BRAND: Brand recognition and attitude norms database

Carolina Raffaelli; Elena Bocchi; Zachary Estes; James S Adelman

doi:10.3758/s13428-024-02525-x

. 2024 Dec 16;57(1):17. doi: 10.3758/s13428-024-02525-x

BRAND: Brand recognition and attitude norms database

Carolina Raffaelli ^1,^✉, Elena Bocchi ², Zachary Estes ^2,³, James S Adelman ⁴

PMCID: PMC11649726 PMID: 39681772

Abstract

Research involving brands has increased substantially in recent decades. However, no extensive and free dataset of consumer responses to branding stimuli exists. The present research develops and validates such a dataset, which we call the Brand Recognition and Attitude Norms Database (BRAND). BRAND is the most comprehensive set of methodologically transparent, freely available, research-relevant consumer responses to branding stimuli, with measures of familiarity (awareness), liking (attitudes), and memory (recognition) of more than 500 top brands and their logos, spanning 32 industries. BRAND includes 5,356 primary datapoints aggregated from 244,400 raw datapoints (i.e., individual familiarity, liking, and memory responses) collected from 2000 US-resident consumers in 2 years (i.e., 2020 and 2024). The data exhibit good reliability, face validity, external validity, robustness across samples and time, cross-validity, and discriminant validity. BRAND can be broadly useful for testing hypotheses involving responses to brands, and for selecting stimuli in any study involving brands or logos. Thus, BRAND can facilitate research not only in consumer behavior and psychology but also in several related academic disciplines (e.g., economics, management, marketing).

Keywords: Brand attitudes, Brand awareness, Brand logos, Brand recognition, Consumer responses

Introduction

Brands are ubiquitous and highly differentiated. They dominate markets ranging from the most basic goods (e.g., Charmin toilet paper) to the most advanced services (e.g., Goldman Sachs investments), and they differentiate along many dimensions of positioning, from price (e.g., economy versus premium; Geyskens et al., 2010) to personality (e.g., sincere versus exciting; Aaker, 1997). Consumers, in turn, purchase brands for many reasons and use them in myriad ways. For instance, some consumers purchase brands out of functional necessity, whereas others buy brands for hedonic pleasure (Hirschman & Holbrook, 1982; Voss et al., 2003). Moreover, some consumers use brands to develop their own identity (McCracken, 1986), whereas others use brands to signal their identity to others (Han et al., 2010). There is simply no denying that brands are prevalent and important in contemporary life and culture. With the prominence of brands in modern societies increasing dramatically, so has scholarly research on brands increased substantially in recent years.

Aside from the brand's name, the next most important brand element is the logo, which is a visual symbol representing the brand. A brand’s logo is an extremely important brand element because it affects brand attitudes (Janiszewski & Meyvis, 2001). The visual properties of logos, such as whether it is complete or incomplete (Hagtvedt, 2011), framed or open (Fajardo et al., 2016), static or dynamic (Cian et al., 2014), angular or circular (Jiang et al., 2016), symmetric or asymmetric (Luffarelli et al., 2019a, 2019b), can affect consumers’ inferences about and evaluations of the brand. In fact, logos can also impact brand authenticity and commitment and, ultimately, the brand’s financial performance (Luffarelli et al., 2019a, 2019b; Park et al., 2013). Like branding more generally, scholarly research on logos has also increased substantially.

Given the ubiquity and importance of brands and logos, and given the growth in research on them, we find it surprising that there is no extensive and freely available dataset of consumer responses to brands and logos. There is no comprehensive dataset that researchers can use to test whether some property generally predicts brand attitudes. Nor is there any dataset to assist researchers in selecting brands or logos that are matched on some properties (e.g., familiarity) and vary on others (e.g., memorability). The present research develops and validates such a dataset, which we call the Brand Recognition and Attitude Norms Database, or BRAND. Before describing the dataset, we first review the current state of research on brands and logos, in order to identify how BRAND can facilitate research in consumer behavior, economics, marketing, and psychology.

Branding research: State of the art

Some of the most important research on branding has been purely theoretical (Keller, 1993) or qualitative (Brown et al., 2003). Such studies can provide insightful overviews or an in-depth understanding of consumer responses to brands. The present research, however, focuses on quantitative studies of branding that use regression or experiment designs. In this section, we briefly review how quantitative research on brands and logos is typically conducted, and the datasets that are available to facilitate such research.

Empirical approaches

Just as in other areas of research, quantitative studies of branding usually take one of two basic approaches. One approach, which we call regression-generalization studies, is to demonstrate that some factor of interest predicts some brand-related outcome (e.g., brand attitudes; sales). Such studies typically use regression analyses to generalize an effect across a large number of branding stimuli, while statistically accounting for other relevant factors. Regression-generalization studies are common in brand-based research. For example, in their seminal investigation of consumer responses to logos, Henderson and Cote (1998) measured a broad range of design attributes (e.g., naturalness, elaborateness) for 195 logos. They then conducted regression analyses to identify which of those factors best predicted consumers’ attitudes toward and memory of the logos.

A second approach to branding research, which we refer to as experimental-proof studies, is to demonstrate that, under some conditions at least, some factor of interest affects some response to branding stimuli (e.g., logo memory; purchase intentions). Essentially, such studies employ experimental designs to provide an existence proof that some property of brands can affect some attitude or behavior, while methodologically controlling other relevant factors. Experimental-proof studies are extremely common in brand-based research, and they typically include only one or a few brands. For instance, in a field experiment, Walsh et al. (2010) used Adidas and New Balance brands to investigate the effect of logo design on brand attitudes.

Extant datasets

Several lists of “top brands” are produced regularly, with some limited amount of data freely available in them. We summarize these brand lists in Table 1. Perhaps the most famous is Interbrand’s “Best Global Brands” list, and Forbes, Kantar (BrandZ), Tenet, and Brand Finance produce similar lists. Most of these lists include not only brand rankings but also some proprietary measure of “brand value,” which is based largely on financial performance (e.g., revenues) but which typically also incorporates an estimate of brand equity (e.g., consumer attitudes). See the Appendix for descriptions of these brand value calculations.

Table 1.

Extant brand datasets

Source	Report	Brands (N)	Key measures
Brand Finance	US 500	500	Rank, value
BrandZ (Kantar)	Top 100 Most Valuable US Brands	100	Rank, value, contribution
Forbes	World's Most Valuable Brands	100	Rank, value, revenues
Interbrand	Best Global Brands	100	Rank, value
Tenet	Top 100 Most Powerful Brands	100	Rank

Open in a new tab

Researchers often use these existing datasets to select brands for use in their studies or to test hypotheses about brands. For instance, Labrecque and Milne (2012) and Luffarelli et al., (2019a, 2019b) utilized the Interbrand list to study the relationship between logo characteristics and brand personality. Van der Lans et al. (2014) examined brand alliances using brands from the Interbrand and BrandZ lists. Johansson et al. (2012) and Madden et al. (2006) used Interbrand brands to represent high-equity brands, and investigated the impact of brand equity on stock returns, respectively. Chu and Keh (2006) explored strategic investments' effects on brand value with the Interbrand list, while Harjoto and Salas (2017) and Gherghina and Simionescu (2015) investigated corporate social responsibility and brand value using Interbrand and Brand Finance values, respectively. Similar examples abound.

Although the extant datasets have proven useful for some research purposes, they also suffer several critical limitations. First, extant datasets tend to include relatively few brands. Most brand datasets, such as Interbrand’s list, include only 100 brands. Those lists thus provide little statistical power for testing hypotheses about brands. Second, because such datasets are proprietary, they tend to provide relatively little detail about the methods used to sample the brands, to generate the data, and/or calculate the brand values. Thus, due to a lack of transparency, academic researchers are generally unable to assess the extent to which the data are suitable for their purposes, much less whether they are reliable and valid. Third, although most brand lists do include some more or less direct measure of brand equity (i.e., consumer perceptions and/or opinions), the currently available brand datasets are based mostly on financial performance in terms of sales, revenues, and so on. Those datasets lack some measures that are of great interest to academic researchers, such as brand awareness, attitudes, and recognition. Fourth, although some more consumer-focused brand lists are available from branding consultancies, they tend to be costly. For instance, the BrandAsset Valuator (BAV) includes equity measures such as brand knowledge, brand personality, and usage frequency, but those data are sold at costs on the magnitude of tens of thousands of dollars, which is beyond the reach of most academic research budgets. A final limitation of extant datasets is that they only include evaluations of the brand. There are currently no rankings or evaluations of other influential brand elements, such as logos.

In sum, brand-based research has increased dramatically in recent years, utilizing both the regression-generalization and the experimental-proof approaches. As explained below, a large dataset of consumer responses to brands could benefit both of those empirical approaches to brand-based research. Although a few datasets of brand information do exist, they include relatively few brands, their evaluation methodologies tend to be opaque, they lack measures that are of interest to many academic researchers, and those that do include measures of consumer equity tend to be financially inaccessible to academic researchers, and they lack different brand elements such as logos. The BRAND database addresses those limitations.

BRAND

The present research aims to generate and validate an extensive set of freely available, methodologically transparent, research-relevant consumer responses to brands. Specifically, we created a dataset of the familiarity, liking, and memorability of more than 500 top brands and their logos. We refer to this dataset as the Brand Recognition and Attitude Norms Database, or BRAND. We collected BRAND data initially in 2020 (i.e., BRAND 2020) and again in 2024 (i.e., BRAND 2024).

What is it?

BRAND is the most comprehensive dataset of frequently researched consumer responses to branding stimuli. As shown in Table 1, the Brand Finance list of top brands is the most extensive freely available list, including 500 brands. We therefore based our BRAND database on this Brand Finance list. Although the Brand Finance list is nominally focused on “US brands,” it nonetheless includes many top global brands. For instance, 57 of the 100 brands on the Forbes (2020) “World’s Most Valuable Brands” list also appeared in the Brand Finance (2019) list, and hence in BRAND 2020. In addition to the brands themselves, we also collected consumer responses to the brands’ logos, because logos critically affect brands’ financial performance (Luffarelli et al., 2019a, 2019b; Park et al., 2013) and because they are a common topic of research in their own right. Thus, BRAND 2020 includes 500 top brands and their logos, for a total of 1000 stimuli.

Researchers often study brand awareness, brand attitudes, and brand recognition, as these measures are key indicators of brand equity (Keller, 1993). For instance, Kent and Allen (1994) and Campbell and Keller (2003) investigated the effects of brand awareness on advertising effectiveness, Percy and Rossiter (1992) examined the relationship between brand awareness and brand attitudes in the context of advertising, and Keller (1987) assessed the effects of advertising on brand attitudes and brand recognition. We therefore measured those three consumer responses in our dataset. We assessed awareness via a measure of familiarity, attitudes via a measure of liking, and recognition via a memory test.

BRAND 2024 is an update and extension of BRAND 2020. It includes all 492 surviving brands from BRAND 2020, that is, the 500 original brands minus eight brands that ceased operating before 2024 (e.g., Twitter). It also includes 97 new entrants to the Brand Finance list in 2023 (e.g., X). BRAND 2024 includes familiarity and liking ratings, but not memory. BRAND 2024 therefore consists of the familiarity and liking of 589 brands and their logos. Overall then, the BRAND dataset includes 597 brands: 500 from BRAND 2020, and 97 that appear in BRAND 2024 but not in BRAND 2020.

Our data collection procedures closely followed those typically employed in cognitive psychology and psycholinguistics, where open sharing of large datasets of attitudinal and behavioral responses to words (cf. brands) and images (cf. logos) is common (e.g., Andrade et al., 2023; Balota et al., 2007; Dan-Glauser & Scherer, 2011; Muraki et al., 2023; Stoinski et al., 2023; Winter et al., 2024).

What is it for?

BRAND has two primary uses: hypothesis testing and stimulus selection. BRAND provides a standardized basis for hypothesis testing via the regression-generalization approach. In fact, BRAND can be useful for retrieving dependent variables, predictor variables, and/or control variables. Regarding dependent variables, any researcher who wishes to test whether some factor predicts recognition of, attitudes toward, or memorability of brands or logos has simply to merge a measure of the given factor with the measures in BRAND. For example, if one hypothesizes that consumers like brand names with a certain phonetic property (e.g., Lowrey & Shrum, 2007), one need only code whether each of the brand names in BRAND has that phonetic property and then use those coded values to predict brand attitudes within BRAND (see Study 2). If one hypothesizes that circular logos are liked more than square logos (e.g., Jiang et al., 2016), then they can simply collect ratings of how rounded or squared the BRAND logos are and test whether those ratings predict consumer attitudes toward the logos in BRAND.

Regarding predictor variables, BRAND can also be used to test whether recognition of, attitudes toward, or memorability of brands or logos predicts some other outcome. Want to know whether logo attitudes predict sales (e.g., Park et al., 2013)? Simply retrieve attitudes toward the logos in BRAND, retrieve those brands’ financial performance data from Compustat, merge the two datasets, and run a regression. Want to know whether well-known logos counterintuitively elicit worse memory (cf. Blake et al., 2015)? Retrieve logo familiarity ratings and logo memory scores from BRAND to find out (see Study 1).

BRAND can also be useful for control variables. Suppose a researcher investigates whether the introduction of a new brand extension increases sales of the brand’s other offerings (Balachander & Ghose, 2003; Swaminathan et al., 2001). Given that some brands – but not others – introduce a new brand extension during the period of study, the researcher could retrieve brand attitudes from BRAND as a control factor, in order to ensure that the effect of a brand extension on sales is not simply due to a confound of highly liked brands being more likely to introduce brand extensions. Alternatively, the researcher could test whether brand attitudes moderate the presumed effect.

Second, BRAND provides a tool for stimulus selection in experimental-proof studies. In fact, just as in regression-generalization studies, BRAND can be useful for both manipulations and controls here also in experimental-proof studies. In terms of manipulations, for instance, when investigating product choices between famous brands and lesser-known brands (e.g., Thoma & Williams, 2013), the familiarity ratings in BRAND could be used to identify more and less known brands. To investigate how brand attitude affects word of mouth, one may consult BRAND to identify brands that vary in liking but are matched on familiarity and memorability. In terms of controls, BRAND may be even more broadly useful. If an experimenter investigates the influence of repetition on liking and memory of logos (Janiszewski & Meyvis, 2001; Van Grinsven & Das, 2016), they could consult BRAND to identify logos that are matched on familiarity, liking, and memorability outside the context of the experimental repetitions. If one wishes to test whether advertisements with celebrity endorsers distract attention away from the brand (Erfgen et al., 2015), they could use BRAND to identify brands that are matched for familiarity and liking, so that any difference in brand memory would be attributable to the presence or absence of a celebrity endorser, rather than to familiarity or liking of the brands.

Thus, whether used as dependent variables (e.g., logo memorability), predictor variables (e.g., brand attitude), or control variables (e.g., brand familiarity), BRAND is ideal for hypothesis testing via regression-generalization studies. BRAND is also ideal for stimulus selection in experimental-proof studies, allowing researchers to easily identify brands or logos that are matched on some consumer responses (e.g., familiarity) and/or that vary on other consumer responses (e.g., attitudes).

Who is it for?

Most obviously, BRAND is useful for branding researchers. Any researcher who studies brands and/or logos can benefit from using BRAND to test hypotheses about awareness, attitudes, and/or recognition, to control for those consumer responses when testing hypotheses about other behaviors or outcomes, and to select well-controlled stimuli in experimental-proof studies of brands and logos.

BRAND may also benefit other marketing researchers who do not specifically investigate branding. Essentially any researcher who uses brands or logos as stimuli can benefit. For instance, advertising researchers can use BRAND to identify suitable brands (e.g., moderate attitudes) to use in their ads. A social media researcher who investigates the impact of corporate tweets on earned media impressions can retrieve brand attitudes from BRAND and include them as a control factor in regression analyses. A marketing strategy scholar who studies the effect of new product launches on revenues can use BRAND to retrieve consumers’ prior attitudes toward the given brands as a potential moderator of the effect. And so on.

BRAND can also benefit researchers in psychology. For instance, academic psychologists sometimes use branding stimuli to test theoretical predictions and/or validate findings with more naturalistic stimuli. Buttle and Westoby (2006) used brand logos (e.g., Apple, Camel, Nike) to investigate the phenomenon of repetition blindness in visual perception, and Streicher and Estes (2015) used Coca-Cola bottles and Red Bull cans to demonstrate that tactile processing facilitates visual perception. Khan et al. (2013) found that more conservative people were more likely to prefer established national brands, and Walasek and Brown (2015) showed that status goods, such as designer brands (e.g., Ralph Lauren), are searched more frequently in US states with high income inequality than in more equal states.

BRAND may also prove useful for researchers in several other related fields, including not only similar disciplines such as management (e.g., Wood, 2000) and economics (e.g., Head & Mayer, 2019) but also less similar disciplines such as sociology (e.g., Hamilton, 2012) and cultural anthropology (e.g., Foster, 2007). Thus, BRAND can facilitate research both within and beyond consumer behavior and psychology.

Overview of studies

We generated and validated BRAND across three studies. Study 1 generated and validated BRAND 2020. Conducted in 2020 on the Brand Finance (2019) US 500 list, this study included measures of the familiarity, liking, and memorability of 500 top brands and their logos. Analyses established the validity and reliability of the dataset in several ways. Study 2 then tested the external validity and utility of BRAND by replicating and generalizing a prior result from the branding literature. Specifically, using fictitious brand names, Lowrey and Shrum (2007) found that consumers like brand names with back vowels (e.g., “Nallen”) more than those with front vowels (e.g., “Nillen”). Study 2 reveals that this effect also holds across a much larger set of real brands in BRAND 2020, demonstrating the validity and utility of BRAND. Finally, Study 3 generated and validated BRAND 2024. Conducted in 2024 on the Brand Finance (2023) US 500 list, this study included measures of the familiarity and liking of 589 brands and their logos. BRAND 2024 includes all surviving brands and logos from BRAND 2020 (i.e., brands that ceased trading between 2020 and 2024 were removed), plus all new entrants on the 2023 Brand Finance list (i.e., brands that appeared in Brand Finance’s 2023 list but not on their 2019 list). Thus, the comprehensive BRAND dataset, integrating both 2020 and 2024, includes a total of 597 brands and logos. This encompasses the initial 500 from BRAND 2020, supplemented by an additional 97 brands that were introduced in BRAND 2024 and were not present in 2020. All stimuli, all surveys, and all data1 are freely available at https://researchbox.org/1892.

Study 1: BRAND 2020

BRAND 2020 includes measures of familiarity, attitudes, and memorability of 500 top brands and their logos, for a total of 3000 aggregated datapoints (i.e., 500 brands and 500 logos × 3 measures). Those 3000 datapoints are aggregated from approximately 150,000 responses from approximately 1200 US-resident consumers. Each respondent evaluated 50–100 brands or logos, and each datapoint is aggregated across approximately 30–60 consumer responses.

Methods

Participants

One thousand two hundred six respondents from the Prolific online research panel participated for pay. Respondents’ ages ranged from 18 to 81, with a mean of 32 years (SD = 12); 54.0% self-identified as female, 43.6% as male, and 2.4% chose not to say. We used a screening function in Prolific to restrict eligibility to users who reported current residence in the U.S., and we subsequently retrieved demographic data from Prolific confirming that nearly all participants (94.4%) self-reported residence in the US (“prefer not to say” = 5.6%). We excluded two participants who reported residence outside the US.

Data were collected from four independent groups. Of the 1204 US-resident participants, approximately half (N = 605) rated their familiarity with and liking of the stimuli (henceforth familiarity-liking group), and after excluding five participants who failed an attention check (see Procedure), 600 valid participants remained in this group. The other half of the participants (N = 599) instead rated their liking and then completed a memory test for the stimuli (liking-memory group), and after excluding five participants who reported a problem with the presentation of logos (see Procedure) and two participants who completed the survey twice, 592 valid participants remained in this group. Within each of those two groups, each participant judged either the brands or their logos, but not both. Of the 600 valid participants in the familiarity-liking group, approximately half rated the brands (N = 297) and half rated the logos (N = 303). Furthermore, of the 592 valid participants in the liking-memory group, approximately half rated the brands (N = 293) and half rated the logos (N = 299). Table 2 summarizes the basic demographic characteristics of the respondents in each group.2

Table 2.

Respondent characteristics (after exclusions) in BRAND 2020

			Age				Gender
Group	Stimuli	N	Min	Max	M	SD	Female	Male
Familiarity-Liking	Brands	297	18	79	30.57	9.98	61%	39%
Familiarity-Liking	Logos	303	18	67	30.46	10.69	61%	39%
Liking-Memory	Brands	293	18	81	32.83	13.26	50%	50%
Liking-Memory	Logos	299	18	78	34.90	11.94	49%	51%
	Overall	1192	18	81	32.19	11.47	55%	45%

Open in a new tab

2.4% of respondents who chose not to report their age and 2.52% who chose not to report their gender are excluded from those respective statistics in the table

Stimuli

We selected the Brand Finance (2019) US 500 because it was the most extensive list of “top brands,” with a total of 500 brands spanning 32 industries, from aerospace to utilities (see Table 3).3 This Brand Finance list uses the “Royalty Relief” method of brand valuation, which is defined as the “net economic benefit that a licensor would achieve by licensing the brand in the open market” (Brand Finance, 2019, p. 30). That is, brand value is derived not only from the brand’s financial performance but also from its brand equity (i.e., consumers’ subjective valuation of the brand). For instance, in the apparel category, Victoria’s Secret and Old Navy generated approximately equal revenues in 2019, but Victoria’s Secret had a far higher brand value due to consumers’ more positive perceptions of the brand.

Table 3.

Brand characteristics in Brand Finance, 2019 and 2023

		Brand Finance 2019 Rank			Brand Finance 2023 Rank
Industry	Example	N	M	SD	N	M	SD
Aerospace & Defense	Boeing	10	209	155	8	197	111
Airlines	Delta	6	194	164	4	126	50
Apparel	Nike	17	260	130	14	279	124
Auto Components	Lear Corp	0	-	-	1	445	-
Automobiles	Ford	11	249	146	11	212	135
Banking	Wells Fargo	31	218	147	32	194	136
Beers	Budweiser	5	262	146	5	292	146
Car Rental Services	Enterprise	3	282	150	5	310	136
Chemicals	Dow	5	291	132	5	329	102
Commercial Services	Deloitte	27	266	159	32	257	150
Cosmetics & Personal Care	Johnson's	27	258	126	19	261	98
Engineering & Construction	General Electric	22	258	114	22	263	125
Exchanges	CME	2	448	74	3	454	54
Food	Purina	24	322	132	30	330	138
Healthcare	UnitedHealthcare	27	257	150	27	219	140
Hotels	Hilton	11	300	124	7	244	116
Household Products	Scotts Miracle-Gro	0	-	-	1	454	-
Insurance	GEICO	15	240	152	19	263	162
IT Services	Accenture	9	199	153	5	268	202
Leisure & Tourism	Royal Caribbean	6	321	75	8	413	65
Logistics	UPS	8	175	152	12	235	164
Media	Disney	20	206	156	25	258	154
Oil & Gas	Chevron	26	286	139	25	286	129
Pharma	Pfizer	10	302	99	8	247	99
Real Estate	CBRE	4	408	56	2	370	60
Restaurants	Starbucks	13	189	129	17	209	147
Retail	Walmart	42	237	151	38	193	134
Soft Drinks	Coca-Cola	11	223	118	10	244	149
Spirits	Jack Daniel's	2	210	25	2	334	93
Tech	Amazon	70	216	154	71	213	152
Telecoms	AT&T	10	168	154	8	174	171
Tires	Dunlop	2	379	8	2	420	22
Tobacco	Marlboro	14	285	126	13	276	120
Utilities	Exelon	10	404	70	9	400	76

Open in a new tab

Because the Brand Finance report does not include the brand logos, we manually and extensively searched for and collected each brand’s logo. For every brand, we first retrieved the logo displayed on the US version of its corporate website. However, we observed that many brands use multiple logos, raising the question of which of them we should include in our dataset. For instance, Facebook uses different logos on its webpage (which includes the full name “Facebook”) and its mobile app (which includes only the letter “f”). Given the pragmatic constraint that we could include only one logo for each brand, we conducted a secondary search for alternative logos, and we developed a set of decision rules to identify a single logo for each brand that we would include in our dataset. This procedure was challenging and cumbersome, as described next.

For every brand, we searched for alternative logos that may appear on the product, in the retail channel (e.g., on packaging), in the app store, or in other points of contact with consumers. Most brands (60% of them) used the same logo consistently across those various consumer touchpoints. However, 200 brands (40%) used different logos in different touchpoints, as in the Facebook example above. After reviewing those discrepant cases, we established a set of rules to determine which logo to include in our dataset. Those rules, and the number of cases decided by each rule, are described in Table 4. After implementing seven rules summarized in the table, we were left with 42 “miscellaneous” cases that could not be determined by any of the other rules. In those cases, two of the authors independently judged which was the brand’s primary logo, based on another review of each brand’s various touchpoints. In case of disagreement, another author cast a tie-breaking judgment.

Table 4.

Logo selection rules

		N
	Selection Rule	2020	2024	Description
Consistent logo use		300	321	The brand uses the same logo consistently across various consumer touchpoints
Inconsistent logo use		200	277	The brand uses different logos across touchpoints
	App	11	15	If one version is the app icon, then take the version that is not the app icon
	Color	44	82	If the two versions display different colors, then take the one that is most prominent in the company's website
	Orientation	28	48	If the two versions of the logo are differently oriented, then take the one on the product or the mode on Google Images
	Movie	4	5	If one version is a stylized logo displayed in movies (i.e., for film-related brands), then take the stylized version
	Part	33	40	If one version is part of the other, then take the most complete version
	Slogan	30	31	If one version includes the slogan, then take the one without the slogan
	Restaurant	8	13	If one version is found on the restaurant signboard, then take the one on the signboard
	Miscellaneous	42	43	If none of the rules above applied, then the logo was selected by consensus among authors

Open in a new tab

Whenever possible, we retrieved the brand’s logo without a background, unless it was an integral part of the logo (e.g., the Goldman Sachs logo consists of the name inside a blue square). We resized all logos to have similar dimensions by positioning them at a minimum distance from the edges of a standardized frame, and we saved them as 400 × 225 resolution portable network graphic (png) files.

Procedure

This research was approved by the research ethics committee at the host university. After providing informed consent, participants in the familiarity-liking group read the following attention check: “How much attention are you willing and able to dedicate to this task? If you select VERY LITTLE, then we will have to reject your work” (options: very little; a whole lot). The data of all participants who selected “very little” were excluded from analyses (N = 5).

The study included 1000 stimuli in total (500 brands + 500 logos). To minimize fatigue and demotivation, each participant evaluated either 50 brands or 50 logos (between-participants). The 500 brands were randomly allocated among ten lists of 50 brands, and we similarly created ten lists of 50 logos. Each participant was randomly assigned to one of those lists, and the 50 stimuli within the assigned list were presented in random order.4

Participants were informed that “We are investigating consumer opinions about [brands/logos].” Participants assigned to evaluate logos were additionally informed that “A logo is a visual symbol representing a brand…Please evaluate the logo (i.e., the visual symbol), rather than the brand that it represents.” Participants in the familiarity-liking group rated the stimuli in terms of familiarity (“Please rate to what extent you are familiar with this [brand/logo];” 1 = not familiar at all; 7 = very familiar) and liking (“Please rate to what extent you like or dislike this [brand/logo];” 1 = dislike; 7 = like). Each stimulus was presented on a separate page, with both rating scales appearing below the brand name or logo.

Participants in the liking-memory group completed two experimental phases. In the exposure phase, participants read the same instructions but were additionally informed that the study would last approximately 10–12 min, and also read the following: “Please do not start this study unless you are able to complete the entire task without a break. If you take a break in the middle, you may be timed out.” Participants then rated their liking of 50 stimuli, exactly as in the familiarity-liking group, except without rating familiarity. Participants were not informed that there would be a second phase of the study. Upon completion of the exposure phase, however, participants immediately began a test phase, which consisted of a recognition memory test. They read the following interim instructions:

“You have completed the first part of the study. Please now begin the second part of the study without taking a break. If you take a break, you may be timed out. In this second (and final) part of the study, we will show you 100 [brands/logos]. Some of these [brands/logos] will be ones that you saw in the first part of the study (i.e., "old"), whereas others will be [brands/logos] that you did not see in the first part (i.e., "new"). Please click 'old' if you think that you saw the [brand/logo] in the previous part of the study, or click 'new' if you think you didn't.”

In addition to the 50 (old) stimuli that the participant evaluated in the exposure phase, participants were also shown 50 (new) stimuli from one of the other lists. Each list of 50 stimuli was yoked with another list of 50 stimuli, so that all stimuli appeared approximately equally often as an old stimulus and as a new stimulus. Each stimulus appeared on a separate page, above the “old” and “new” response options, and the 100 stimuli appeared in random order. The recognition memory task was modeled after Cortese et al. (2010).

Finally, at the end of the study, participants in the liking-memory group were asked an additional question. Because a few participants in the familiarity-liking group commented in a feedback textbox that sometimes the logos took a while to load or occasionally did not load at all, in this liking-memory task we added a question at the end of the study to assess whether this problem had occurred. Participants were asked “When you were doing the experiment, did all [logos/brands] clearly appear on your screen?” Response options were “Yes, all [logos/brands] clearly appeared” and “No, some [logos/brands] never appeared.” Participants who indicated that some logos or brands did not appear were then asked “How many [logos/brands] do you think you have missed?”, and they typed their estimate into a textbox. Only 14 participants (i.e., 2.4% of the sample) indicated that some stimuli did not appear, and those 14 participants’ mean estimate of the number of stimuli that did not appear was 9 (SD = 13).5 The data of all participants who reported missing ten or more stimuli were excluded from analyses (N = 5).

Results

Data aggregation

In the BRAND dataset, the brand is the unit of analysis. For each measure of each brand (e.g., familiarity with the Nike logo), we aggregated across all participants’ responses. For continuous measures (i.e., familiarity and liking ratings), these were averages. For categorical measures (i.e., memory judgments), these were proportions. Each brand was rated for familiarity by approximately 30 participants (min = 29, max = 31, SD = 0.64). Because both groups of participants (i.e., familiarity-liking and liking-memory) rated their liking of the brands, each brand received about twice as many liking ratings: approximately 59 participants (min = 58, max = 60, SD = 0.78). And because each brand appeared as an “old” memory stimulus for some participants and as a “new” stimulus for others, each brand also received memory judgments by approximately 59 participants (min = 58, max = 60, SD = 0.80). Analogously, each logo received approximately 30 familiarity ratings (min = 28, max = 32, SD = 1.27), 60 liking ratings (min = 56, max = 63, SD = 2.48), and 60 memory judgments (min = 57, max = 62, SD = 1.72).6

Memory measures

We analyzed memory performance via signal detection theory, with four primary measures: A hit occurs when an “old” stimulus is correctly identified as old; a miss occurs when an old stimulus is incorrectly judged as new; a false alarm occurs when a new stimulus is incorrectly judged as old; and a correct rejection occurs when a new stimulus is correctly identified as new. One simple and common method for converting those primary measures into an accuracy score, which we refer to as corrected accuracy (Stanislaw & Todorov, 1999), is to subtract the false alarm rate from the hit rate (i.e., Hits–False Alarms). This score essentially reflects the likelihood of correctly recognizing an old stimulus, while also correcting for response bias (captured here by the false alarm rate). A similar but more formal measure, d’, subtracts the z-score of the false alarm rate from the z-score of the hit rate (i.e., Z_Hits – Z_{False Alarms}), thus providing a standardized accuracy measure in SD units. We include all six of these memory measures in the BRAND 2020 dataset. However, because corrected accuracy and d’ were extremely correlated (brands: r = 0.98, p < 0.001; logos: r = 0.97, p < 0.001), in all other text, tables, and figures, we report corrected accuracy (Hits–False Alarms) as our primary measure of recognition memory, and we do not discuss d’ further.

Distributions

Descriptive statistics of the familiarity ratings, liking ratings, and memory scores for the brands and logos are reported in Table 5 and histograms of the six focal measures are shown in Fig. 1. Brand familiarity ratings ranged from 1.00 (Parker-Hannifin, Lam Research) to 6.90 (Target), with a mean of 3.76 (SD = 1.92). Notably, however, the distribution is bimodal (Fig. 1). Brand liking ratings ranged from 2.22 (Marlboro) to 6.17 (Netflix), with a mean of 4.35 (SD = 0.57). As shown in Fig. 1, the distribution was normal. Brand memorability (corrected accuracy) scores ranged from 0.03 (ADT) to 0.90 (Olive Garden, Safeway, Dick’s Sporting Goods, Lincoln), with a mean of 0.62 (SD = 0.14), and were normally distributed. Logo familiarity ratings ranged from 1.00 (CA Technologies) to 6.97 (Amazon, Ford, Walmart), with a mean of 4.18 (SD = 2.09). As with brands, logo familiarity shows a bimodal distribution with peaks near 1 and 7 (Fig. 1). Logo liking ratings ranged from 2.72 (Arrow Electronics) to 5.83 (FedEx), with a mean of 4.39 (SD = 0.61), and were normally distributed. Logo memorability scores ranged from 0.24 (HCL) to 0.94 (GEICO), with a mean of 0.66 (SD = 0.12), and were normally distributed.

Table 5.

Descriptive statistics of measured variables in BRAND 2020

Stimuli	Measure	N	Min	Max	M	SD	Skew
Brands	Familiarity	500	1.00	6.90	3.76	1.92	0.01
	Liking	500	2.22	6.17	4.35	0.57	0.48
	Hits	500	0.46	0.98	0.80	0.11	– 0.65
	Misses	500	0.02	0.54	0.20	0.11	0.65
	False Alarms	500	0.02	0.83	0.18	0.09	1.50
	Correct Rejections	500	0.17	0.98	0.82	0.09	– 1.50
	Corrected Accuracy	500	0.03	0.90	0.62	0.14	– 0.59
	d'	500	0.12	3.30	1.88	0.55	0.03
Logos	Familiarity	500	1.00	6.97	4.18	2.09	– 0.11
	Liking	500	2.72	5.83	4.39	0.61	0.00
	Hits	500	0.44	0.98	0.81	0.09	– 0.42
	Misses	500	0.02	0.56	0.19	0.09	0.42
	False Alarms	500	0.02	0.40	0.15	0.07	0.56
	Correct Rejections	500	0.60	0.98	0.85	0.07	– 0.56
	Corrected Accuracy	500	0.24	0.94	0.66	0.12	– 0.40
	d'	500	0.62	3.70	2.06	0.53	0.33

Open in a new tab

Fig. 1 — Distributions of the familiarity, liking, and memorability (corrected accuracy) of brands and their logos in BRAND 2020

Reliability

Recall that two independent groups of participants rated their liking of the BRAND 2020 stimuli: Some participants rated familiarity and liking, whereas others rated liking and then completed a memory test. We therefore used those independent sets of liking ratings to assess test–retest reliability. Brand-liking ratings by the familiarity-liking group and the liking-memory group correlated strongly, r = 0.79, p < 0.001. Logo liking ratings by the two groups of participants also correlated strongly, r = 0.77, p < 0.001. Thus, the reliability of BRAND 2020 was good.

Correlations among measures

Table 6 displays the correlations among the measures of BRAND 2020 (see lines 1-6). As expected, the six measures all correlated positively with each other. Here, we highlight two observations that are theoretically and practically meaningful. First, more familiar brands were liked more (r = 0.73) than less familiar brands, and more familiar logos were also liked more (r = 0.72) than less familiar logos. These results corroborate the mere exposure effect (Janiszewski & Meyvis, 2001; Zajonc, 1968): Increasing exposures to a stimulus tends to improve attitudes toward that stimulus. Second, more familiar brands were remembered better (r = 0.53) than less familiar brands, and more familiar logos were also better remembered (r = 0.33) than less familiar logos. These results corroborate the repetition effect (Hintzman, 1976; Janiszewski et al., 2003): Increasing exposures to a stimulus tend to improve recognition of that stimulus. Given that brand familiarity and logo familiarity were non-normally distributed, we also replicated these analyses with the nonparametric Spearman rank-order correlation, and the results were extremely similar, including significant positive correlations among all variables.

Table 6.

Correlations (Pearson r) among variables in BRAND 2020. * p < .05, ** p < .01, *** p < .001

	1	2	3	4	5	6	7	8	9
1. BRAND: Brand Familiarity	—
2. BRAND: Brand Liking	.73***	—
3. BRAND: Brand Memorability	.53***	.42***	—
4. BRAND: Logo Familiarity	.95***	.67***	.51***	—
5. BRAND: Logo Liking	.69***	.60***	.37***	.72***	—
6. BRAND: Logo Memorability	.32***	.21***	.41***	.33***	.28***	—
7. Brand Finance: Rank (N = 500)	– .37***	– .22***	– .19***	– .39***	– .22***	– .11*	—
8. Brand Finance: Value (N = 100)	.45***	.23*	.31**	.43***	.39***	.08	– .94***	—
9. BrandZ: Contribution (N = 95)	.40***	.35***	.35***	.29**	.29**	.33**	.09	.26*	—

Open in a new tab

Validation (Brand Finance)

BRAND 2020 includes 500 brands, ranked by Brand Finance from best to worst in terms of brand value. To validate the scores of BRAND 2020, we examined their correlations with brand rank in Brand Finance (2019; see Table 6). Note that stronger brands have numerically low ranks (e.g., 1st), so negative correlations are expected. Brand Finance ranks brands according to an integration of their financial performance (e.g., revenues) with consumers’ “goodwill” or attitude toward the brand (i.e., brand equity). Thus, highly ranked brands should be well liked. Moreover, given that brands’ advertising expenditures are typically proportional to their financial position (Danenberg et al., 2016), highly ranked brands should also be highly familiar and memorable. Indeed, brand rank correlated negatively with brand familiarity (r = – 0.37), brand liking (r = – 0.22), and brand memorability (r = – 0.19).

Prior research suggests that more likable logos can improve firm performance in terms of customer loyalty and sales (Park et al., 2013). This should manifest as a correlation between logo liking and brand rank. Moreover, due to highly ranked brands’ relatively higher advertising expenditures (Danenberg et al., 2016), one might expect those brands’ logos to also be more familiar and memorable. Indeed, brand rank correlated negatively with logo familiarity (r = – 0.39), logo liking (r = – 0.22), and logo memory (r = – 0.11). Thus, as expected, top consumer brands (which have numerically low ranks) and their logos elicited higher familiarity, liking, and memorability than lesser consumer brands. These results support the validity of the measures in BRAND 2020.

In addition to providing brand ranks for all 500 brands, Brand Finance also provides brand values for the top 100 brands, where brand value is a proprietary measure combining financial performance data with brand equity measures (see Appendix). As a robustness check, we tested whether the measures in BRAND 2020 predict those brand values. To correct for substantial skew (3.38), the brand values were log-transformed, producing a more normal distribution (skew = 1.03). Note that because stronger brands have higher brand values, positive correlations are expected. Indeed, the positive correlations between the measures in BRAND 2020 and the brand values (see Table 6, line 8) mirrored the negative correlations between BRAND 2020 and brand rank (Table 6, line 7). These results support the robustness of BRAND’s predictive validity.

Cross-validation (BrandZ)

The brand ranks and values in the Brand Finance list, like in most other brand lists, derive from a weighting of financial performance data and brand equity measures. Because those datasets (see Table 1) are proprietary, academic researchers are not privy to how financial performance and brand equity are weighted in the brand value calculations (see Appendix for general descriptions), nor are the brand equity measures freely available to researchers. A notable exception is BrandZ, which releases a measure of brand contribution, which in turn is some combination of consumers’ perceptions of “meaning,” “differentiation,” and “salience.” Although academic researchers are not privy to exactly how brand contribution is measured or calculated – again, because it is proprietary – it nonetheless appears conceptually similar to the measures in BRAND. We therefore used those BrandZ contribution scores to cross-validate the measures in BRAND 2020. Of the 100 brands in BrandZ (2020), 95 were also in Brand Finance (2019), and hence BRAND 2020. As expected, the correlations with brand contribution were positive and significant (see Table 6, line 9), and they generally compared well to BRAND’s correlations with the Brand Finance ranks and values. These results further support the validity of the measures in BRAND 2020.

Predicting memory

As the correlations reported above reveal, familiarity predicted liking and memory (Table 6). Familiarity and liking were measured in one sample, while liking and memory were measured in a separate sample, ensuring that the same participants never assessed both familiarity and memory. We also tested, via linear regression, whether familiarity and liking collectively predicted memory. Together brand familiarity and brand liking explained 28.2% of the variance in brand memory, F(2, 497) = 97.43, p < 0.001. The two predictors were sufficiently independent, VIF = 2.11, but only brand familiarity significantly predicted brand memory, β = 0.48, t = 8.65, p < 0.001. Regarding the logos, together, logo familiarity and logo liking explained 11.3% of the variance in logo memory, F(2, 497) = 31.77, p < 0.001. Collinearity was again moderate but acceptable, VIF = 2.08, and only logo familiarity significantly predicted logo memory, β = 0.27, t = 4.51, p < 0.001. Thus, although familiarity and liking both correlated with memory, familiarity appears to be the better predictor of memory. Likability, in contrast, appears not to affect the memorability of brands or logos. This result corroborates the recent finding that the likability and the memorability of brand elements (e.g., slogans) often diverge (Hodges et al., 2024).

Predicting Brand Finance (2019) rank from BRAND 2020

We also tested the extent to which our BRAND measures collectively predicted brand rank. To address the unacceptably high collinearity between brand familiarity and logo familiarity (VIF = 13.14), we conducted two different analyses. The first analysis was a regression that included all six measures of BRAND 2020. In this analysis, we discuss the explanatory power of the BRAND measures collectively, but we refrain from interpreting the coefficients of the predictors individually (due to their collinearity). In the first block of this model, brand familiarity, brand liking, and brand memory collectively explained 14.5% of the variance in brand rank, F(3, 496) = 28.03, p < 0.001. When logo familiarity, logo liking, and logo memory were entered in a second block, they explained an additional 2.0% of the variance in brand rank, F(6, 493) = 16.23, p < 0.001. This latter result corroborates the finding that logo attitudes can affect brand equity (Park et al., 2013).

In a second analysis, we addressed collinearity by excluding logo familiarity from the model, in order to render the coefficients of the other five BRAND predictors interpretable. Collinearity among the five predictors was moderate but acceptable, all VIF < 3.04. Results are shown in Table 7. Importantly, only brand familiarity significantly predicted brand rank, β = – 0.49, t = – 6.79, p < 0.001. The unstandardized coefficient (B = – 36.93) indicates that for each point that a brand increases on the seven-point brand familiarity scale, it jumps an average of 37 places up the brand rankings.

Table 7.

BRAND 2020 measures predict Brand Finance (2019) rank. ** p < .01, *** p < .001

Predictor	B	SE	t
Constant	206.60	75.37	2.74	**
Brand Familiarity	– 36.93	5.44	– 6.79	***
Brand Liking	25.89	15.71	1.65
Brand Memory	6.44	51.81	0.12
Logo Liking	13.51	14.07	0.96
Logo Memory	10.21	55.97	0.18

Open in a new tab

R² = 14.67%, F(5, 494) = 16.98, p < .001

In contrast, neither brand nor logo liking nor memorability significantly predicted brand rank. These findings support the common marketing assumption that brand exposure is key for consumer acceptance. How likable or memorable that exposure is appears to have little or no effect on brand performance.

Subgroup analysis

Finally, we also examined whether men and women, and younger and older respondents, produced similar distributions and results. As shown in Figs. 2 and 3, men and women produced similar distributions, with significant correlations between judgments by men and by women in brand familiarity (r = 0.91, p < 0.001), brand liking (r = 0.80, p < 0.001), brand memory (r = 0.40, p < 0.001), logo familiarity (r = 0.92, p < 0.001), logo liking (r = 0.74, p < 0.001), and logo memory (r = 0.24, p < 0.001).

Fig. 2 — Distributions of the familiarity, liking, and memorability (corrected accuracy) of brands and their logos across genders in BRAND 2020

Fig. 3 — Distributions of the familiarity, liking, and memorability (corrected accuracy) of brands and their logos across ages in BRAND 2020

To compare younger and older respondents, we calculated the median age of the entire sample (29.0 years), and we categorized individuals below the median as “younger” and those above as “older.” Younger and older respondents produced similar distributions of BRAND measures, with significant correlations in brand familiarity (r = 0.93, p < 0.001), brand liking (r = 0.79, p < 0.001), brand memory (r = 0.30, p < 0.001), logo familiarity (r = 0.92, p < 0.001), logo liking (r = 0.76, p < 0.001), and logo memory (r = 0.24, p < 0.001).

We also replicated the analysis testing whether the BRAND 2020 measures predict the Brand Finance (2019) ranks, separately for female and male respondents, and for younger and older respondents. See Tables 8, 9, 10 and 11 for results. The basic result obtained from the whole sample, whereby brand familiarity in BRAND significantly predicted Brand Finance rank (reported in the preceding section), successfully replicated within each of these four subgroups of participants.

Table 8.

BRAND 2020 measures predict Brand Finance (2019) rank in the female subgroup. *** p < .001

Predictor	B	SE	t
Constant	230.96	65.33	3.54	***
Brand Familiarity	– 32.89	5.13	– 6.41	***
Brand Liking	17.90	14.33	1.25
Brand Memory	31.50	42.75	0.74
Logo Liking	14.44	12.58	1.15
Logo Memory	– 23.20	50.27	– 0.46

Open in a new tab

R² = 12.47%, F(5, 494) = 14.08, p < .001

Table 9.

BRAND 2020 measures predict Brand Finance (2019) rank in the male subgroup. *** p < .001

Predictor	B	SE	t
Constant	276.81	69.84	3.96	***
Brand Familiarity	– 35.13	4.69	– 7.49	***
Brand Liking	20.79	14.20	1.46
Brand Memory	– 27.49	37.73	– 0.73
Logo Liking	5.33	12.54	0.43
Logo Memory	7.77	38.02	0.20

Open in a new tab

R² = 16.78%, F(5, 494) = 19.92, p < .001

Table 10.

BRAND 2020 measures predict Brand Finance (2019) rank in the younger subgroup. * p < .05, *** p < .001

Predictor	B	SE	t
Constant	155.09	68.51	2.26	*
Brand Familiarity	– 41.46	5.33	– 7.77	***
Brand Liking	27.59	14.45	1.91
Brand Memory	57.97	39.33	1.47
Logo Liking	20.16	11.65	1.73
Logo Memory	12.00	51.06	0.24

Open in a new tab

R² = 15.89%, F(5, 494) = 18.67, p < .001

Table 11.

BRAND 2020 measures predict Brand Finance (2019) rank in the older subgroup. *** p < .001

Predictor	B	SE	t
Constant	324.00	73.15	4.43	***
Brand Familiarity	– 27.63	4.68	– 5.91	***
Brand Liking	13.84	13.74	1.01
Brand Memory	– 20.83	37.90	– 0.55
Logo Liking	– 3.43	12.91	– 0.27
Logo Memory	– 0.81	38.14	– 0.02

Open in a new tab

R² = 12.86%, F(5, 494) = 14.58, p < .001

We conclude that the measures in BRAND 2020 are relatively stable across men and women, and across younger and older respondents. However, researchers interested in testing for more nuanced differences across respondents’ gender or age can do so with BRAND: The data file includes a pivot table function that researchers can use to specify sample characteristics, thereby allowing researchers to select specific subsamples (e.g., women aged 40–60 years) and to compare one subsample to another (for more detailed description on the pivot table, please refer to the “Data availability” section after Study 3).

Discussion

Study 1 developed and validated BRAND 2020. Both men and women, and younger and older respondents, produced similar distributions of familiarity ratings, liking ratings, and memory scores of 500 top US brands and their logos. The liking measure in BRAND 2020 exhibited good test–retest reliability, and the correlations among measures in BRAND 2020 successfully replicated some classic effects (e.g., mere exposure). Those measures also successfully predicted the brand ranks and brand values in Brand Finance (2019), as well as the “brand contribution” in BrandZ (2020), thus supporting the predictive validity of BRAND 2020.

Study 2: Conceptual replication of Lowrey and Shrum (2007)

Having established the validity and reliability of BRAND 2020, we next tested its utility for brand-related research by attempting to replicate an effect from the branding literature. We chose to replicate Lowrey and Shrum’s (2007) theoretically influential and practically important finding that the sounds (i.e., phonemes) in a brand name affect consumers’ brand attitudes. Specifically, Lowrey and Shrum tested the effects of front and back vowels (i.e., /i/ vs. /ɔ/, as in Gimmel vs. Gommel) on consumers’ attitudes toward brands with specific attributes (e.g., slowness or heaviness). We similarly tested whether the brands in BRAND 2020 with front versus back vowels differ in brand attitudes (i.e., brand liking in BRAND 2020).

Our study differed from Lowrey and Shrum’s (2007) original study in four key respects. First, whereas Lowrey and Shrum used only one front vowel and one back vowel (i.e., /i/ vs. /ɔ/), our analysis includes all front vowels (/i/, /ɪ/, /ɛ/, /æ/) and all back vowels (/u/, /ʌ/, /ʊ/, /ɑ/, /ɔ/), providing a more comprehensive test of the vowel effect. Second, whereas Lowrey and Shrum contrasted front and back phonemes internal to the name (i.e., not the first or last phoneme in the name), we instead contrast initial phonemes (i.e., the first phoneme in the name). This may be important because initial phonemes predict word valence better than internal phonemes (Adelman et al., 2018), and the same may be true of brand names. Third, whereas Lowrey and Shrum used fictitious brand names, our analysis instead uses real brand names, testing the vowel effect with greater ecological validity. Fourth, whereas Lowrey and Shrum examined a limited set of product categories for which specific attributes (e.g., slowness or heaviness) are important, our analysis instead includes brands from a wide variety of industries (see Table 4), thereby testing the generality of the vowel effect on brand attitudes.