Content-Type: text/html
"Who Was 'Shadow'?" The Computer Knows:
Using Grammar-Program Statistics in Content Analyses Finally
May Solve This Civil War Riddle
and Other Writing Mysteries
By Barbara G. Ellis, Ph.D.
and Steven J. Dick, Ph. D.
Mass Communication Department
P.O. Box 90335
McNeese State University
Lake Charles LA 70609-0335
318-475-5290
FAX 318-475-5291
Abstract
The object of this study was to use the statistics-documentation portion
of a word-processing program's grammar-check feature as a final,
definitive and objective tool for content analyses performed in
qualitative investigations to irrevocably rule out John H. Linebaugh,
John B. Dumble, or John Reid McClanahan as being the Civil War
special
correspondent who wrote for many newspapers under the pseudonym of
"Shadow". An enduring and intriguing mystery for journalism historians,
"Shadow's" identity has never been discovered although Linebaugh was
a
leading candidate for a time. On the Appeal, Linebaugh was an
editorial
writer and, briefly, the "special" who wrote as "Ashantee;" Dumble
was
an assistant editor and McClanahan, the senior proprietor, senior
editor
and chief editorial writer.
The investigators were two professors from the Mass Communication
Department of Louisiana's McNeese State University, Drs. Barbara G.
Ellis, associate professor, and Steven J. Dick, assistant professor.
"Who Was 'Shadow'?" The Computer Knows:
Using Grammar-Program Statistics in Content Analyses Finally
May Solve This Civil War Riddle
and Other Writing Mysteries
One of the intriguing, unsolved mysteries for some specialists in Civil War
journalism has been deducing who hid behind the pseudonym of "Shadow" in a
period when Confederate free-lance war correspondents originally used pen
names to avert harassment, military confinement, or publishers'
proscriptions against "double-dipping". Their postings, focus,
knowledgeability, and writing styles were well familiar to millions of
readers, much like World War II's Ernie Pyle. They were circulation
builders to publishers who could afford "stars" like Felix de Fontaine
("Personne," "Quel Qu'un"), Peter Wellington Alexander ("P.W.A.," "A,"
"Sallust"), Samuel Chester Reid, Jr. ("Sparta," "Ora," "290"), and
"Shadow," whose identity went to the grave with the many publishers who
paid for his incisive commentaries about military affairs,
essentially on
the Western front with the Army of Tennessee and commanders Braxton
Bragg,
Joe Johnston, and John B. Hood. [1]
These forerunners of syndicated columnists had friends and foes--President
Jefferson Davis to Confederate legislators, privates to generals
(of both
armies) such as Braxton Bragg who alternated between cultivating a
private
press agent like publisher John Forsythe of the Mobile Daily
Advertiser and
Register and locking up lesser lights like soldier-correspondent William
W. Screws. [2]
As journalism historian Frank Luther Mott commented about such enforced
anonymity: "[It] had one effect not contemplated: it made the pen
names of
many war 'specials' nationally famous." [3]
Scholars like J. Cutler Andrews ferreted out the identities of fifty-five
who chose colorful bylines ranging from "Mint Julep,"
"N'Importe," and
"Dixie" to "B," "Grapeshot," and "Se De Kay." But he and other
investigators were frustrated and confounded about the famed "Shadow". [4]
"Shadow" was well informed, plain spoken, and as ubiquitous as the
Confederate cavalry's Scarlet Pimpernel, General Nathan Bedford
Forrest.
His column appeared first with The Memphis Daily Appeal, in June
1863, then
the Mobile Register and Advertiser--and probably was well pirated. [5]
Historian Thomas H. Baker was among many who concluded that John H.
Linebaugh was "Shadow." [6] Linebaugh was the Appeal's bloviating
Poe-like
stringer "Ashantee," who eschewed trenches and fled the field when
the
first shell hit Chattanooga in 1863Dhighly uncharacteristic of
"Shadow" or
others. [7] Baker may have been influenced by CommercialAppeal
writer George
Sisler who in 1957 penned an undocumented historical journal article
about
Bragg's jailing Linebaugh for treason. To underpin the article's
thrust
that Linebaugh was a long-time thorn in Bragg's side, he doctored
documentation and credited Linebaugh, a fiftyish civilian, not only with
"Shadow's" work, but that of six other correspondents: "Leigh,"
"Ramrod,"
"Harvey," "Waverly," "Special," all in the army, and "Wanderer." [8]
Andrews's far more credible qualitative investigation of Linebaugh's life
ruled him out as "Shadow," but narrowed suspects to Stephen
Tillinghast
Hammond, Dr. Fielding Travis Powell, Albert Roberts, and, especially,
Henry
Watterson. Andrews' surrender note to peers said: ". . . I leave the
resolution of this fascinating and as yet unsolved problem of identity
to
any other historians who may be interested." [9]
The computer, in general, and the document-statistics portion from certain
grammar programs, in particular, now make it possible for
scholars in many
disciplines to provide the final testDdefinitive and objective
quantitative
content analysisDto extensive qualitative content analyses aimed either at
solving age-old riddles about authors like "Shadow" or in characterizing
writing styles.
The inclusion of quantitative measurement by the computer in content
analyses, eliminating subjectivity save for excerpt selections, can
provide
substantial underpinnings to qualitative efforts; and it is available to
anyone with a word-processing program that includes a grammar check.
In the past year alone, many academicians have seen the potential of
fine-tuned programs such as Que Software's Rightwriter and Lifetree
Software's Correct Grammar, for they quantify words used per sentence,
sentences per paragraph, number of syllables, and at least two Flesch
readability scales.
This particular study utilized the computerized grammar statistics to
underpin an already overwhelming case built on qualitative evidence
to
prove the hypotheses that Linebaugh was not "Shadow" and, secondly,
that
he did not write any of the other columns Sisler attributed to him.
[10] The
project was a byproduct of a larger, ongoing study connected to a
book on
the Appeal's wartime travels. That study's aim was to determine
which
editorials were written by the newspaper's senior partner and
editor-in-chief, John Reid McClanahan, during 1864-65.
Literature Search
The literature concerning quantitative analysis to determine authorship is
remarkably slim when contrasted with the monumental body of work
since the
Eighteenth century, principally conducted in an array of disciplines
from
censorship, psychology and the social sciences to politics,
espionage, and
marketing effectiveness. [11] Literary analyses of known author's
styles
scrutinized by quantitative methods certainly have been ongoing since
Sherman's 1888 investigation of sentence length in literature [12] and
Markov's
1913 study of Pushkin's Eugene Onegin [13] and, in the last half of this
century by Miles examination of poetic diction in 1946,[14] Schorer's
1949
study of the works of Jane Austen, Emily Bront , and George Eliot [15]
and
Parrish and Painter's concordance compilations for the writings of
Matthew
Arnold and William Butler Yeats. [16]
Judging from the conclusions drawn in many of these studies, investigators
appear to have embraced the view that the words reveal the
author, a
perception noted long ago by Lee (and thousands of editors):
The real revelation of the writer (as of the artist) comes in a far subtler way
than by . . .autobiography; and comes despite all
effort to elude it; . . . .For
what the writer does communicate is his temperament,
his organic personality,
with its preferences and aversions, its pace and
rhythm and impact and balance,
its swiftness or languor . . . and this he does
equally whether he be rehearsing
veraciously his own concerns or inventing someone
else's. [17]
Thus, there is Garraty noting president Woodrow Wilson's predilection for
adjectives [18] and, in 1940, Boder's measuring the emotional
instability of
Ralph Waldo Emerson and William James according to a ratio of
adjectives to
verbs [19] Both studies, combined with Busemann's system of dividing active
ideas by qualitative ones and the prevalent use of Flesch's
readability
measures, lay the groundwork in modalities helpful to the
determination of
authorship.
One interesting sidelight for journalists to the foregoing is that the
nation's key readability experts were hired by the wire services in
the
1940s as consultants to improve writing standardsDRudolph Flesch and
W. A.
Danielson for the Associated Press and Robert Gunning for United
Press. [20]
Perhaps because prior to the computer, a content analysis was laborious,
tedious, expensive, and subjective, few investigators had the
inclination
or energies to apply it to determination of authorship where far
greater
pitfalls lay in wait than, say, with a study about poor semantics
relationship to crankiness and paranoia [21] , or those on language in
annual
reports, collective-bargaining agreements or insurance policies.
[22]
One significant pitfall in validity of data, listed years ago by Berelson,
was sampling.[23] Samples can either be too small or, if they
are monumental,
Garraty's caveat about subjectivity can adversely impact a study's
merit as
it involves decisions as to what excerpts to include, categorization,
accurate coding, weighting of subjects, to say nothing of cultural
writing
styles,[24] authors employing several distinct styles to suit
publishers and
the potential for the subject to employ ghost writers; if teachers
traditionally have been reluctant to charge a pupil with plagiarism even
with a preponderance of damning evidence, what scholar is likely to
venture
the hypothesis that John F. Kennedy's Profiles in Courage was ghost
written?
Such inhibiting factors may have had bearing on why only two stellar
investigations on determination of authorship are perennially cited:
Spurgeon's 1935 work on William Shakespeare, and Mosteller and Wallace's
1964 efforts on the Federalist Papers. Both were pre-computer age
and
done with the laborious, tedious, expensive, and subjective methods
of
manual computationDplus the potential for selection flaws in sampling.
Nevertheless, Spurgeon's research generally is accorded the stature of
being the singular pioneer qualitative and quantitative content
analysis on
author determination. Essentially, her aim was to characterize the oeuvre
of Shakespeare, but, secondarily, to verify his authorship of certain
works.
She sought to end the four-hundred year old controversy that much credited
to the Bard was written by Francis Bacon, Christopher Marlowe,
Ben Jonson,
George Chapman, Thomas Dekker, or Philip Massinger. [25]
Her measurements involved comparative frequencies and types of imagery. [26]
When Mosteller and Wallace, in 1964, set out to sort out the
portions of
the Federalist Papers written by Alexander Hamilton and James
Madison, it
would appear they faced the same challenges and somewhat similar
methodology. They had the additional advantage of considerable writing
samples for that study.
Like forensics experts in science or police investigation, the three
researchers buttressed their findings by integrating qualitative with
quantitative research, for they knew their disciplines had legions of
doubters; that their discoveries have been largely ignored would seem
to
attest to that concern [27] perhaps because the researchers
perceived this
additional step would support their venture into what is largely
forensic
work familiar to detectives. The plethora of readability formulas
available todayDFlesch, Gunning, Cloze, McLaughlin, et al.Dand both the
computer and grammar-check programs would have enhanced validity and
simpified their exhaustive and exhausting efforts. [28]
Computer-aided analyses have been in existence almost from the instant
in the 1950s the Berelsons recognized the value of Univac and had
graduate assistants punch data into IBM cards. With the advent of
software that performed word counts and applied the Flesch reading test,
Sebeok and Zeps and their computerized examination of four thousand
Cheremis folk stories opened the floodgates to this new research tool.
[29]
By 1962, Stone and Bales, et al. had designed the General Inquirer
system for the social science and psychology fields that, with its
special set of dictionaries, could do counts, retrieve data, and
tabulate words. [30] Two years later, Sedelow and Sedelow were
presenting
a landmark paper on computational stylistics at an IBM Literary Data
Processing Conference. [31]
Almost simultaneously, software designers were producing programs like
WORDS with functions of sorting, editing, and statistical
analysis,
essentially geared for psychotherapy. For literary research there was
VIA (Verbally Indexed Associations) which also sorted words, grouped
data by root and could perform frequency counts; among its initial
projects was an analysis of Hamlet. [32] Then came fine-tuning with
1975
programs like TEXAN which analyzed syllables. [33]
Today, researchers have progressed far beyond word counts and wrestling
with homographs and overcoming ambiguity in project design. [34]
Investigators like Danielson, Lasorsa and Im have been using computers
to pioneer comparative writing styles of books and newspapers to
reveal
the decline of readability; they appear to be the first to utilize
the
grammar-check feature, their analysis based solely on the Flesch
Reading
Ease scores. [35]
To respond to essentially the same tacit criticisms undoubtedly received
by Spurgeon, Mosteller and Wallace, they argued: "Objections to
the use
of mechanistic formulas to evaluate how easy it is to read a piece
of
prose have existed since the introduction of the first equation and
probably will continue to persist, regardless of how sucessfully these
mathematical equations perform. Since no better methods for
assessing
readability have been offered, we are more willing to accept the
illumination offered by these formulas than to curse the darkness." [36]
However, they did not underpin their quantitative research with
qualitative analysis, as did Spurgeon; nor did they use this new method
of measurement to determine authorship. Yet the ability to do the
tedious investigative work of Spurgeon now is available to any scholar
with a laptop computer and software containing a grammar-check
feature;
or to use more than one to underpin quantitative findings, as was
done
in this project.
Methods
Data from this period are incomplete and fragmented. The ongoing war and
the passing years have destroyed most of the best evidence.
There will be
no overwhelming piece of evidence in this study to identify
"Shadow." A
mixture of methods was necessary, and this study used three. First,
primary sources identified the location and activities of the major
suspects in this decades-old puzzle. Second, writing styles were
analyzed
qualitatively. Finally, computer analyses of known writing samples
helped
quantitatively to identify differences in writing styles. No method
alone
was enough to identify "Shadow" positively. Andrews' challenge
still
stands. But the investigators sought to positively rule out Linebaugh
quantitatively as Andrews once did qualitatively.
The qualitative methods to refute Sisler's assumptions about Linebaugh's
work utilized prior to the application of quantitative
measurements of
Linebaugh, "Shadow," "Leigh," "Waverly," and "Wanderer," basically
involved
several modalities beyond what the literature furnished about Linebaugh's
life.
Initially, there was an examination of his filings as "Ashantee" and those
attributed to him from Bragg's Florida campaigns of 1861 and
General John
B. Hood's surrender of Atlanta to General William T. Sherman in
September
1864. This content analysis focused on story type (feature or
hard-news),
assignment posting, topics, viewpoint, analogies and examples,
diction,
sentence length, dateline, and method of transmission (the penurious
McClanahan permitted only the privileged to use the telegraph).
Research also entailed a comparison of Sisler's presentation of Linebaugh's
purported excerpts to the original stories in The Memphis Daily Appeal and
a literature search to determine Bragg's relationships with individual
reporters in terms of harassment, litigation, incarceration, or
cultivation.
This sizable body of evidence strongly indicated Linebaugh wrote only the
"Ashantee" articles and that significant and intentional errors
permeated
Sisler's 1957 work, an article subsequently used perhaps by many
scholars
for conclusions about The Memphis Daily Appeal's 1862-65 hegira
around the
South as it fled Grant and Sherman's armies; for nearly four
decades, it
may have significantly contributed to many of the myths surrounding
that
famous "run".
Andrews's most compelling argument in his qualitative efforts to rule out
Linebaugh as being "Shadow" is contained in an entire chapter of
his
definitive reportorial study The South Reports the Civil War. He noted
that "Shadow's" column in the Mobile Register on May 28, 1864
identified
himself as "the captain of a company of Confederate pikemen at
Nashville
after the fall of Fort Donelson in February 1862." This ruled out
McClanahan and Dumble, both well over conscription age and definitely tied
to the Appeal's production as it moved from Memphis to Grenada,
Mississippi, to Jackson, then to Atlanta and, finally, to Montgomery.[37]
Linebaugh was fifty-one in the following year[38] when Bragg had him
imprisoned
for treason, and when released, he raged: "_as a citizen I was not subject
to military arrest_" [39]
Andrews also pointed out that Linebaugh drowned October 26, 1864 in the
Alabama River, yet: "Exactly one month and a day after Linebaugh's
death
the Mobile Register published another letter from "Shadow" that was
dated
at Florence, Alabama on November 23, 1864." What Andrews failed to
do was
a content analyses of the writing of both men, where the style,
vocabularies, subject and attitudes are totally dissimilar. Linebaugh, a
defrocked Episcopal priest, [40]was a pretentious showoff, fond of
studding his
filings with foreign expressions ("voltigeur," "lucus anon luccende"[41]),
arcane historical allusions, and preachments; he repelled the
Appeal's
thousands of Army readers he was supposed to be attracting with this
typical sentence:
That this occultation was to be expected, as the necessary prelude, nay,
presage of our independence was the prophesy of the
thoughtful and it is
only those who knew not what revolution meant, or supposed
it was a
transition from one dream of criminal or inglorious repose to
another,
those who knew nothing of freedom in its aspects of dignity,
and were
willing to take servitude, even the degraded servitude of
superimposed
Yankee domination if it were gilded with wealth, or attended
by an
emasculate or traitorous enjoyment of delegated prosperity, in
preference
to the noble enjoyment of mental, moral, and political
independenceDthe
independence of a free-born heart, mind and willDwho are
cast down; so much
cast down as to be now willing to make terms of peace upon condition of
preserving a mess of miserable potage. [42]
By contrast, "Shadow" never used foreign expression, elevated diction,
religious, historical, or literary allusions. His readers had no
difficulty with:
The hills surrounding this war-like village are being fortified, of course.
Perhaps the purpose is to keep the men employed, more than for any other
design. Chattanooga is already defensible enough,
naturally. Besides
there is not much ! kellhoo! that it will ever be attacked.
No point on
the border is safer from raids. Its surroundings are
admirably adapted for
successful defense. The river in its bend from the base of Lookout
mountain, almost encircles it, leaving but a short space of
perhaps half a
mile or more from one bend to the other, which might be
traced with a line
of earthworks, which would render it a complete fortress.
The hills of
Chattanooga, ascending from the river, are higher than those
of Vicksburg,
and command the opposite bank for miles back to the base of
the Cumberland
range. [43]
Computer Analysis
The goal of this stage was to quantitatively identify difference in
writing styles. To eliminate bias, the analysis was performed by a
separate researcher D one who had not read the original works. The
computer could quickly and blindly perform standard readability
analysis.
Although, primary evidence and previous literature pointed to Linebaugh,
the tests performed on the data also included Dumble and
McClanahan as a
test of the method. Known writing samples of each person were
collected
and entered into computer text files.[44] Samples were defined as
whole
articles known to be written by each subject (headlines omitted). To
meet the requirements of readability tests, samples less than 100
words
were dropped. Two other samples were dropped because poor copies
made
it impossible to determine intended punctuation.
Keep as Monaco 9
Correct Grammar Readability Analysis
Document: Sample Results
Paragraphs: 7
Sentences: 49 (7.0 per paragraph)
Words: 1093 (22.3 per sentence)
Letters: 4908 (4.4 per word)
Syllables per 100 words: 148
Passive sentences: 13 (26 % passive)
Long sentences: 0 (0 % long)
Misspelled words: 0 (100 % correct)
Sentences fixed: 0 (100 % correct)
Sentences hard to read: 0 (100 % correct)
Flesch Reading Ease score: 58.7 (Fairly Easy)
Grade level required: 10 (85 % of U.S. adults)
Flesch-Kincaid grade level: 10.6
Gunning Fog Index: 13.6
The program Correct Grammar for the Macintosh (version 3.0.1, _ 1992,
Writing Tools Group) was used for the analysis. The program was
set to
analyze readability only and to produce a results file for each
sample
(see example above**). A macro program was used to copy data from
the
results files into a Microsoft Excel Spreadsheet.[45]
The next step was to pick measurements. Three readability scores were
chosen: Flesch Reading Ease; Flesch-Kincaid Grade Level; and
Gunning-Fog
Index. These scores measured readability in different slightly
different ways.[46] In addition, four other measures were found to be an
important difference from a preliminary qualitative analysis:
Letters-per-word; Words-per-sentence; Sentences-per-paragraph, and the
Percent of Passive sentences.
A simple t-test (two-tailed, assuming unequal variances) was performed
in Excel on each measure to determine likelihood that the writing
samples came from different authors. In each case, the null hypotheses
was "The writing samples came from the same author."
There are important limitations to the quantitative analysis.
First, it was impossible to get writing samples from everyone in the
Confederacy's press corps*. So it is always possible that someone
else
could have been "Shadow." The most that this study could do was to
show
that certain authors had distinctly different writing styles from
"Shadow."
Second, known writing samples and documentation of authorship have been
lost over the years. We acknowledge that the number of samples
for this
study is too small to be completely reliable. At best, these
numbers
only give an indication of who had a roughly similar writing style.
More samples could have been created by taking multiple samples of 100
words, from each known article. This option was rejected because it
would have given too much weight to longer articles.
Third, the readability formulas are using mid-1900 standards to judge
mid-1800 writing. Some of the tests depend on the counting
syllables
that may have been different or unrecognizable in writing of this
period. Test results will be useful to the point that they give
relative difficulty and show consistent differences across measures.
Finally, there may be other reasons why "Shadow's" writing is different
from the subjects. The author could have intended "Shadow's"
works for
a different audience and intentionally wrote differently.
"Shadow's"
stories may have been produced for a different purpose (e.g., hard
news
versus commentaries) and accidentally been written differently. Or
the
author may have tried to hide the source of "Shadow's" work by
altering
the style.
Results
Appendix A has the complete results. This section will only summarize
findings. The three readability measures yielded consistent
results.
Each measure identified "Shadow" as an easier read than Dumble,
Linebaugh, or McClanahan. In all cases but one, the t-test found the
difference significant at the 0.01 level of reliability.
Table 1
Readability Test Results
Flesch Reading
Ease Scale
Flesch-Kincaid
Grade Level
Gunning-Fog
Grade Level
"Shadow"
53.2
11.2
13.2
McClanahan
51.1
12.7
*
15.6
*
Dumble
48.7
*
12.9
*
15.7
*
Linebaugh
43.0
*
14.5
*
16.9
*
*Indicates significantly different from "Shadow" at a 0.01 level of reliability
Despite the reading ease scores, the data suggest that "Shadow" used
longer words than the subjects. McClanahan used the shortest words
(4.5
letters/word) on the average with a difference that was significant
at
the 0.01 level. Dumble's words were only slightly longer (4.6
letters/word) and still significantly different (0.02 level). It was
not possible to show a significant difference between Linebaugh's and
"Shadow's" word length (4.7 and 4.8 characters per word,
respectively).
Figure 1
Average Letters per Word[47]
[--- Pict Graphic Goes Here ---]
Figure 2
Average Words per Sentence
[--- Pict Graphic Goes Here ---]
"Shadow" may have used longer words, but had fewer of them per sentence
(see Figure 2). "Shadow's" sentences were significantly shorter
than
the subjects (all at the 0.01 level). While "Shadow's" work averaged
only 21.9 words per sentence, Linebaugh composed at a rate of 29.4
words
per sentence. The sentences of McClanahan and Dumble were also
longer
with a somewhat more compact 26.8 and 26.1 words per sentence,
respectively.
Again, "Shadow" shows significantly (all at the 0.01 level) more economy
in sentences per paragraph. "Shadow's" 3.7 sentences per
paragraph were
downright Spartan compared to Dumble's 18.0. McClanahan and
Linebaugh
used nearly double the number of sentences per paragraph (6.8 and
6.3
respectively) as "Shadow."
Figure 3
Average Sentences per Paragraph
[--- Pict Graphic Goes Here ---]
Unlike other measures, no significant differences were found in the use
of passive voice. Percent Passive ranged from 31.6% for
McClanahan to
39.6% for Dumble. "Shadow" (35.6%) and Linebaugh (34.9) were nearly
identical. Wide variation in the use of passive voice made this
measure
less discriminating D particularly considering the small sample
size.
Figure 4
Average Percent Passive Sentences
[--- Pict Graphic Goes Here ---]
Discussion
The quantitative analysis is cold and unfeeling compared to the emotion
of the authors' words. In this study, quantitative analysis
disclosed
some significant difference that may have been missed otherwise.
Clearly, "Shadow's" writing style differed from Linebaugh, once
considered to be a prime suspect. The differences are just as clear
with Dumble and McClanahan. From these data, it is time to broaden
the
search for "Shadow."
In general, the quantitative data suggests that "Shadow" wrote simply.
The suggested grade level for "Shadow's" work was lower than the
subjects. "Shadow" used shorter sentences and shorter paragraphs.
"Shadow's" longer word length may have been due to a greater use of
proper nouns (e.g., Rappahannock and gubernatorial) and titles. The
simpler writing style may suggest a less educated, less able author,
but
the passive voice measure discounts that somewhat. "Shadow" was
able to
communicate in a crisp writing style, yet not use passive voice any
more
than other authors.
Recommendations for future use of the method
Readability analysis is not definitive. It is simply one tool out of
many for the historian and should be used as such. Small sample
size
and the possibility of deception on the original players make it less
reliable than quantitative analysis in other areas of research.
Still,
when properly applied, it can provide previously unavailable
evidence.
With desktop computers and grammar checking programs, it can be
quick
and simple.
The person wishing to do readability analysis may consider the following
suggestions. First, keep in mind that the computer is best at
doing
redundant repetitive tasks. A good macro program and a properly
written
macro can save the researcher (or an assistant) hours of
mind-numbing
work.
Second, data produced by the computer should never have to be typed into
the computer. Re-keying data introduces human error and rounding
errors. The computer does not misplace a decimal or mind holding
endless fractions that the human operator would quickly round-off. In
Figure 1 above, the y-axis the graphing program remembered fractions
long forgotten by the researcher and produced a more accurate graph.
Third, choose your programs carefully for ability and inter-operability.
These days there is no reason why data should not be transferable from
one program to another. This project was done almost entirely on
Microsoft products (Word, Works, Excel). Simple transfer of data from
word processing to data analysis save time and trouble.
For this project, a stand-alone grammar checking program was found to be
more useful than those embedded in the word processor. Even
though
Microsoft Word (version 5.1a) uses Correct Grammar as its grammar
checking subroutine, the separate program Correct Grammar offered more
choices and the ability to output data into text files.
Finally, it is vital to get as many writing samples as possible. It is
difficult to get reliable statistics from small samples.
Ideally, the
researcher must find a consistently large set of writing samples for
all
subjects.
Conclusion
No one knows for certain yet who "Shadow" was, but the quantitative content
analyses, coupled with the qualitative research data, strongly suggests
that this "special" correspondent definitely was not Linebaugh.
Further,
the inclusion of McClanahan and Dumble in the quantitative portion
demonstrates how authorship determination can be accomplished by
journalism
historians. Andrews posited that Henry Watterson, later the long-time
proprietor and editor of the Louisville Courier-Journal, was the
"most
likely" candidate. [48] Watterson had been on the staff of the New
York Times
and the Washington Daily States by the time he joined the
Confederate army;
before the war ended, he had been an editor on the Chattanooga Rebel, and
the Montgomery Daily Mail. [49] Thus, there are more than
sufficient
samplesDlengthy or briefDof this period of his writings that may be
compared with "Shadow's" contributions to the Appeal and, chiefly, the
Mobile Register and Advertiser. [50]
So identifying "Shadow" is a mystery still as ripe as when a frustrated
Andrews threw down that challenge to the next generation of
academic
sleuths. Indeed, the computer makes determination of authorship or
authors' characteristics possible for many other disciplines than
journalism history.
Discoveries to age-old riddles today are at the researchers' fingertips and
can be made in the comfort of their offices or homes, resting as they do
on statistical data furnished by the grammar-check portions of an
array of
word-processing programs.
Endnotes
Appendix A
Flesh Reading Ease
Avg
Stdv
N
t-test
Shadow
53.2
3.3
9
McClanahan
51.1
13.5
65
0.1914
Dumble
48.7
7.0
25
0.0047
Linebaugh
43.0
9.6
20
0.0000
Flesh-Kincaid Grade Level
Avg
Stdv
N
t-test
Shadow
11.2
0.8
9
McClanahan
12.7
3.9
65
0.0018
Dumble
12.9
1.9
25
0.0002
Linebaugh
14.5
2.7
20
0.0000
Gunning- Fog Index
Avg
Stdv
N
t-test
Shadow
13.2
0.9
9
McClanahan
15.6
4.2
65
0.0000
Dumble
15.7
2.1
25
0.0000
Linebaugh
16.9
2.8
20
0.0000
Letters per
Word
Avg
Stdv
N
t-test
Shadow
4.8
0.2
9
McClanahan
4.5
0.2
65
0.0019
Dumble
4.6
0.1
25
0.0173
Linebaugh
4.7
0.2
20
0.1831
Words per
Sentence.
Avg
Stdv
N
t-test
Shadow
21.9
2.4
9
McClanahan
26.8
8.8
65
0.0002
Dumble
26.1
4.5
25
0.0005
Linebaugh
29.4
6.1
20
0.0000
Sentences. per
Paragraph
Avg
Stdv
N
t-test
Shadow
3.7
1.3
9
McClanahan
6.8
4.9
65
0.0000
Dumble
18.0
8.4
25
0.0000
Linebaugh
6.3
4.8
20
0.0005
% Passive
Avg
Stdv
N
t-test
Shadow
35.6
12.2
9
McClanahan
31.6
23.2
65
0.4063
Dumble
39.6
16.1
25
0.4073
Linebaugh
34.9
19.2
20
0.9017
[1] Frank L
uther Mott, A
merican Journ
alism: A Hist
ory of Newspa
pers in the U
nited
States Thr
ough 250 Year
s 1690 to 194
0 (New York:
Macmillan, 19
49), 330, 336
, 338; Jean
Folkerts
and Dwight L
. Teeter, Voi
ces of a Nati
on: A History
of the Media
in the Unite
d
States, 1
st ed. (New Y
ork: Macmilla
n, 1989), 210
, 213.
[2] J
udith Lee Hal
lock, Braxton
Bragg and Co
nfederate Def
eat, (Tuscalo
osa, Ala.: Th
e
U
niversity of
Alabama Press
, 1991) 2: 89
, 216, 231; J
. Cutler Andr
ews, TheSouth
Reports
th
eWar (Princet
on: Princeton
University P
ress, 1970),
248-51, 236.
[3] Mott, Am
erican Journa
lism, 338.
[
4] Andrews, S
outh Reports
War , 543-547
. Their iden
tities were,
respectively,
Capt.
Theodoric
Carter who w
rote for the
Chattanooga D
aily Rebel an
d the Montgom
ery Daily
Mail; Albe
rt J. Street,
Savannah Rep
ublican, Memp
his Daily App
eal, Mobile D
aily
Ad
vertiser and
Register; Joh
n R. Thompson
, Memphis Dai
ly Appeal; Wi
lliam D. Barr
,
Memph
is Daily Appe
al; W. B. Gal
breath, Memph
is Daily Appe
al; and Charl
es D. Kirk,
Chattano
oga Daily Reb
el, Memphis D
aily Appeal,
Louisville Da
ily Courier,
Augusta Daily
Consti
tutionalist.
[5] Andrews,
South Report
s War, 248-51
; The Memphi
s Daily Appea
l, 6, 12, 13,
17, 20,
25, 27,
30 June 1863
; 1, 3, 4, 1
4, 15, 16 Jul
y 1863.
[6]
Thomas H. Ba
ker, "Refugee
Newspaper: T
he Memphis Da
ily Appeal, 1
862-1865," Jo
urnal
of Souther
n History,29,
335 (Novembe
r 1966).
[7]
The Memphis
Daily Appeal,
18, 20 July
1863; 3l Aug
. 1863; 5, 7
,8, 12, 16,17
, 22,
September
1863; ibid.,
6, 12, 13, 17
, 20, 25, 27,
30 June 1863
; 1, 3, 4, 14
, 15, 16
July 1863.
[8] George
Sisler, "The
Arrest of a M
emphis Daily
Appeal War Co
rrespondent o
n Charges
of Tre
ason," West T
ennessee Hist
orical Societ
y Papers ,11,
76-92 (1957)
.
[9] Andrew
s, South Repo
rts War, 543-
47; "Leigh" w
as a gunner w
ith Col. John
W. A.
Sanford's
battery comm
anded by Brig
.-Gen. James
R. Chalmers;
"Ramrod" was
a
soldi
er-correspond
ent covering
Northeast Mis
sissippi (The
Memphis Dail
y Appeal, 14,
16
Oct
ober 1862); "
Harvey" was i
nfantry Capta
in J. Harvey
Mathes, (Bake
r, The Memphi
s
Comme
rcial Appeal:
The History
of a Southern
Newspaper [
Baton Rouge:
Louisiana Sta
te
Univ
ersity Press,
1971),199; "
Waverly" was
in the cavalr
y (The Memphi
s Daily Appea
l,
May-
June 1864, pa
ssim); and "S
pecial," was
with Gen. Wil
liam B. Bate'
s division (T
he
Memp
his Daily App
eal, 16, 18 J
une 1864).
[
10] Some of t
he qualitativ
e research wa
s based on An
drews's curso
ry primary da
ta about
Li
nebaugh's lif
e and that wr
iter's commu
nications wit
h Confederate
Vice Preside
nt
Alexander H.
Stephens, and
items concer
ning his wher
eabouts and d
eath that app
eared in
the Appeal.
However, th
e major effor
t of this pro
ject's chief
investigator
centered on
a minute qua
litative cont
ent analyses
of Linebaugh'
s "Ashantee"
columns, the
knowledge
o
f Appeal mana
gement and ed
itorial polic
ies, and the
familiarity w
ith assignmen
t
beats
, content, sc
ope and writi
ng styles of
the newspaper
's dozens of
war
cor
respondents.
For instanc
e, Sisler cla
imed that Lin
ebaugh began
to cover Brag
g for the App
eal in the
1861 Florida
campaign, beg
inning with t
he raid again
st Santa Rosa
Island, then
the
Kentucky ca
mpaign, and,
finally, the
Confederate a
rmy's retreat
from Dalton
to Atlanta.
None of tha
t data are tr
ue.
Sisler
stated that L
inebaugh sent
exclusive di
spatches from
Florida to t
he Appeal wit
h
the datel
ine of Oct.
8, 1861. The
newspaper ha
d no correspo
ndents' filin
gs on this
engag
ement, but di
d reprint dis
patches from
the Mobile A
dvertiser and
Register as
well
as tho
se from the
Press Associa
tion, the New
Orleans Delt
a, the Richmo
nd Dispatch.
The Kent
ucky, Middle
and East Tenn
essee hard-ne
ws dispatches
from Tupelo
to Knoxville
and
Nashville in
1862 were re
prints from t
he Mobile Adv
ertiser and R
egister.
Feature-st
ory treatment
of that camp
aign, exclusi
ve to the App
eal, was pro
vided by
"Leigh" who
se style gene
rally was boy
ishly simple
and personal,
a departure
from the
detached, e
litist rococo
of "Ashantee
".
"Leigh,"
not Linebaug
h, covered Br
agg's epic 71
5-mile march
from Tupelo,
Miss. to the
bat
tle at Munfor
dville, Ky.,
but Sisler cr
edited his Oc
t. 16, 1863 d
ispatch to
Linebaugh
. He did the
same thing w
ith "Leigh's"
and "Ramrod'
s" columns ab
out the
Perryville b
attle. Sisle
r also attrib
uted to Lineb
augh one of "
Shadow's" fil
es on
B
ragg's comman
d appointment
of both the
armies of Ken
tucky and Mis
sissippi (The
Memphis
Da
ily Appeal, 1
5 July 1863).
One of "
Leigh's" best
eyewitness a
ccounts was t
he battle of
Murfreesboro
(which Sisler
said Lineb
augh wrote);
at one point
this Mississ
ippian's abil
ity to fire h
is cannon
was im
peded because
he was splat
tered with th
e brains of a
fellow artil
lerist (The
Memphis
Daily Appeal,
22 January 1
863) hardly i
n keeping wit
h Linebaugh's
aversion to
battlef
ields .
As
"Leigh" bega
n to indicate
disillusionm
ent with Brag
g, Sisler als
o added that
column
to L
inebaugh's st
ring; he did
the same with
two addition
al filings by
both "Leigh"
and
"Shado
w" on Bragg's
inertia in r
e-attacking R
osecrans at M
urfreesboro (
The Memphis
Dail
y Appeal, 12
June 1863).
Linebaugh'
s debut in th
e Appeal was
June 27, 1863
following th
e June 15 ann
ouncement
that t
he newspaper
had hired "an
intelligent
reporter" to
cover "the fr
ont" of the A
rmy
of Tenn
essee, especi
ally from hea
dquarters; "S
hadow's" firs
t column appe
ared on June
6, 1863
); for McClan
ahan to use t
wo pseudonyms
for one writ
er seems as u
nlikely as hi
s
using
two correspo
ndents for th
e same beat,
given his pub
lishing coura
ge and common
sense.
What was mo
re likely was
that "Shadow
" became tro
ublesome to M
cClanahan, wa
s
dissa
tisfied with
the Appeal or
received a m
ore attractiv
e offer from
Mobile publis
her
Joh
n Forsythe.
Or that he to
ok the assign
ment temporar
ily when the
Appeal first
settled
in Atlanta w
ith the under
standing that
McClanahan w
ould find a r
eplacement at
summer
's end. Curi
ously, Sisler
never mentio
ned Linebaugh
's pseudonym
of "Ashantee"
;
inste
ad, he identi
fied him as "
Shadow". But
both bylines
appear in th
e issues of J
une
27,
June 30, and
July 16 with
"Shadow's" r
eadable, sold
iery style pr
oviding a dis
tinct
c
ontrast from
Linebaugh's a
ttempts to re
plicate that
of Addison an
d Steele. It
was
en
tirely possib
le, of course
, for a publi
sher to prote
ct an outspok
en correspond
ent by
phasing in a
new pseudonym
as he phases
out another;
"Shadow's" l
ast dispatch
to the
Appeal was Ju
ly 16, 1863.
The appearan
ce of three s
tories by "As
hantee" from
Charles
ton and Augus
ta at the ver
y time an App
eal editorial
ist said he w
as visiting a
n
ailin
g daughter in
North Alabam
a does hint o
f that kind o
f deliberate
legerdemain b
y
manag
ement in tryi
ng to conceal
that Linebau
gh had desert
ed his post f
ollowing his
first t
aste of war.
Moreover, co
nsidering "Sh
adow's" popul
arity and "As
hantee's" lac
k
of it
, it would ha
ve made littl
e business se
nse to the as
tute McClanah
an to phase o
ut
the
"Shadow" byl
ine for "Asha
ntee".
"Sha
dow" continu
ed to file fo
r the Mobile
Advertiser an
d Register af
ter Linebaugh
's
resignation f
rom the Appea
l in February
1864. Lineb
augh's death
by drowning o
n July
26, 1864 (And
rews, South R
eports War, 5
44) makes Sis
ler's claim o
f Atlanta cov
erage
p
roblematic.
Further, qua
litative cont
ent analysis
showed Lineba
ugh's exhibit
ionistic tend
encies: a
p
enchant for f
oreign expres
sions, histor
ical allusion
s, complex vo
cabulary, and
wo
rdiness demon
strated in ne
cklaces of pr
epositional p
hrases. Such
a writing st
yle was
so
singular that
it was appar
ent who was w
riting some o
f the Appeal'
s editorials
during
the winter of
1863-64. By
contrast, "S
hadow's" work
is devoid of
any of these
patter
ns; he presen
ted facts and
events in th
e unadorned s
tyle of hard
news and was
equally
as direct in
drawing conc
lusions from
such data.
In terms of
journalistic
skills, for a
ll of his lit
erary pretens
ions, Linebau
gh's leads
too often sta
rted with "th
e news from .
. ." (The Me
mphis Daily A
ppeal, 31 Aug
ust,
1863; 1, 3,
5 September
1863) while "
Shadow" almo
st unfailingl
y used the "f
ive-W"
lead(The Memp
his Daily App
eal, 12 June,
17 July 1863
). Unlike "S
hadow's" tota
l focus
on the front
and obvious
knowledge of
military tact
ics and topog
raphy, Lineba
ugh's
t
opics until h
is employment
seemed threa
tened (when h
e used troop-
movement info
rmation
provided by
a high-rankin
g friend) gen
erally were o
n theoretical
strategies,
Greek
m
ythology, Wel
lington minut
ia, or religi
ous admonitio
ns (The Memph
is Daily Appe
al, 5
S
eptember; 17
July 1863).
Linebaugh als
o lacked news
sense as whe
n he failed t
o report
hi
s interview w
ith the Jacks
onian era's D
uff Green, on
e-time editor
of the Unite
d
State
s Telegraph (
22 August; 7
September 186
3) or bolted
under fire fr
om a major sc
oop
in
the initial b
ombardment of
Chattanooga
(The Memphis
Daily Appeal,
21 August 18
63).
"
Shadow" also
lacked Lineb
augh's ascerb
ic meanness,
demonstrated
when the latt
er
beli
ttled a Georg
ia gubernator
ial candidate
(5 September
1863).
[11]
Krippendorff
dates conten
t analysis fr
om the Swedis
h state churc
h analyzing t
he
Songs of Zion
for heresy i
n the 18th ce
ntury (Klaus
Krippendorff,
Content Anal
ysis: An
Introductio
n to Its Meth
odology (Beve
rly Hills: Sa
ge Publicatio
ns, 1980), 13
.
[12] L. A.
Sherman,"Som
e Observation
s Upon the Se
ntence-Length
in English P
rose,"
Universit
y Studies of
the Universit
y of Nebraska
, Vol. 1, No.
2, (1888), 1
19-130.
[13]
Krippendorff
, Content Ana
lysis, 13.
[
14] Josephine
Miles, The V
ocabulary of
Poetry (Los A
ngeles: Unive
rsity of Cali
fornia
Press, 19
46), 112.
[1
5] M. Schore
r, "Fiction a
nd the 'Matri
x of Analogy'
," Kenyon Rev
iew, 11, 539-
60.
[16] S.
M. Parrish, e
d., A Concord
ance to the P
oems of Matth
ew Arnold (It
haca, N.Y.:
Corn
ell Universit
y Press, 1959
), 216; S. M.
Parrish and
J. A. Painter
, A Concordan
ce to
the P
oems of W. B.
Yeats (Ithac
a, N.Y.: Corn
ell Univerity
Press, 1963)
, 155.
[17]
Vernon Lee, T
he Handling o
f Words and O
ther Studies
in Literary P
sychology
(Londo
n: J. Lane, 1
923), 155.
[
18] Ithiel de
Sola Pool, e
d., Trends in
Content Anal
ysis (Urbana,
Ill.: Univer
sity of
Illinois
Press, 1959)
, 215.
[19]
D. P. Boder,
"The Adjectiv
e-Verb Quotie
nt: A Contrib
ution to the
Psychology of
La
nguage," Psyc
hological Rec
ord, 3, 310-3
43.
[20] Wer
ner J. Severi
n and James W
. Tankard, Jr
., Communicat
ion Theories:
Origins,
Method
s, and uses i
n the Mass Me
dia, 3d Ed. (
New York: Lon
gman, 1988),
118.
[21] Po
ol, Trends in
Content Anal
ysis, 177.
[
22] Severin a
nd Tankard, 1
20-24.
[23]
Bernard Berel
son, Content
Analysis in C
ommunication
Research (Gle
ncoe, Ill.: T
he
Free Press, 1
952), 175.
[24] Pool, Tr
ends in Conte
nt Analysis,
187.
[25] Ca
roline F. E.
Spurgeon, Sha
kespeare's Im
agery and Wha
t it Tells Us
(Boston:
Beacon
Press, 1985)
, Appendices
II-IV.
[26]
Spurgeon, Sha
kespeare'sIma
gery, passim.
[27] Berels
on, Content A
nalysis, 22.
[28] G. Wayn
e Shamo, "Pre
dicting Sylla
ble Count By
Computer, " J
ournalism Qua
rterly,
52, 344-
46 (Summer 19
75).
[29] T.
A. Sebeok an
d V. J. Zeps,
"An Analysis
of Structure
d Content Wi
th Applicatio
n
o
f Electronic
Computer Rese
arch in Psych
olinguistics,
" Language an
d Speech, 1,
181-93.
[30]
P. J. Stone,
R. F. Bales,
J. Z. Namenw
irth, and D.
M. Ogilvie, "
The General
Inqu
irer: A Compu
ter System fo
r Content Ana
lysis and Ret
rieval Based
on the Senten
ce as
a Uni
t of Informat
ion," Behavio
ral Science,
7, 484-94.
[
31] Sally Y.
Sedelow, W. A
. Sedelow, Jr
., and T. Rug
gles, "Some P
arameters for
Co
mputational S
tylistics: Co
mputer Aids t
o the Use of
Traditional C
ategories in
Stylist
ic Analysis.
In Proceedin
gs of the IBM
Literary Dat
aProcessing
Conference (
Yorktown
Heights, N.
Y. : IBM, 196
4), 211-29.
[32] Ole R. H
olsti, Conten
t Analysis fo
r the Social
Sciences and
Humanities (R
eading,
Mass. :
Addison-Wesle
y Publishing
Company), 154
-55.
[33] Sh
amo, "Predict
ing Syllable
Count By Comp
uter," 344.
[34] Holsti,
Content Analy
sis, 191.
[3
5] Wayne A. D
anielson, Dom
inic L. Lasor
sa and Dae S.
Im, "Journal
ists and Nove
lists:
A Study o
f Diverging S
tyles," Journ
alism Quarter
ly, 69, 436-4
45 (Summer 19
92).
[36] Da
nielson, Laso
rsa, and Im,
"Journalists
and Novelists
," 442.
[37]
Andrews, Sou
th Reports Wa
r, 544. When
Dumble petit
ioned to get
his 1862-1865
back
wages
, he testifie
d he had been
on the Appea
l from June 8
, 1862 to Apr
il 15, 1865
(Pet
ition of John
B. Dumble, M
cKnight v. Di
ll, Shelby Co
unty Chancery
Court, Tenne
ssee,
Jan.
26, 1869, 619
-20).
[38] 1
850 Louisiana
Census, East
Baton Rouge
Parish, p. 17
7.
[39] The
Memphis Daily
Appeal, 5 N
ovember, 1863
.
[40] Journ
al of Proceed
ings of the T
hirty-Fourth
Annual Conven
tion of the P
rotestant
Episco
pal Church in
the Diocese
of Georgia, M
ay 8, 1856, 1
9; One Hundre
d Years of Li
fe:
Emmanue
l Church (Ath
ens, Ga.: Emm
anuel Church,
November 194
3), 11-12.
[
41] The Memph
is Daily Appe
al, 25 July 1
863; 1 Septem
ber 1863.
[4
2] Ibid., 31
August 1863.
Linebaugh ha
d only been s
tringing for
five days bef
ore he
remarked
obviously to
criticism "My
Letters may
have seemed t
oo discursive
for the
general tas
te, but_ " an
d then launch
ed into a typ
ically length
y sentence th
at was 102
words lon
g (The Memphi
s Daily Appea
l, 20 July,18
63)
[43] Ibi
d., 17 July 1
863.
[44] Or
iginally, the
samples were
entered into
one large fi
le. A simple
macro progra
m
(
AutoMac III)
was used to s
plit the file
into individ
ual sample fi
les. A macro
program
au
tomates repet
itive tasks a
nd keystrokes
which can be
performed by
any programs
. In
t
his case, the
macro copied
the sample,
pasted to a n
ew document,
saved and clo
sed the
new document
under a new
name.
[45]
At this stag
e, macro prog
ram deleted u
nnecessary in
formation and
placed tabs
bet
ween data poi
nt. The mac
ro program op
erated within
Microsoft Wo
rd, version 4
.0.
Mo
st spreadshee
t and databas
e programs ca
n import data
when data po
int are separ
ated by
tab
s, and record
s are separat
ed by carriag
e returns.
[
46] The Flesc
h Reading Eas
e Score is ba
sed on the nu
mber of words
in each sent
ence,
and the av
erage number
of syllables
per word. Th
e Flesch-Kinc
aid system at
tempts to
represent
readability a
s a school gr
ade level. Th
e Flesch-Kinc
aid formula h
as become a
standard
required by
the U.S. Gove
rnment and th
e military (D
OD MIL-M-3878
4B). The
Gunning Fo
g index consi
ders sentence
length but e
mphsizes word
length. Bak
er, Robert
and Dave
Johnson, Corr
ect Grammar f
or the Macint
osh (San Fran
cisco: Lifetr
ee Software
Inc.) Do
cumentation ,
46-47.
[47
] Note: All f
igures were p
roduced by Mi
crosoft Excel
prior to dat
a rounding.
The
unusual Y-ax
is notations
are the resul
t of rounding
errors but m
ore accuratel
y reflect
relationsh
ips.
[48]
Andrews, Sout
h Reports War
, 547.
[49]
Folkerts and
Teeter, Voice
s of Nation,
245-46; Andre
ws, South Rep
orts War,
544-54
7.
[50] Ibid
., 545.