3 does report a precisely estimated elasticity of spot prices with respect to station density; however, this result is not robust to addition of mean household income to the specification in column 4.

16

TABLE A5: DETERMINANTS OF ADVERTISING SPOT PRICES

Dependent Variable: Log of MSA TV Spot

Price (1) (2) (3) (4) Log of Population per sq. mile 0.611*** 0.603*** 0.654*** 0.528*** (0.0814) (0.0829) (0.0904) (0.106) Log of BP Share of All Stations -0.488 -0.192 -0.113 (0.449) (0.466) (0.432) Log of Gas Market HHI 0.725 0.305 0.126 (0.688) (0.660) (0.660) Log of Stations per sq. mile -3.355** -2.162 (1.418) (1.499) Log of Mean Household Income 0.736*** (0.246) Constant 1.978*** 1.996*** 1.797*** -5.799** (0.520) (0.520) (0.548) (2.465) Observations 91 91 91 91 R-squared 0.505 0.516 0.536 0.573 Avg. Spot Price 273.9 Spot Price S.D. 270.1 Notes: All variables are measured at the MSA level. The table reports OLS estimates on the relationship between MSA TV spot prices and various MSA characteristics. Spot prices are computed using Kantar Ad$pender data. We use OPIS data to compute (1) the BP share of all stations, (2) gasoline market Herfindahl-Hirschman Index (HHI) and (3) stations (non-BP) per square mile. We use Census data for population and income measures. Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1

17

Notes on Tables A6-A7: Advertising Results Robustness to Additional Controls

Table A6 provides the results of repeating the specification of Table 5 with added controls for BP

stations’ market share, defined as the share of stations in a DMA in our sample selling BP-branded gasoline

in the pre-spill period. The market share has a mean (median) of 9.1% (7.8%), and a standard deviation of

9.2 percentage points. Similarly, Table A7 provides the results of repeating the specification of Table 5

with added controls for the density of competing gasoline stations, defined as the number of non-BP gas

stations in our sample divided by the number of square miles in a given zip code. This measure of density

has a mean (median) of 0.79 (0.45) non-BP stations per square mile and a standard deviation of 1.01.

The results indicate that there is no change in the estimated price difference coefficient on the

interaction of DMA-level BP ad spending and being a BP station after including market share or station

density controls. The coefficient for advertising impact on sales remains imprecisely estimated in both

specifications. The results from Table A6 further suggest that the oil spill affected BP prices significantly

more in areas with lower pre-spill BP market share. The predicted oil spill impact on BP prices in markets

with a one-standard deviation higher pre-spill BP advertising is approximately equal to the predicted oil

spill impact in markets with a 2.4 percentage point higher pre-spill BP station share. (Note that a standard

deviation of advertising expenditures is $3.4 million.) The results in Table A7 show that there is no

detectable impact of (non-BP) station density on the oil spill impact on BP prices or quantities.

18

TABLE A6: ROBUSTNESS TO CONTROLS FOR BP MARKET SHARE (1) (2) (3) (4) (5) (6)

First Stage Second Stage First Stage Second Stage

BP Adspend BP*(BP Adspend Price Diff. BP Adspend

BP*(BP Adspend Sales Diff.

VARIABLES Demeaned Demeaned) Demeaned Demeaned) BP -0.288* 0.911*** -0.026*** -0.229 0.968*** -0.026 (0.167) (0.074) (0.005) (0.177) (0.079) (0.020) Green Index -0.198*** 0.000 0.005*** -0.205*** 0.000 -0.002 (0.022) (0.010) (0.001) (0.024) (0.011) (0.003) BP*(Green Index) 0.055 -0.143*** -0.006*** 0.098 -0.107*** 0.010 (0.077) (0.034) (0.002) (0.081) (0.036) (0.009) Income, Demeaned 0.005* -0.000 -0.000 0.006** 0.000 0.000 (0.003) (0.001) (0.000) (0.003) (0.001) (0.000) BP*(Income, Demeaned) 0.014** 0.019*** 0.001*** 0.014* 0.020*** -0.002** (0.007) (0.003) (0.000) (0.007) (0.003) (0.001) BP market share, Demeaned 24.044*** 0.000 -0.354*** 24.063*** -0.000 -0.266** (0.874) (0.386) (0.027) (0.916) (0.410) (0.114) BP*(BP market share, Dm.) -25.039*** -0.995 0.419*** -25.505*** -1.442* 0.462** (1.604) (0.708) (0.044) (1.689) (0.757) (0.190) Spot TV Ad Price, Demeaned 0.010*** -0.000 0.010*** -0.000 (0.000) (0.000) (0.000) (0.000) BP*(Spot TV Ad Price, Dm.) 0.004*** 0.014*** 0.003*** 0.014*** (0.000) (0.000) (0.000) (0.000) Ad spending, Demeaned 0.000 0.001 (0.000) (0.002) BP*(Ad spending, Demeaned) 0.003*** -0.001 (0.001) (0.003) Constant 1.200*** 0.000 0.045*** 1.198*** -0.000 -0.004 (0.074) (0.033) (0.002) (0.078) (0.035) (0.009) Observations 5,002 5,002 5,002 4,582 4,582 4,582 R-squared 0.728 0.817 0.122 0.730 0.817 0.005 Notes: Source: OPIS, Sierra Club, R.L. Polk, the U.S. Green Building Council, and U.S. Census. The sample is restricted to stations with available data on Green Index, household income, and BP advertising expenditures. The estimates mirror those of Table 5, with added controls for BP’s pre-spill market share, defined as fraction of stations in the DMA in our sample selling BP-branded gasoline (mean 0.091). Columns (1) and (2) report the first stage results for the ‘price effects’ regression; Columns (4)-(5) do so for the ‘sales effect’ regressions, and Columns (3) and (6) report the resulting IV regression results. The specification controls for green index, demeaned median household income, BP market share, and instruments for demeaned cumulative BP advertising expenditures during the 'Beyond Petroleum' campaign years for the BP Corporation, BP fuels, and environmental issues. Expenditures are in $millions, with mean $1.5 and std. $3.4 mil. The instruments are the metropolitan-area average TV spot advertising price (across industries) over period 2007-2008, and the spot price interacted with a BP dummy. The Green Index is 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 2000 U.S. $thousands. Significance at 1%***, 5%** and 10%*.

19

TABLE A7: ROBUSTNESS TO CONTROLS FOR STATION DENSITY (1) (2) (3) (4) (5) (6)

First Stage Second Stage First Stage

Second Stage

BP

Adspend BP*(BP Adspend

BP Adspend

BP*(BP Adspend

VARIABLES Demeaned Demeaned) Price Diff. Demeaned Demeaned) Sales diff. BP 1.157*** 0.677*** -0.044*** 1.180*** 0.698*** -0.029**

(0.112) (0.045) (0.003) (0.117) (0.048) (0.012) Green Index -0.258*** 0.000 0.005*** -0.272*** 0.000 -0.002

(0.025) (0.010) (0.001) (0.026) (0.011) (0.003) BP*(Green Index) 0.163** -0.094*** -0.007*** 0.212** -0.060* 0.011

(0.083) (0.034) (0.002) (0.087) (0.036) (0.009) Income, Demeaned -0.000 -0.000 0.000 0.002 0.000 0.000

(0.003) (0.001) (0.000) (0.003) (0.001) (0.000) BP*(Income, Demeaned) 0.007 0.007** 0.000** 0.006 0.008** -0.002** (0.008) (0.003) (0.000) (0.008) (0.003) (0.001) Station Density, Demeaned -0.042 0.000 -0.001 -0.001 0.000 -0.000 (0.046) (0.018) (0.001) (0.048) (0.020) (0.005) BP*Station Density, Demeaned -0.690*** -0.732*** 0.001 -0.708*** -0.709*** -0.013 (0.133) (0.054) (0.003) (0.138) (0.057) (0.013) Spot TV Ad Price, Demeaned 0.012*** -0.000 0.012*** 0.000

(0.000) (0.000) (0.000) (0.000) BP*(Spot TV Ad Price, Dm.) 0.003*** 0.015*** 0.003*** 0.015***

(0.000) (0.000) (0.000) (0.000) Ad spending, Demeaned -0.001*** -0.000

(0.000) (0.001) BP*(Ad spending, Demeaned) 0.004*** 0.000

(0.001) (0.002) Constant -0.480*** -0.000 0.069*** -0.482*** -0.000 0.014*** (0.045) (0.018) (0.001) (0.048) (0.020) (0.005) Observations 5,002 5,002 5,002 4,582 4,582 4,582 R-squared 0.689 0.824 0.075 0.692 0.824 0.003 Notes: Source: OPIS, Sierra Club, R.L. Polk, the U.S. Green Building Council, and U.S. Census. The sample is restricted to stations with available data on Green Index, household income, and BP advertising expenditures. The estimates mirror those of Table 5 with added controls for the density of non-BP gas stations per square mile at the zip code level (mean 0.79). Columns (1) and (2) report the first stage results for the ‘price effects’ regression; Columns (4)-(5) do so for the ‘sales effect’ regressions, and Columns (3) and (6) report the resulting IV regression results. The specification controls for green index, demeaned median household income, BP market share, and instruments for demeaned cumulative BP advertising expenditures during the 'Beyond Petroleum' campaign years for the BP Corporation, BP fuels, and environmental issues. Expenditures are in $millions, with mean $1.5 and std. $3.4 mil. The instruments are the metropolitan-area average TV spot advertising price (across industries) over period 2007-2008, and the spot price interacted with a BP dummy. The Green Index is 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 2000 U.S. $thousands. Significance at 1%***, 5%** and 10%*

20

Notes on Tables A8-A9: Robustness to Spot TV Only

The results discussed in Section 4.2 of the text focus on BP advertising expenditures since we are

aggregating over many forms of advertising media (e.g., television or print). Alternatively, our data also

allow us to conduct our analysis by focusing on television advertisements only. Table A8 presents results

for demeaned BP Spot TV advertising expenditures, and Table A9 focuses on Spot TV units (in hundreds

of ads) during the 'Beyond Petroleum' campaign years for the BP Corporation, BP fuels, and environmental

issues. Columns 1 and 2 report OLS results for station prices and quantity sold, respectively. Columns 5

and 8 provide the second-stage results where we use metro area television spot prices to instrument for BP

advertising units. The results for station prices (in column 5) again show that advertising helped mitigate

the impact of the oil spill: an additional 100 TV advertising units above the mean increased station prices

by 0.1 or 0.3 cents per gallon (OLS and IV, resp., Table A9). An additional $1 million in spot TV advertising

expenditures increased BP stations’ prices after the spill by 0.4 or 0.6 cents per gallon (OLS and IV, resp.,

Table A8). The impact on quantities is not precisely estimated which mirrors the result we obtain for all

advertising expenditures.

Notes on Tables A10-A11: Robustness to Controlling for Alternative Forms of Advertising Section 4.2.2 of the text explains that there are two possible issues that may alter the interpretation

of our results. First, it may be the case that during-spill advertising is correlated with pre-spill advertising.

To address this concern, we show that the effect of pre-spill BP advertising is robust to controlling for

advertising during the oil spill. Second, an additional concern is that other forms of advertising may have

affected consumer demand for BP stations, particularly local and ancillary product advertising (e.g., for

individual BP service stations and their convenience stores). To address this issue, we create an additional

measure to control for these other types of advertising. The ad measures are specifically constructed as

follows: Step 1: We use all Kantar advertising data for 2000-2008 for which BP is listed as ‘Ultimate

Owner.’ Step 2: We drop all advertisements for which the ‘advertiser’ (entity paying the ad) is clearly not

related to BP or BP gas stations, namely Arco and individual Arco stations as well as Amoco and individual

Amoco stations (as these are excluded from the analysis), Castrol and Castrol brands (Lube Express), and

a handful of other entities mainly related to BP chemicals manufacturing. Step 3: As previously noted, our

core corporate advertising measure includes all ads for (i) BP Corporation, (ii) BP fuels and oils, and (iii)

explicitly environmental advertisements such as for solar systems or explicit ‘Beyond Petroleum’

announcements run during 2000-2008. Step 5: All remaining ads are included in our new control variable,

21

consisting of advertisements related to BP-affiliated convenience stores and products, individual service

stations, ancillary product services, and miscellaneous items such as BP credit cards. As a suggestive test

of the importance of the Beyond Petroleum corporate branding has green advertising per se, we interact

these different advertising measures with a dummy variable for whether stations are located in “green zips,”

defined as zip codes whose green index scores above the median. The results are displayed in Table A11.

Column (1) replicates the benchmark advertising results. Column (2) adds local and ancillary product

advertising measures. Column (3) repeats the benchmark results with the green zip dummy instead of the

green index variable as measure for environmental preferences, and with green zip interactions. Finally,

Column (4) adds interactions with local and ancillary product advertising. The results confirm that the

estimated protective benefit of our core corporate branding measure is robust to controlling for other BP

station-related advertising. In addition, though noisy, the point estimates suggest that the impact of the

likely environmentally-themed core corporate advertising was larger at stations in high-green-preference

markets, whereas the impact of local and ancillary product ad spending was stronger in low-green-

preference markets.

22

TABLE A8: ROBUSTNESS CHECK: SPOT TV ADVERTISING EXPENDITURES

(1) (2) (3) (4) (5) (6) (7) (8)


Second Stage

VARIABLES Price Diff. Sales Diff. BP Ad

Units, Dm. BP*(BP Ad Units, Dm.) Price Diff.

BP Ad Units, Dm.

BP*(BP Ad Units, Dm.) Sales Diff.

BP -0.040*** -0.029*** 0.563*** 0.139*** -0.042*** 0.570*** 0.144*** -0.026** (0.003) (0.011) (0.065) (0.026) (0.003) (0.067) (0.027) (0.011) Green Index 0.006*** -0.001 -0.141*** -0.000 0.005*** -0.147*** 0.000 -0.002 (0.001) (0.003) (0.014) (0.006) (0.001) (0.015) (0.006) (0.003) BP*(Green Index) -0.007*** 0.009 -0.080 -0.222*** -0.006*** -0.065 -0.212*** 0.010 (0.002) (0.008) (0.049) (0.019) (0.002) (0.051) (0.020) (0.008) Income, Demeaned -0.000 0.000 -0.007*** 0.000 0.000 -0.006*** -0.000 0.000 (0.000) (0.000) (0.002) (0.001) (0.000) (0.002) (0.001) (0.000) BP*(Income, Demeaned) 0.001*** -0.002** 0.016*** 0.009*** 0.001*** 0.015*** 0.009*** -0.002** (0.000) (0.001) (0.005) (0.002) (0.000) (0.005) (0.002) (0.001) Ad Spend, Demeaned 0.001** 0.001 -0.002*** -0.000 (0.000) (0.001) (0.000) (0.002) BP*(Ad Spend, Demeaned) 0.004*** -0.000 0.006*** -0.001 (0.001) (0.003) (0.001) (0.003) Spot TV Ad Price, Demeaned 0.008*** 0.000 0.008*** 0.000 (0.000) (0.000) (0.000) (0.000) BP*(Spot TV Ad Price, Dm.) 0.002*** 0.010*** 0.002*** 0.010*** (0.000) (0.000) (0.000) (0.000) Constant 0.066*** 0.013*** -0.424*** -0.000 0.068*** -0.426*** 0.000 0.014*** (0.001) (0.004) (0.026) (0.010) (0.001) (0.028) (0.011) (0.004) Observations 5,088 4,662 5,002 5,002 5,002 4,582 4,582 4,582 R-squared 0.076 0.002 0.755 0.864 0.070 0.757 0.865 0.003 Shea's Partial R-squared 0.691 0.794 0.692 0.794 Angrist-Pischke F-Stat 9426 18183 8631 16542 AP F-Stat p-value 0.000 0.000 0.000 0.000

Sources: OPIS, Sierra Club, the U.S. Green Building Council, the U.S. Census and Kantar Media. The dependent variable is price difference in columns (1) and (5), and log-quantity difference in columns (2) and (8). The specification controls for Green Index, demeaned median household income, and demeaned BP Spot TV advertising spending (in millions of US$) during the 'Beyond Petroleum' campaign years for the BP Corporation, BP fuels, and environmental issues (mean 0.7, std. 2.2). The price difference is the average net price in the during-spill period minus that in the pre-spill period. The log-quantity is the log average quantity in the during-spill period minus that in the pre-spill period. Columns (3)-(4) and (6)-(7) provide the first-stage results for IV regressions with demeaned average spot TV advertising price as instrument. We calculate the Green Index by summing 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. Zip-code income is in 2000 US$. Standard errors in parentheses. ** p<0.01, * p<0.05.

23

TABLE A9: ROBUSTNESS CHECK: SPOT TV ADVERTISING UNITS

(1) (2) (3) (4) (5) (6) (7) (8)


Second Stage

VARIABLES Price Diff. Sales Diff. BP Ad

Units, Dm. BP*(BP Ad Units, Dm.) Price Diff.

BP Ad Units, Dm.

BP*(BP Ad Units, Dm.)

Sales Diff.

BP -0.041*** -0.031** 7.337*** 5.361*** -0.051*** 7.419*** 5.441*** -0.022

(0.003) (0.012) (0.414) (0.196) (0.003) (0.426) (0.205) (0.014)

Green Index 0.005*** -0.001 -0.801*** -0.000 0.005*** -0.836*** -0.000 -0.002

(0.001) (0.003) (0.090) (0.043) (0.001) (0.095) (0.046) (0.003)

BP*(Green Index) -0.006*** 0.009 -0.556* -1.356*** -0.005** -0.411 -1.247*** 0.010

(0.002) (0.008) (0.312) (0.147) (0.002) (0.323) (0.155) (0.009)

Income, Demeaned 0.000 0.000 -0.003 0.000 0.000 0.002 0.000 0.000

(0.000) (0.000) (0.011) (0.005) (0.000) (0.012) (0.006) (0.000)

BP*(Income, Demeaned) 0.001*** -0.002** 0.145*** 0.142*** 0.000 0.141*** 0.142*** -0.001*

(0.000) (0.001) (0.029) (0.014) (0.000) (0.030) (0.014) (0.001)

Ad Spend, Demeaned -0.000*** 0.000 -0.001*** -0.000

(0.000) (0.000) (0.000) (0.001)

BP*(Ad Spend, Demeaned) 0.001*** -0.000 0.003*** -0.001

(0.000) (0.001) (0.000) (0.001) Spot TV Ad Price, Demeaned 0.023*** 0.000 0.023*** 0.000

(0.001) (0.000) (0.001) (0.000) BP*(Spot TV Ad Price, Dm.) 0.001 0.024*** 0.001 0.024***

(0.001) (0.001) (0.001) (0.001)

Constant 0.067*** 0.013*** -1.975*** 0.000 0.068*** -1.978*** 0.000 0.014***

(0.001) (0.004) (0.168) (0.079) (0.001) (0.177) (0.085) (0.004)

Observations 5,088 4,662 5,002 5,002 5,002 4,582 4,582 4,582

R-squared 0.072 0.002 0.392 0.523 0.063 0.399 0.523 0.003

Shea's Partial R-squared 0.306 0.332 0.314 0.332

Angrist-Pischke F-Stat. 1714 1935 1608 1748

AP F-Stat p-value 0.000 0.000 0.000 0.000 Sources: OPIS, Sierra Club, the U.S. Green Building Council, the U.S. Census and Kantar Media. The dependent variable is price difference in columns (1) and (5), and log-quantity difference in columns (2) and (8). The specification controls for Green Index, demeaned median household income, and demeaned BP Spot TV advertising units (in hundreds) during the 'Beyond Petroleum' campaign years for the BP Corporation, BP fuels, and environmental issues (mean 7.45, std. 10.8). The price difference is the average net price over during-spill period minus the average net price over pre-spill period. The log-quantity is the log average quantity over during-spill period minus the log average quantity over pre-spill period. Columns (3)-(4) and (6)-(7) provide the first-stage results for IV regressions with demeaned average spot TV advertising price as instrument. We calculate the Green Index by summing 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. Zip-code income is in 2000 US$. Standard errors in parentheses. Significance at 1%***, 5%** and 10%*.

24

TABLE A10: ROBUSTNESS TO CONTROLLING FOR DURING-SPILL ADS

(1) (2) (3) (4)

VARIABLES Price Diff. Price Diff. Price Diff. Price Diff.

BP -0.0418** -0.0444** -0.0293* -0.0298**

(0.00279) (0.00277) (0.0114) (0.0115)

Green Index 0.00552** 0.00445** -0.00136 -0.00157

(0.000604) (0.000603) (0.00253) (0.00256)

BP*(Green Index) -0.00722** -0.00580** 0.00954 0.00987

(0.00207) (0.00204) (0.00848) (0.00850)

Income, Demeaned 0.00000 0.000108 0.000275 0.000286

(0.000000) (0.000000) (0.000307) (0.000308)

BP*(Income, Demeaned) 0.000525** 0.000401* -0.00167* -0.00170*

(0.000191) (0.000190) (0.000781) (0.000786)

Pre-Spill Ad spending, Demeaned -0.000323 0.00156** 0.000114 0.000472

(0.000245) (0.000292) (0.00101) (0.00122)

BP*(Pre-Spill Ad spending, Demeaned) 0.00343** 0.00257** 0.000177 0.000149

(0.000479) (0.000582) (0.00195) (0.00241)

During-Spill Ad spending, Demeaned -0.00744** -0.00142

(0.000646) (0.00271)

BP*(During-Spill Ad spending, Demeaned) 0.00329** 0.000000

(0.00137) (0.00569)

Constant 0.0668** 0.0696** 0.0135** 0.0140**

(0.00106) (0.00107) (0.00442) (0.00453)

Observations 5,088 5,088 4,662 4,662

Adjusted R-squared 0.074 0.099 0.002 0.001

Sources: OPIS, Sierra Club, the U.S. Green Building Council, the U.S. Census and Kantar Media. The dependent variable is price difference in columns (1)-(2) and log-quantity difference in columns (3) and (4). The specification controls for Green Index, demeaned median household income, and demeaned BP advertising expenditures during the 'Beyond Petroleum' campaign years for the BP Corporation, BP fuels, and environmental issues, and during the BP oil spill from May-October 2010. The price difference is the average net price over during-spill period minus the average net price over pre-spill period. The log-quantity is the log average quantity over during-spill period minus the log average quantity over pre-spill period. We calculate the Green Index by summing 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. Zip-code income is in 2000 US$. Standard errors in parentheses. Significance at 1%***, 5%** and 10%*.

25

TABLE A11: CORE CORPORATE VS. OTHER ADVERTISING AND GREEN ZIP TRIPLE INTERACTIONS

(1) (2) (3) (4)

VARIABLES Price Diff. Price Diff. Price Diff. Price Diff.

BP -0.042*** -0.039*** -0.034*** -0.031*** (0.003) (0.003) -0.004 (0.004) Green Index 0.006*** 0.006*** (0.001) (0.001) BP*(Green Index) -0.007*** -0.007*** (0.002) (0.002) Green Zip Dummy 0.003 0.002 (0.002) (0.002) BP*(Green Zip Dummy) -0.013** -0.013** (0.005) (0.006) Income, Demeaned 0.000 0.000 0.000*** 0.000*** (0.000) (0.000) (0.000) (0.000) BP*(Income, Demeaned) 0.001*** 0.001*** 0.000 0.000 (0.000) (0.000) (0.000) (0.000) Green Ad Spending -0.000 -0.000 -0.001*** -0.001*** (0.000) (0.000) (0.000) (0.000) BP*(Corporate Ad Spending) 0.003*** 0.002*** 0.003*** 0.001 (0.000) (0.001) (0.001) (0.002) BP*(Corporate Ad Spending)*(Green Zip) 0.002* 0.002 (0.001) (0.002) Local/Ancillary Product Ad Spending -0.000 0.001** (0.000) (0.000) BP*(Local/Ancillary Product Ad Spending) 0.003 0.003 (0.002) (0.003) BP*(Local/Ancil. Ad Spending)*(Green Zip) -0.002 (0.004) Constant 0.067*** 0.067*** 0.065*** 0.065*** (0.001) (0.001) (0.002) (0.002) Observations 5,088 5,088 5,422 5,422 R-squared 0.075 0.076 0.063 0.065

Data sources: OPIS and Kantar Media. Dependent variable is the individual station's price difference which is defined as the average net price over the during-spill period minus the average net price during the pre-spill period. The advertising measures control for demeaned BP advertising expenditures during the Beyond Petroleum campaign years (2000-2008). “Corporate” advertising includes ads related to the BP Corporation, BP fuels, and environmental issues. “Local/Ancillary Product” advertising includes other BP service station related ads such as for convenience stores and products and individual service stations. The “Green Zip Dummy” equals one for stations in zip codes whose green index measure is above the median. Column (1) replicates the benchmark specification. Column (2) adds local/ancillary product ad spending. Column (3) uses the Green Zip Dummy instead of the Green Index to measure environmental preferences, and adds a benchmark interaction. Column (4) adds local/ancillary product ad spending and interactions. Significance at 1%***, 5%** and 10%*.

26

Section 3: Details and supporting materials

OPIS Data Details and Sample Construction We filter the price data at the zip code level according to the following criteria.

1. We begin with the daily price observations for each store from 2007 to October 2010.42 We then remove store-weeks without at least five days’ worth of price observations. This removes about 10 percent of observations from the raw data.

2. Next, we require that each store have at least 3 years’ worth of weekly observations. To further ensure the consistency of our stores, we also flag large one-day changes in prices indicative of an error in data ( “Twinkie effect”) in the price data and drop stores that are particularly affected by this error. Specifically, for each store we record the first and last day of operation in the data and require that each store have non-Twinkie price observations for at least 80 percent of these possible days.

3. With the remaining stores, we filter the data at the zip code level, keeping zips that have at least 5 distinct stores. We also require that each zip code have at least one observation (from at least one store) for every week from 2007-2010.

The above creates a list of usable zip codes from the pricing data. We have similar restrictions on the stores and zip codes used from the weekly quantity data as detailed below.

1. We begin with weekly quantity data from 2009 to December 2010. Within the weekly store quantity observations, we drop any store that is absent from the data for 3 months or more at some point in our data.

2. From this set of stores, we construct z-scores for each store’s quantity by quarter. (We allow each store to have two extreme values by setting the two highest z-scores to missing). Next, we filter the data at the zip code level by removing any zip code and all its stores if that zip code has at least one store with a z-score below -3.0 or above 3.0 in any quarter of the data.

3. From this remaining set of stores, we drop any zip code that has fewer than 5 distinct stores.

4. Finally, we filter the data again to drop zip codes with implausibly high variation in quantity sold. We do this by computing the mean and standard deviation for quantity sold in each zip code. Next, we compute the ratio of the standard deviation over the mean. Calculating the mean of this ratio, we drop all zip codes above the mean.

The remaining zip codes comprise our list of usable zip codes from the quantity data. For the proceeding analyses, we restrict the data to observations from zip codes that meet the above criteria in

42 In our updated data, we have observations that extend up to March 2011. Using all of our price data (which span January 2007 to March 2011) and filtering based on various density criteria at the zip code level does not affect the main results presented in this paper.

27

both the price and quantity data. In total, this yields 1,338 usable zip codes. Note that we pick good zip codes and re-introduce the “bad” stations within those zip codes for the analysis presented in the paper.

TABLE A12: NUMBER OF STATIONS ACROSS SAMPLE CUTS

Price Data Qty Data Both

# # #

Stores in OPIS Raw Data 135,973 119,631 118,813

Stores Located in "Good Zips" 15,825 13,865 13,795

Stores Located in "Good Zips" and Not ARCO

14,167 12,575 12,519

Stores Located in "Good Zips", Not ARCO and Not BP Competitor

7,503 6,735 6,709

Stores Located in "Good Zips", Not ARCO, Not BP Competitor and Have Demographic Info

7,406 6,648 6,622

Date post:	02-Mar-2021
Category:	Documents
Upload:	others
View:	1 times
Download:	0 times