Selling, General, and Administrative expense (SGA)-Based Metrics in Marketing: Conceptual and Measurement Challenges

Abstract

Many studies use variables from the Compustat database to measure various marketing constructs, yet no clear guidelines detail which metrics correspond with which constructs. Justifications rest mainly on the ready availability of easy-to-use measures that seem related to a particular construct. As a result, various metrics have been utilized to capture the same construct, and the same metric, such as selling, general, and administrative expenses (SGA), has been applied to capture vastly different constructs. But using SGA inappropriately can lead to biased estimates, questionable support for the hypotheses, and potentially misleading implications for research and practice. To test the validity of SGA for multiple relevant marketing and sales constructs, this study gathers data on benchmark variables from alternative data sources and applies a multitrait-multimethod (MTMM) approach. Results show that in general, SGA has been applied too liberally in marketing contexts; SGA is an appropriate operationalization only for some constructs. This article provides guidelines for the proper conceptualization and operationalization of marketing constructs.

Full text

1
Selling, General, and Administrative expense (SGA)-Based Metrics in Marketing:
Conceptual and Measurement Challenges
Annette Ptok
Rupinder P. Jindal
Werner Reinartz
[April 17, 2018]
ABSTRACT
Many studies use variables from the Compustat database to measure various marketing
constructs, yet no clear guidelines detail which metrics correspond with which constructs.
Justifications rest mainly on the ready availability of easy-to-use measures that seem related to a
particular construct. As a result, various metrics have been utilized to capture the same construct,
and the same metric, such as selling, general, and administrative expenses (SGA), has been
applied to capture vastly different constructs. But using SGA inappropriately can lead to biased
estimates, questionable support for the hypotheses, and potentially misleading implications for
research and practice. To test the validity of SGA for multiple relevant marketing and sales
constructs, this study gathers data on benchmark variables from alternative data sources and
applies a multitrait-multimethod (MTMM) approach. Results show that in general, SGA has been
applied too liberally in marketing contexts; SGA is an appropriate operationalization only for
some constructs. This article provides guidelines for the proper conceptualization and
operationalization of marketing constructs.
Keywords: Validation, Content validity, Construct validity, Selling, General, and Administrative
expense (SGA), Compustat, Marketing–accounting interface

2
To understand the impact of marketing and sales force activities on firm performance,
vast literature exists in marketing strategy and management that employs constructs ranging from
simple advertising spending to complex strategic marketing capabilities. As the Marketing
Science Institute (MSI 2016, p. 6) acknowledges, “making every dollar count is a marketing
imperative for all organizations. To do so requires a keen understanding of all the different brand-
building and sales-generating activities an organization may choose to engage in.” This
imperative is challenging though; few sources provide easy, cost-effective access to reliable data
across companies that capture these activities in detail. Companies protect such data closely
because they can reveal underlying strategies. Faced with this paucity of representative data,
scholars are forced to overlook the complexity of marketing constructs and their conceptual and
operational requirements in favor of achieving measurement objectives. But when studies do not
fully define or conceptualize the marketing constructs they use, it results in ambiguity and
contradiction in their meaning and measures (Varadarajan 2010).
Given the lack of alternatives, research has heavily relied on one particular source,
Compustat, which has become the go-to source for scholars interested in studying and comparing
brand-building and sales performance across organizations. This database reports on publicly
traded companies that, due to fiscal regulations, must disclose their earnings and expenditures on
various items. Compustat’s reporting is based on more than 300 items from annual income
statements, balance sheets, statements of cash flows, and supplemental data about publicly traded
companies in the United States and Canada (Wharton 2016). There are, however, no clear
guidelines on matching various marketing constructs to metrics from Compustat. In particular,
researchers have relied extensively on Compustat's selling, general, and administrative expense
(SGA) metric to capture a diverse number of constructs including marketing spending, sales
intensity, advertising intensity, and marketing assets. A reason for SGA’s prolific use is its

3
comprehensive nature – it “aggregates all costs incurred in the regular course of business except
costs associated with the production of goods and services” (Standard and Poor’s 2013, p. 269).
As a result, it tends to have one or more items that may intuitively relate to the construct a
researcher wants to capture. Nonetheless, this rationale rests on little more than the availability of
an easy-to-use measure that appears intuitive. This characterization applies to several Compustat
metrics, and thus, various metrics often serve to capture the same construct too. For example, in
addition to SGA, some studies use advertising expense to assess marketing spending. Using these
metrics to operationalize marketing constructs brings together two vastly different domains of
accounting and marketing. These domains differ in the common knowledge of how various
constructs should be defined and which variables can be applied, and in what ways, to measure
them. Before using any such variable, one should conscientiously seek to deduce theoretical
constructs, which is a prerequisite for empirical measurement, and then test the validity of their
operationalization (MacKenzie 2003). Not doing so can lead to biased estimates, questionable
support for hypotheses, and potentially misleading implications for research and practice.
Our objective is to provide a conceptual assessment of commonly used marketing and
sales constructs and an empirical assessment of alternative measures. Specifically, we address the
following three research questions:
RQ1. Which marketing and sales constructs have been measured using SGA?
RQ2. Is SGA a valid measure for these constructs? Are there alternative measures for these
constructs that may be equally or more valid?
RQ3. What guidelines can be developed for choosing between SGA and these alternative
measures?
In turn, we make several contributions to literature. First, this article provides a structured
overview of the widespread use of SGA in marketing strategy literature. Considering the disparity

4
in SGA-based operationalizations, this compilation of the status quo is overdue. Second, by
spanning the boundary between the accounting and marketing domains, we integrate frequently
neglected knowledge from accounting into marketing strategy. Specifically, we address the
conceptual breadth of each marketing construct and its operationalization using accounting-based
measures, which helps differentiate the constructs that can be measured optimally using SGA
from those that cannot. We thus demonstrate the importance of proper conceptualization of a
construct and the validation of its subsequent operationalization. Third, we add to marketing
theory and practice by deducing guidelines for appropriate operationalization of several
marketing and sales constructs. In so doing, we ensure a better understanding of the scope of
Compustat for marketing research and accordingly generate guidelines for employing available
information. These insights can improve the validity of research findings and their implications
for managers. Table 1 provides an overview of our research process.
----------------Insert Table 1 about here ----------------
Conceptual Framework
The misuse of SGA to capture various marketing and sales constructs has increased over
the past two decades in both marketing and management fields (figure 1). To find studies that
adopted this measure, we searched the EBSCO online research database after 1995, but limited
our search to 22 well-recognized peer-reviewed journals in the fields of marketing and
management such as Academy of Management Journal, Academy of Management Review,
Journal of Marketing, Journal of Marketing Research, Journal of the Academy of Marketing
Science, and Strategic Management Journal (see Web appendix 1 for the complete list of
journals). We also reviewed the reference lists of identified articles for other relevant sources. In
total, we identified 87 articles that used SGA or its modifications to operationalize one or more
marketing or sales constructs (see table 2 for a summary of operationalizations; see Appendix 1

5
for construct-wise list of articles). The constructs differ in their contextual reference and
complexity, explaining financial performance measures such as brand equity, (abnormal) stock
market returns, market value, productivity, and profitability. In turn, these constructs have been
used to perform benchmarking analyses, judge managerial ability, allocate resources, and study
firm performance.
----------------Insert Figure 1 about here ----------------
Our literature review revealed substantial variation in the emphasis placed on precise
construct definitions, as well as the general lack of validation. Imprecise definitions increase the
likelihood of misaligned or misspecified operationalizations, as manifest in the use of SGA to
operationalize diverse, wide-ranging constructs, such as marketing assets, marketing resources,
marketing capabilities, advertising intensity, sales intensity, and marketing spending. Considering
that SGA comprises 29 cash outflow items (see Web Appendix 2), it would be difficult to draw a
direct link between it and the various marketing and sales constructs. These outflow items reflect
many different constructs but most of the items are irrelevant to any particular construct (Enache
and Srivastava 2018). These items capture diverse firm activities, well beyond the functions of
sales and marketing. If categorized according to Porter’s value chain framework (Porter 1985),
two-thirds of the items relate to support activities, such as infrastructure and human resource
management. Only one-third of them pertain to primary activities, including marketing and sales
functions. Furthermore, only three items—advertising expenses, commissions, and marketing
expenses—directly relate to these functions (Standard & Poor’s 2013), and they account for only
a small proportion of SGA. For example, between 1997 and 2015, across all companies in
Compustat, aggregate advertising expenses accounted for less than 12% of SGA, whereas rental
expenses made up 13%, and R&D expenses accounted for 17%. Whereas the use of a composite
variable to measure a marketing construct implies that the estimated effects and resulting

6
strategies pertain to the relevant marketing items it contains, the composition of SGA suggests
that the effects instead could be related to one or more support activities required for operations.
Firms with similar SGA values could differ wildly in the size of various items. Thus, a detailed
analysis is needed to examine the validity of SGA for measuring marketing and sales constructs.
----------------Insert Table 2 about here ----------------
Table 2 summarizes the operationalizations of marketing and sales constructs based on
SGA, revealing both the constructs and the multiple measures employed to capture them.
Broadly, 11 major constructs have been operationalized using three key variables from
Compustat: SGA, advertising expense (ADV), and research and development expense (R&D).
This table also illustrates the arbitrary use of SGA. To take an example, SGA measures marketing
spending in several studies (Dutta et al. 1999, 2005; Narasimhan et al. 2006; Sarkees et al. 2014),
but a modification of this metric, “SGA – R&D” has been applied for the same purpose in several
other studies (Luo 2008; Dinner et al. 2009; Bharadwaj et al. 2011; Kurt and Hulland 2013). In
addition to inconsistency in the operationalization of a particular construct, multiple constructs
are often captured using the same operationalization. For example, in addition to marketing
spending, marketing assets (Balsam et al. 2011), marketing intensity (Krishnan et al. 2009),
marketing efficiency (Lin et al. 2014), and marketing capabilities (Luo et al. 2005) have been
measured using SGA too. Yet these constructs are clearly distinct from one another, so SGA
cannot possibly serve as a valid measure for all of them. This arbitrary use of SGA has led to
multiple operationalizations of a single construct and similar operationalizations of multiple
constructs. In each case, the operationalization may not sufficiently match the construct.1
1 Sometimes, use of SGA has been justified by intuitive reasoning. For example, because SGA budgets may be
interpreted as a sign of financial resources of a firm, SGA appears to be a good proxy of marketing resources. Such
operationalization suffers from lack of proper validation and can be hit or miss. Intuitively, there may be equally
good or better proxies available within Compustat. For example, marketing resources which imply items such as
cash, customer loyalty, brand equity, and patents could be measured using more direct and conceptually relevant
measures such as “goodwill” or “total intangible assets”. One could even employ “working capital” or “cash and

7
In Figure 2, we bring together marketing and sales constructs and accounting variables.
The figure depicts how cash outflows are treated as per accounting standards in Compustat, and
the various marketing constructs that have been measured using SGA. Accounting differs
markedly from marketing in its treatment of cash outflows. Marketing usually treats them as
generic, but accounting has a set of specific rules based primarily on the timing of returns from
outflows (Hansen 1990). Cash outflows that do not generate future economic returns are treated
as expenses in income statements; those that generate future economic returns are capitalized as
assets in the balance sheet and depreciate over time. Expenses also can be divided further into
broad subcategories, such as the cost of goods sold (COGS), SGA, and other expenses. Similarly,
assets comprise two broad subcategories, tangible and intangible.
----------------Insert Figure 2 about here ----------------
On the basis of their conceptual properties, we categorize the marketing constructs in
Figure 2 as either accounting or operating in nature, which ideally would be captured with
accounting or operating measures, respectively. Accounting measures are “reflections of past or
short-term financial” (Gentry and Shen 2010, p. 514) activity that “rely upon financial
information reported in income statement, balance sheet and statements of cash flow” (Carton
and Hofer 2006, p. 61). They are “generally expressed as values, ratios or percentages” (Carton
and Hofer 2006, p. 63). Constructs that are shorter-term, relatively more objective, and primarily
concerned with financial activity, such as marketing spending, are conducive to such measures.
Operating measures instead “represent how the organization is performing on non-financial
issues.… Most of the measures in this category require primary data from management in the
form of their assessment of own performance” (Carton and Hofer 2006, p. 62). They do not
short-term investments” or “cash,” which are conceptually aligned to, and better capture, the resources a firm has
available to cover its expenses. Of course, to choose the right operationalization one needs to establish content and
construct validity which we propose later.

8
appear in the income statement, balance sheet, or cash flow statement. Constructs such as
marketing capabilities, which are longer-term, relatively more subjective, and concerned with
non-financial performance, are more appropriate for such measures (Moorman and Day 2016).
This categorization provides a basis for relating the constructs to Compustat metrics and
assessing their conceptual validity. Definitions of all constructs appear in table 2.
Research Design
To be valid, a measure should assess “the magnitude and direction of (1) all of the
characteristics and (2) only the characteristics of the construct it is purported to assess” (Peter
1981, p. 134). Simply put, “a measure is valid if it measures what it is supposed to measure”
(Heeler and Ray 1972, p. 361). We analyze the appropriateness and validity of SGA for each
construct using a two-step approach for establishing content and construct validity (Table 1).
Content validity pertains to the conceptual adequacy of the proposed measure for capturing the
construct’s domain characteristics (DeVellis 2012). We test the content validity of the baseline
constructs (spending, assets, resources, and capabilities) with respect to SGA by deriving a set of
decision criteria. Adequate fit between SGA and each construct, according to these decision
criteria, is a necessary condition for validation. If content validity exists, we move on to further
testing for construct validity at the operational level. Construct validity is “the vertical
correspondence between a construct, which is at an unobservable conceptual level, and a
purported measure of it, which is at an operational level” (Peter 1981, p. 134). The tests for
construct validity use the multitrait-multimethod (MTMM) approach. We test SGA against a set
of benchmark variables that are relatively purer and obtained from other data sources (e.g.,
Advertising Age, Selling Power, and balance sheet information in Compustat): measured media
spending, estimated unmeasured spending, number of salespeople, goodwill, and other intangible
assets.

9
For our study, the differences between a concept, construct, and variable are critical (see
Web Appendix 3). A concept is “a bundle of meanings or characteristics associated with certain
events, objects, conditions, situations” (Emory and Cooper 1991, p. 51). Constructs combine two
or more simple concepts, especially if the idea “to convey is not directly subject to observation”
(Emory and Cooper 1991, p. 51). Concepts and constructs operate at the theoretical level;
variables, on the other hand, operate at an empirical level. A variable “is a symbol to which
numerals or values are assigned” (Kerlinger 1986, p. 27 cf. Emory and Cooper 1991). Multiple
labels sometimes are used across different contexts to refer to the same entity though. For
example, when referred to as a construct, SGA conveys a broader sense of operating expenses
measured by several manifest variables. When referred to as a variable, it represents the measure
within Compustat, manifest in nature and applied to approximate, either partly or fully, one or
more constructs.
Testing for Content Validity
To start, a “clear and concise conceptual definition of the focal construct” (MacKenzie
2003, p. 323) is required to capture the characteristics of its domain. A set of decision criteria can
specify the nature of a construct and demarcate it from other, related constructs. In line with
academics’ call for rigor and relevance (Kumar 2016), we suggest five decision criteria to
determine each construct in terms of its theoretical and managerial aspects. These criteria
encompass three dimensions of a construct – conceptual, operational, and managerial. Two
criteria capture a construct’s conceptual properties in terms of the domain of its definition and
level of its abstraction. Two other criteria define a construct’s operational or measurement
requirements according to the time horizon and level of objectivity or subjectivity. The last
criterion places the construct in the overall managerial context reflecting its business focus. We
consider fit on the two criteria capturing a construct’s conceptual properties to be necessary for

10
content validity; fit on any one of the three dimensions of a construct though is not sufficient by
itself to establish content validity.
In our framework, the domains of the constructs’ definitions enable us to categorize them
as either accounting or operating. As we noted earlier, constructs that are shorter-term, relatively
more objective, and primarily concerned with financial performance (e.g., marketing spending)
are accounting in nature, whereas those that are longer-term, relatively more subjective, and
concerned with non-financial performance (e.g., marketing capabilities) are operating in nature.
The level of abstraction of a construct denotes the divergence between its conceptual and
operational scope and influences the ease with which it can be measured (Nunnally 1978;
Viswanathan 2005). Constructs vary from low abstraction (simple to measure, e.g., advertising
spending) to high abstraction (difficult to measure, e.g., marketing capability). Time horizon is
the degree to which a construct is attributable to a specific operating period (Katsikeas et al.
2016). For example, marketing spending is short-term, but marketing assets, which generate
future economic value beyond a particular period, are long-term. The level of objectivity classifies
the construct at an operational level according to the type of measures needed, that is, manifest or
latent (Katsikeas et al. 2016). Constructs such as marketing capabilities include high proportions
of subjective judgment, so they have relatively low objectivity; their measurement depends
largely on qualitative assessments. Constructs such as marketing spending, which primarily
depend on the level of expenses, instead have high objectivity. Finally, the business focus of a
construct determines whether it is strategic (broader and abstract in scope; longer timeframe;
higher involvement of senior management) or tactical (narrow and specific in scope; shorter
timeframe; lower involvement of senior management) (Shapiro 1989; Brink et al. 2006;
Casadesus-Masanell and Ricart 2010). Marketing spending might be considered tactical, because
it aims to achieve specific, short-term subgoals that contribute to the ultimate business goal (e.g.,

11
firm performance). Marketing capabilities instead would be more strategic in nature. With these
five decision criteria, we define and demarcate the constructs, according to both research and
practice perspectives. Only when an adequate fit on these criteria establish content validity do we
proceed with the process of establishing construct validity. We consider fit on the two necessary
criteria capturing conceptual properties alone, or with additional criteria, adequate for content
validity. However, if the two necessary criteria are not met then content validity is not established
and there is no justification to conduct a construct validation exercise (see figure 3). This
‘decision rule’ is not confined to this research domain alone; rather, this rule should apply to
testing the validity of any metric – whether in marketing or in other disciplines.
----------------Insert Figure 3 about here ----------------
Testing for Construct Validity
We test whether an operationalization corresponds to the underlying construct it aims to
measure. Construct validity consists of convergent and discriminant validity. Convergent validity
indicates the degree to which different measures of the same construct are in agreement whereas
discriminant validity indicates the degree to which measures of different constructs are distinct
(Bagozzi 1994). We assess construct validity using the MTMM matrix (Campbell and Fiske
1959; Churchill 1979; Bagozzi 1994). The MTMM matrix offers a “framework for developing
measure validation from available or easily obtainable generated data” (Heeler and Ray 1972, p.
363), relying on the analysis of correlations among several variables measured by different
techniques. Thus, alternative operationalizations can be compared to see how well they measure
the same construct (e.g., SGA from Compustat vs. a benchmark metric obtained from an
alternative source) (Table 3). The alternative data source should provide relatively purer and less
biased information about the construct of interest.
----------------Insert Table 3 about here----------------

12
The main diagonal of the MTMM matrix (labeled I in Table 3) consists of the reliability
correlations, derived from the correlation of a measure of trait (construct) with itself in a test–
retest situation. In our study context, this diagonal consistently takes a value of 1, because all the
data were obtained from secondary sources that are subject to consistent, regulated accounting
data reporting standards (Carton and Hofer 2006).
For construct validity, the MTMM method includes several requirements. Specifically,
convergent validity requires that the entries in the validity or, monotrait-heteromethod (measures
of the same trait obtained by different methods) diagonal (labeled III in Table 3) are significantly
different from zero and sufficiently large. Discriminant validity is demonstrated by the
divergence of the measure of interest from other measures not “measuring the same variable or
concept” (Heeler and Ray 1972, p. 362). For this consideration, the MTMM approach uses three
criteria. First, correlations in each cell of diagonal III should be greater than the correlations in its
column and row in the heterotrait-heteromethod (measures for different traits obtained by
different methods) cells (labeled IV in Table 3). This minimum requirement simply means that
the correlation between two different measures of the same variable “should be higher than the
correlations obtained between that variable and any other variable having neither trait nor method
in common” (Bagozzi 1994, p. 22). Second, the correlations in diagonal III should be greater than
those in the heterotrait-monomethod (measures for different traits obtained by the same method)
cells (labeled II in Table 3). This more stringent requirement suggests that the correlations of
different measures of a trait should be greater than correlations among traits that have methods in
common. That is, “a variable correlates higher with an independent effort to measure the same
trait than with measures designed to get at different traits which happen to employ the same
method” (Bagozzi 1994, p. 22). Third, if the matrix contains information on more than two traits,

13
the same pattern of trait interrelationship should appear in all heterotrait triangles, for both the
monomethod and the heteromethod blocks.
Data
Data Sources
We obtained data from three sources: Compustat, Advertising Age, and Selling Power.
Compustat provides data for companies publicly listed in the United States or Canada; the
“Compustat North America Fundamentals Annual” data set comprises annual, worldwide,
company-level information on expenses such as SGA, advertising, and R&D, as well as on assets
such as goodwill and intangible assets. We obtained 19 years of data (1997–2015) from
Compustat. To ensure proper application of the validation approach, we excluded all observations
with zero or missing values for our key variables of interest. It is unlikely that any company has
zero annual expenses on SGA and advertising expenses; a zero value likely implies that either the
company did not disclose the value or Compustat failed to register it. Compustat reports a
missing value (blank cell) if it is unable to obtain a value (Standard and Poor’s, personal
correspondence).
Advertising Age and Selling Power provide benchmark data to judge the validity of the
SGA-based metrics. 2 Advertising Age provides annual, company-level data on the marketing
expenses of 200 leading companies in the U.S. and 100 leading companies worldwide. Selling
Power tracks the 500 U.S.-based companies that employ the largest sales forces. It provides
2 We also considered other data sources (e.g., Ebiquity, PIMS, Hoover) of benchmark variables but found them
unsuitable. For example, Ebiquity reports data at the country level only, and its consultants advised us against
aggregating these country-level data to obtain worldwide data. PIMS provides information at the strategic business
unit level for participating companies, so it likewise is unsuitable. Hoover does not include any information related to
marketing spending but rather provides qualitative information about big players only.

14
annual, company-level information on the number of salespeople in the United States. These two
sources thus offer purer and less biased benchmark information on the variables of interest. 3
For construct validation, we needed to match data across the different sources. We started
with 20,365 observations from Compustat and 1,900 observations from Advertising Age (100
observations per year for 1997–2015). More than half of the companies listed in Advertising Age
(worldwide data set) are not listed in the U.S. or Canada and thus not included in Compustat,
even though they advertise in these countries. Due to missing or zero values on focal variables in
Compustat, matching the data from these two sources left us with 506 observations. After
removing extreme outliers, we retained 499 observations, which constitute Sample 1.4,5 Almost
two-thirds of the observations are from companies earning their revenue predominantly from
B2C market, 5 percent of observations are from predominantly B2B firms while the remaining
are from firms that cater significantly to both markets. It represents 73 unique companies all of
which spend heavily on marketing communication (a key criterion for their inclusion in the
Advertising Age database). The data range from one to nineteen years for individual companies,
with an average of about seven years for each company. In this sample of active advertisers with
high spending, advertising expenses account for about 23% of SGA.
3 We empirically validated the benchmark measures from these alternative sources by collecting data from annual
reports of public and private companies. We thank an anonymous reviewer for this suggestion. We note here that
these benchmark measures provide purer information on the three focal variables only – advertising expense,
promotional expense, and salesforce expense. Whether these measures are also better than SGA at capturing any
particular marketing construct depends on both content and construct validity.
4 Outliers can have significant influences on correlation coefficients, so extreme outliers should be removed
(Schwertman et al. 2004). We used Tukey’s (1977) formula: lower fence: Quartile 1 – 3*(Quartile 3 – Quartile 1);
upper fence: Quartile 3 + 3*(Quartile 3 – Quartile 1). All values outside the fences were removed, which reduced the
number of observations to 499. As we explain with our robustness checks, including these extreme outliers still
provided similar results.
5 There could be a potential sample selection bias as certain firms/industries may be overly represented in Advertising
Age than in Compustat. We conducted propensity score matching to check if the smaller sample size used in the
empirical analysis is representative of the broader sample drawn from Compustat. The results present no evidence of
sample selection bias. The details of the matching procedure are available in web appendix 4.We thank an
anonymous reviewer for this suggestion.

15
Next, we matched the data from Compustat with data from Selling Power to obtain
Sample 2. We started with 3,500 observations (500 observations per year for 2009–2015) from
Selling Power. When matched with 7,539 observations from Compustat for this time period, and
after excluding outliers and observations with missing or zero values, we were left with 409
observations, which constituted Sample 2. Almost 43 percent of the observations are from
companies earning their revenue predominantly from B2B markets. The rest are divided almost
evenly between predominantly B2C firms and firms that cater significantly to both markets.
Interestingly, a large number of B2C firms are in the pharmaceutical industry that is known to
employ large salesforces to target physicians. It represents 86 unique companies with the largest
sales forces (the key criterion for their inclusion in the Selling Power database). These data range
over time periods from one to seven years for individual companies, with an average of about five
years for each company.
Variables
The set of variables from Compustat used for construct operationalization includes
selling, general, and administrative expenses (SGA), advertising expenses (ADV), and research
and development expenses (R&D). These variables are the most frequently employed in
marketing literature, so they represent variables of interest in terms of construct validation. We
test them against the benchmark variables derived from Advertising Age, Selling Power, and
Compustat itself. The benchmark variables, as reliable alternative measures of specific constructs,
consist of measured media spending, estimated unmeasured spending, the number of people
employed in sales functions, total intangible assets, goodwill, and other intangible assets.
Variables and their data sources are listed in Table 4.
----------------Insert Table 4 about here----------------

16
Beyond the definitions in Table 4, a few additional notes are necessary in relation to
selected variables. Specifically, measured media spending spans 19 media channels and is
reported at both the worldwide level (100 companies every year) and the U.S. level (200
companies every year). A company must have “measured-media spending in at least three of the
four major regions—defined as the US and Canada; Asia Pacific; Europe, Middle East, and
Africa; and Latin America” to qualify for entry in the worldwide list (Advertising Age 2016a). In
addition, estimated unmeasured spending, or the estimate of spending on sources that are not
included in the measured media category (Advertising Age 2016b), is reported only for the U.S.
market (200 companies). To compare it against the global Compustat data, we needed to obtain a
worldwide measure of estimated unmeasured spending. For this we calculated the ratio of
measured media spending of 100 companies at the worldwide level to their measured media
spending in the United States. With the assumption that this ratio should hold for estimated
unmeasured spending too, we applied it to obtain worldwide estimated unmeasured spending
from the information available for the 100 U.S. companies. As we explain with our robustness
checks subsequently, we allowed for divergence of ±33% from these calculated values. Finally,
the information on the estimated number of salespeople refers to 500 U.S. companies (Selling
Power 2016). This variable is reported at the U.S. level only. To compare it with Compustat data
at the worldwide level, we referred to each company’s annual reports and other business
publications between 2009 and 2015 to get information on their total sales (in U.S. dollars)
worldwide and in the United States. We calculated this ratio, then multiplied the number of U.S.
salespeople with this number to impute the number of salespeople worldwide. Similar to
estimated unmeasured spending, we again allowed for a divergence of up to ±33% from these
calculated values.

17
The descriptive statistics for all the variables are in Table 5, Panels A (Sample 1) and B
(Sample 2).
----------------Insert Table 5 about here ----------------
Results
Our validation approach consists of both conceptual and empirical assessments.
Conceptual Assessment (Content Validity)
We apply the five decision criteria to identify constructs that are conceptually aligned
with SGA (Table 6). As a construct, SGA provides a period-defined expense and thus could be
categorized as accounting in its domain and short-term in nature. The ease of tracking the various
components of SGA indicates a low level of abstraction and a high level of objectivity. Moreover,
SGA is relatively tactical in business focus; its primary role is to support the firm’s overall
business activities.
The baseline construct spending thus is conceptually well-aligned with SGA, in that it
represents expenses and is composed of cash outflows on several items.6 However, SGA has only
moderate fit with assets. Tangible assets include property, plants, and equipment; intangible
assets refer to items such as customer loyalty, brand equity, and patents. Both types can have
tremendous impacts on firm performance. Although SGA and assets align on the necessary
decision criteria (domain of definition and level of abstraction), they exhibit less alignment on the
other three (time horizon, objectivity, and business focus). Nonetheless, we apply an empirical
analysis to validate SGA as a measure of spending and assets. Regarding the five benchmark
6 Marketing spending, as used in the study for validation of SGA as a measure, has two subconstructs – advertising
spending and promotional spending. Arguably, marketing spending on some activities such as advertising may
bestow relatively longer-term benefits compared with spending on other activities such as promotions. However,
considered in a comparative perspective, the spending construct is relatively short-term when compared with, say,
the assets construct. Also, marketing literature that has used SGA – a short-term accounting variable – to measure
spending has implicitly considered it short-term.

18
variables, similar to SGA, three of the benchmark variables (measured media spending, estimated
unmeasured spending, and number of salespeople) seem conceptually well-aligned with
spending. Therefore, we use these variables to check the construct validity of spending. Two
benchmark variables (goodwill and other intangible assets from balance sheet information in
Compustat) instead are conceptually well-aligned with assets and thus serve as the benchmark
variables for the construct validity assessment of assets.
----------------Insert Table 6 about here----------------
Resources and capabilities (as well as exploitation, a subconstruct of marketing
capability; Vorhies et al. 2011) are not aligned with SGA. They differ consistently on the
conceptual, operational, and managerial dimensions. Resources and capabilities address operating
performance whereas SGA is an accounting indicator. The greater intangibility of resources and
capabilities also demands qualitative and subjective judgments, or a high level of abstraction and
low level of objectivity. Resources and capabilities are strategic and develop over time, such that
they are longer-term in their time horizon. All the decision criteria thus reiterate the incongruence
of these constructs with SGA. Because the necessary condition for content validity is not
satisfied, we establish that SGA is an inadequate operationalization for resources and capabilities.
In stark and worrisome contrast, many studies have used SGA for this purpose.
In summary, SGA seems conceptually aligned with spending and assets (and thus with
efficiency and intensity), and it fulfills the necessary condition for content validation. However,
SGA comprises 29 items that cover a broad range of distinct activities, so we still need to test for
construct validity. Only 3 of the 29 items—ADV, commissions, and marketing expenses—relate

19
directly to selling and marketing cash outflows. Thus, we empirically examine the suitability of
SGA to measure these and other constructs next.7
Empirical Results
Construct Validity of Marketing Spending. In prior literature, spending on marketing
communication (often referred to simply as marketing spending) has been measured using
different variables available in Compustat, such as ADV, SGA, and its modifications (SGA –
ADV, SGA – R&D). This spending consists of two distinct subconstructs (or traits, in MTMM
nomenclature): advertising spending and promotional spending. We thus consider two different
scenarios for construct validation. In the first, we assume advertising spending is measured by
ADV whereas promotional spending is measured by SGA or one of its modifications. In the
second scenario, we switch them such that promotional spending is measured by ADV whereas
advertising spending is measured by SGA or one of its modifications. We test these measures
against two benchmarks from Advertising Age, measured media spending and estimated
unmeasured spending. On the basis of its composition, measured media spending clearly captures
advertising spending, whereas estimated unmeasured spending captures promotional spending.
We correlate these two benchmark measures with ADV and SGA (or one of its modifications) in
an MTMM format, which yields 4 MTMM matrices in each scenario.8 In all these matrices, the
Compustat data represent method 1 for obtaining data, and the Advertising Age data represents
7 We note the difference between marketing and sales functions, which are often organized and executed in different
organizational departments and treated differently. Marketing involves activities to start and maintain a customer
relationship (van Triest et al. 2009), such as advertising and promotional efforts, which generate customer awareness
and establish brand preference. Sales seeks to stimulate actual purchases through sales force activities such as
negotiations over price and delivery (Kotler and Rackham 2006).
8 In addition to the two common modifications of SGA (SGA – ADV, SGA – R&D), we test another modification
(SGA – ADV – R&D) to check if SGA has any significant marketing-related component, beyond ADV and R&D,
which may justify its use as a measure of marketing constructs. Thus, Scenario 1 includes four MTMM matrices:
advertising spending measured using ADV whereas promotional spending measured using SGA, SGA – ADV, SGA
– R&D, or SGA – ADV – R&D, respectively. Scenario 2 also uses four matrices, with promotional spending
measured as ADV whereas advertising spending measured using each of the four SGA-based metrics.

20
method 2. The results for the first MTMM matrix (ADV measures advertising spending whereas
SGA measures promotional spending) are in Table 7, Panel A.
For convergent validity, coefficients in the validity diagonal should be significantly
different from zero and high enough to warrant further investigation. In MTMM 1, although both
coefficients are statistically significant, the coefficient for trait 1, measured using ADV (.74), is
much higher than the one for trait 2, measured using SGA (.34). For discriminant validity, a
validity coefficient should be higher than the values in its column and row in the heterotrait-
heteromethod cells. For example, the correlation between ADV and measured media spending
should be higher than the correlations between ADV and estimated unmeasured spending or SGA
and measured media spending (which have neither traits nor methods in common). This condition
is fulfilled for trait 1 measured using ADV (.74 > .69; .74 > .38) but not for trait 2 measured using
SGA (.34 < .38; .34 < .69). Furthermore, the validity coefficient should be higher than all
coefficients in the heterotrait-monomethod cells. For example, the correlation between ADV and
measured media spending should be higher than the correlations between measured media
spending and estimated unmeasured spending or ADV and SGA. This condition is again fulfilled
only for trait 1 measured using ADV (.74 > .64; .74 > .52) and not for trait 2 measured using
SGA (.34 < .52; .34 < .64). Overall, the results suggest that only ADV fulfills the conditions of
convergent and discriminant validity for measuring advertising spending; SGA does not fulfill
these conditions for measuring promotional spending. The similar MTMM matrices for the
modifications of SGA (i.e., SGA – ADV, SGA – R&D, SGA – ADV – R&D) provide similar
results (see Table 7, Panel B for results of all four matrices 1–4). That is, none of the SGA-based
measures fulfill conditions of construct validity to measure promotional spending.
In the second scenario, we switched the measures so that ADV measures promotional
spending whereas SGA measures advertising spending. Neither ADV nor SGA, or any of its

21
modifications, fulfills the conditions this time. Thus, ADV offers a good measure of advertising
spending and a partial measure of total marketing spending, but SGA fails to capture marketing
spending or any of its subconstructs. The conceptual relationship of spending with intensity and
efficiency allows us to extrapolate the results for marketing communication spending to
marketing intensity and efficiency too.
----------------Insert Table 7 about here----------------
Construct Validity of Marketing Assets. In line with our adopted definition of a
marketing asset (i.e., as noted previously, a “customer-focused measure of the value of the firm
(and its offerings) that may enhance the firm’s long-term value”; Rust et al. 2004, p. 78),
marketing usually focuses on intangible forms, such as customer relationships, brand equity, and
patents. We therefore subsume marketing investments under assets. Following accounting
standards, assets are recorded on the balance sheet, but commonly used measures of investments
or assets, such as ADV and SGA and its modifications (SGA – ADV, SGA – R&D), appear in
the income statement. We thus validate the measures from the income statement against two
entries from the balance sheet that capture intangible assets: goodwill and other intangible assets.
For validation purposes, the two subconstructs of assets are perceptual assets, such as
customer relationships and brand equity, and intellectual assets, such as property rights, including
“patents, trademarks, registered designs and copyrights” (Kristandl and Bontis 2007, p. 1519).
Similar to our tests of the validity of marketing spending measures, we consider two scenarios. In
the first, we assume perceptual assets are measured by ADV whereas intellectual assets are
measured by SGA or one of its modifications. In the second, we switch them, such that
intellectual assets are measured by ADV whereas perceptual assets are measured by SGA or one
of its modifications. We test these measures against goodwill and other intangible assets, as
reported in the balance sheet. Goodwill captures perceptual assets well; other intangible assets

22
capture intellectual assets. We correlate these two benchmark measures with ADV and SGA (or
one of its modifications) in an MTMM format, yielding a total of three MTMM matrices for each
scenario.9 In all these matrices, the income statement is designated method 1 for obtaining data,
and the balance sheet is method 2. The results of the first MTMM matrix (ADV measuring
perceptual assets, SGA measuring intellectual assets) are in Table 8, Panel A. Then in Panel B,
we report the results for all three matrices (5–7) in scenario 1. Convergent and discriminant
validity analyses indicate that neither ADV nor SGA-based measures from the income statement
are valid measures of the two subconstructs of marketing assets.
----------------Insert Table 8 about here----------------
Construct Validity of Sales Force Spending. Sources of data on sales force spending
usually do not split this construct into multiple traits, which makes it difficult to apply an MTMM
approach (which needs a minimum of two traits from each data collection method) to validate this
construct. We rely instead on bivariate correlations, which “describe the degree of relationship
between two variables” (Nunnally 1978, p. 121). Correlation of the number of salespeople with
SGA (0.67) is positive and statistically significant (see Table 5, Panel B). This correlation stays
significant when we exclude ADV and R&D from SGA; in fact the correlation increases when we
exclude ADV from SGA (0.69). Thus SGA, and especially its modification SGA – ADV, seems
to represent sales force spending relatively well.
Table 9 provides a summary of all the constructs, their operationalizations, benchmark
variables used for construct validation, and empirical tests.
----------------Insert Table 9 about here----------------
9For the three MTMM matrices in Scenario 1, perceptual assets are measured using ADV in each case whereas
intellectual assets are measured using SGA, SGA – ADV, or SGA – R&D. Scenario 2 also includes three matrices in
which intellectual assets are always measured using ADV whereas perceptual assets use the three SGA-based
metrics.

23
Robustness Checks
We conducted several checks to test the robustness of our results. First, the MTMM
methodology relies on arithmetic differences in the magnitudes of the correlation coefficients.
One might question the statistical significance of these differences. Using a method proposed by
Steiger (1980), we thus test for the statistical equality or inequality of correlation coefficients. To
check equality, we considered pairs of correlation coefficients in which two pairs share one
variable in common (Steiger 1980). These correlation coefficients were converted into z-scores,
using Fisher's r-to-z transformation, which we applied to compute the asymptotic covariance of
the estimates. These quantities were then used in an asymptotic z-test. The results for marketing
spending from Sample 1 indicate that ADV and SGA are not equally correlated with measured
media spending (z = 11.31, p < .01) or estimated unmeasured spending (z = 9.91, p < .01). In
addition, ADV and the various modifications of SGA were not equally correlated with measured
media spending or estimated unmeasured spending. Considering their pairwise correlation
coefficients, ADV appears to be an appropriate measure for marketing spending, but SGA and its
modifications are not. The results for sales force spending from Sample 2 further indicate that
ADV and SGA are not equally correlated with salespeople (z = 10.91, p < .01); ADV and the
various modifications of SGA are not equally correlated with salespeople either. The pairwise
correlation coefficients suggest that SGA – ADV represents sales force expenses well.
Second, we had removed extreme outliers from our samples (i.e., values above or below
three times the interquartile range; Dattero et al. 1991). To check whether retaining the outliers
would have led to different conclusions, we re-estimated all the MTMM matrices with the full
data set. The results remained substantively similar. Another argument suggests that even
moderate outliers might bias the conclusions, so we also re-estimated the matrices after removing

24
the moderate outliers (i.e., 1.5 times the interquartile range). The results again were substantively
similar.
Third, differences in companies’ performance might influence how well the metrics from
Compustat reflect various constructs. Thus, we performed several median splits of our data set,
according to high and low values of the ratios of various variables of interest: SGA to sales, ADV
to sales, R&D to sales, goodwill to sales, other intangibles to sales, and assets to sales. The results
across both high and low groups for all these splits remain substantively similar to those based on
the entire data set and strongly support our initial MTMM findings (see Web appendix 5).
Fourth, our data did not provide worldwide values for estimated unmeasured spending or
number of salespeople, so we had to impute these values, and the imputations might not capture
the true values. To check the robustness of these results, as we noted previously, we allowed for a
divergence of up to ±33% of the calculated values. For both variables, we generated three
additional series, at 20%, 25%, and 33% divergence levels. For example, for estimated
unmeasured spending, we allowed the imputed values to vary randomly in either direction by
20%, which produced the first series. Then we used this series in our analysis, to determine if the
results changed significantly. We repeated this exercise for 25% and 33% for both variables. The
results were substantively similar.
Fifth, in addition to our validity analysis, we considered the reasoning used in prior
studies to justify the use of SGA and its modifications to measure marketing constructs. A high
correlation between ADV and SGA is the most common justification, yet without appropriate
conceptual and empirical assessment, this reasoning is not based on sound logic. Web appendix 6
provides an overview of correlations between SGA and some of its components, available
separately in the income statement. This comparison shows that SGA is highly correlated not
only with ADV (.70) but also with other expenses, such as R&D (.65), rental expenses (.74), and

25
pension and retirement expenses (.66). Even if these components were removed from SGA, the
remainder still correlates highly with these components. It even is highly correlated with
unrelated variables reported in the income statement; for example, the correlation between SGA
and the cost of goods sold (COGS), which provides information about a company’s expenses for
producing goods and services, is .80. Going solely by the size of the correlations, if SGA is an
appropriate operationalization for advertising spending, it would be an even better
operationalization of COGS. The two have little conceptual overlap though. Thus, SGA cannot be
considered an adequate proxy for every item represented by its 29 components. Conceptual
validity is necessary to establish before correlation should even be considered.
Discussion
A broad literature review of marketing and management journals reveals that SGA from
Compustat has been used to operationalize several marketing- and sales-related constructs. This
widespread, inconsistent use of SGA points to potential problems related to an inadequate
conceptualization and operationalization. With a measurement validation approach, we seek to
assess the level of congruence between the constructs and measures, using data from Compustat,
Advertising Age, and Selling Power.
Although a conscientious conceptualization is a prerequisite of construct validation,
research studies that rely on SGA frequently overlook this crucial step. Such gaps arise in other
areas of research too; for example, nine out of ten studies of marketing performance fail to
provide clear conceptual definitions before attempting their operationalizations (Katsikeas et al.
2016). Operationalization without proper conceptualization, or without proper empirical
validation, can result in over- or underestimation of the effects of focal constructs. The
inconsistent use of SGA across multiple constructs also challenges the validity of their estimated
effect sizes. Identical operationalizations of different constructs imply that the attribution of

26
estimated effects to specific constructs may be erroneous and lead to inaccurate managerial
implications that hinder decision-making effectiveness. For example, an erroneous allocation of
budget to marketing and sales activities could hinder the effective use of various marketing and
sales levers to improve firm performance.
Our empirical analysis shows that SGA is inadequate for a number of constructs that it is
commonly used to operationalize. Although a focal construct, marketing spending, is
conceptually aligned with SGA, our empirical results show that SGA and its modifications are
not valid operationalizations of marketing spending or its subconstructs. Marketing-related cash
outflows are only a small component of SGA. Thus, studies using SGA to measure marketing
communication spending or its subconstructs might have inferred incorrect influences of these
expenditures. Our results suggest that ADV from Compustat, which is equally easily available, is
a satisfactory measure of advertising spending and at least a partial measure of total marketing
spending. Furthermore, SGA is ill-suited to measure complex constructs such as marketing
capabilities, which instead require multidimensional, latent variable approaches to capture the
transformation of cash outflows into competitive advantages.
Regarding marketing assets, our conceptual and empirical results indicate that neither
ADV nor SGA (or any of its modifications) is satisfactory. Goodwill and other intangible assets,
two variables equally easily available from Compustat, are better measures. For sales force
spending, the results provide evidence of a strong overlap between the benchmark measure,
number of sales force employees, and SGA-based metrics, especially SGA – ADV. Therefore,
SGA appears valid for measuring sales force spending, in line with the general nature of selling,
general, and administrative cash outflows. The proportion of sales expenses, in terms of
commissions and salaries, constitutes a large component of SGA. Beyond validation, the results
affirm the expected distinction between marketing and sales constructs. Sales force spending does

27
not have a significant overlap with advertising or promotional spending, which are key
components of marketing communication spending. Thus, SGA is not an appropriate
operationalization for marketing and sales at the same time. We summarize the construct and
measure fits in Figure 4.
----------------Insert Figure 4 about here----------------
Guidelines for Using SGA
From our theoretical and empirical analysis, we derive guidelines for researchers
interested in using SGA to operationalize marketing and sales constructs. These guidelines can
help build coherent knowledge about the conceptualization of constructs in general and their
operationalization using SGA in particular.
Ascertain Conceptual Congruence between Construct and Measure. Our review of
marketing and management literature reveals frequent subpar construct definitions. Studies often
fail to define or delineate constructs before operationalizing them, often based solely on cross-
references or contextual examples. The use of ambiguous definitions (for example, defining a
construct as a consequence or cause of other concepts and constructs) or pseudo-definitions (i.e.,
specifying a construct merely with an enumeration of examples) can lead to misspecifications
(MacKenzie 2003). Imprecise or insufficient specification of the construct domain and content
also may lead to their over- or underestimation, causing potential errors in the effect estimates
due to incongruence between the construct and its measure. This problem also makes the results
incomparable across studies and inhibits their synthesis, which is critical for cumulative
knowledge building (Katsikeas et al. 2016). Both the complexity of a construct and the required
adequacy of the measure to fit that complexity should be taken into account and be reflected in
the measurement variable. Any dissonance can severely bias the estimation results and their
inferences. Researchers thus would do well to derive precise definitions, embedding their focal

28
constructs into a broader (organizational) context. Then they can develop evaluative frameworks
to assess the validation of constructs on conceptual and operational levels. Such frameworks help
reveal which facets of a construct should be considered when choosing variables for its
operationalization in empirical research.
Avoid Using SGA as an All-Encompassing Measure and Test for Construct Validity.
Many of the 29 cash outflow items that occur over the regular course of business and constitute
SGA have little direct link to marketing functions. At a conceptual level, using SGA as a measure
of a construct reduces the multifaceted variable to one component; at an operational level though,
it necessarily remains an aggregate of 29 disparate items. This clear discrepancy somehow takes a
backseat when researchers use SGA or any of its modifications as an all-encompassing measure
for so many distinct constructs. Still, our results suggest that SGA can be adapted to match some
constructs relatively well, by removing certain outflow items such as ADV and R&D. The key is
to remove unrelated cash outflow items to increase the variance explained and reduce estimation
errors related to the focal construct. Even in this case, SGA and its modifications should be tested
for validity with respect to a benchmark variable before being used to operationalize a construct.
The benchmark variable can be obtained from a distinct data source that provides relatively purer
and unbiased information, sometimes even from Compustat itself. For example, a benchmark
variable that measures marketing assets already is available in the balance sheet.
Avoid Justifications Based on Data Unavailability by Considering Alternative
Sources. Compustat in general and SGA in particular are popular sources, because of their clear
advantages: easy availability and cross-industry, firm-specific data across several time periods.
However, scholars cannot ignore their limitations. The variables are too broad to provide precise
measures, so they introduce measurement error, potential model misspecification, and biased
estimates. To suggest SGA is adequate for construct operationalization solely because valid

29
measures are not available is not appropriate or accordant with a measurement philosophy that
seeks to reduce errors and obtain precise estimates. Following precedents of inadequate
operationalizations in existing research simply passes on the measurement biases from one study
to the next. Instead, researchers should either redefine the construct, to bring it more in line with
available measures, or obtain an adequate measure from other data sources that provide less noisy
variables and better capture the focal construct. Either approach is preferable to forcing an
inadequate variable on a construct with which it is not sufficiently aligned. Admittedly, these
approaches may reduce sample sizes; compared with Compustat, the alternative sources such as
Advertising Age and Selling Power are limited in their coverage. However, their measures can
explain more of the variance of the focal construct, which leads to more precise measurements.
Overall, we believe that SGA has been utilized too liberally in marketing. Of course, researchers
have to trade-off the generalizability of findings from publicly available data against the precision
and specificity of findings from private data, based on their research goals. As we show though,
for several marketing-related constructs, more valid measures may be available within
Compustat.
Following these guidelines can help improve measurement validity on both conceptual
and operational levels. Current literature is characterized by different operationalizations for the
same construct, as well as the same operationalization for different constructs. Our proposed
guidelines may help researchers determine the appropriateness of measures for underlying
constructs, which would improve conceptual completeness, operational consistency, estimations
of true effect sizes, and comparison and replication of results. Overall, this study is a first step
toward establishing common knowledge about the use of accounting-based variables in marketing
research.

30
Considering the critical importance of marketing and sales force–related decisions, this
study has implications for managers too. Marketing spending is a small component of SGA, so
decisions based on its use as a measure might lead to inappropriate marketing strategies and
misdirected budget allocations. The use of proper measures will provide true effect sizes and help
assess crucial performance indicators that provide a basis for strategic decisions. By using proper
measures, managers can better allocate their budgets and justify their decisions. They also gain a
reliable approach for benchmarking their performance, according to appropriately aligned
measures.
Limitations and Further Research
Although this research contributes to an enhanced understanding of the use of SGA-based
metrics to measure marketing and sales constructs, our empirical analysis has a few limitations
that suggest avenues for further study. First, our data come from multiple industries, but we did
not consider potential industry-specific differences. Compustat reveals some differences in the
composition of items included in SGA for specific industries. Continued research could explore
these differences, in terms of the construct validity across industries. Studies that classify
operating constructs using industry-specific characteristics would also enrich fundamental
marketing knowledge. Second, our study highlights several performance-related constructs, such
as capabilities and marketing exploitation that remain under-researched and insufficiently
defined, in terms of their conceptualization and operationalization. We confined our study to
baseline constructs and their accounting-based measures, but further research should define more
complex constructs and derive valid operationalizations for them too. Third, it would be
interesting to study if a certain portion of R&D spending could be considered as contributing to
brand building especially in industries such as technology and healthcare. Empirically, research
in this domain has taken an all-or-none approach to R&D – either using SGA as is or removing

31
R&D from SGA. Future research may attempt to arrive at, say, a proportion of R&D expense that
could be considered as related to marketing when measuring marketing constructs. Fourth, recent
work has looked at obtaining required measures from SGA by employing the relationships of its
components with some aggregate measures such as revenues and then apportioning SGA on that
basis (Enache and Srivastava 2018). Future research may consider replicating such an approach
in a marketing context. Fifth, we validated measured media spending and number of salespeople
as benchmark variables based on actual data from annual reports of companies but were not able
to do so for estimated unmeasured spending due to data unavailability. Future research could
consider collecting data on promotional spending directly from firms to validate this benchmark
variable from Advertising Age. Sixth, results indicate that sales force spending is better captured
by SGA – ADV compared with SGA – ADV – R&D. This could imply heavier representation of
firms that may have relatively lower R&D expense. Also, the use of Advertising Age data may
make the findings more relevant for B2C settings. Seventh, we relied on an MTMM approach for
our empirical validation. This approach has some limitations though such as absence of clear
standards to determine when a particular criterion has been met. Future research may consider
other alternative techniques. Finally, the common use of accounting data sources by marketing
researchers suggests the need to build more knowledge at the interface of these two domains.
Variables from accounting need to be linked clearly with marketing constructs. For example,
coordination spending is a manifest construct applied in marketing, but it is not consistently
derived from Compustat. Additional research might build on our approach to establish guidelines
for establishing strong reasoning to support such constructs and improve the consistency of their
measurement. Relatedly, scholars that have used SGA as a measure in the past should replicate
their studies with benchmark or alternative measures. Besides helping clarify any mixed results or

32
bringing expected theoretical relationships of interest to surface, doing so could provide another
means of empirical validation.

33
References
Advertising Age (2016a). About global marketers 2015. Retrieved August 18, 2016, from
http://adage.com/datacenter/globalmarketers2015/
Advertising Age (2016b). Methodology for 200 leading national advertisers, 2016 ed. Retrieved
July 6, 2016, from http://adage.com/article/datacenter/methodology-200-leading-national-
advertisers-2016-ed/304581/
Ambler, T., Kokkinaki, F., Puntoni, S., & Riley, D. (2001). Assessing market performance: the
current state of metrics. Working paper, London Business School, Centre of Marketing.
Bagozzi, R. P. (1994). Measurement in marketing research: basic principles of questionnaire
design, in Principles of marketing research, ed. Bagozzi, R. P., Cambridge, MA: Blackwell,
1–49.
Balsam, S., Fernando, G.D., & Tripathy, A. (2011). The impact of firm strategy on performance
measures used in executive compensation. Journal of Business Research, 64, 187–193.
Barney, J. B. (1991). Firm resources and sustained competitive advantage. Journal of
Management, 17, 99–120.
Barney, J. B., & Arikan, A. (2001). The resource-based view: origins and implications. In M.
Hitt, R. Freeman, & J. Harrison (Eds.), Handbook of strategic management (pp. 124–185).
Oxford, UK: Blackwell.
Bharadwaj, S. G., Tuli, K. R., & Bonfrer, A. (2011). The impact of brand quality on shareholder
wealth. Journal of Marketing, 75, 88–104.
Bragg, S. M. (2010). Cost reduction analysis: tools and strategies. New York: Wiley.
Brink, D. V. D., Odekerken-Schröder, G., & Pauwels, P. (2006). The effect of strategic and
tactical cause-related marketing on consumers’ brand loyalty. Journal of Consumer
Marketing, 23, 15–25.
Campbell, D. T., & Fiske, D. W. (1959). Convergent and discriminant validation by the
multitrait-multimethod matrix. Psychological Bulletin, 56, 81–105.
Carton, R. B., & Hofer, C. W. (2006). Measuring organizational performance. In R. B. Carton &
C. W. Hofer (Eds.). Cheltenham: Edward Elgar.
Casadesus-Masanell, R., & Ricart, J. E. (2010). From strategy to business models and onto
tactics. Long Range Planning, 43, 195–215.
Churchill, G. A., Jr. (1979). A paradigm for developing better measures of marketing constructs.
Journal of Marketing Research, 16, 64–73.

34
Dattero, R., White, E. M., & Janson, M. A. (1991). Methods for the identification of data outliers
in interactive SQL. Journal of Database Administration, 2, 7–18.
Day, G. S. (1994). The capabilities of market-driven organizations. Journal of Marketing, 58, 37–
52.
DeVellis, R. F. (2012). Scale development. Thousand Oaks, CA: Sage Publications.
Dinner, I. M., Mizik, N., & Lehmann, D. (2009). The unappreciated value of marketing: the
moderating role of changes in marketing and R&D spending on valuation of earnings
reports. Marketing Science Institute Special Report.
Dutta, S., Narasimhan, O., & Rajiv, S. (1999). Success in high-technology markets: is marketing
capability critical? Marketing Science, 18, 547–568.
Dutta, S., Narasimhan, O., & Rajiv, S. (2005). Conceptualizing and measuring capabilities:
methodology and empirical application. Strategic Management Journal, 26, 277–285.
Emory, W., & Cooper, D. R. (1991). Business research methods. Homewood, IL: Irwin.
Enache, L., & Srivastava, A. (2018). Should intangible investments be reported separately or
commingled with operating expenses? New evidence. Management Science, In press.
Gentry, R. J., & Shen, W. (2010). The relationship between accounting and market measures of
firm financial performance: how strong is it? Journal of Managerial Issues, 22, 514–530.
Hansen, D. R. (1990). Management accounting. Boston: PWS-KENT Publishing Company.
Heeler, R. M., & Ray, M. L. (1972). Measure validation in marketing. Journal of Marketing
Research, 9, 361–370.
Katsikeas, C. S., Morgan, N. A., Leonidou, L. C., & Hult, G. T. M. (2016). Assessing
performance outcomes in marketing. Journal of Marketing, 80, 1–20.
Kerlinger, F. (1986). Foundations of behavioral research. Fort Worth, TX: Harcourt Brace
Jovanovich.
Kotler, P., & Rackham, N. (2006). Ending the war between sales and marketing. Harvard
Business Review, 84, 1–14.
Kozlenkova, I. V., Samaha, S. A., & Palmatier R. W. (2014). Resource-based theory in
marketing. Journal of the Academy of Marketing Science, 42, 1–21.
Krishnan, H. A., Tadepalli, R., & Park, D. (2009). R&D intensity, marketing intensity, and
organizational performance. Journal of Managerial Issues, 21, 232–244.
Kristandl, G., & Bontis, N. (2007). Constructing a definition for intangibles using the resource
based view of the firm. Management Decision, 45, 1510–1524.

35
Kumar, V. (2016). My reflections on publishing in Journal of Marketing. Journal of Marketing,
80, 1–6.
Kurt, D., & Hulland, J. (2013). Aggressive marketing strategy following equity offerings and
firm value: the role of relative strategic flexibility. Journal of Marketing, 77, 57–74.
Lin, C., Tsai, H., & Wu, J. (2014). Collaboration strategy decision-making using the Miles and
Snow typology. Journal of Business Research, 67, 1979–1990.
Luo, X. (2008). When marketing strategy first meets Wall Street: marketing spendings and firms’
initial public offerings. Journal of Marketing, 72, 98–109.
Luo, Y., Zhao, J. H., & Du, J. (2005). The internationalization speed of e-commerce companies:
an empirical analysis. International Marketing Review, 22, 693–709.
MacKenzie, S. B. (2003). The dangers of poor construct conceptualization. Journal of the
Academy of Marketing Science, 31, 323–326.
March, J. G. (1991). Exploration and exploitation in organizational learning. Organization
Science, 2, 71–87.
Moorman, C., & Day, G. S. (2016). Organizing for marketing excellence. Journal of Marketing,
80(6), 6–35.
MSI (2016). 2014-2016 research priorities. Retrieved May 7, 2016, from
http://www.msi.org/research/2014-2016-research-priorities/
Narasimhan, O., Rajiv, S., & Dutta, S. (2006). Absorptive capacity in high-technology markets:
the competitive advantage of the haves. Marketing Science, 25, 510–524.
Nath, P., Nachiappan, S., & Ramanathan, R. (2010). The impact of marketing capability,
operations capability and diversification strategy on performance: a resource-based view.
Industrial Marketing Management, 39(2), 317–329.
Nunnally, J. C. (1978). Psychometric theory. In R. R. Wright & M. Gardner (Eds.). New York:
McGraw-Hill.
Peter, J. P. (1981). Construct validity: a review of basic issues and marketing practices. Journal
of Marketing Research, 18, 133–145.
Porter, M. E. (1985). Competitive advantage. New York: The Free Press.
Raithel, S., Sarstedt, M., Scharf, S., & Schwaiger, M. (2012). On the value relevance of customer
satisfaction: multiple drivers and multiple markets. Journal of the Academy of Marketing
Science, 40(4), 509–525.

36
Rust, R. T., Ambler, T., Carpenter, G. S., Kumar, V., & Srivastava, R. K. (2004). Measuring
marketing productivity: current knowledge and future directions. Journal of Marketing, 68,
76–89.
Sarkees, M., Hulland, J., & Chatterjee, R. (2014). Investments in exploitation and exploration
capabilities: balance versus focus. Journal of Marketing, 22, 7–23.
Schwertman, N. C., Owens, M. A., & Adnan, R. (2004). A simple more general boxplot method
for identifying outliers. Computational Statistics and Data Analysis, 47, 165–174.
Selling Power (2016). Selling Power 500 largest sales forces. Retrieved September 23, 2016
from http://www.sellingpower.com/content/article/index.php?a=10505/selling-power-
500/largest-sales-forces/2015&page=1/
Shapiro, C. (1989). The theory of business strategy. RAND Journal of Economics, 20, 125–137.
Standard and Poor's (2003). Standard and Poor’s Compustat user’s guide. New York: McGraw-
Hill.
Standard and Poor's (2013). Standard and Poor’s Compustat user’s guide. Retrieved January 27,
2014 from http://www.batd.eu/debodt/downloads/compustat_user_all.pdf
Steiger, J. H. (1980). Tests for comparing elements of a correlation matrix. Psychological
Bulletin, 87, 245–251.
van Triest, S., Bun, M. J. G., van Raaij, E. M., & Vernooij, M. J. A. (2009). The impact of
customer-specific marketing expenses on customer retention and customer profitability.
Marketing Letters, 20, 125–138.
Tukey, J. W. (1977). Exploratory data analysis. Pearson Education: United States.
Varadarajan, R. (2010). Strategic marketing and marketing strategy: domain, definition,
fundamental issues and foundational premises. Journal of the Academy of Marketing
Science, 38, 119–140.
Viswanathan, M. (2005). Measurement error and research design. Thousand Oaks, CA: Sage
Publications.
Vorhies, D. W., Orr, L. M., & Bush, V. D. (2011). Improving customer-focused marketing
capabilities and firm financial performance via marketing exploration and exploitation.
Journal of the Academy of Marketing Science, 39, 736–756.
Wharton (2016). Fundamental annuals data list. Retrieved November 24, 2016 from https://wrds-
web.wharton.upenn.edu/wrds/ds/compm/funda/index.cfm?navId=84

37
Table 1. Research design and validation approach
Research question Process step Level of analysis
1. Which marketing and sales constructs have been
measured using SGA?
1. Initial literature overview and analysis of the use of
SGA
2. Integration of literature to link the domains of
marketing and accounting
2. Is SGA a valid measure for the constructs? Are
there alternative measures for these constructs
that are equally or more valid?
3. Measurement validity
I. Content validity
a) Domain of definition
b) Level of abstraction
c) Time horizon
d) Level of objectivity
e) Business focus
II. Construct validity
a) Multitrait-multimethod (MTMM) matrix
b) Bivariate correlation matrix
Conceptual level
Qualitative validation
Empirical level
Quantitative validation
3. What guidelines can be developed for choosing
between SGA and alternative measures?
4. Development of guidelines

38
Table 2. SGA-based operationalization of marketing and sales constructs and subconstructs
Construct / Subconstruct Definition Studies Using the Operationalization
SGA ADVa SGA –
R&D
SGA –
R&D –
ADV
SGA +
R&D +
ADV
 SGA expense 12
 Sales (force) spendingb The amount of money spent on sales force activities to
stimulate purchases, such as “prospecting, defining needs,
preparing and presenting proposals, negotiating contracts, and
implementing the sale” (Kotler and Rackham 2006, p. 11).
7 1
 Marketing &
administrative spending
1
 Coordination spending 2 2
 Marketing spendingb “The total amount of money spent by a firm in all its marketing
related activities” (Nath et al. 2010, p. 322).
13 1 5
 Advertising spending 5
 Promotional spending 1
 Marketing assetsb,c “Customer-focused measures of the value of the firm (and its
offerings) that may enhance the firm's long-term value” (Rust et
al. 2004, p. 78).
5 1 1
 Marketing intensityb,d “Effects caused by marketing investments that, for instance,
enable a firm to build a strong brand name and to intensify its
relationship with its most valuable customers” (Raithel et al.
2012, p. 515).
2 4
 Advertising intensity 1
 Sales intensity 3
 Marketing efficiencyb,d Represents a “performance outcome viewed relative to the
resources consumed” (Katsikeas et al. 2016, p. 5); it features
growth, including changes in cash inflows or outflows (Ambler
et al. 2001; Carton and Hofer 2006).
3 1

39
 Marketing resourcesc “Tangible and intangible assets firms use to conceive of and
implement their strategies” (Barney and Arikan 2001, p. 138 cf.
Kozlenkova et al. 2014). They must be valuable, rare,
inimitable, and non-substitutable (Barney 1991).
1 1
 Marketing capabilityc “Complex bundles of skills and collective learning, exercised
through organizational processes that ensure superior
coordination of functional activities” (Day 1994, p. 38). They
differ from resources in that whereas resources are monetarily-
driven assets (tangible or intangible) that determine the
organization’s input factors, capabilities are its skills to use
these input factors.
6 1
 Marketing exploitationc,d Linked to capabilities, such that it refers to “the refinement and
extension of existing competencies, technologies and
paradigms” (March 1991, p. 85)
2
 Discretionary spendinge 1
 Fixed expensee 5
 Customer relationship
specific investment
1
Notes: SGA is selling, general, and administrative expenses; ADV is advertising expenses; and R&D denotes research and development
expenses.
aStudies that use the variable along with SGA are counted.
bConstructs ideally measured using accounting measures.
cConstructs ideally measured using operating measures.
dIntensity, efficiency, and exploitation represent higher-level constructs, comprised of one or more of baseline constructs (spending,
assets, resources, and capabilities) and distinct only in their objectives. Their validation thus depends on the validation of the baseline
constructs; no separate tests are conducted for them.
eDiscretionary spending and fixed expenses do not have a specific contextual meaning in terms of business operations. They are
influenced less by changes in the firm’s activity level (Hansen 1990); discretionary spending can even be eliminated without affecting
organizational profitability immediately (Bragg 2010). Depending on the objective, they thus can be applied to various functions such as
advertising and R&D.

40
Table 3. Multitrait-multimethod (MTMM) matrix
Method 1 (Data Source 1) Method 2 (Data Source 2)
Trait 1 Trait 2 Trait 1 Trait 2
Variable 1 Variable 2 Variable 3 Variable 4
Method 1
(Data Source 1)
Trait 1 Variable 1 I: 1.00
Trait 2 Variable 2 II: Heterotrait-
monomethod I: 1.00
Method 2
(Data Source 2)
Trait 1 Variable 3 III: Monotrait-
heteromethod
IV: Heterotrait-
heteromethod I: 1.00
Trait 2 Variable 4 IV: Heterotrait-
heteromethod
III: Monotrait-
heteromethod
II: Heterotrait-
monomethod I: 1.00
Notes:
For convergent validity, correlation coefficients in III should be significantly different from 0 and
should be sufficiently large.
For discriminant validity, correlation coefficients in III should be larger than in IV and correlation
coefficients in III should be larger than in II.

41
Table 4. Data sources, variables, and descriptions
Variable Description
Data source: Compustat
SGA (Selling, general, and
administrative expense)
All operating expenses (other than those directly related to
production) incurred in the regular course of business.
ADV (Advertising expense) The cost of advertising media (radio, TV, newspapers, and
periodicals) and promotional expenses. It does not include
other selling and marketing expenses.
R&D (Research and
development expense)
All costs related to the development of new products or
services. It does not include market research or market testing
activities, or routine or periodic alterations to existing
products, manufacturing processes, and other ongoing
operations.
Goodwill Value assigned to long-term perceptual assets (e.g., brand
name, client relationships, and employee morale), which
increase the earning potential of the company.
Other intangible assets Intellectual assets such as patents and rights, which have a
monetary value for the company.
Total intangible assets Sum of goodwill and other intangible assets
Data source: Advertising Age (2016a, 2016b)
Measured media spending Estimated annual spending across 19 media: TV (broadcast
network TV, spot TV, syndicated TV, and network cable TV),
radio (network, national spot, and local), magazines
(consumer magazines, Sunday magazines, local magazines,
and B2B magazines), newspapers (local and national),
Spanish-language media (magazines, newspapers and TV
networks), outdoor, internet (excluding paid search and
broadband video), and free-standing inserts.
Estimated unmeasured
spending
Estimates of spending on direct marketing, promotion, co-
operative marketing, coupons, catalogs, product placement,
events, and unmeasured forms of digital media (e.g., display,
paid search, video, and social media).
Total marketing spending Sum of measured media spending and estimated unmeasured
spending
Data source: Selling Power (2016)
Number of salespeople Estimated number of people employed in sales functions
Notes: These measures are in millions of dollars, except for the number of salespeople, which is
measured in thousands. Definitions of the Compustat variables are available in Standard and
Poor’s (2003).

42
Table 5. Descriptive statistics and correlations
A. Sample 1: Match of Compustat and Advertising Age data sets (N = 499)
Variable
a Mean S.D. Min. Max. 1 2 3 4 5 6 7 8 9 10 11 12
1 SGA 11197 9421 11 56733 1
2 SGA – ADV 9161 8617 10 55133 .99 1
3 SGA – R&D 8515 7375 11 56733 .96 .96 1
4 SGA – ADV – R&D 6787 6889 10 55133 .93 .95 .99 1
5 ADV 1773 1309 1 8000 .52 .41 .45 .29 1
6 R&D 2815 3223 0 12540 .71 .70 .51 .44 .51 1
7 Total intangible assets 12455 22431 0 225278 .46 .41 .44 .40 .37 .22 1
8 Goodwill 7081 11697 0 104568 .44 .39 .43 .39 .32 .21 .94 1
9 Other intangibles 6250 13074 0 120710 .39 .35 .40 .35 .33 .15 .94 .77 1
10 Total marketing
spending
2165 1439 277 8554 .40 .32 .34 .22 .79 .48 .24 .20 .22 1
11 Measured media
spending
1263 910 43 4984 .38 .30 .32 .20 .74 .45 .21 .18 .18 .93 1
12 Estimated unmeasured
spending
902 674 20 3723 .34 .28 .31 .20 .69 .43 .22 .17 .22 .88 .64 1
Notes: Correlations greater than .09 (absolute value) are significant at the .05 level. Extreme outliers were removed before obtaining these
statistics (Schwertman et al. 2004). We identified values far outside the data set using the Tukey (1977) formula – lower fence: Quartile 1
– 3*(Quartile 3 – Quartile 1); upper fence: Quartile 3 + 3*(Quartile 3 – Quartile 1). All values outside the fences were eliminated from the
data set.
aMeasured in millions of U.S. dollars.

43
B. Sample 2: Match of Compustat and Selling Power data sets (N = 409)
Variablea Mean S.D. Min. Max. 1 2 3 4 5 6
1 SGA 5296 7759 70 39697 1
2 SGA – ADV 4712 6924 68 36425 .99 1
3 SGA – R&D 3750 5695 59 37967 .97 .96 1
4 SGA – ADV – R&D 3166 4870 58 34695 .95 .96 .98 1
5 ADV 584 1384 .30 9729 .68 .56 .70 .54 1
6 Number of salespeople 5617 7024 .46 31401 .67 .69 .63 .65 .33 1
Notes: Correlations greater than .10 (absolute value) are significant at the .05 level. Extreme
outliers were removed before obtaining these statistics (Schwertman et al. 2004). We identified
values far outside the data set using the Tukey (1977) formula – lower fence: Quartile 1 –
3*(Quartile 3 – Quartile 1); upper fence: Quartile 3 + 3*(Quartile 3 – Quartile 1). All values
outside the fences were eliminated from the data set.
aMeasured in millions of U.S. dollars except the number of salespeople which is measured in
thousands.

44
Table 6. Conceptual analysis results
Conceptual Dimension Operational Dimension Managerial
Dimension
Construct/Variable Domain of
Definition
Level of
Abstraction
Time
Horizon
Level of
Objectivity
Business Focus
Spending Accounting Low Short-term High Strategic/Tactical
Assets Accounting/
Operating
Medium Long-term Medium Strategic
Resources Operating High Long-term Low Strategic
Capabilities Operating High Long-term Low Strategic
SGA expense Accounting Low Short-term High Tactical
Measured media
spending
Accounting Low Short-term High Strategic/Tactical
Estimated unmeasured
spending
Accounting Low Short-term High Strategic/Tactical
Salespeople Quantitative
(Accounting)
Low Short-term High Strategic/Tactical
Goodwill Accounting Medium Long-term Medium Strategic
Other intangible assets Accounting Medium Long-term Medium Strategic

45
Table 7. Construct validation for marketing communication spending
A. MTMM 1 results
Method 1
(Compustat)
Method 2
(Advertising Age)
MTMM 1 (Suitability of SGA to
measure promotional spending)
Trait 1
(Advertising
spending)
Trait 2
(Promotional
spending)
Trait 1
(Advertising
spending)
Trait 2
(Promotional
spending)
ADV SGA
Measured
media
spending
Estimated
unmeasured
spending
Method 1
(Compustat)
Trait 1 ADV 1
Trait 2 SGA .52** 1
Method 2
(Advertising
Age)
Trait 1 Measured
media spending .74** .38** 1
Trait 2
Estimated
unmeasured
spending
.69** .34** .64** 1

46
B. Overview of results from MTMM matrices 1–4
Trait 1
(Advertising spending)
Trait 2
(Promotional spending)
MTMM 1 ADV SGA
Convergent validity .74** .34** X
Discriminant validity
1
st
condition
2
nd
condition
.74 > .69 .34 < .38 X
.74 > .38 .34 < .69
.74 > .64 .34 < .52 X
.74 > .52 .34 < .64
MTMM 2 ADV SGA – ADV
Convergent validity .74** .28** X
Discriminant validity
1
st
condition
2
nd
condition
.74 > .69 .28 < .30 X
.74 > .30 .28 < .69
.74 > .64 .28 < .41 X
.74 > .41 .28 < .64
MTMM 3 ADV SGA – R&D
Convergent validity .74** .31** X
Discriminant validity
1
st
condition
2
nd
condition
.74 > .69 .31 < .32 X
.74 > .32 .31 < .69
.74 > .64 .31 < .45 X
.74 > .45 .31 < .64
MTMM 4 ADV SGA – ADV – R&D
Convergent validity .74** .20** X
Discriminant validity
1
st
condition
2
nd
condition
.74 > .69 .20 ≈ .20 X
.74 > .20 .20 < .69
.74 > .64 .20 < .29 X
.74 > .29 .20 < .64
Notes: As shown in the table, neither SGA nor any of its modifications is a good measure of promotional
spending. If the measures were switched, still neither SGA nor any of its modifications will be a good
measure of advertising spending.
**p < .01 (two-tailed).

47
Table 8. Construct validation for marketing assets
A. MTMM 5 results
Method 1
(Income statement from
Compustat)
Method 2
(Balance sheet from
Compustat)
MTMM 5 (Suitability of SGA to measure
intellectual assets)
Trait 1
(Perceptual
assets)
Trait 2
(Intellectual
assets)
Trait 1
(Perceptual
assets)
Trait 2
(Intellectual
assets)
ADV SGA Goodwill
Other
intangible
assets
Method 1
(Income statement
from Compustat)
Trait 1 ADV 1
Trait 2 SGA .48** 1
Method 2
(Balance sheet
from Compustat)
Trait 1 Goodwill .31** .45** 1
Trait 2 Other intangible
assets .36** .42** .77** 1

48
B. Overview of results from MTMM matrices 5–7
Trait 1
(Perceptual assets)
Trait 2
(Intellectual assets)
MTMM 5 ADV SGA
Convergent validity .31** .42**
Discriminant validity
1
st
condition
2
nd
condition
.31 < .36 X .42 < .45 X
.31 < .45 .42 > .36
.31 < .77 X .42 < .48 X
.31 < .48 .42 < .77
MTMM 6 ADV SGA – ADV
Convergent validity .31** .39**
Discriminant validity
1
st
condition
2
nd
condition
.31 < .36 X .39 < .40 X
.31 < .40 .39 > .36
.31 < .77 X .39 > .31 X
.31 ≈ .31 .39 < .77
MTMM 7 ADV SGA – R&D
Convergent validity .31** .42**
Discriminant validity
1
st
condition
2
nd
condition
.31 < .36 X .42 < .43 X
.31 < .43 .42 > .36
.31 < .77 X .42 > .33 X
.31 < .33 .42 < .77
Notes: None of the measures from the income statement are good measures of marketing assets. This
finding is consistent even if measures were switched, such that ADV measured intellectual assets and
SGA measured perceptual assets. The sample size of this analysis is 395 data points, because we excluded
observations with missing values for other intangible assets.
**p < .01 (two-tailed).

49
Table 9. Summary of construct validation
Construct
Category
Construct and Operationalization Benchmark Variables Empirical Test for
SGA or modificationsa
Spending Construct: Marketing Spending
(Subconstructs: Advertising Spending;
Promotional Spending)
1. SGA
2. SGA – ADV
3. SGA – R&D
Measured media
spending, Estimated
unmeasured spending
MTMM 1
MTMM 2
MTMM 3
Robustness check
SGA – ADV – R&D
MTMM 4
Construct: Sales Force Spending
1. SGA
2. SGA – ADV – R&D
Number of salespeople
Bivariate correlations
Bivariate correlations
Robustness check
SGA – ADV
Bivariate correlations
Assets Construct: Marketing Assets
(Subconstructs: Perceptual assets; Intellectual
assets)
1. SGA
2. SGA – R&D
3. SGA; ADVb
Goodwill, other
intangible assets
MTMM 5
MTMM 6
MTMM 5
Robustness Check
SGA – ADV
MTMM 7
Efficiency Construct: Marketing Efficiency
(based on Marketing Spending)
1. SGA; ADVb
MTMM 1
Intensity Construct: Marketing Intensity
(Subconstruct: Advertising intensity)
(based on Marketing Spending)
1. SGA
2. SGA – R&D
MTMM 1
MTMM 3
Construct: Sales Intensity
(based on Sales Force Spending)
1. SGA
Bivariate correlations
Notes: For each construct, we show only subconstructs or measures that have been employed in
previous literature. If an operationalization had been expressed as a ratio, our analysis focuses
only on the component (numerator or denominator) that explicitly includes the measure of
interest.
aOnly MTMM matrices under scenario 1 have been depicted. There are 4 matrices under scenario
2 for marketing spending and 3 matrices under scenario 2 for marketing assets which have not
been included in the table.
bWe used SGA along with ADV to measure a specific construct in this case.

50
Figure 1. Number of studies employing SGA as a measure in marketing research (from 1987 to
2017)
0
2
4
6
8
10
12
14
1987 1990 1993 1996 1999 2002 2005 2008 2011 2014 2017
Number of studies in the year
Year
Correct Usage Incorrect Usage

51
Figure 2. Treatment of cash outflows in Compustat; and accounting vs. operating nature of marketing constructs
Notes: All constructs and subconstructs in the rectangle with dashed bold lines have been measured using SGA in one or more studies
Cash Outflow
Operating Expenses Assets
Expensed Capitalized
Tangible Assets Intangible Assets
Marketing
Capability
Constructs ideally measured
with Accounting Measures
Constructs ideally measured
with Operating Measures
Cost of Goods
Sold (COGS)
Other Operating
Expense
Marketing
Spending
Marketing
Assets
Marketing
Resources
Marketing
Exploitation
Sales
Intensity
Accounting Treatment
of Cash Outflows in
Compustat Data
Sales force
Spending
Selling, General,
and Administrative
Expense (SGA)
Advertising
Expense
Promotion
Expense
Marketing
Intensity Marketing
Efficiency
Treatment of cash outflows in Accounting Theoretical relationships

52
Figure 3. Decision criteria for content validation
{Criteria depicted with solid lines indicate
2 necessary conditions for content validity}
{Decision ‘rule’: Proceed only if content
validity has been established}
Domain of
definition
Level of
abstraction
Time
horizon
Level of
objectivity
Business
focus
Content
validity
Construct
validity

53
Figure 4. Decision tree
Notes: A checkmark in the top line means that SGA is a valid measure for the construct; a cross means that it is not a valid measure. A checkmark
below the marketing and sales constructs indicates which alternative measures are valid or better suited. Marketing intensity, marketing efficiency,
sales intensity, and marketing exploitation are constructs comprised of one or more of the baseline constructs (expenses, assets, resources, and
capabilities), differing only in their measurement objective. The validation of these constructs thus follows from their respective baseline constructs.
Marketing resources and marketing capability require industry-specific or even firm-specific measurement approaches, predominantly based on
qualitative operationalizations. Finally, both operating and accounting measures are needed to capture marketing assets in total.
Constructs ideally measured
with Accounting Measures
Marketing
Spending
Constructs ideally
measured with
Operating Measures
Sales
Intensity
Selling, General and
Administrative
Expense (SGA)
Sales force
Spending
Marketing
Efficiency
Marketing
Intensity
 (Measured Media spending,
Estimated Unmeasured Spending)
SGA
–
ADV
~ADV
Advertising
Spending
Promotional
Spending
ADV
Estimated Unmeasured
Spending
Measured Media
Spending
Marketing
Assets
Marketing
Resources

Intangible Assets
Marketing
Capability
Marketing
Exploitation

54
Appendix
Appendix 1. Use of SGA in marketing and management literature (1995–2016)
Concept/Construct Operationalization Authors
Marketing assets /
investments
SGA

Balsam, Fernando, and Tripathy 2011
 Banker, Mashruwala, and Tripathy 2014
 Borah and Tellis 2014
 Kotha, Rajgopal, and Rindova 2001
 Hornig and Fischer 2013
SGA – R&D
SGA – R&D – ADV
SGA; ADV

Lee and Chang 2014
 Enache and Srivastava 2018

Hornig and Fischer 2013
Marketing expense SGA

Bentley, Omer, and Sharp 2012
(Denominator: Sales)
 Dinner 2011
 Dutta, Narasimhan, and Rajiv 1999, 2005
 Sarkees, Hulland, and Chatterjee 2014
 Corona 2009, 2014
 Cook, Maulth, and Spaeth 2007
 Habib 2017
 Higgins, Omer, and Phillips 2015
 Nam and Kannan 2014
 Narasimhan, Rajiv, and Dutta 2006
 Nath, Nachiappan, and Ramanathan 2010 (as
one operationalization variable)
 Raassens, Wuyts, and Geyskens 2014
(Denominator: Assets)
 Snyder 2009
 Swaminathan and Moorman 2009
 Kalaignanam et al. 2013
SGA – R&D

Dinner, Mizik, and Lehmann 2009
 Luo 2008
 Kurt and Hulland 2013
 Bharadwaj, Tuli, and Bonfrer 2011
 Shin, Sakakibara, and Hanssens 2008
Sales (force)
expense
SGA

Koku 2011
 Kumar 1999
 Wuyts, Dutta, and Stremersch 2004
 Mhatre, Joo, and Lee 2014
 Achrol and Seo 2011
 Lin, Lee, and Hung 2006
 Sarkees and Luchs 2011
SGA – ADV – R&D

Kim and McAlister 2011

55
SGA expense SGA

Achrol 2012
 Ailawadi, Borin, and Farris 1995
 Bayus, Erickson, and Jacobson 2003
 Bell and Gordon 1999
 Boulding and Christen 2008
 Efendi et al. 2013
 Foster and Gupta 1994
 Huang, Seow, and Shangguan 2011
 Kalwani and Narayandas 1995
 Moorman, Du, and Mela 2005
 Mottner and Smith 2009
 Poston and Grabski 2001
 Rangan and Bell 1998
 Rego, Morgan, and Fornell 2013
 Rust and Huang 2012
Advertising expense SGA

Collins and Han 2004
 Demerjian, Lev, and McVay 2012
 Ding, Stolowy, and Tenenhaus 2007
 Wiles 2007
Promotional expense SGA

Vinod and Rao 2000
Marketing and
administration
expense
SGA

Lévesque, Jogleklar, and Davies 2012
Sales and general
expense
SGA

Mittal et al. 2005
Discretionary
expense
SGA + ADV + R&D

Ho, Liu, and Ouyang 2012
Marketing capability SGA

Bahadir, Bharadwaj, and Srivastava 2008
 Patwardhan 2014
 Cheng et al. 2008
 Lee and Rugman 2012
 Luo, Zhao, and Du 2005
 Rugman and Sukpanich 2006
SGA – R&D

Darroch and Miles 2011
Sales capability SGA

Boyd and Brown 2012
Marketing resource SGA

Cook, Moult, and Spaeth 2007 (Denominator:
Sales)
Marketing resource
intensity
SGA – R&D

Raassens, Wuyts, and Geyskens 2014
(Denominator: Assets)
Marketing intensity SGA
SGA – R&D

Krishnan, Tadepalli, and Park 2009
(Denominator: Sales)
 Raithel et al. 2012 (Denominator: Assets)
 Dinner, Mizik, and Lehmann 2009
(Denominator: Assets)

56

Mizik and Jacobson 2007 (Denominator:
Assets)
 Mizik 2010 (Denominator: Assets)
Sales intensity SGA

Berman et al. 1999 (Denominator: Sales)
 Siddharthan and Kumar 1990 (Denominator:
Sales)
 Nair and Selover 2012 (Denominator: Sales)
Advertising intensity SGA

Grubaugh 1987 (Denominator: Sales)
Marketing efficiency SGA

Cook, Moult, and Spaeth 2007 (Denominator:
Sales)
 Lin, Tsai, and Wu 2014 (Denominator: Sales)
SGA; ADV

Morgan and Rego 2009 (Denominator: Sales)
Marketing
exploitation
SGA

Sarkees, Hulland, and Chatterjee 2014
(Numerator: Sales)
 Bentley, Omer, and Sharp 2013
(Denominator: Sales)
Coordination
expense
SGA
SGA – R&D

Lee et al. 2014
 Lee et al. 2015
 Ray, Wu, and Konana 2009
 Im, Grover, and Teng 2013
SGA – ADV – R&D
– software – bad debt
– pension and
retirement

Shin 1999 (for manufacturing industries)
SGA – ADV – bad
debt – pension and
retirement

Shin 1999 (for non-manufacturing industries)
SGA – ADV–
pension and
retirement

Shin 1999 (for finance industry)
Fixed expense SGA

Bruton, Keels, and Scifres 2002
 Gaspar and Massa 2006
 Mitra and Chaya 1996 (Denominator: Sales)
 Bharadwaj 2000 (Denominator: Sales)
 Haleblian and Finkelstein 1993
(Denominator: Sales)
Customer
relationship-specific
investments
SGA

Irvine, Park, and Yildizhan 2016

57
Bibliography for Appendix
Achrol, R. S. (2012). Slotting allowances: a time series analysis of aggregate effects over three
decades. Journal of the Academy of Marketing Science, 40(5), 673–694.
Achrol, R. S., & Seo, J. H. (2011). In marketing channel theory and slotting allowances: an
empirical analysis using quantile regression. American Marketing Association, 286–295.
Ailawadi, K. L., Borin, N., & Farris, P. W. (1995). Market power and performance: a cross-
industry analysis of manufacturers and retailers. Journal of Retailing, 71(3), 211–248.
Bahadir, S. C., Bharadwaj, S. G., & Srivastava, R. K. (2008). Financial value of brands in
mergers and acquisitions: is value in the eye of the beholder? Journal of Marketing, 72(6),
49–64.
Balsam, S., Fernando, G. D., & Tripathy, A. (2011). The impact of firm strategy on performance
measures used in executive compensation. Journal of Business Research, 64(2), 187–193.
Banker, R. D., Mashruwala, R., & Tripathy, A. (2014). Does a differentiation strategy lead to
more sustainable financial performance than a cost leadership strategy? Management
Decision, 52(5), 872–896.
Bayus, B. L., Erickson, G., & Jacobson, R. (2003). The financial rewards of new product
introductions in the personal computer industry. Management Science, 49(2), 197–210.
Bell, D. E., & Gordon, D. S. (1999). The king-size company. Journal of Interactive Marketing,
13(1), 66–86.
Bentley, K. A., Omer, T. C., & Sharp, N. Y. (2012). Business strategy, financial reporting
irregularities, and audit effort. Contemporary Accounting Research, 30(2), 780–817.
Berman, S. L., Wicks, A. C., Kotha, S., & Jones, T. M. (1999). Does stakeholder orientation
matter? The relationship between stakeholder management models and firm financial
performance. Academy of Management Journal, 42(5), 488–506.
Bharadwaj, A. (2000). A resource-based perspective on information technology capability and
firm. MIS Quarterly, 24(1), 169–196.
Bharadwaj, S. G., Tuli, K. R., & Bonfrer, A. (2011). The impact of brand quality on shareholder
wealth. Journal of Marketing, 75(5), 88–104.
Borah, A., & Tellis, G. J. (2014). Make, buy, or ally? Choice of and payoff from announcements
of alternate strategies for innovations. Marketing Science, 33(1), 114–133.
Boulding, W., & Christen, M. (2008). Disentangling pioneering cost advantages and
disadvantages. Marketing Science, 27(4), 699–716.
Boyd, D. E., & Brown, B. P. (2012). Marketing control rights and their distribution within
technology licensing agreements: a real options perspective. Journal of the Academy of
Marketing Science, 40(5), 659–672.
Bruton, G. D., Keels, J. K., & Scifres, E. L. (2002). Corporate restructuring and performance: an
agency perspective on the complete buyout cycle. Journal of Business Research, 55(9), 709–
724.

58
Cheng, M.-Y., Lin, J.-Y., Hsiao, T.-Y., & Lin, T. W. (2008). Censoring model for evaluating
intellectual capital value drivers. Journal of Intellectual Capital, 9(4), 639–655.
Collins, C. J., & Han, J. (2004). Exploring applicant pool quantity and quality: the effects of early
recruitment practice strategies, corporate advertising, and firm reputation. Personnel
Psychology, 57, 685–717.
Cook, V. J., Moult, W., & Spaeth, J. (2007). Marketing meets finance, Working Paper, 1–48.
Corona, R. (2009). Is Costco better than Walmart? A comparative analysis based on enterprise
marketing efficiency. Working Paper, 1–34.
Corona, R. (2014). A comparative analysis of major US retailers based on enterprise marketing
efficiency. Global Journal of Business Research, 8(4), 25–40.
Darroch, J., & Miles, M. P. (2011). A research note on market creation in the pharmaceutical
industry. Journal of Business Research, 64(7), 723–727.
Demerjian, P., Lev, B., & McVay, S. (2012). Quantifying managerial ability: a new measure and
validity tests. Management Science, 58(7), 1229–1248.
Ding, Y., Stolowy, H., & Tenenhaus, M. (2007). R&D productivity: an exploratory international
study. Review of Accounting and Finance, 6(1), 86–101.
Dinner, I. M., Mizik, N., & Lehmann, D. (2009). The unappreciated value of marketing: the
moderating role of changes in marketing and R&D spending on valuation of earnings
reports. Marketing Science Institute Special Report.
Dinner, I. M., (2011). The interpretation of marketing actions and communications by the
financial markets. Dissertation Thesis, 1-152.
Dutta, S., Narasimhan, O., & Rajiv, S. (1999). Success in high-technology markets: is marketing
capability critical? Marketing Science, 18(4), 547–568.
Dutta, S., Narasimhan, O., & Rajiv, S. (2005). Conceptualizing and measuring capabilities:
methodology and empirical application. Strategic Management Journal, 26(3), 277–285.
Efendi, J., Kinney, M. R., Smith, K. T., & Smith, L. M. (2013). Marketing supply chain using
B2B buy-side e-commerce systems: does adoption impact financial performance. Working
Paper, 1–32.
Enache, L., & Srivastava, A. (2018). Should intangible investments be reported separately or
commingled with operating expenses? New evidence. Management Science, In press.
Foster, G., & Gupta, M. (1994). Marketing, cost management, and management accounting.
Journal of Management Accounting Research, 6, 63–77.
Gaspar, J.-M., & Massa, M. (2006). Idiosyncratic volatility and product market competition. The
Journal of Business, 79(6), 3125–3152.
Grubaugh, S. G. (1987). Determinants of direct foreign investment. Review of Economics and
Statistics, 69(1), 149–152.
Habib, A. (2017). Business strategy, overvalued equities, and stock price crash risk. Research in
International Business and Finance, 39, 389–405.

59
Haleblian, J., & Finkelstein, S. (1993). Top management team size, CEO dominance, and firm
performance: the moderating roles of environmental turbulence and discretion. Academy of
Management Journal, 36(4), 844–863.
Higgins, D., Omer, T. C., & Phillips, J. D. (2015). The influence of a firm’s business strategy on
its tax aggressiveness. Contemporary Accounting Research, 32(2), 674–702.
Ho, L.-C. J., Liu, C.-S., & Ouyang, B. (2012). Bloated balance sheet, earnings management, and
forecast guidance. Review of Accounting and Finance, 11(2), 120–140.
Hornig, T., & Fischer, M. (2013). Validating financial brand equity metrics: how useful are brand
valuation methods? Dissertation Thesis, 103–156.
Huang, R., Seow, G. S., & Shangguan, J. Z. (2011). Intangible investments and the pricing of
corporate SGA expenses. Journal of Business & Economic Studies, 17(2), 67–77.
Im, K. S., Grover, V., & Teng, J. T. C. (2013). Do large firms become smaller by using
information technology? Information Systems Research, 24(2), 470–491.
Irvine, P. J., Park, S. S., & Yildizhan, C. (2016). Customer-base concentration, profitability, and
the relationship life cycle. American Accounting Association, 91(3), 883–906.
Kalaignanam, K., Kushwaha, T., Steenkamp, J.-B. E. M., & Tuli, K. R. (2013). The effect of
CRM outsourcing on shareholder value: a contingency perspective. Management Science,
59(3), 748–769.
Kalwani, M. U., & Narayandas, N. (1995). Relationships: do they pay off for supplier firms?
Journal of Marketing, 59(1), 1–16.
Kim, M., & McAlister, L. M. (2011). Stock market reaction to unexpected growth in marketing
expenditure: negative for sales force, contingent on spending level for advertising. Journal
of Marketing, 75(7), 68–85.
Koku, P. S. (2011). On the connection between R&D, selling expenditures, and profitability in
the pharmaceutical industry revisited. Journal of Strategic Marketing, 19(3), 273–283.
Kotha, S., Rajgopal, S., & Rindova, V. (2001). Reputation building and performance: an
empirical analysis of the top-50 pure internet firms. European Management Journal, 19(6),
571–586.
Krishnan, H. A., Tadepalli, R., & Park, D. (2009). R&D intensity, marketing intensity, and
organizational performance. Journal of Managerial Issues, 21(2), 232–244.
Kumar, P. (1999). The impact of long-term client relationships on the performance of business
service firms. Journal of Service Research, 2(1), 4–18.
Kurt, D., & Hulland, J. (2013). Aggressive marketing strategy following equity offerings and
value: the role of relative strategic flexibility. Journal of Marketing, 77(9), 57–74.
Lee, I. H., & Rugman, A. M. (2012). Firm-specific advantages, inward FDI origins, and
performance of multinational enterprises. Journal of International Management, 18(2), 132–
146.

60
Lee, J., & Chang, Y. B. (2014). Interplay between internal investment and alliance specialization
in R&D and marketing. Industrial Marketing Management, 43(5), 813–825.
Lee, J., Sridhar, S., Henderson, C. M., & Palmatier, R. W. (2015). Financial performance effect
of customer-centric structure on long-term financial performance. Marketing Science, 34(2),
250–268.
Lévesque, M., Joglekar, N., & Davies, J. (2012). A comparison of revenue growth at recent-IPO
and established firms: the influence of SG&A, R&D and COGS. Journal of Business
Venturing, 27(1), 47–61.
Lin, C., Tsai, H.-L., & Wu, J.-C. (2014). Collaboration strategy decision-making using the miles
and snow typology. Journal of Business Research, 67(9), 1979–1990.
Lin, B.-W., Lee, Y., & Hung, S.-C. (2006). R&D intensity and commercialization orientation
effects on financial performance. Journal of Business Research, 59(6), 679–685.
Luo, X. (2008). When marketing strategy first meets Wall Street: marketing spendings and firms’
initial public offerings. Journal of Marketing, 72(9), 98–109.
Luo, Y., Zhao, J. H., & Du, J. (2005). The internationalization speed of e-commerce companies:
an empirical analysis. International Marketing Review, 22(6), 693–709.
Mhatre, N., Joo, S.-J., & Christopher Lee, C. (2014). Benchmarking the performance of
department stores within an income elasticity of demand perspective. Benchmarking: An
International Journal, 21(2), 205–217.
Mitra, S., & Chaya, A. K. (1996). Analyzing Cost-effectiveness of organizations: the impact of
information technology spending. Journal of Management Information Systems, 13(2), 29–
57.
Mittal, V., Anderson, E. W., Sayrak, A., & Tadikamalla, P. (2005). Dual emphasis and the long-
term financial impact of customer satisfaction. Marketing Science, 24(4), 544–555.
Mizik, N. (2010). The theory and practice of myopic management. Journal of Marketing
Research, 47(8), 594–611.
Mizik, N., & Jacobson, R. (2007). Myopic marketing management: evidence of the phenomenon
and its long-term performance consequences in the SEO context. Marketing Science, 26(3),
361–379.
Moorman, C., Du, R., & Mela, C. F. (2005). The effect of standardized information on firm
survival and marketing strategies. Marketing Science, 24(2), 263–274.
Morgan, N. A., & Rego, L. L. (2009). Brand portfolio strategy and firm performance. Journal of
Marketing, 73(1), 59–74.
Mottner, S., & Smith, S. (2009). Wal-Mart: supplier performance and market power. Journal of
Business Research, 62(5), 535–541.
Nair, A., & Selover, D. D. (2012). A study of competitive dynamics. Journal of Business
Research, 65(3), 355–361.

61
Nam, H., & Kannan, P. K. (2014). Informational value of social tagging networks. Journal of
Marketing, 78(7), 21–40.
Narasimhan, O., Rajiv, S., & Dutta, S. (2006). Absorptive capacity in high-technology markets:
the competitive advantage of the haves. Marketing Science, 25(5), 510–524.
Nath, P., Nachiappan, S., & Ramanathan, R. (2010). The impact of marketing capability,
operations capability and diversification strategy on performance: a resource-based view.
Industrial Marketing Management, 39(2), 317–329.
Patwardhan, A. M. (2014). A partial theory of holistic firm-level marketing capability: an
empirical investigation. Journal of Management and Marketing Research, 16(8), 1–46.
Poston R., & Grabski S. (2001). Financial impacts of enterprise resource planning
implementations. International Journal of Accounting Information Systems, 2(2), 271–294.
Raassens, N., Wuyts, S., & Geyskens, I. (2014). The performance implications of outsourcing
customer support to service providers in emerging versus established economies.
International Journal of Research in Marketing, 31(3), 280–292.
Raithel, S., Sarstedt, M., Scharf, S., & Schwaiger, M. (2012). On the value relevance of customer
satisfaction: Multiple drivers and multiple markets. Journal of the Academy of Marketing
Science, 40(4), 509–525.
Rangan, V. K., & Bell, M. (1998). Dell online. Journal of Interactive Marketing, 12(4), 63–86.
Ray, G., Wu, D., & Konana, P. (2009). Competitive environment and the relationship between IT
and vertical integration. Information System Research, 20(4), 585–603.
Rego, L. L., Morgan, N. A., & Fornell, C. (2013). Reexamining the market share-customer
satisfaction relationship. Journal of Marketing, 77(9), 1–20.
Rugman, A., & Sukpanich, N. (2006). Firm-specific advantages intra-regional sales and
performance of multinational enterprises. The International Trade Journal, 20(3), 355–382.
Rust, R. T., & Huang, M.-H. (2012). Optimizing service productivity. Journal of Marketing,
76(2), 47–66.
Sarkees, M. E., & Luchs, R. (2011). Stochastic frontier estimation in international marketing
research: exploring untapped opportunities. Measurement and Research Methods in
International Marketing (22), 99–114.
Sarkees, M. E., Hulland, J., & Chatterjee, R. (2014). Investments in exploitation and exploration
capabilities: balance versus focus. Journal of Marketing Theory and Practice, 22(1), 7–23.
Siddharthan, N. S., & Kumar, N. (1990). The determinants of inter-industry variations in the
proportion of intra-firm trade: the behaviour of US multinationals. Weltwirtschaftliches
Archive, 126(3), 581–591.
Shin, N. (1999). Does information technology improve coordination ? An empirical analysis.
Logisitcs Information Management, 12(1/2), 138–144.
Shin, H. S., Sakakibara, M., & Hanssens, D. M. (2008). Marketing and R&D investment of
leader vs. follower. Working Paper, 1–39.

62
Snyder, S. (2009). Marketing and R&D complementarity in the pharmaceutical industry.
Working Paper, 1–17.
Swaminathan, V., & Moorman, C. (2009). Marketing alliances, firm networks, and firm value
creation. Journal of Marketing, 73(9), 52–69.
Vinod, H. D., & Rao, P. M. (2000). R&D and promotion in pharmaceuticals: a conceptual
framework and empirical exploration. Journal of Marketing Theory and Practice, 8(4), 10–
20.
Wiles, M. A. (2007). The effect of customer service on retailers’ shareholder wealth: the role of
availability and reputation cues. Journal of Retailing, 83(1), 19–31.
Wuyts, S., Dutta, S., & Stremersch, S. (2004). Portfolios of interfirm agreements in technology-
intensive markets: consequences for innovation and profitability. Journal of Marketing,
68(2), 88–100.

63
Web Appendices
Web Appendix 1. Alphabetical list of journals searched to find studies using SGA to capture
marketing-related constructs (1995–2016)
1. Academy of Management Journal
2. Academy of Management Review
3. British Journal of Management
4. European Journal of Marketing
5. Industrial Marketing Management
6. International Journal of Research in Marketing
7. Journal of Economics and Management Strategy
8. Journal of International Management
9. Journal of International Marketing
10. Journal of Management
11. Journal of Management Studies
12. Journal of Marketing
13. Journal of Marketing Research
14. Journal of Public Policy & Marketing
15. Journal of Retailing
16. Journal of Service Research
17. Journal of the Academy of Marketing Science
18. Management Science
19. Marketing Letters
20. Marketing Science
21. Quantitative Marketing and Economics
22. Strategic Management Journal

64
Web Appendix 2. 29 items in SGA and their classification according to Porter’s value chain
framework
Porter’s value chain framework and relevant SGA items
Porter’s idea of value chain, based on the process view of organizations, considers an
organization as a system made up of subsystems each with input, transformation process, and
output activities (Porter 1985). Each activity, aimed at creating value for the customers,
involves the acquisition and consumption of resources including capital. SGA, which tracks the
flow of capital to various business activities, thus can be viewed with a value chain lens. Value
chain framework divides the chain of activities into two main categories: primary activities and
support activities. Primary activities include activities related to inbound logistics, operations,
outbound logistics, marketing & sales, and service. Support activities include activities related
to procurement, human resource management, technological development, and infrastructure.
Below, we have categorized the 29 SGA items into relevant primary and support activities in
the value chain framework.
Primary activities along with relevant SGA items:
 Inbound logisticsa
 Operations
1. Operating expenses when a separate Cost of Goods Sold (COGS) figure is given and no
Selling, General, and Administrative Expense (SGA) figure is reported
2. Research and development (R&D) expense
3. Amortization of R&D costs
4. R&D companies’ company-sponsored R&D
 Outbound logistics
5. Delivery expenses
6. Freight-out expense
 Marketing & sales
7. Advertising expense (ADV)
8. Commissions
9. Marketing expense
 Servicea
Support activities along with relevant SGA items:
 Procurementa
 Human resource management
10. Directors’ fees and remuneration
11. Financial service industries’ labor, occupancy and equipment, and related expenses
12. Labor and related expenses (including salary, pension, retirement, profit sharing,
provision for bonus and stock options, employee insurance, and other employee benefits
when reported below a gross profit figure)
13. Severance pay (when reported as a component of SGA)
14. Stock-based compensation when reported below a gross profit figure

65
 Technological development
15. Engineering expense
 Infrastructure
16. Accounting expense
17. Bad debt expense (provision for doubtful accounts)
18. Corporate expense
19. Foreign currency adjustments when included by the company
20. Indirect costs when a separate COGS is given
21. Legal expense
22. Parent company charges for administrative services
23. Recovery of allowance for losses
24. State income tax when included by the company
25. Research revenue that is less than 50% of total revenues for 2 years
26. Strike expense
27. Extractive industries’ lease rentals or expense, delay rentals, exploration expense, R&D
expense, geological and geophysical expenses, drilling program marketing expenses,
and carrying charges on nonproducing properties
28. Restaurants’ preopening and closing costs
29. Retail companies’ preopening and closing costs and rent expense
aSGA contains no item that is relevant to this activity.

66
Web Appendix 3. Concepts, Constructs, and Variables
As noted in the main text, the differences between a concept, a construct, and a variable
are critical. A concept is “a bundle of meanings or characteristics associated with certain events,
objects, conditions, situations” (Emory and Cooper 1991 p. 51). A construct, which is relatively
more complex, is “an image or idea specifically invented for a given research and/or theory-
building purpose” (Emory and Cooper 1991, p. 51). Constructs combine two or more simple
concepts, especially if the idea or image intended “to convey is not directly subject to
observation” (Emory and Cooper 1991, p. 51). Precise definitions help clarify and measure both
concepts and constructs. Good definitions in turn must meet the criteria of specificity, clarity,
consistency, and distinctiveness (MacKenzie 2003). Specificity requires that the construct be
defined in “a sufficiently precise manner” (MacInnis 2011, p. 141). Clarity indicates that the
definition is unambiguous. Consistency and distinctiveness demand that the definition is aligned
with prior research and clearly separated from other constructs (MacKenzie 2003). However,
“there are few empirical referents by which to confirm that an operational definition really
measures what we hope it does,” such that “when measurements by two different definitions
correlate well, it supports the view that they are measuring the same concept” (Emory and
Cooper 1991, p. 54). Measurements by two different definitions do not correlate well if one or
both of them is not a true identifier or if different partial meanings of the concepts are being
measured (Emory and Cooper 1991). This caution holds for both concepts and constructs.
Although concepts and constructs are not sharply demarcated and are often used
interchangeably, they both differ markedly from variables. Concepts and constructs operate at the
theoretical level; variables operate at an empirical level. A variable “is a symbol to which
numerals or values are assigned” (Kerlinger 1986 p. 27 cf. Emory and Cooper 1991). Variables
can be manifest and thus directly observable or latent and hypothetical, such that they must be
approximated by manifest variables (Whiteley and Kite 2013). The measurement of constructs
that rely on latent variables thus may suffer from some measurement error due to the
approximation (DeVellis 2012). It is noteworthy that multiple labels may be used in different
contexts to refer to the same entity. As we noted in the main text, when it is referred to as a
construct, SGA conveys a broader sense of operating expenses measured by several manifest
variables. But when it is referred to as a variable, it represents a measure within Compustat that is
manifest in nature and applied to approximate, either partly or fully, one or more constructs.

67
Web Appendix 4. Propensity Score Matching
We conducted propensity score matching to check that the smaller sample size used in
the empirical analysis is representative of the broader sample drawn from Compustat. To
calculate propensity scores [Caliendo and Kopeinig 2008; Dehejia and Wahba 2002], we used
inclusion in our sample 1 as the “treatment” variable. [Note: Observations had been included in
sample 1 solely based on whether information on them was also available in Advertising Age.]
All 465 observations included in the original sample 1 (treated units from hereon) were given
value of 1 whereas the remaining 18,154 observations (control units from hereon) were given
value 0 for the treatment variable. We estimated a probit model with “treatment” as the
dependent variable. Given that we don’t know the predictors determining which observations
undergo treatment, we cast a wide net to capture any potential influence of characteristics such as
revenue, earnings, assets, liabilities, equity, dividends, and expenses (Caliendo and Kopeinig
2008). Specifically, we included the following variables from Compustat as covariates: (1) net
sales, (2) net income, (3) non-operating income, (4) retained earnings, (5) earnings before interest
and taxes, (6) gross total of property, plant, and equipment, (7) total current assets, (8) total
current liabilities, (9) total invested capital, (10) total long-term debt, (11) total treasury stock,
(12) total common equity, (13) common stock, (14) debt in current liabilities, (15) book value per
share, (16) cash on hand, (17) common dividends, (18) pension and retirement expense, and (19)
rental expense.
The region of common support for the calculated propensity scores ranges from 0.006 to
0.8. A total of 13 blocks ensured that the mean propensity score is not different for the two
groups of observations in each block. We conducted a total of 247 t-tests to check that the
explanatory variables were balanced across the two groups within each block (Dehejia and
Wahba 2002).
Next, we followed three different methods of matching based on propensity scores:
nearest-neighbor matching, stratification matching, and kernel matching (Becker and Ichino
2002). First, nearest-neighbor matching is a straightforward method which takes each treated unit
and searches for the observation with the closest propensity score in the control group. Although
all treated units find matches, some of these matches may be poor because for some treated units
the nearest neighbor in the control group may have a very different propensity score. Second,
stratification method divides the common support region of the propensity score into a set of
intervals such that within each interval, the two groups of observations have on average the same
propensity score. It calculates the impact within each interval by taking the mean difference in
outcomes between treated and control observations. Doing so, however, may discard some
treated units if no control is available in their block. Finally, kernel method matches all the
treated units with a weighted average of all controls with weights that are inversely proportional
to the distance between the propensity scores of treated and control observations. Thus, it is a
relatively more robust matching strategy.

68
Table WA.1 Results of Propensity Score Matching
SGA SGA – ADV SGA – R&D
Nearest-neighbor matching
Number treated 465 465 465
Number matched 1061 1061 1061
Difference in average values of two groups

1898

2657 2354
standard error 1154 1539 1308
t-value

1.64

1.73 1.80
Stratification matching
Number treated 359 359 359
Number matched 12598 12598 12598
Difference in average values of two groups

426

1147 1055
standard error 1065 1044 1158
t-value

0.40

1.10 0.91
Kernel matching
Number treated 465 465 465
Number matched 18154 18154 18154
Difference in average values of two groups 661

160 3
standard error 872 870 918
t-value 0.76

0.18 <0.01
In our study’s context, we don’t have a conventional outcome variable conditional on the
treatment. The key variable of our interest, SGA, does not depend on inclusion of an observation
in Advertising Age. We are primarily interested in checking if the data in sample 1 is
representative of the broader data drawn from Compustat when it comes to average value of our
focal variable. We used three different variables, which have been used in construct validation for
marketing spending in our study, as outcome variables in these matching exercises: SGA, SGA –
ADV, and SGA – R&D.
As results in table WA.1 show, there are no significant differences across the two groups
for any of the outcome variables in any of the matching methods. Thus, the evidence indicates
that the smaller sample size used in the empirical analysis is sufficiently representative of the
broader sample drawn from Compustat.

69
Web Appendix 5. Robustness checks for marketing spending validation
MTMM
Median splits 1 2 3 4
SGA
ADV SGA – ADV ADV SGA – R&D
ADV SGA – ADV – R&D ADV
SGA/Sales Low
X X X X
High
X X X X
ADV/Sales Low
X X X X
High
X X X X
R&D/Sales Low
X X X X
High
X X X X
Goodwill/Sales Low
X X X X
High
X X X X
Other intangibles/Sales Low
X X X X
High
X X X X
Assets/Sales Low
X X X X
High
X X X X
Notes: The results of similar robustness checks for the validation of marketing assets indicate that neither ADV nor SGA (or its
modifications) sufficiently capture the construct or its subconstructs (perceptual and intellectual assets).

70
Web Appendix 6. Correlations of SGA with other variables from Compustat, 1997 to 2015 (N = 20,365 listwise)
Mean S. D. Min. Max. 1 2 3 4 5 6 7 8 9 10
1 SGA 898 3561 .14 96915 1
2 ADV 87 370 0 6144 .70 1
3 R&D 152 772 0 12282 .65 .69 1
4 PR 32 199 0 6795 .66 .60 .56 1
5 RENT 56 193 0 5025 .74 .58 .41 .67 1
6 COGS 2521 13391 0 355913 .80 .61 .45 .57 .65 1
7 SGA – ADV 811 3313 .12 94415 .99 .64 .62 .64 .74 .79 1
8 SGA – R&D 746 3116 .14 96915 .98 .63 .49 .62 .75 .80 .99 1
9 SGA – PR 865 3433 .12 95763 .99 .69 .64 .63 .73 .80 .99 .98 1
10 SGA – RENT 842 3419 .02 94415 .99 .70 .65 .65 .72 .80 .99 .98 .99 1
Notes: COGS denotes cost of goods sold; PR denotes pension and retirement expenses; and RENT denotes rental expenses. All correlation
coefficients are significant at .01 level.

71
Bibliography for Web Appendices
Becker, S. O. & Ichino, A. (2002). Estimation of average treatment effects based on propensity
scores. The Stata Journal, 2(4), 358–377.
Caliendo, M. & Kopeinig, S. (2008). Some Practical Guidance for the implementation of
propensity score matching. Journal of Economic Surveys, 22(1), 31–72.
Dehejia, R. H. & Wahba, S. (2002). Propensity score matching methods for nonexperimental
causal studies. Review of Economics and Statistics, 84(1), 151–61.
DeVellis, R. F. (2012). Scale Development. In R. F. DeVellis (Ed.). SAGE Publications.
Emory, W., & Cooper, D. R. (1991). Business Research Methods. In W. Emory & D. R. Cooper
(Eds.). Homewood: Irwin.
Kerlinger, F. (1986). Foundations of behavioral research. In F. Kerlinger (Ed.). Fort Worth, TX:
Harcourt Brace Jovanovich.
MacInnis, D. J. (2011). A framework for conceptual contributions in marketing. Journal of
Marketing, 75, 136–54.
MacKenzie, S. B. (2003). The dangers of poor construct conceptualization. Journal of the
Academy of Marketing Science, 31, 323–326.
Porter, M. E. (1985). Competitive advantage. New York: The Free Press.
Whitely, B. E., & Kite, M. E. J. (2013). Principles of research in behavioral science. In B. E.
Whitely & M. E. J. Kite (Eds.). New York and London: Routledge.