Corporate Reputation Review Volume 11 Number 4

Academic Research

Predicting Firm Reputation Through Content

Analysis of Shareholders’ Letter

John Geppert
Finance Department, University of Nebraska – Lincoln, Lincoln, NE, USA

Janice E. Lawrence
School of Accountancy, University of Nebraska – Lincoln, Lincoln, NE, USA
Program in Business Ethics, University of Nebraska – Lincoln, Lincoln, NE, USA

ABSTRACT INTRODUCTION
In this study, we use content analysis to meas- ‘A CEO’s words are powerful storytelling
ure the usefulness of the narrative content of the tools, fashioning opinions and offering an
chairman’s letter to the shareholders as a pos- important point of view’ (Amernic and
sible proxy for corporate reputation. Firms with Craig, 2007). Such begins a recent paper
both high and low corporate reputations are pointing out the power of CEO-speak.
identified using three prominent rating services. The CEO’s letter to shareholders is imme-
Using content analysis, significant variations in diately accessible to billions on corporate
word choice, word length and word variety are websites. According to the authors, this
found between these two groups. The firms with increased corporate scrutiny has created
a high corporate reputation use less varied, demands for strategies to communicate
shorter and more concrete words than the low clearly with a consistent message. Their sug-
reputation firms. These high reputation firms gestions to improve the words CEOs use
concentrate on realism with a matter-of-fact type in communicating with the public are an
of style. Based on these variables, a logit regres- indication of the belief that language can
sion successfully classifies 81 percent of our study manipulate attitudes.
firms according to perceived reputation. The re- Corporate reputation, while communi-
sults are further tested by matching unranked cating valuable information, is also not un-
corporations to our study firms. The logit values ambiguously quantified. Therefore, any link
for the matched firms are statistically different discovered between management communi-
from those of the study groups. We attribute cations and corporate reputation is useful.
this to content differences in the letter to the Previous research evidence suggests that dis-
shareholders. This significant association be- cretionary narrative disclosures, such as the
tween narrative content and corporate reputation president’s or chairman’s letter to the share-
suggests that management uses the shareholders’ holders, while not receiving much attention,
letter as a tool of reputation management. contain important and useful information
Corporate Reputation Review (2008) 11, 285–307. (Abrahamson and Amir, 1996; Abrahamson
doi:10.1057/crr.2008.32 and Park, 1994; Smith and Taffler, 2000).
The content of the shareholders letter is Corporate Reputation Review,
KEYWORDS: CEO letter; content analysis; not mandated or regulated. Management Vol. 11, No. 4, pp. 285–307
© 2008 Palgrave Macmillan,
reputation; shareholder letter controls the content of these disclosures, so 1363-3589

www.palgrave-journals.com/crr/ 285
 Predicting Firm Reputation Through Content Analysis of Shareholders’ Letter

these disclosures are directly influenced by observing a corporation’s (management’s) re-

the image management wants to convey.The sponse to items such as environmental impact,
objective of this study is to apply content employee concerns, stockholder inquiries and
analysis techniques to determine whether other aspects of social responsibility impacting
the language used in the shareholders’ letters the firm. Bennett and Kottasz (2000) find at
is systematically different between high and least 16 different definitions of corporate
low reputation firms. reputation in the literature.
Existing independent rankings of per- Barnett et al. (2006) note that, while there
ceived corporate reputation are used to iden- has been an explosion of research in corpo-
tify companies with high and low corporate rate reputation, a precise definition of cor-
reputations. Content analysis of letters to the porate reputation remains lacking. In an
shareholders allows us to determine any sig- attempt to provide a common vernacular,
nificant differences in the letters between they use a cluster analysis method to iden-
companies with high and low rankings. tify three distinct clusters of corporate repu-
We then determine whether discretionary tation definitions.These are labeled awareness,
communications are consistent with formal assessment and asset value. The key feature of
measures of reputation, in the sense that they definitions falling under the awareness cluster
are predictive of the reputation ratings. We is the lack of associating a positive or nega-
find that linguistic style does vary across high tive value with the firm on any dimension.
and low reputation firms and that the dif- The concept underlying awareness is simply
ferences can be used to correctly classify that stakeholders are familiar with the firm’s
firms as high or low reputation with 81 per- existence. The second cluster, assessment, is
cent accuracy. Our findings provide guidance chiefly characterized by definitions pertain-
to chairmen in crafting their writing styles ing to evaluation, judgment and references
so that they can achieve the highest signal to esteem made by firm outsiders (eg, is the
of reputation in their letter to the sharehold- firm perceived as reputable, as a good cor-
ers. In addition, our findings provide evi- porate citizen?). The final cluster of defini-
dence to researchers on linguistic factors that tions, asset value, refers to reputation as
are associated with reputation. something of market value to the firm, such
The remainder of this paper is organized as brand name capital.
as follows. The second section presents the We posit that the reputation concepts
theoretical framework and develops the hy- most likely influenced by the chairman’s let-
potheses. The third section describes the re- ter relate to the assessment and asset value
search method, followed by a discussion of clusters. Since the chairman’s letter is distrib-
the results in the fourth section. The fifth uted to the shareholders in the company’s
section examines the impact of financial per- annual report, the target audience is already
formance on the predictive power of the familiar with the company’s existence and,
content of the shareholders letter, and the therefore, awareness is not the chief reputa-
final section presents the discussion, implica- tion concept influenced by the letter. Barnett
tions and conclusions of the findings. et al. (2006) argue that reputation as an asset
value is a result of a firm’s visible actions and
THEORETICAL FRAMEWORK responses to external events. The asset value
of a firm’s reputation derives from manage-
Corporate Reputation ment of product quality, use of human
Corporate reputation is based on the percep- resources and the resulting financial
tion of multiple factors by those outside the performance. The chairman can influence
company. Reputations can be built through the asset value of a firm’s reputation in his

286 Corporate Reputation Review Vol. 11, 4, 285–307 © 2008 Palgrave Macmillan 1363-3589
 Geppert and Lawrence

letter to the shareholder through two mech- transactions, the letter to the shareholder can
anisms. First, the chairman can highlight the contain forecasts and discussions of informa-
firm’s achievements with a discussion of the tion sometimes described as ‘soft.’ Abraham-
firm’s (recent) past actions and their results. son and Amir (1996: 1159) claim the letter
Secondly, the chairman can affect the asset is interesting because of this ‘relative freedom
value aspect of the firm’s reputation simply in choosing the information reported … and
by announcing that the firm will undertake the lack of restrictions on the way the in-
some investor desired activity, even if the formation is reported.’ Their results suggest
activity is not subsequently implemented. that the letter contains useful information
Westphal and Zajac (1995, 1998) refer to this about the corporation’s future and not just
type of behavior as ‘symbolic management’ past performance. For example, Smith and
of investors. The letter to the shareholders is Taffler (2000: 625) find the letter is highly
one avenue available to the CEO for such associated with future bankruptcy. Their
symbolic management. research design uses a z-score based on
In contrast to asset value, which looks at financial information to sort the firms into
actions and outcomes, assessment reputation bankrupt and nonbankrupt firms. Then, us-
is nearly synonymous with corporate image. ing a lexicon of 168 words with evaluative,
In managing corporate image, it is important potency, activity and manageability thematic
that the public perceive that the firm’s com- components, Smith and Taffler estimate a
munications are reliable. Greyser (1999) ar- stepwise Fisher discriminant analysis on the
gues that central to the link between two groups. The majority of the model’s
company behavior and reputation is firm explanatory power is attributed to words
credibility. Credibility requires stakeholders related to financial variables, including prof-
to perceive that the message conveyed is it and loss references. They find, however,
transparent (and implicitly truthful). Chris- that nonfinancial statement words also con-
tiansen (2002) argues that firms are expect- tribute significantly to the classification. As
ed to contribute to transparency and a such, they conclude ‘such unaudited narra-
primary conduit for transparency is com- tive disclosures contain important infor-
munication with stakeholders. The chair- mation associated with the future of the
man’s letter has the potential to affect the company and are not just reporting on past
perceived credibility and transparency of the performance.’ Because the Smith and Taffler’s
firm through the chairman’s language style research design initially uses financial ratios
and word choice in accordance with the as- to form the bankrupt and nonbankrupt
sessment mechanism of reputation. firms, it does not allow direct testing of the
marginal contribution of the word choice
Letter to the Shareholders over financial ratio information in the
The chairman’s letter is a unique component same regression. In a separate study, how-
of a firm’s annual report. The letter to the ever, Tennyson et al. (1990) specifically
shareholders (referred to, henceforth, as the test whether the narrative information con-
letter) is an unaudited narrative. It is not a tains incremental explanatory information
required document and, therefore, its con- on bankruptcy that is not captured by finan-
tent is not regulated. There are no standards, cial ratios.Tennyson et al. (1990) include nar-
regulations or content requirements for the rative information along with five financial
letter.The CEO is free to select the informa- ratios in a logit regression predicting bank-
tion included in the letter and the manner ruptcy. They find that the narrative variable
in which it is presented. Unlike the financial remains significant even in the presence of
statements, which present the results of past the financial variables. We perform a similar

© 2008 Palgrave Macmillan 1363-3589 Vol. 11, 4, 285–307 Corporate Reputation Review 287
 Predicting Firm Reputation Through Content Analysis of Shareholders’ Letter

analysis and discuss the relationship between We anticipate the answer to these research
financial performance and reputation in a questions will be yes. Yet, as this is an ex-
later section. ploratory study, we hold no other precon-
The chairman has substantial influence ceptions about the results. These hypotheses
over the content of the letter. Ingram and are posited in the null form as follows:
Frazier (1983: 49) view the annual report as
a platform for preaching (management’s)
H1o: The content of the shareholders’ letter of
philosophies. Michalisin et al. (2004) point
corporations ranked highly in terms of rep-
out that management commonly uses public
utation does not differ from the content of
relations firms to help create the content of
letters of corporations with low rankings.
the annual report text. But Salancik and
Meindl (1984: 243) find evidence that the H2o: The content of the letter to the sharehold-
same individual chairman’s writing style and ers is not useful in classifying corporations
word use remains similar when employed at according to their reputation rankings.
different companies. This implies direct in-
volvement and influence on the letter’s style
and content by the chairman. METHODS
We posit that the CEO is the ultimate
spokesman for a company and that he or she Data Sources
projects a ‘tone’ for the company with respect The heterogeneous nature of definitions of
to factors such as customer relations, environ- corporate reputation outlined by Barnett
mental impact and social responsibility. Given et al. (2006) makes it difficult, if not impos-
Salancik and Meindl’s findings about the sta- sible, to unambiguously classify a firm as high
bility of a CEO’s writing style and word or low reputation. There are aspects of rep-
choice, we believe that these writing style dif- utation, however, that are shared by three of
ferences directly impact the firm’s perceived the most popularly cited ratings of corporate
credibility and transparency and are associated reputation and we discuss below why we
with firm’s assessment reputation. believe that these shared features capture
A significant association between the nar- the assessment and asset value aspects of a
rative content of the shareholder’s letter and firm’s reputation.
corporate reputation would be consistent The classification of firms in the study
with the hypothesis that management uses the group as having either high or low reputa-
letter to the shareholders as a tool to affect tion comes from three different sources:
the asset value and assessment value of firm (1) Fortune Magazine’s Most Admired Com-
reputation through symbolic management, panies 2002, (2) Harris Interactive 2002
perceived credibility and transparency. Corporate Reputation Survey and (3) Busi-
ness Ethics Magazine’s 100 Best Corporate
Hypotheses Citizens Report. These classifications of
This study will examine two research ques- firm reputation and responsibility are based
tions: (1) Can content analysis of the CEO’s on perceived ratings of corporations on
letter to the shareholders identify significant varying dimensions. These rankings have
differences between common characteristics been used in previous research (Stanwick, A
of corporations ranked highly (or lowly) and Stanwick, SD, 2003).
as to reputation? (2) Can the significant Fortune Magazine’s Most Admired Com-
differences (if any) be used to differentiate panies study surveys over 16,000 top finan-
between corporations ranked highly (or cial analysts, executives and directors from
lowly) as to reputation? eligible companies (with at least $1.3bn in

288 Corporate Reputation Review Vol. 11, 4, 285–307 © 2008 Palgrave Macmillan 1363-3589
 Geppert and Lawrence

revenue) to identify corporations with the environment, Social responsibility, Financial per-
strongest reputations. Raters are asked to formance and Vision and leadership.
evaluate each firm on each of eight attributes Business Ethics magazine evaluates corpo-
on a scale of zero (‘Poor’) to ten (‘Excellent’). rations’ ethical responsibility each year and
The eight attributes are: (1) Innovativeness, (2) lists the 100 best citizens. This ranking is
Employee Talent, (3) Use of Corporate Assets, (4) based on the firms’ responsibility toward a
Social Responsibility, (5) Quality of Manage- variety of stakeholders including the envi-
ment, (6) Financial Soundness, (7) Long-term ronment and the community. They start
Investment Value and (8) Quality of Product/ with the KLD Research and Analysis, Inc.
Services. Social responsibility affects the wid- rankings of social commitment to different
est number of stakeholders outside the stakeholder groups. (KLD evaluates the
company and is the source of ratings used Russell 1,000 and the Domini stock index.)
in this study. To these KLD evaluations, Business Ethics
Harris Interactive’s Corporate Reputation adds its own review of a firm’s treatment
Survey is our second data source. Harris In- of the following: environment, community,
teractive is an online market-research firm employees, diversity (of employees and board)
which developed the survey and study in and customers. These evaluations are used to
conjunction with the Reputation Institute. calculate a numerical score for each firm.
These rankings are based on an annual on- Table 1 describes how the asset value/
line survey of people’s perceptions of cor- assessment aspects of corporate reputation
porate reputations. Members of the public are included in the rankings by these three
first nominate firms for both best and worst sources.1 The asset value focus of reputation
categories. Detailed online surveys are then is captured by criteria related to financial
sent to over 21,000 members of the public performance, product quality and use of
to rate and rank the 60 firms most frequent- human resources. The survey concepts related
ly nominated. Over 600 respondents rate to these three factors for each rating firm are
each company on 20 attributes grouped into shown in Panel A of Table 1. The asset value
six factors such as Emotional appeal (trust in aspect of corporate reputation is emphasized
the company), Product and services, Workplace in both the Fortune and Harris rankings.

Table 1: Aspects of Reputation Used in Rankings

Fortune Magazine Harris Business Ethics

Panel A: asset value aspects used

Financial performance Use of corporate assets, financial Financial performance None
soundness, long-term investment
value
Product quality Quality of product/services Product and services None
Human capital Innovativeness, employee talent, Vision and leadership None
quality of management

Panel B: assessment aspects used

Judgment and Social responsibility Emotional appeal Environment,
references to esteem (trust in the company), community,
workplace environment, employees,
social responsibility diversity

© 2008 Palgrave Macmillan 1363-3589 Vol. 11, 4, 285–307 Corporate Reputation Review 289
 Predicting Firm Reputation Through Content Analysis of Shareholders’ Letter

The assessment aspect of corporate reputa- will be labeled ‘high firms,’ while the ‘low
tion is included in criteria used by all three firms’ label will be applied to those listed as
rating sources, although it is emphasized in having low corporate reputation.
the Business Ethics rankings. We believe that High Reputation Sample. To identify firms
the word choice of the chairman’s letter with an extremely high reputation, we search
can affect the asset value aspect of reputation first for firms with repeat listings in the top
through symbolic management and the 100 Best Corporate Citizens, as rated by Busi-
assessment aspect of reputation by affecting ness Ethics. We identify five firms that are listed
the perceived credibility and transparence in the top 100 Best Corporate Citizens three
of the firm. years in a row. These are included in our study
set as high reputation firms. (Note that for
The Data three of these firms we include the 2000 and
The study group consists of 39 firms, each 2001 letter, resulting in a total of eight letters
classified as either ‘high’ or ‘low’ reputation from Business Ethics magazine’s listings.)
firms. The classifications sources of the data We chose the Social Responsibility at-
are summarized in Table 2. tribute on Fortune’s Most Admired Compa-
The determinants of what constitutes a nies list because it measures a firm attribute
high reputation firm are tenuous at best, so that impacts a broad measure of stakeholders.
there will necessarily be disagreement with We believe that firms that are viewed as good
any classification scheme. It is easier to find corporate citizens will be looked upon fa-
firms classified as high, rather than as low, vorably and, thus, have a high reputation.The
reputation firms. Presumably for legal reasons, top ten ranked firms in Social Responsibil-
it is more common to find published lists of ity are included in our study.
firms that are deemed to be admired or with The top ten corporations from the Harris
a high corporate reputation than it is to find Interactive Survey are identified. After elim-
firms judged in a published list as unethical inating firms also listed by another source,
or having a low corporate reputation. For this seven of these firms remain and are includ-
research study, the firms listed, according to ed in our high reputation firm sample.
the three sources described above, as socially Low Reputation Sample. The sample of low
responsible with high corporate reputations reputation firms includes seven of the bottom
ten firms on Fortune’s Social Responsibility
ratings and seven of the bottom ten on Harris’s
Table 2: Classification Sources for
Sample Firms reputation survey. The letter is not obtainable
for the other bottom ranked firms. Business
Classification source No. of high No. of low Ethics magazine lists only the top firms and,
reputation reputation therefore, does not identify any firms we can
firms firms include in the low reputation sample.
The above procedure results in a final
Fortune’s Most Admired 10 7 study sample of 39 firms, 25 firms viewed
Companies of 2002 as extremes at the high end of the responsible/
Harris Interactive 2002 7 7 reputation spectrum and 14 firms viewed as
Corporate Reputation extremes at the low end of the responsible/
Survey reputation spectrum.
Business Ethics Corporate 8 0
Responsibility Report
Content Analysis
Total in sample 25 14 We perform a content analysis to identify
any systematic differences in the letter

290 Corporate Reputation Review Vol. 11, 4, 285–307 © 2008 Palgrave Macmillan 1363-3589
 Geppert and Lawrence

between firms classified as high reputation quality of a text in terms of the richness of
firms and those classified as low. its vocabulary.
Content analysis covers a wide array of The variety index is the total number of
topics and techniques. Roughly speaking, different words used in a text divided by the
content analysis is a structured process used total number of words used in the text. A
to find patterns that occur in any form of high score indicates a speaker’s avoidance of
communication. Examples include counting overstatement and a preference for precise,
the number of violent acts in a movie, molecular statements. In addition, Carpenter
looking at trends in color choice in print (1986) argues that apprehensive or cautious
advertisements or identifying isolationist demeanor leads to a greater word variety use.
themes in a political candidate’s speeches. Several studies (Zhou et al., 2002; Colwell
For our analysis, we use the narrative of et al., 2002; Dulaney, 1982) find that a
the letter found in the annual report of higher variety index is associated with de-
each corporation. ception. Conversely, a more relaxed style
Our DICTION 5.0 software analysis is results in a lower variety index and is inter-
performed on only the text of the letter. preted to be more truthful. We posit that the
Graphics and tables are ignored. Numbers variety index will be related to corporate
referred to in the text are included in the reputation through the assessment reputation
analysis with the idea that if specific numbers mechanism since the variety index is associ-
are being mentioned, the chairman believes ated with deception and thus transparency.
these are in some way different from num- Recall that Greyser (1999) states that cred-
bers summarized in a table. The analysis is ibility is the central link between company
conducted at the individual word level. That behavior and public perception. Perceived
is, grammatical classifications such as clauses credibility of the chairman may, thus, be en-
or parts of speech are ignored. The DIC- hanced by using a more relaxed writing style
TION 5.0 software assigns each word to a and a lower variety index.
concept classification or theme.These themes Table 3 presents a summary of the variety
are described in the following section. In indices for the high and low firm categories.
content analysis literature, this approach falls We test whether, on average, the letter from
under the heading of conceptual analysis, the firms with the best reputations use more
which measures the appearance and frequen- or less variety of words than the firms with
cy of concepts represented by individual the worst reputations. Table 4 presents a t-test
words. This word level approach has been of the equality of means for the variety indi-
useful in prior research on truthfulness (Bond ces for the two groups. We find the high
and Lee, 2005; Newman et al., 2003). reputation firms use a more homogeneous
word choice within each annual report. Their
RESULTS average variety index is 0.42 compared to
0.50 for the low reputation firms. The great-
Variety Index er variety index for the low reputation firm
There are 4,737 unique words that are used annual reports may reflect a tendency for low
at least once in a letter by at least one of reputation firms to use different terms
the firms. The first feature we calculate in throughout the text for the same concept.
the content analysis of each annual report is Alternatively, the low reputation firms may
a variety index. The variety index was one be less focused in their narrative and cover
of the first measures used in content analy- more topics. The variety index would be
sis to determine the heterogeneity of a text’s high in both cases but it is not a refined
wording. It is a linguistic indicator for the enough measure to distinguish between the

© 2008 Palgrave Macmillan 1363-3589 Vol. 11, 4, 285–307 Corporate Reputation Review 291
 Predicting Firm Reputation Through Content Analysis of Shareholders’ Letter

Table 3: Variety Indices

Low reputation firms High reputation firms

Range Mean Range Mean

Total word count 210–1113 768.79 69–1510 1024.26

Total no. of different words 141–514 361.29 276–614 427.65
Variety index 0.39–0.67 0.50 0.30–0.52 0.42

Table 4: t-Test of Greater Word Variety of TION 5.0 uses a lexicon of 10,000 words to
Low Reputation Firms divide a text into five major semantic features:
Activity, Optimism, Certainty, Realism and
t-Test: two sample assuming Low High Commonality. These five features are com-
unequal variances reputation reputation posed of combinations of 35 sub-features.
firms firms
Table 5 lists the DICTION 5.0 classifications.
DICTION 5.0 analyzes texts in 500 word
Mean variety index 0.50 0.42
blocks. The resulting DICTION score repre-
Variance 0.01 0.00
Observations 14 25 sents the number of times each word (per
500 word text length) from 1 of the 35 sub-
Hypothesized mean 0 features appears in the text. These sub-feature
difference scores are then aggregated to form the five
d.f. 21 major thematic categories. The aggregation
t Stat 2.81 process is simply the sum of various sets of
P(T ⭐ t) one-tail 0.00 the sub-features. According to Li (2006), very
t Critical one-tail 1.72 few studies examine the text of publicly avail-
P(T ⭐ t) two-tail 0.01 able corporate documents; instead the major-
t Critical two-tail 2.08 ity of the analysis has been on the quantitative
variables released by the companies.While the
content of the letters appears similar with re-
two cases. We interpret both characterizations spect to the topics covered, we hypothesize
as lending to the reader’s perception that there that there will be subtle differences in the style
is less truthfulness in the text and thus a less and tone of the letters.We use DICTION 5.0
favorable reputation for the firm. software because it was designed to identify
and isolate aspects of linguist tone. At this stage
Content Analysis Based on Thematic of our research, we do not have a specific
Scores hypothesis on what dimensions the letters of
The variety index only considers a writing low and high reputation firms will differ.
style feature and not a thematic characteristic. We perform t-tests on the average DIC-
To summarize the thematic character TION scores to discover any differences in
of each text, we use the DICTION 5.0 the themes of letters from the high and low
Text-Analysis software to classify the words of reputation firms. Table 6 shows the results of
each letter into thematic categories. The fol- the t-tests. For the 35 sub-features, the t-test
lowing category descriptions are based shows significant differences between high
the DICTION 5.0 Text-Analysis software, and low reputation firms in the themes of
version 5.0. Refer to the users’ manual for ‘Inspiration,’ ‘Present Concerns,’ ‘Concrete-
additional information (Hart, 2000). DIC- ness,’ ‘Variety’ and ‘Complexity.’ For the five

292 Corporate Reputation Review Vol. 11, 4, 285–307 © 2008 Palgrave Macmillan 1363-3589
 Geppert and Lawrence

Table 5: DICTION 5.0 Thematic Categories and Sub-Categories

Major categories
Activity Optimism Certainty Realism Commonality

Sub-features
Numerical terms Satisfaction Cognition Concreteness Liberation
Ambivalence Inspiration Passivity Past concern Denial
Self-reference Blame Spatial terms Centrality Motion
Tenacity Hardship Familiarity Rapport
Leveling terms Aggression Temporal terms Cooperation
Collectives Accomplishment Present concern Diversity
Praise Communication Human interest Exclusion

Calculated sub-features
Insistence Embellishment Variety Complexity

major categories, the only significant differ- to the performance interpretation, we believe
ence is in the ‘Realism’ score. the Present Concerns score, along with the
The ‘Inspiration’ score is determined by the ‘Concrete’ score, indicates a straightforward,
presence of words from a set of 122 words. matter-of-fact style in the narrative of high
These words convey desirable characteristics reputation firms that is unrelated to perform-
such as honesty, virtue and self-sacrifice; qual- ance. As an example of a matter-of-fact style,
ities such as courage, mercy and dedication; one high reputation firm’s report states:
and ideals such as patriotism, success, education
As a leading provider of low-cost mort-
and justice. These words constitute 1.82 per-
gage capital for home buyers to finance
cent of the words used by the group of high
their homes, our firm is at the center of
reputation firms compared to 1.37 percent
the housing industry, one of the strongest
used by low reputation firms. The ‘Present
growth sectors in America.
Concerns’ score is based a set of 269 words,
mostly present tense verbs. The words are not The statement is clear, easy to interpret and
topic specific, but relate to general activity.The to the point.2
high reputation firms use more of these terms The most dramatic difference in the
than the low firms (1.75 per 500 words vs sub-categories between high and low repu-
1.40). It may be possible that high reputation tation firms is with the ‘Concreteness’
firms use present tense because they feel they score. The Concreteness score is derived
can state what they do, whereas low reputation from a set of 745 words with no thematic
firms implicitly accept that they are not very similarity other than tangibility and
successful by implying what they are attempt- materiality.
ing to do. In our sample, the low reputation The low reputation firms score higher in
firms, however, have current return to five year the Variety and Complexity features. These are
trailing average return that are larger than their features of the words or text rather than the-
matched counterparts and larger than the high matic characteristics.We have already discussed
reputation firms. As a result, instead of omitting that the low reputation firms use a greater
any mention of the past, these low reputation variety of words and the DICTION 5.0 anal-
firms should contrast the past and current ysis reconfirms this feature (a score of 0.42 vs
results, noting their improvement. In contrast 0.50). The ‘Complexity’ score is simply the

© 2008 Palgrave Macmillan 1363-3589 Vol. 11, 4, 285–307 Corporate Reputation Review 293
 Predicting Firm Reputation Through Content Analysis of Shareholders’ Letter

Table 6: t-Tests of Thematic Categories with Significant Differences between High and
Low Reputation Firms

DICTION 5.0 Average high Average low p-Value, two sided t, p-Value, two sided t,
category equal variance unequal variance

Sub-features
Inspiration 1.82 1.37 0.09 0.08
Present concern 1.75 1.40 0.08 0.06
Concreteness 4.12 2.74 0.02 0.01
Variety 0.42 0.50 0.00 0.01
Complexity 5.09 5.25 0.09 0.06

Major category score

Realism 46.46 44.45 0.01 0.00

average number of characters per word. On With the exception of the Human inter-
average, the low reputation firms use longer est score, the high reputation firms score
words and words with less concrete (more higher in each of the sub-features that con-
than one) meanings. Zhou et al. (2002) posit tribute positively to the Realism score (the
that the high variety index associated with first bracketed term). In addition, the high
deception results from superfluous or mean- reputation firms score lower on the sub-
ingless language when the writer wants to features that contribute negatively to the
give the impression of completeness. For Realism score (the second bracketed term).
example, one low reputation firm’s report A qualitative summary of the DICTION
states: 5.0 content analysis indicates that the set of
high reputation firms on average use less
… we remain committed to being good
variety of words, shorter words and more
stewards of our asset base and to taking
concrete words than the set of low reputa-
action regarding under-performing assets
tion firms. These firms concentrate on real-
wherever possible.
ism with a matter-of-fact type of style.
The meaning of the sentence is difficult to To help validate the automated procedure,
determine and it seems to be mere filler, we had all of the letters read by a knowl-
rather than useful information. edgeable, but unbiased (unaware of the firm
The only major thematic score that is dif- classification results or our intended use of
ferent for the two groups is the ‘Realism’ her judgments) reader. The letters of the
category. High reputation firms score higher sample firms were randomized. The reader
for the Realism feature. The Realism theme was then asked to read each letter and
represents tangible, immediate concerns of indicate on a five-point Likert scale her
every day life.The realism score is an additive perception of the following three issues:
function of eight of the sub-features where (1) the extent of word variety used, (2) the
concreteness of the chosen words (in terms
Realism = [Familiarity + Spacial of multiple meanings) and (3) the complex-
ity of the writing. We find that the human
+ Temporal + Present concern
evaluator did not rate high and low reputa-
+ Human interest + Concrete] tion firms statistically different on the vari-
−[ Past concerns + Complexity] ety and concreteness dimensions. Her rating

294 Corporate Reputation Review Vol. 11, 4, 285–307 © 2008 Palgrave Macmillan 1363-3589
 Geppert and Lawrence

of complexity, however, matches the auto- The logit model correctly classifies 11of
mated procedure. We find that she rates the the 14 firms a priori classified as low reputation
writing style of low reputation firms as more firms (coded as 0) and misclassifies 3. Of the
complex and the difference is statistically 25 firms a priori classified as high reputation
significant at the 10 percent level.3 firms (coded 1), the logit correctly classifies
These results allow us to reject the null 21 and incorrectly classifies 4.The overall cor-
H1 and conclude that the content/writing rect classification performance is 82.1 percent.
style of the letter to shareholders of corpo- A naïve classification would be only 64.1 per-
rations ranked highly in terms of corporate cent correct (always classify as high reputation
reputation does differ significantly from the (25/39)). Henriksson and Merton (1981) de-
content/writing style of letters of corpora- velop a metric for evaluating binary classifica-
tions with low rankings. tions. Briefly, their method as it applies to our
study is as follows:
Classification of Firms Based on Let:
Thematic Content
The next stage in our content analysis is to see N1 = number of High reputation firms = 25
whether we can use the sub-feature and the-
matic differences to classify firms as high or N2 = number of Low reputation firms = 14
low reputation using a logit regression model. n1 = number of correct classifications of High
For the logit model, the dependent variable is reputation firms = 21
the 0/1 classification of each firm, 1 for high
and 0 for low reputation firms. From the pre- n2 = number of incorrect classifications of Low
vious comparisons, we use the ‘Variety,’ ‘Com- reputation firms = 3
plexity’ and ‘Concreteness’ indices as
independent variables in the logit. The param- Table 8: Logit Classification Results
eter estimates are given in Table 7. The esti-
mated signs are as expected from the previous Predicted Predicted Predicted
results. Higher Variety and Complexity scores low high correctly
lead to a greater probability of a low reputation reputation reputation (%)
classification, while a higher Concrete score
leads to a greater probability of a high reputa- Actual low 11 3 78.6
tion classification. Firms with logit values of reputation
Actual high 4 21 84.0
greater than 0.5 are classified as high reputation
reputation
firms. Those with logit values of less than 0.5
are classified as low reputation firms. Table 8 The cut value is 0.5
reports the classification results. Overall correct prediction: 82.1 percent

Table 7: Logit Estimates

Variables B SE Wald d.f. Significance Exp(B)

Step 1a: variety − 35.52 16.08 4.88 1 0.03 0.00

Complex − 7.88 3.17 6.17 1 0.01 0.00
Concrete 0.60 0.35 2.97 1 0.09 1.82
Constant 59.71 23.83 6.28 1 0.01 8.6E+25
a
Variables entered on Step 1: variety, complex, concrete

© 2008 Palgrave Macmillan 1363-3589 Vol. 11, 4, 285–307 Corporate Reputation Review 295
 Predicting Firm Reputation Through Content Analysis of Shareholders’ Letter

6 and not variations in the economic environ-

Low High
5 ment, industry membership or firm size, we
use a matched pairs approach. A control group
4 Misclassified High Misclassified Low
of firms is created by selecting a matching
# of Firms 3 firm for each of our study firms from
2 COMPUSTAT. These control group firms
1
are unranked and are matched to the study
firms on sales (for size), the three-digit SIC
0
0.00 0.17 0.33 0.50 0.67 0.83 1.00
code (for industry), and date (for economic
Logit Value environment) to control for these variables.
We use the estimated logit model to clas-
Figure 1: Logit classification
sify the matched control group. We have no
information available on the reputation of the
A measure of classification accuracy is control group and, thus, no a priori reason to
given by believe these matched firms will fall into high
or low reputation categories in any system-
n1 N 2 − n2 atic way. Consequently, the distribution of
K= +
N1 N2 logit scores for the control-matched firms
should be uniformly distributed on 0–1.
The logit predictions for all of the firms
K varies from 0 (always wrong) to 2 (perfect (study high, low and matched control) groups
classification). can be found in Tables 9a and 9b; those firms
The naïve classification (always classify classified as high are in Table 9a with those
high) leads to a value for K of K = (25/25) classified as low in Table 9b.
+ (14 − 14)/14 = 1 or always classify low as Figure 2 presents the actual distribution
K = (0/25) + (14 − 0)/14 = 1. of the logit values for the study high reputa-
The logit results in K = (21/25) + (14 − 3)/14 tion firms and their matching control firms.
= 1.63 or an improvement of 63 percent The 45-degree line illustrates the distribu-
over a naïve classification. Figure 1 reports tion from a uniform variate. Ideally, the high
the details of the classification. reputation firms would all have logit values
The high reputation firms are much of 1.0 (or at least values greater than 0.5 to
skewed toward the right indicating extreme be correctly classified). Figure 2 shows that
separation between high and low reputation 84 percent are classified correctly, as only 16
firms. The low reputation firms are more percent of the high reputation firms have
evenly distributed in the less than 0.5 cutoff logit values less than 0.5. The matched firms
range. The highlighted portions indicate the also have a tendency toward logit values
logit misclassifications. greater than 0.5, but the likelihood is not as
great as in the sample of high reputation
Comparison of Logit Results for Study firms. Thus, we conclude that the letter’s
and Control Group content does contain information beyond
The logit results indicate that the thematic that which can be attributed to size, industry
content and writing style is significantly dif- membership and economic environment
ferent between our high and low reputation and can be used to classify firms.
firms. These differences can be used to clas- When we examine the distribution of log-
sify the firms with a considerable degree of it values for the sample of low reputation firms
accuracy. To further substantiate that the dif- and their matching firms, we find the differ-
ferences are related to the reputation ratings ence even more pronounced. This is illustrated

296 Corporate Reputation Review Vol. 11, 4, 285–307 © 2008 Palgrave Macmillan 1363-3589
 Geppert and Lawrence

Table 9: Firms Classified by Logit Value as (a) High Reputation and (b) Low Reputation

Firm type Concreteness Variety Complexity Logit value

(a) High reputation

Control firm 4.06 0.48 4.71 1.00
High reputation firm 2.56 0.56 4.28 1.00
High reputation firm 8.41 0.56 4.79 1.00
Control firm 3.40 0.46 4.88 1.00
Control firm 8.49 0.47 5.23 1.00
Control firm 6.36 0.47 5.08 1.00
High reputation firm 2.99 0.48 5.07 0.99
Control firm 3.81 0.48 5.07 0.99
Control firm 3.72 0.53 4.84 0.99
Control firm 3.79 0.52 4.92 0.99
High reputation firm 8.35 0.55 5.13 0.99
High reputation firm 2.44 0.51 4.88 0.99
High reputation firm 4.13 0.55 4.85 0.99
Control firm 5.30 0.54 5.05 0.98
Control firm 2.60 0.51 4.99 0.98
Control firm 2.07 0.52 4.92 0.98
High reputation firm 3.03 0.49 5.13 0.98
Control firm 2.32 0.53 4.91 0.97
High reputation firm 5.36 0.49 5.34 0.97
High reputation firm 4.15 0.52 5.14 0.96
Control firm 6.69 0.51 5.39 0.96
High reputation firm 7.96 0.53 5.40 0.96
Control firm 3.20 0.51 5.14 0.95
High reputation firm 2.16 0.58 4.77 0.94
Control firm 2.88 0.49 5.27 0.92

Low reputation firma 3.43 0.52 5.20 0.91

Control firm 4.58 0.54 5.20 0.91
High reputation firm 3.20 0.53 5.17 0.89
Control firm 3.17 0.53 5.17 0.89
High reputation firm 5.44 0.56 5.22 0.87
High reputation firm 2.96 0.58 4.95 0.87
High reputation firm 4.39 0.56 5.15 0.87
High reputation firm 4.86 0.51 5.42 0.86
Control firm 2.33 0.53 5.14 0.86
High rep firm 4.06 0.60 4.97 0.84
Control firm 2.71 0.53 5.20 0.82
High reputation firm 2.72 0.56 5.10 0.78
Control firm 3.28 0.55 5.19 0.78
Control firm 4.31 0.54 5.33 0.75
High reputation firm 3.31 0.53 5.30 0.75

Low reputation firma 4.29 0.55 5.31 0.71

High reputation firm 4.10 0.62 4.99 0.70
Control firm 1.71 0.66 4.63 0.69

© 2008 Palgrave Macmillan 1363-3589 Vol. 11, 4, 285–307 Corporate Reputation Review 297
 Predicting Firm Reputation Through Content Analysis of Shareholders’ Letter

Table 9: Continued

Firm type Concreteness Variety Complexity Logit value

Control firm 3.20 0.60 5.02 0.68

Control firm 4.05 0.54 5.36 0.67
Control firm 2.48 0.56 5.16 0.66

Low reputation firma 6.65 0.61 5.30 0.55

Control firm 3.18 0.55 5.32 0.54

(b) Low reputation

Control firm 4.40 0.56 5.42 0.44
Control firm 2.62 0.63 4.98 0.42
Low rep firm 1.95 0.57 5.20 0.42
Control firm 2.93 0.55 5.37 0.41
Low reputation firm 1.74 0.65 4.83 0.40
Low reputation firm 2.70 0.50 5.59 0.38

High reputation firma 3.80 0.57 5.36 0.38

High reputation firma 3.47 0.55 5.44 0.35
Control firm 1.93 0.54 5.37 0.35

High reputation firma 2.30 0.55 5.36 0.34

Low reputation firm 3.31 0.55 5.44 0.33
Control firm 5.12 0.66 5.10 0.30
Low reputation firm 2.47 0.62 5.09 0.28
Low reputation firm 2.05 0.54 5.42 0.27
Low reputation firm 1.16 0.60 5.09 0.27
Control firm 0.99 0.61 5.06 0.23
Low reputation firm 2.75 0.59 5.29 0.22

High reputation firma 2.32 0.58 5.35 0.16

Control firm 3.84 0.60 5.41 0.13
Low rep firm 2.68 0.58 5.44 0.10
Control firm 2.20 0.66 5.09 0.07
Control firm 3.36 0.54 5.74 0.06
Control firm 2.86 0.57 5.59 0.05
Low reputation firm 1.27 0.70 4.98 0.03
Low reputation firm 2.03 0.69 5.25 0.01
Control firm 1.91 0.69 5.40 00.0
a
Misclassified firms in bold type

in Figure 3. Approximately 80 percent of the tain information beyond that which can be
low reputation firms have logit values less than explained by variations in size, industry and
0.5, compared to only approximately 40 per- economic environment and can be used to
cent of the corresponding matched firms. classify firms.
Again, based on this difference in distribution, Tables 10 and 11 present the results of a
we conclude that the letter’s content does con- matched t-test on the mean logit scores of

298 Corporate Reputation Review Vol. 11, 4, 285–307 © 2008 Palgrave Macmillan 1363-3589
 Geppert and Lawrence

1.0
0.9
0.8
0.7
Logit Value

0.6 Test High Reputation

0.5
Matched High Reputation
0.4
0.3
0.2
0.1
0.0
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
Percentage of Firms with lower Logit Values

Figure 2: Distribution of logit values for high reputation firms and matched firms

1.0
0.9
0.8
0.7
Logit Values

0.6 Test Low Reputation

0.5
Matched Low Reputation
0.4
0.3
0.2
0.1
0.0
0.00 0.10 0.20 0.30 0.40 0.50 0.60 0.70 0.80 0.90 1.00
Percentage of Firms with lower Logit Values

Figure 3: Distribution of logit values for low reputation firm and matched firms

the two cases. In both cases, the low reputa- and low reputation firms. In addition, the
tion and their matched sample distributions content can be used (in-sample) to classify
are statistically different at better than the 10 firms with a high degree of accuracy.
percent level. These results allow us to reject Previous empirical literature (Roberts and
the null H2 and conclude that the content Dowling, 2002) shows, however, a positive
of the letter can be used to classify corpora- relationship between reputation and finan-
tions according to their corporate reputation cial performance, and this causality is bidi-
rankings. rectional. Roberts and Dowling decompose
reputation into components orthogonal to
ADDITIONAL ANALYSIS prior financial performance (measured as
ROA) and find that reputation contains
Corporate Reputation and Financial a distinct impact separate from financial
Performance performance. In this section, we examine
The evidence presented in the prior sections whether returns from the study and
supports the notion that the content of the matched groups are statistically different, and
letters to the shareholders differs for high whether financial performance incorporates

© 2008 Palgrave Macmillan 1363-3589 Vol. 11, 4, 285–307 Corporate Reputation Review 299
 Predicting Firm Reputation Through Content Analysis of Shareholders’ Letter

Table 10: Statistical Significance of the performing well verses those performing
Differences in Figure 2 poorly in the year discussed in the letter.
Stockholders are also concerned with a
t-Test: paired two sample Test high Matched firm’s current performance relative to its past
for means reputation firms performance. We examine two measures of
financial performance: (1) the total return in
Mean logit score 0.81 0.73
the year of the shareholders’ letter, and (2)
Variance 0.06 0.11
the percentage change in the current year’s
Observations 25 25
Pearson correlation 0.16 return from the average of the previous five
year’s returns.
Hypothesized mean 0
difference Return Results
d.f. 24 Table 12a compares returns from the 25 high
t Stat 1.06 reputation firms to 22 firms from the
P(T ⭐ t) one-tail 0.15 same industries matched on size and time
t Critical one-tail 1.71 period. The high reputation firms have a
P(T ⭐ t) two-tail 0.30 lower annual average return (0.17) compared
t Critical two-tail 2.06 to (0.30) that for the matched firms. The
difference in mean returns is not statistically
different from 0. To control for industry
Table 11: Statistical Significance of the effects, Table 12b uses a matched pair t-test.
Differences in Figure 3 The mean return of the matched firms is
higher (0.18 vs 0.31), and the difference is
t-Test: paired two sample Test low Matched significant at the 0.08 level. The return dif-
for means reputation firms ference supports the thesis that high reputa-
tion is a separate construct from performance.
Mean logit score 0.34 0.62 The matched high reputation firms per-
Variance 0.06 0.13 formed better, but as seen in Table 9, they
Observations 14 14
have lower average logit values. Tables 12a
Pearson Correlation − 0.41
and 12b contrast with the outcome in
Hypothesized mean 0 Table 10 which reports that the logit values
difference for the high reputation firms are statistically
d.f. 13 greater than those of their matched firms.
t Stat − 2.05 Tables 13a and 13b compare the return for
P(T ⭐ t) one-tail 0.03 firms in the low reputation group with their
t Critical one-tail 1.77 matched counterparts.The mean returns from
P(T ⭐ t) two-tail 0.06 the low reputation group and their matched
t Critical two-tail 2.16 firms are insignificantly different, using either
a two-sample or paired two-sample t-test.
Again, this contrasts sharply with the results
the predictive power of the content of the in Table 11, where the mean logit value for
letters to the shareholders in classifying high the low reputation group is nearly half that
and low reputation firms. of its matched group.

Financial Performance Measures Relative Performance Results

It is reasonable to surmise that the language While absolute return is certainly important
used in the letter would differ for firms to shareholders, they are also interested in

300 Corporate Reputation Review Vol. 11, 4, 285–307 © 2008 Palgrave Macmillan 1363-3589
 Geppert and Lawrence

Table 12: Comparison of Mean Returns Table 13: Comparison of Mean Returns
of High Reputation Firms and Matched of Low Reputation Firms and Matched
Firms: (a) t-Test and (b) Paired t-Test Firms: (a) t-Test and (b) Paired t-Test

(a) t-Test (a) t-Test

t-Test: two sample assuming Study high Matched t-Test: two sample assuming Study low Matched
unequal variances reputation firms unequal variances reputation firms

Mean return 0.17 0.30 Mean return 0.05 0.21

Variance 0.11 0.21 Variance 0.40 0.51
Observations 23 19 Observations 10 11
Pearson correlation
Hypothesized mean 0
difference Hypothesized mean 0
d.f. 32 difference
t Stat − 1.06 d.f. 19
P(T ⭐ t) one-tail 0.15 t Stat − 0.54
t Critical one-tail 1.69 P(T ⭐ t) one-tail 0.30
P(T ⭐ t) two-tail 0.30 t Critical one-tail 1.73
t Critical two-tail 2.04 P(T ⭐ t) two-tail 0.60
t Critical two-tail 2.09
(b)Paired t-test
t-Test: paired two sample Study high Matched (b)Paired t-test
for mean reputation firms
t-Test: paired two sample Study low Matched
Mean return 0.18 0.31 for means reputation firms
Variance 0.13 0.22
Mean return 0.11 0.33
Observations 18 18
Variance 0.59 0.78
Pearson correlation 0.60
Observations 7 7
Pearson correlation 0.39
Hypothesized mean 0
difference
Hypothesized mean 0
d.f. 17
difference
t Stat − 1.46
d.f. 6
P(T ⭐ t) one-tail 0.08
t Stat − 0.63
t Critical one-tail 1.74
P(T ⭐ t) one-tail 0.28
P(T ⭐ t) two-tail 0.16
t Critical one-tail 1.94
t Critical two-tail 2.11
P(T ⭐ t) two-tail 0.56
t Critical two-tail 2.45
how their company is doing relative to its
recent past.We calculate a measure of relative
performance as follows: Tables 14a and 14b compare the relative per-
formance of high reputation firms with their
Relative performance = matched counterparts. While the matched
Current return − Average of firms performed relatively better than the
high reputation firms, the difference is not
previousfive year’s returns statistically significant.
Average of previous Tables 15a and 15b repeat the comparison
five year’s return for the low reputation firms and their matched

© 2008 Palgrave Macmillan 1363-3589 Vol. 11, 4, 285–307 Corporate Reputation Review 301
 Predicting Firm Reputation Through Content Analysis of Shareholders’ Letter

Table 14: Comparison of Relative Returns Table 15: Comparison of Relative Returns
of High Reputation Firms and Matched of Low Reputation Firms and Matched
Firms: (a) t-Test and (b) Paired t-Test Firms: (a) t-Test and (b) Paired t-Test

(a) t-Test (a) t-Test

t-Test: two sample Study high Matched t-Test: two sample assuming Study low Matched
assuming unequal reputation firms unequal variances reputation firms
variances
Mean relative return 7.64 0.29
Mean relative 0.07 1.17 Variance 658.27 18.46
return Observations 10 11
Variance 3.53 66.19
Observations 23 19 Hypothesized mean 0
difference
Hypothesized mean 0 d.f. 9
difference t Stat 0.90
d.f. 20 P(T ⭐ t) one-tail 0.20
t Stat − 0.58 t Critical one-tail 1.83
P(T ⭐ t) one-tail 0.28 P(T ⭐ t) two-tail 0.39
t Critical one-tail 1.72 t Critical two-tail 2.26
P(T ⭐ t) two-tail 0.57
t Critical two-tail 2.09 (b) Paired t-test

(b)Paired t-test t-Test: paired two sample for Study low Matched
means reputation firms
t-Test: paired two Study high Matched
sample for means reputation firms Mean relative return 11.56 0.95
Variance 927.74 28.92
Mean relative 0.24 1.23 Observations 7 7
return Pearson correlation − 0.05
Variance 3.98 70.01
Observations 18 18 Hypothesized mean 0
Pearson correlation 0.06 difference
d.f. 6.00
Hypothesized mean 0 t Stat 0.90
difference P(T ⭐ t) one-tail 0.20
d.f. 17 t Critical one-tail 1.94
t Stat − 0.50 P(T ⭐ t) two-tail 0.40
P(T ⭐ t) one-tail 0.31 t Critical two-tail 2.45
t Critical one-tail 1.74
P(T ⭐ t) two-tail 0.62
t Critical two-tail 2.11
Impact of Performance on Predictive
Power of Content Variables
firms. The low reputation firms have, on Finally, we check to see whether the
average, better relative performance than return performance information nullifies the
their matched group. Again, this result in predictive power of the content variables.
conjunction with Table 10 shows that Table 16 shows the change in logit results
even though the low reputation firms had when return and relative performance are
better performance, they had lower logit included as independent variables. First
values. note that neither the return nor the relative

302 Corporate Reputation Review Vol. 11, 4, 285–307 © 2008 Palgrave Macmillan 1363-3589
 Geppert and Lawrence

Table 16: Logit Estimates with Return and Relative Return

Variables B SE Wald d.f. Significance

Variety − 46.18 20.82 4.92 1 0.03

Complexity − 4.59 2.92 2.47 1 0.12
Concrete 0.66 0.47 1.95 1 0.16
Constant 53.23 24.59 4.69 1 0.03
Return − 1.54 1.35 1.30 1 0.25
Relative return − 0.05 0.07 0.51 1 0.48

performance coefficients are statistically dif- ture. In this they define three aspects of cor-
ferent from 0. Also note that coefficients on porate reputation: awareness, assessment and
the three other independent variables are es- asset value. While far from complete, these
sentially the same as in Table 7. The two categories go far in providing a unified
notable differences are in significance of the framework for discussion and analysis. Aware-
‘concrete’ variable (which drops from 0.09 ness is simply the idea that the market knows
to 0.16) and the significance of the ‘com- of the firm without making any judgment
plexity’ variable (which drops from 0.01 of the quality or value of the firm. The asset
to 0.12) with the addition of these two value aspect of corporate reputation refers to
financial variables. the dollar value of a firm’s reputation. For
The analysis of the return performance example, this asset value aspect would refer
and the content analysis suggests that per- to the market value Coca Cola would place
formance and the content analysis variables on its corporate name. The assessment cate-
contain distinct information.This conclusion gory refers to the judgments made about the
must be somewhat tempered due to the sur- company, such as whether the company is
vival bias present in the low reputation firms. perceived as ‘honest, ethical, socially respon-
The low reputation firms in our sample ex- sible, etc.’ We believe that the chairman’s let-
perience a greater degree of bankruptcy than ter to the shareholders has the potential to
our sample of high reputation firms. Since influence the asset value and assessment aspects
the bankrupt low reputation firms drop from of a firm’s reputation.
the sample, this biases the expected returns The chairman’s letter can potentially affect
from low reputation firms upward (since the asset value reputation of the firm through
we are only looking at the more successful symbolic management as discussed in West-
surviving low reputation firms). It is possible phal and Zajac (1995, 1998). Stockholders
that the spurious increase in the low reputa- respond to the announced desired firm activ-
tion firms’ returns blurs the relation between ity even if it is not subsequently undertaken.
reputation and return. Westphal and Zajac comment that symbolic
management is most effective when the en-
DISCUSSION AND IMPLICATIONS vironment is uncertain or ambiguous. The
chairman’s letter typically covers broad aspects
Discussion of the firm’s recent experiences and plans
Research into the concept of ‘corporate about future activity. Uncertainty about the
reputation’ is hampered by the lack of a well future business environment lends itself well
agreed upon definition. To deal with this, to symbolic management. The letter could
Barnett et al. (2006) develop a taxonomy of influence shareholders’ perception of firm
corporate reputation from the prior litera- value by outlining future firm undertakings,

© 2008 Palgrave Macmillan 1363-3589 Vol. 11, 4, 285–307 Corporate Reputation Review 303
 Predicting Firm Reputation Through Content Analysis of Shareholders’ Letter

for example, ‘we are encouraged by the that the linguistic features (as opposed to
progress we’re making in our alternative fuels topic) do differ between high and low rep-
division.’ Further, Westphal and Zajac argue utation firms. High reputation firms use less
that justifying language can be used to temper complex (shorter), less varied and more con-
or preempt an anticipated negative event, for crete words. These linguistic features are
example, ‘planned layoffs will allow us to be- consistent with greater credibility. We sur-
come a leaner and more focused company.’ mise that the differences among the high and
The chairman’s letter thus can potentially low reputation firms conveyed in the chair-
change the value of the firm’s reputation man’s letter operate through the assessment
through symbolic management even if the reputation category, even though the aggre-
physical assets and actions of the firm remain gate nature of the rankings does not allow us
unchanged. to test this mechanism directly.
While the asset value aspect and sym-
bolic management affect reputation through Implications
‘what is said,’ we believe the assessment aspect While this study is primarily exploratory, our
chiefly affects reputation through the per- results establish that linguistic style is associ-
ceived credibility of the speaker. Greyser ated with measures of reputation that are
(1999) remarks that credibility is the key distinct from financial performance. We pos-
feature linking firm behavior and public per- it that linguistic style can affect reputation
ceptions. Several linguistic features have been through symbolic management and through
found to be associated with deception. perceived credibility. Symbolic management
Among these, Zhou et al. (2002) find that a would act primarily on the asset value cate-
greater word variety increases perceived de- gory of reputation as defined by Barnett et
ception. Presumably, a deceptive narrative al. (2006); perceived credibility would act on
uses more filler language to give the impres- the assessment category of reputation.
sion that it contains more meaningful con- Our research suggests a possible relationship
tent than in actuality. Both the automated between symbolic management and asset value
content analysis and human readers deter- reputation.We propose that firms whose letters
mined that low reputation rated firms used focus on future activity are targeting asset value
more complex writing styles. The low repu- reputation. One could test this by developing
tation firms also used greater variety and less a unique coding system that specifically targets
concrete words. In contrast, a matter-of-fact symbolic management. Symbolic management
concrete writing style is associated with often works by addressing future-intended ac-
greater truthfulness/credibility (Zhou et al., tivities. Verb tense structure and word choices
2002; Colwell et al., 2002; Dulaney, 1982). A related to future intentions (eg, seek to, strive
matter-of-fact concrete writing style can in- to, anticipate) may relate to asset value reputa-
crease the assessment aspect of firm reputa- tion. Under this proposition, firms that rate
tion by increasing perceived truthfulness and highly by Fortunes’ categories ‘Financial Per-
credibility of the chairman. The three ratings formance, Product Quality and Human Capi-
sources we use, Fortune Magazine, Harris tal’ would be expected to use more symbolic
Interactive and Business Ethics magazine use management and words related to future ac-
criteria that span all three aspects of corpo- tivities. Finding a link between asset value rep-
rate reputation. utation and symbolic management would be
While using the overall rankings does not useful for chairmen and financial analysts.
allow us to separate how the content and Chairmen could specifically focus their narra-
linguistic styles affect the relative importance tive on future activities as tool to increase the
of the three reputation categories, we do find firm’s perceived reputation.

304 Corporate Reputation Review Vol. 11, 4, 285–307 © 2008 Palgrave Macmillan 1363-3589
 Geppert and Lawrence

Our results relating the linguistic style of the chairman. We identify significant dif-
characteristics of variety and complexity to ferences in the letters between firms ranked
reputation suggest that perceived credibility highly and those ranked lowly according to
affects reputation. One could test the corporate reputation. A content analysis finds
link among credibility, linguistic style and significant variations in (1) word choice, (2)
reputation by having human subjects word length and (3) word variety between
separately rate high and low reputation firm the firms classified as having a high reputa-
narratives for credibility and truthfulness tion and those having a low reputation. The
and compare these human ratings to ratings firms with a high corporate reputation, on
based on linguistic characteristics such as the average, use less variety of words, shorter
variety and complexity indices. Under our words and more concrete words than the set
hypothesis, firms whose chairman’s letters of low reputation firms. These linguistic fea-
are rated as credible by human subjects tures are associates with perceived truthful-
should also be rated highly by the rating ness and credibility. High reputation firms
agencies and use low variety and low com- concentrate on realism with a matter-of-fact
plexity words. If complexity is negatively type of style.
related to reputation, then simplifying the Using concrete, complexity and variety
writing style could improve the perception variables, a logit regression successfully
of reputation. Letters conveying the highest classifies approximately 82 percent of our
level of reputation possible would use a con- study firms according to their perceived
crete, matter-of-fact style, and avoid unnec- reputation. The results are further tested us-
essary filler language or lengthy word ing unranked firms, matched to our study
choices. group of firms on the basis of industry
These results are also interesting from an membership, firm size and economic condi-
auditor’s perspective. Being able to use the tions. The logit ratings for the matched firms
letter’s narrative to detect credibility would are statistically different from the high and
be useful as a sorting mechanism. Firms whose low reputation sample firms. We contend
chairman’s letters are identified as less credible these differing logit values are a result
could have their financial statements more of content differences in the letter to the
closely examined for deception. Such useful- shareholders.
ness could be tested in future research. As with any statistical relationship, the di-
rection of causality is not determined. Our
CONCLUSIONS analysis does show that the high and low
In this study, we use content analysis to reputation firms do use different writing
measure any differences between the CEO’s styles. Previous studies have shown that
letter to shareholders of high and low repu- the styles differences we find are related
tation firms. The usefulness of the narrative to perceived credibility and truthfulness. In
content of the letter in classifying firms addition, we find the language differences
according to their reputation leads us to are not associated with current or past firm
believe the letter is associated with the per- performance.
ceived reputation of the firm. We believe While our results are consistent with the
the linguistic style of the letter affects the notion that Shareholder Letter word choice
perception of the firm’s credibility and con- influences perceived reputation, a caveat
sequently the assessment aspect of corporate on the direction of causality is warranted.
reputation. Word choice might also be a reflection of
The style and content of this letter is not the reputation the chairman perceives the
regulated and is under the direct influence firm to have. For example, if the chairman

© 2008 Palgrave Macmillan 1363-3589 Vol. 11, 4, 285–307 Corporate Reputation Review 305
 Predicting Firm Reputation Through Content Analysis of Shareholders’ Letter

believes that people appreciate what the firm Carpenter, R. (1986) ‘The statistical profile of
does, then he may feel free to write in a language behavior with Machiavellian intent while
experiencing caution and avoiding self incrimina-
matter-of-fact style about the firm’s accom- tion’, Annuals of New York Academy of Sciences, 606(1),
plishments as accomplishments. When a firm 5–17.
with a mediocre reputation has no accom- Christiansen, L.T. (2002) ‘Corporate communication:
plishments to claim, the chairman may resort The challenge of transparency’, Corporate Communi-
to writing about future positive intentions cations: An International Journal, 7(3), 162–168.
rather than current accomplishments. In the Colwell, K., Hiscock, C. and Memon, A. (2002) ‘In-
terviewing techniques and the assessment of state-
latter case, the causality runs from reputation ment credibility’, Applied Cognitive Psychology, 16,
to word style and not word style influencing 287–300.
reputation. Distinguishing the causal direc- Dulaney Jr., E. (1982) ‘Changes in language behavior
tion between word choice and reputation as a function’, Human Communication Research, 9(1),
poses an interesting question for future 75–82.
Greyser, S. (1999) ‘Advancing and enhancing corporate
research.
reputation’, Corporate Communications: An Interna-
NOTES tional Journal, 4(4), 177–181.
1 We do not include the awareness category because Hart, R.P. (2000) DICTION 5.0 Text-Analysis User
each survey requires that the survey participants be Manual, Digitext, Inc., Austin, TX, USA, pp. 32–37.
aware of the firms they are rating. Consequently, Henriksson, R.D. and Merton, R.C (1981) ‘On
the reputation ratings would only spuriously be market timing and investment performance. An
related to awareness. equilibrium theory of value for market forecasts’,
2 In addition to the clarity of the statement, the letter Journal of Business, University of Chicago Press, 54(3),
gives verifiable statistics as to why the firm is a lead- 363–406.
ing provider and why the housing industry is one of Ingram, R.W. and Frazier, K.B. (1983) ‘Narrative
the strongest growth sectors of the economy. The com- disclosures in annual reports’, Journal of Business
bination of clarity and verifiable statements adds to Research, 18(2), 614–622.
perceived credibility and reputation. Li, F. (2006) ‘Do stock market investors understand
3 We do not report these results in the paper, but they the risk sentiment of corporate annual reports?’,
are available from the authors upon request. Working Paper, Sephen M. Ross School of Business,
University of Michigan.
REFERENCES Michalisin, M., Karau, S. and Tangpong, C. (2004) ‘The
Abrahamson, E. and Amir, E. (1996) ‘The information effects of performance and team cohesion on
content of the president’s letter to shareholders’, attribution: A longitudinal simulation’, Journal of
Journal of Business Finance and Accounting, 23(8), Business Research, 57, 1108–1115.
1157–1182. Newman, M., Berry, D. and Richards, J. (2003)
Abrahamson, E. and Park, C. (1994) ‘Concealment of ‘Lying words: Predicting deception from linguistic
negative organizational outcomes: An agency theo- styles’, Personality and Social Psychology Bulletin, 29(2),
ry perspective’, Academy of Management Journal, 37(5), 665–675.
1302–1334. Roberts, P.W. and Dowling, G.R. (2002) ‘Corporate
Amernic, J. and Craig, R. (2007) ‘Improving reputation and sustained superior financial perform-
CEO-speak’, Journal of Accountancy, 203(1), ance’, Strategic Management Journal, 12, 1077–1093.
65–66. Salancik, G.R. and Meindl, J.R. (1984) ‘Corporate at-
Barnett, M., Jermier, J. and Lafferty, B. (2006) ‘Corpo- tributions as strategic illusions of management con-
rate reputation:The definitional landscape’, Corporate trol’, Administrative Science Quarterly, 29, 238–254.
Reputation Review, 9(1), 26–38. Smith, M. and Taffler, R. (2000) ‘The chairman’s state-
Bennett, R. and Kottasz, R. (2000) ‘Practitioner ment: Content analysis of discretionary narrative
perceptions of corporate reputation: An empirical disclosures’, Accounting, Auditing & Accountability Jour-
investigation’, Corporate Communications: An Interna- nal, 13(5), 624–646.
tional Journal, 5(4), 224–234. Stanwick, P. and Stanwick, S.D. (2003) ‘CEO and
Bond, G. and Lee, A. (2005) ‘Language of lies in pris- ethical reputation:Visionary or mercenary?’, Manage-
on: Linguistic classification of prisoners’ truthful and ment Decision, 41(10), 1050–1057.
deceptive natural language’, Applied Cognitive Psy- Tennyson, M.B., Ingram, R.W. and Dugan, M.T. (1990)
chology, 19, 313–329. ‘Assessing the information content of narrative

306 Corporate Reputation Review Vol. 11, 4, 285–307 © 2008 Palgrave Macmillan 1363-3589
 Geppert and Lawrence

disclosures in explaining bankruptcy’, Journal of Busi- reforms and shareholder reactions’, Administrative
ness Finance and Accounting, 17(3), 391–410. Science Quarterly, 43, 127–153.
Westphal, J. and Zajac, E. (1995) ‘Accounting for the ex- Zhou, L., Twitchell, D., Qin, T., Burgoon, J. and
planation of CEO compensation: Substance and sym- Nunamaker Jr., J. (2002) ‘An exploratory study into
bolism’, Administrative Science Quarterly, 40, 283–308. deception detection in text-based computer-medi-
Westphal, J. and Zajac, E. (1998) ‘The symbolic man- ated communication’, Proceedings of the 36th Hawaii
agement of stockholders: Corporate governance International Conference on System Sciences.

© 2008 Palgrave Macmillan 1363-3589 Vol. 11, 4, 285–307 Corporate Reputation Review 307
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.