Versant Writing Test

TM

Test Description and Validation Summary

Table of Contents
Table of Contents .......................................................................................................................... 1
1. Introduction ............................................................................................................................... 4
2. Test Description ....................................................................................................................... 4
2.1 Workplace Emphasis .............................................................................................................................................................. 4
2.2 Test Design ............................................................................................................................................................................... 4
2.3 Test Administration ................................................................................................................................................................ 5
2.4 Test Format .............................................................................................................................................................................. 5
Part A: Typing ............................................................................................................................................................................ 5
Part B: Sentence Completion ................................................................................................................................................. 6
Part C: Dictation ....................................................................................................................................................................... 7
Part D: Passage Reconstruction ............................................................................................................................................. 8
Part E: Email Writing ................................................................................................................................................................ 9
2.5 Number of Items ..................................................................................................................................................................... 9

3. Test Construct ........................................................................................................................ 10

3.1 Facility in Written English .................................................................................................................................................... 10
3.2 The Role of Context ............................................................................................................................................................ 12
3.3 The Role of Memory ............................................................................................................................................................ 13
4. Content Design and Development ....................................................................................... 13
4.1 Vocabulary Selection............................................................................................................................................................. 13
4.2 Item Development ................................................................................................................................................................ 13
4.3 Item Prompt Recording ....................................................................................................................................................... 14
4.3.1 Voice Distribution......................................................................................................................................................... 14
4.3.2 Recording Review ......................................................................................................................................................... 14
5. Score Reporting ...................................................................................................................... 15
5.1 Scoring and Weighting.......................................................................................................................................................... 15
5.2 Score Use ................................................................................................................................................................................ 17

6. Data Resources for Scoring Development ........................................................................... 17

6.1 Data Collection ...................................................................................................................................................................... 17
6.1.1 Native Speakers ............................................................................................................................................................. 18
6.1.2 Non-Native Speakers ................................................................................................................................................... 18
6.2 Expert Human Rating ........................................................................................................................................................... 18

7. Validation ................................................................................................................................. 19
7.1 Validity Study Design ............................................................................................................................................................ 19
7.1.1 Validation Sample .......................................................................................................................................................... 19
7.2 Structural Validity .................................................................................................................................................................. 20
7.2.1 Descriptive Statistics .................................................................................................................................................... 20
7.2.2 Standard Error of Measurement ............................................................................................................................... 21
7.2.3 Test Reliability ............................................................................................................................................................... 21
7.2.4 Dimensionality: Correlations among Subscores .................................................................................................... 21
7.2.5 Machine Accuracy ......................................................................................................................................................... 22
7.2.6 Differentiation among Known Populations ............................................................................................................. 23

2013 Pearson Education, Inc. or its affiliate(s).

Page 2 of 30

7.3 Concurrent Validity .............................................................................................................................................................. 24

7.3.1 Versant Writing Test and TOEIC ............................................................................................................................. 24
7.3.2 Versant Writing Test and CEFR Level Estimates .................................................................................................. 25

8. Conclusion ............................................................................................................................... 28
9. About the Company ............................................................................................................... 28
10. References ............................................................................................................................. 29

2013 Pearson Education, Inc. or its affiliate(s).

Page 3 of 30

1. Introduction
Pearsons Versant Writing Test, powered by Ordinate technology, is a computer-based assessment
instrument which is designed to measure how well a person can handle workplace English in written
form. The Versant Writing Test is intended for adults 18 years of age and older and takes about 40
minutes to complete. Because the Versant Writing Test is delivered automatically by the Versant
testing system, the test can be taken at any time, from any location on a computer. A human examiner is
not required. The computerized scoring allows for immediate and objective results that are reliable and
correspond well with traditional measures of English language proficiency.
The Versant Writing Test measures facility in written English in the workplace context. Facility is defined
as how well a person can understand spoken or written English and respond in writing appropriately on everyday
and workplace topics at a functional pace. Versant Writing Test scores provide reliable information that
can be applied to placement, qualification and certification decisions by academic institutions, businesses
and government agencies. The test is also appropriate for monitoring progress as well as measuring
instructional outcomes. (The Versant English Test1 is also available if it is necessary to evaluate spoken
English.)

2. Test Description
2.1 Workplace Emphasis
The Versant Writing Test is designed to measure the candidates ability to understand and use English in
workplace contexts. The test does not target language use in one specific industry (e.g., banking,
accounting, travel, health care) or job category (e.g., shop clerks, accountant, tour guide, nurse) because
assessing the candidates English ability in such specific domains requires both English ability and content
knowledge, such as subject matter knowledge or job-specific terminology. Rather, the Versant Writing
Test is intended to assess how well and how efficiently the candidate can process written English on
general topics such as scheduling, commuting, and training that are commonly found in the workplace
regardless of industry or job category.

2.2 Test Design

The Versant Writing Test has five tasks: Typing, Sentence Completion, Dictation, Passage
Reconstruction, and Email Writing. All items in the Versant Writing Test elicit responses from the
candidate that are analyzed automatically. These item types provide multiple, fully independent
measures that underlie facility in written English, including sentence comprehension and construction,
passive and active vocabulary use, and appropriateness and accuracy of writing. Because more than one
item type contributes to many of the subscores, the use of multiple item types strengthens score
reliability.
The Versant Writing Test score report is comprised of an Overall score and five diagnostic subscores:
Grammar
Vocabulary
Voice & Tone
Organization
Reading Comprehension
1

For more information about the Versant English Test, please refer to Versant English Test: Test Description and Validation Summary.

2013 Pearson Education, Inc. or its affiliate(s).

Page 4 of 30

The Overall score is a weighted average of the five subscores. Together, these scores describe the
candidates facility in written English in everyday and workplace contexts. As supplemental information,
Typing Speed and Typing Accuracy are also reported on the score report.
The Versant testing system automatically analyzes the candidates responses and posts scores to a
secure website usually within minutes of completing the test. Test administrators and score users can
view and print out test results from ScoreKeeper, a password-protected section of Pearsons website
(www.VersantTest.com).

2.3 Test Administration

The Versant Writing Test is a computer-based test. The computer must have an Internet connection
and Pearsons Computer Delivered Test (CDT) software (available from http://www.versanttest.com/
technology/platforms/cdt/index.jsp).
The candidate must use a microphone headset. It is best practice to provide the candidate with the
test-taking tutorial before the actual testing begins so that the candidate can become familiar with the
test format. (Please contact Pearson for this tutorial.) When the CDT software is launched, the
candidate is prompted to enter a unique Test Identification Number (TIN) using the keyboard. The
candidate is prompted to adjust the volume before beginning the test.
The instructions for each section are spoken by a recorded examiner voice and are also displayed on
the computer screen. Candidates interact with the test system in English, typing their responses using a
keyboard. When the test is finished, the candidate clicks a button labeled, End Test.
The candidate has a set amount of time to respond to each item. A timer can be seen in the upper right
corner of the computer screen. If candidates do not finish a response in the allotted time, their work is
saved automatically and the next item begins. If candidates finish before the allotted time has run out,
they can click a button labeled Next to move on to the next item.

2.4 Test Format

The Versant Writing Test consists of five task types. Each task is introduced with instructions and an
example. The instructions and example for the tasks are spoken in an examiner voice and are also
displayed on the computer screen. It takes approximately 30 to 40 seconds to introduce each task,
after which the screen transitions to the first item.
The following subsections provide brief descriptions of the item types and the abilities required to
respond to the items in each of the five parts of the Versant Writing Test.
Part A: Typing
In the Typing task, candidates see a passage at the top of the computer screen and have 60 seconds to
type the passage exactly as it appears into a box at the bottom of the screen. All passages deal with
different aspects of typical business topics or activities. The passages are relatively simple in structure
and vocabulary and range in length from 120 to 130 words. The SMOG2 Readability Index
(http://www.harrymclaughlin.com/SMOG.htm) was used to identify and refine the readability score for
each passage. SMOG estimates the number of years of education needed to comprehend a passage. The
2

SMOG stands for Simple Measure of Gobbledygook.

2013 Pearson Education, Inc. or its affiliate(s).

Page 5 of 30

algorithm factors in the number of polysyllabic words across sentence samples (McLaughlin, 1969). All
passages have a readability score between 10 and 12, which is at a high school level and can be easily
typed by most educated English speakers with adequate typing skills.
Examples:
Whenever you have a fantastic idea, you should always write it down. If you don't,
it is quite possible that you will forget about it. Many creative people have a pen and
paper close at hand at all times. That way, whenever an interesting thought comes to
them, they are prepared to write it down. Later on, when they have time, they sit
down and read through their list of ideas.
You can benefit from this practice, too. Keeping a notebook full of thoughts is a
great way of understanding yourself better, because it tells you how you think. It
allows you to return to an interesting idea when you have the opportunity to do so.
You might find that you've created something that can change the world forever.
This task has several functions. First, since typing is a familiar task to most candidates, it is a comfortable
introduction to the interactive mode of the written test as a whole. Second, it allows candidates to
familiarize themselves with the keyboard. Third, it measures the candidates typing speed and accuracy.
The Versant Writing Test assumes a basic competence in typing for every candidate. Since it is
important to disambiguate candidates typing skills from their written English proficiency, it is
recommended that test administrators review each candidates typing score. If typing speed is below 12
Words per Minute, and/or accuracy is below 90%, then it is likely that this candidates written English
proficiency was not properly measured due to poor typing skills. The test administrator should take this
into account when interpreting test scores.
Part B: Sentence Completion
In this task, candidates read a sentence that has a word missing, and they supply an appropriate word to
complete the sentence. Occasionally, two adjacent sentences are presented but still only one word is
missing. Candidates are given 25 seconds for each item. During this time, candidates must read and
understand the sentence, retrieve a lexical item to complete the sentence, and type the word in the text
box provided. Sentences range in length from 4 to 30 words. Across all items in this task, candidates
are exposed to sentences with words missing from various parts of speech (e.g., noun, verb, adjective,
adverb) and from different positions in sentences: sentence-initial, sentence-medial, sentence-final.
Examples:
1. I'm sorry but your bill is long past __________.
2. He arrives __________ and is often the first one here.
3. I asked a coworker to take over my __________ because I wasnt feeling well.
It is sometimes thought that fill-in-the-gap tasks (also called cloze tasks) are more authentic when longer
passages or paragraphs are presented to the candidate, as this enables context-inference strategies.
However, research has shown that candidates rarely need to look beyond the immediate sentence in
order to infer the correct word to fill the gap (Sigott, 2004). This is the case even when test designers
specifically design items to ensure that candidates go beyond sentence-level information (Storey, 1997).
Readers commonly rely on sentence-level comprehension strategies partly because the sentence

2013 Pearson Education, Inc. or its affiliate(s).

Page 6 of 30

surrounding the gap provides clues about the missing words part of speech and morphology and partly
because sentences are the most common units for transmission of written communication and usually
contain sufficient context for meaning.
Above and beyond knowledge of grammar and semantics, the task requires knowledge of word use and
collocation as they occur in natural language. For example, in the sentence: The police set up a road
____ to prevent the robbers from escaping, some grammatical and semantically correct words that
might fit include obstacle, blockage or impediment. However, these would seem inappropriate
word choices to a native reader, whose familiarity with word sequences in English would lead them to
expect a word such as block or blockade.
In many Sentence Completion items there is more than one possible correct answer choice. However,
all items have been piloted with native speakers and learners of English and have been carefully reviewed
with reference to content, collocation and syntax. The precise nature of each item and possible answer
choices are quantified in the scoring models.
The sentence completion task draws on interpretation, inference, lexical selection and morphological
encoding, and as such reflects the candidates mastery of vocabulary in use.
Part C: Dictation
In the Dictation task, each item consists of one sentence. When candidates hear a sentence, they must
type the sentence exactly as they hear it. Candidates have 25 seconds to type each sentence. The
sentences are presented in approximate order of increasing difficulty. Sentences range in length from 3
words to 14 words. The items present a range of grammatical and syntactic structures, including
imperatives, wh-questions, contractions, plurals, possessives, various tenses, and particles. The audio
item prompts are spoken with a natural pace and rhythm by various native speaker voices that are
distinct from the examiner voice.
Examples:
1. Theres hardly any paper left.
2. Success is impossible without teamwork.
Corporations and companies are staying current with the latest technologies.

Dictation requires the candidate to perform time-constrained processing of the meanings of words in
sentence context. The task is conceived as a test of expectancy grammar (Oller, 1971). An expectancy
grammar is a system that governs the use of a language for someone who has knowledge of that
language. Proficient listeners tend to understand and remember the content of a message, and not the
exact words used; they retain the message rather than the words that carry the message. Therefore,
when writing down what they have heard, candidates need to use their knowledge of the language either
to retain the word string in short term memory or to reconstruct the sentence that they have
forgotten. Those with good knowledge of English words, phrase structures, and other common
syntactic forms can keep their attention focused on meaning, and fill in the words or morphemes that
they did not attend to directly in order to reconstruct the text accurately (Buck, 2001:78).
The task is a good test of comprehension, language processing, and writing ability. As the sentences
increase in length and complexity, the task becomes increasingly difficult for candidates who are not

2013 Pearson Education, Inc. or its affiliate(s).

Page 7 of 30

familiar with English words and sentence structures. Analysis of errors made during dictation reveals
that the errors relate not only to interpretation of the acoustic signal and phonemic identification, but
also to communicative and productive skills such as syntax and morphology (Oakeshott-Taylor, 1977).
The Dictation task describes one part of the grammar construct that is assessed in the Versant Writing
Test.
Part D: Passage Reconstruction
Passage Reconstruction is similar to a task known as free-recall, or immediate-recall: Candidates are
required to read a text, put it aside, and then write what they can remember from the text. In this task,
a short passage is presented for 30 seconds, after which the passage disappears and the candidate has 90
seconds to reconstruct the content of the passage in writing. Passages range in length from 30 to 75
words. The items sample a range of sentence lengths, syntactic variation and complexity.
Two discourse genres are presented in this task: narrative and email. Narrative texts are short stories
about common situations involving characters, actions, events, reasons, consequences, or results. Email
texts are adapted from authentic electronic communication and may be conversational messages to
colleagues or more formal messages to customers.
In order to accurately reconstruct a passage, the candidate must read the passage presented, understand
the concepts and details, and hold them in short term memory in order to reconstruct the passage.
Individual candidates may naturally employ different strategies when performing the task.
Reconstruction may be somewhat verbatim in some cases, especially for shorter passages answered by
advanced candidates. For longer texts, reconstruction may be accomplished by paraphrasing and
drawing on the candidates own choice of words. Regardless of strategy, the end result is evaluated
based on the candidates ability to reproduce the key points and details of the source passage using
grammatical and appropriate writing. The task requires the kinds of skills and core language
competencies that are necessary for activities such as responding to requests in writing, replying to
emails, documenting events or decisions, summarizing documents, or writing the minutes of meetings.
Examples:
(Narrative) Corey is a taxi driver. It is his dream job because he loves driving cars.
He started the job ten years ago and has been saving up money since then. Soon, he
will use this money to start his own taxi company.
(E-Mail) Thank you so much for being so understanding about our delay of
shipment. It has been quite difficult to get materials from our suppliers due to the
recent weather conditions. It is an unusual circumstance. In any case, we should be
able to ship the products to you tomorrow. In the meantime, if you have any
questions, please feel free to contact me.
The Passage Reconstruction task is held to be a purer measure of reading comprehension than, for
example, multiple choice reading comprehension questions, because test questions do not intervene
between the reader and the passage. It is thought that when the passage is reconstructed in the
candidates mother tongue then the main ability assessed is reading comprehension, but when the
passage is reconstructed in the target language (in this case, English), then it is more an integrated test of
both reading and writing (Alderson, 2000:230). Since the task is a measure of comprehension and
accurate production of sentence-level and paragraph-level writing at functional, workplace speeds,
performance is reflected in the Reading Comprehension and Grammar subscores.
2013 Pearson Education, Inc. or its affiliate(s).

Page 8 of 30

Part E: Email Writing

In this task, candidates are given an opportunity to demonstrate their writing ability using email in
relatively formal, work-related settings. Candidates are presented with a short description of a situation
and must write an email in response to the situation. Possible functions which candidates might
encounter include, but are not limited to: giving suggestions, making recommendations, requesting
information, negotiating a problem, giving feedback, and reporting an event. Candidates are given nine
minutes to read and respond to the situation. Responses of at least 100 words are expected, and those
that are less than 30 words or which are off-topic are assigned the lowest possible score.
Each email situation contains several elements:
the setting or place of work where the correspondence takes place
the addressee to whom the email is to be written, and the relationship between the
candidate and the addressee
the goal or functional purpose of the email
three themes (e.g., suggestions, reasons, or recommendations) which the candidate should address in
his/her response.
Example:
You work for a restaurant. The restaurant's manager, Ms. Johnson wants to reward
her employees for working hard but can't afford to increase salaries at this time.
Write an email to her suggesting three other ways she could reward her staff.
Your suggestions must come from the following three themes:
free lunch
employee discount
vacation days
You should include all three themes. Provide supporting ideas for each of your
suggestions.
Candidates are not expected to generate original content for their responses as the themes to address
are provided for them. However, candidates are required to construct elaborations, supporting ideas or
reasons for each of the themes. In order to fulfill the task, candidates must understand the situation
presented, relate it to their existing knowledge, and synthesize and evaluate the information such that an
appropriate response can be composed. Candidates must be conscious of the purpose of the email,
address each of the themes, and understand the relationship between themselves as the writer and the
intended recipient of the email. Candidates must fully understand the prompt in order to construct an
informative, organized, succinct response with appropriate tone, word choice, and grammatical
accuracy. Therefore, performance on the Email Writing task is reflected in the Grammar, Vocabulary,
Voice & Tone, Organization, and Reading Comprehension subscores.

2.5 Number of Items

In total, 43 items are presented to each candidate in the five separate sections, Parts A through E. In
each section, the items are drawn from a much larger item pool. For example, each candidate is
presented with twenty Sentence Completion items selected quasi-randomly from the pool, so that most
items will be different from one test administration to the next. The Versant testing system selects
2013 Pearson Education, Inc. or its affiliate(s).

Page 9 of 30

items from the item pool taking into consideration, among other things, the items level of difficulty and
its form and content in relation to other selected items. Table 1 shows the number of items presented
in each section.
Table 1. Number of items presented per section.

Task

Presented

A. Typing

B. Sentence Completion

20

C. Dictation

16

D. Passage Reconstruction

E. Email Writing

Total

43

3. Test Construct
3.1 Facility in Written English
For any language test, it is essential to define the test construct, or the skills and knowledge reflected in
the test scores (Bachman, 1990; Bachman & Palmer, 1996). The Versant Writing Test is designed to
measure a candidate's facility in written English in the workplace context, which is how well the person can
understand spoken or written English and respond in writing appropriately on everyday and workplace topics at a
functional pace.
The constructs that can be observed in the candidates performances in the Versant Writing Test are
knowledge of the language, such as grammar and vocabulary, and knowledge of writing conventions,
such as organization and tone. Underlying these observable performances are psycholinguistic skills
such as automaticity and anticipation. As candidates operate with texts and select words for constructing
sentences, those who are able to draw on many hours of relevant experience with grammatical
sequences of appropriate words will perform at the most efficient speeds.
The first concept embodied in the definition of facility is how well a candidate understands spoken or
written English. Both modalities of encoding (listening and reading) are covered in the test. Dictation
exposes candidates to spoken English and the remaining sections present written English that candidates
must read and comprehend within given time limits.
Listening dictation requires segmenting the acoustic stream into discrete lexical items and receptively
processing spoken language forms including morphology, phrase structure and syntax in real-time. The
task simulates use of the same skills that are necessary for many real-life written tasks, such as
professional transcribing, listening to a customer over the telephone and inputting information into an
electronic form, and general listening and note-taking. Buck (2001) asserts that dictation is not so much
an assessment of listening skills, as it is sometimes perceived, but rather an assessment of general
language ability, requiring both receptive and productive knowledge. This is because it involves both
comprehension and (re)production of accurate language.
Reading requires fluent word recognition and problem-solving comprehension abilities (Carver, 1991).
Interestingly, the initial and most simple step in the reading process, word recognition, is what

2013 Pearson Education, Inc. or its affiliate(s).

Page 10 of 30

differentiates native readers from even highly proficient second-language readers (Segalowitz et. al.,
1991). Native readers have massively over-learned words by encountering them in thousands of
contexts, which means that they can access meanings automatically and also anticipate frequentlyoccurring surrounding words.
Proficient language users consume fewer cognitive resources when processing spoken English or
analyzing English text visually, and therefore have capacity available for other higher-level comprehension
processes. Comprehension is conceived as parsing sentences, making inferences, resolving ambiguities,
and integrating new information with existing knowledge (Gough et. al., 1992). Alderson (2000:43)
suggests that these comprehension skills involve vocabulary, discourse and syntactic knowledge, and are
therefore general linguistic skills which may pertain to listening and writing as much as they do to
reading.
By utilizing integrated listening/reading and written response tasks, the Versant Writing Test taps core
linguistic skills and measures the ability to understand, transform and rework texts. After initial
identification of a word, either as acoustic signal or textual form, candidates who are proficient in the
language move on to higher-level prediction and monitoring processes including anticipation.
Anticipation enables faster and more accurate decoding of language input, and also underlies a
candidates ability to select appropriate words when producing text. The key skill of anticipation is
assessed in the Sentence Completion and Passage Reconstruction tasks of the Versant Writing Test as
candidates are asked to anticipate missing words and reconstruct textual messages.
The second concept in the definition of facility in written English is how well the candidate can respond
appropriately in writing. The composition tasks in the Versant Writing Test are designed to assess not
only proficiency in the core linguistic skills of grammatical and lexical range and accuracy, as described
above, but also the other essential elements of good writing such as organization, effective expression of
ideas, and voice. These are not solely language skills but are more associated with effective writing and
critical thinking, and must be learned. Assuming these skills have been mastered in the writers first
language (L1), they may be transferable and applied in the writers L2, if their core linguistic skills in L2
are sufficiently advanced. Skill in organization may be demonstrated by: presenting information in a
logical sequence of ideas; highlighting salient points with discourse markers; signposting when
introducing new ideas; giving main ideas before supporting them with details. When responding to an
email, skill in voice and tone may be demonstrated by: properly addressing the recipient; using
conventional expressions of politeness; showing understanding of the recipients point of view by
rearticulating their opinion or request; and fully responding to each of the recipients concerns.
Because the most widely used form of written communication is email, the Versant Writing Test
directly assesses the ability to compose informative emails with accuracy and correct word choice, while
also adhering to the modern conventions regarding style, rhetoric, and degree of formality for business
settings.
The last concept in the definition of facility in written English is the candidates ability to perform the
requested tasks at a functional pace. The rate at which a candidate can process spoken language, read
fluently, and appropriately respond in writing plays a critical role in whether or not that individual can
successfully communicate in a fast-paced work environment. A strict time limit imposed on each item
ensures that proficient language users are advantaged and allows for discriminating candidates with
different levels of automaticity.
The scoring of the Versant Writing Test is grounded in research in applied linguistics. A taxonomy of
the components of language knowledge which are relevant to writing are presented in a model by Grabe
2013 Pearson Education, Inc. or its affiliate(s).

Page 11 of 30

and Kaplan (1996). Their model divides language knowledge into three types: linguistic knowledge,
discourse knowledge, and sociolinguistic knowledge. These are broadly in line with the Versant Writing
Test subscores of Grammar and Vocabulary (linguistic knowledge), Organization (discourse knowledge),
and Voice & Tone (sociolinguistic knowledge).
Table 2. Taxonomy of Language Knowledge (adapted and simplified from Grabe and Kaplan, 1996:220-221).

1. Linguistic
Knowledge

a. Written code (spelling, punctuation)

b. Phonology and morphology (sound/letter correspondence, morpheme
structure)
c. Vocabulary (interpersonal, academic, formal, technical, topic-specific, nonliteral words and phrases)
d. Syntactic/Structural (syntactic patterns, formal structures, figures of
expression)

2. Discourse
Knowledge

a.
b.
c.
d.
e.

3. Sociolinguistic
Knowledge

a. Functional uses of written language

b. Register and situation (status of interactants; degree of formality; degree of
distance; topic of interaction)
c. Sociolinguistic awareness across languages and cultures

Marking devices (cohesion, syntactic parallelism)

Informational structuring (topic/comment, given/new)
Recognizing main topics
Organizing schemes (top-level discourse structure)
Inferencing (bridging, elaborating)

Aligned with the taxonomy presented in Table 2, linguistic knowledge maps onto a linguistic aspect of
performance in the scoring of the test; whereas discourse and sociolinguistic knowledge relate to a
rhetoric aspect. Comprehension is not mapped explicitly onto the taxonomy because it addresses
language knowledge as opposed to the specific information conveyed by the language. However,
comprehension is recognized as an important factor for facility in written English, and is, therefore,
identified as a unique aspect of the candidates performance in the scoring.
In sum, there are many processing elements required to participate in a written exchange of
communication: a person has to recognize spoken words or words written in an email or text received,
understand the message, formulate a relevant response, and then compose stylistically appropriate
sentences. Accordingly, the constructs that can be observed in the candidates performances in the
Versant Writing Test are knowledge of the language, such as grammar and vocabulary, comprehension
of the information conveyed through the language, and knowledge of writing conventions, such as
organization and tone. Underlying these observable performances are psycholinguistic skills such as
automaticity and anticipation. As candidates operate with texts and select words for constructing
sentences, those who are able to draw on many hours of relevant experience with grammatical
sequences of appropriate words will perform at the most efficient speeds.

3.2 The Role of Context

Grabe and Kaplans taxonomy explains why some of the test material is context-independent (e.g.
Sentence Completion) and some material is context-bound. Scoring related to Linguistic Knowledge,
such as vocabulary, discourse and syntactic knowledge, can be elicited from performance on contextbound material but is more efficiently elicited from performance on context-independent material.

2013 Pearson Education, Inc. or its affiliate(s).

Page 12 of 30

Scoring related to Discourse and Sociolinguistic Knowledge, however, requires context, awareness of
audience, and functional purpose for communication.
Except for the Email Writing task, all items present context-independent material in English. Contextindependent material is used in the test items for three reasons. First, context-independent items
exercise and measure the most basic meanings of words, phrases, and clauses on which contextdependent meanings are based (Perry, 2001). Second, when language usage is relatively contextindependent, task performance depends less on factors such as world knowledge and cognitive style and
more on the candidates facility with the language itself. Thus, the test performance relates most closely
to language abilities and is not confounded with other candidate characteristics. Third, contextindependent tasks maximize response density; that is, within the time allotted for the test, the candidate
has more time to demonstrate performance in writing the language because less time is spent presenting
contexts that situate a language sample or set up a task demand. The Dictation, Sentence Completion
and Passage Reconstruction tasks present context-independent material while the Email Writing task
presents a situation with schema that candidates must attune to, for example, the purpose of the writing
and the relationship between themselves and the intended recipient of the email. In this way, Email
Writing allows for the assessment of the grammar and mechanics of writing, as well as knowledge of the
email genre and the rhetorical and cultural norms for organizing information in emails.

3.3 The Role of Memory

Some measures of language proficiency can be misconstrued as memory tests. Since some Versant
Writing Test tasks involve holding sentences in memory in order to type them, or re-assembling
paragraphs from memory, it may seem that these tasks are unduly influenced by general memory
performance. Note that every Dictation and Passage Reconstruction item on the test was presented to
a sample of educated native speakers of English. If memory, as such, were an overriding component of
performance on the Versant Writing Test tasks, then native English speakers should show greater
performance variation on these items according to the presumed range of individuals memory spans
(see Section 8.2.5 for native-speaker performance). Also, if memory capacity (rather than language
ability) were a principal component of the variation among people performing these tasks, the test
would not correlate so closely with other accepted measures of language proficiency (see Section 7.3.2,
CEFR Level Estimates).

4. Content Design and Development

4.1 Vocabulary Selection
The vocabulary used in the test items was taken from a general English corpus and a business English
word list. The general English corpus was restricted to forms of the 8,000 most frequent words found
in the Switchboard Corpus (Godfrey and Holliman, 1997), a corpus of three million words taken from
spontaneous telephone conversations. The business English word list was restricted to forms of the
3,500 most frequent words found in the University of Cambridge Business English Certificate Preliminary
Wordlist, Barrons 600 Essential Words for the TOEIC, and Oxford Business and Finance words.

4.2 Item Development

The Versant Writing Test items were drafted by trained item writers. All items writers have advanced
degrees or training in applied linguistics, TESOL, or language testing. In general, structures used in the
test reflect those that are used in common everyday or workplace settings. The items employ a wide
range of topics from relatively general English domains to common workplace domains. The item
2013 Pearson Education, Inc. or its affiliate(s).

Page 13 of 30

writers were provided a list of potential topics/activities/situations with regard to the business domain,
such as:
Announcements
Business trips
Complaints
Customer service
Fax/Telephone/E-Mail
Inventory
Scheduling
Marketing/Sales
Item writers were specifically requested to write items so that items would not favor candidates with
work experience or require any work experience to answer correctly. The items are intended to be
within the realm of familiarity of both a typical, educated, native English speaker and an educated adult
who has never lived in an English-speaking country.
Draft items were then reviewed internally by a team of test developers, all with advanced degrees in
language-related fields, to ensure that they conformed to item specifications and English usage in
different English-speaking regions and contained appropriate content. Then, draft items were sent to
external experts on three continents. The pool of expert reviewers included several individuals with
PhDs in applied linguists and subject matter experts who worked as training and recruitment managers
for large corporations. Expert review was conducted to ensure 1) compliance with the vocabulary
specification, and 2) conformity with current colloquial English usage in different countries. Reviewers
checked that items would be appropriate for candidates trained to standards other than American
English.
All items, including anticipated responses for Sentence Completion, were checked for compliance with
the vocabulary specification. Most vocabulary items that were not present in the lexicon were changed
to other lexical items that were in the corpus and word list. Some off-list words were kept and added
to a supplementary vocabulary list, as deemed necessary and appropriate. The changes proposed by the
different reviewers were then reconciled and the original items were edited accordingly.
For an item to be retained in the test, it had to be understood and responded to appropriately by at
least 90% of a reference sample of educated native speakers of English.

4.3 Item Prompt Recording

4.3.1 Voice Distribution
Two native speakers (one male and one female) were selected for recording the spoken prompts in the
Dictation section. A professional male voice recorded the examiner prompts for the test. Recordings
were made in a professional recording studio in Menlo Park, California.
4.3.2 Recording Review
Multiple independent reviews by test developers were performed on all the recordings for quality,
clarity, and conformity to natural conversational styles. Any recording in which reviewers noted some
type of error was either re-recorded or excluded from installation in the operational test.

2013 Pearson Education, Inc. or its affiliate(s).

Page 14 of 30

5. Score Reporting
5.1 Scoring and Weighting
Of the 43 items in an administration of the Versant Writing Test, up to 41 responses are used in the
automatic scoring. The first item responses in the Sentence Completion and Dictation sections are
considered practice items and are not incorporated into the final score.
The Versant Writing Test score report is comprised of an Overall score and five diagnostic subscores
(Grammar, Vocabulary, Organization, Voice & Tone, and Reading Comprehension).
Overall: The Overall score of the test represents the ability to understand English input and
write accurate, appropriate texts at a functional pace for everyday and workplace purposes.
Scores are based on a weighted combination of the five diagnostic subscores. All scores are
reported in the range from 20 to 80.
Grammar: Grammar reflects how well the candidate understands, anticipates and produces a
variety of sentence structures in written English. The score is based on the ability to use
accurate and appropriate words and phrases in meaningful sentences.
Vocabulary: Vocabulary reflects how well the candidate understands and produces a wide
range of words in written English from everyday and workplace situations. The score is based
on accuracy and appropriateness of word use for topic, purpose, and audience.
Organization: Organization reflects how well the candidate presents ideas and information in
written English in a clear and logical sequence. The score is based on the ability to guide readers
through written text and highlight significant points using discourse markers.
Voice & Tone: Voice and Tone reflects how well the candidate establishes an appropriate
relationship with the reader by adopting an appropriate style and level of formality. The score is
based on the writer's ability to address the reader's concern and have an overall positive effect.
Reading Comprehension: Reading reflects how well the candidate understands written
English texts on everyday and workplace topics. The score is based on the ability to operate at
functional speeds to extract meaning, infer the message, and respond appropriately.
Figure 1 illustrates which sections of the test contribute to each of the five subscores. Each vertical
rectangle represents the response utterance from a candidate. The items that are not included in the
automatic scoring are shown in green.

2013 Pearson Education, Inc. or its affiliate(s).

Page 15 of 30

Grammar

Typing

Vocabulary

Voice & Tone

Sentence Completion

Dictation

Organization

Passage
Reconstruction

Reading
Comprehension

Email Writing

40 minutes
Figure 1. Relation of subscores to item types.

Table 3 shows how the five subscores are weighted to achieve an Overall score.
Table 3. Subscore weighting in relation to Versant Writing Test Overall score.

Subscore

Weighting

Grammar

30 %

Vocabulary

30 %

Organization

10 %

Voice & Tone

10 %

Reading Comprehension

20 %

Overall Score

100 %

The subscores are based on several aspects of the candidates performance: a linguistic aspect (the range
and accuracy of word use), a content aspect (the comprehensiveness of the information given), and a
rhetoric aspect (the organization and presentation of information).
The linguistic aspect is informed by the Grammar and Vocabulary subscores. Combined, these two
dimensions account for 60% of the overall score because knowledge of a wide range of words and the
accuracy of their use are the pre-requisites of successful written communication. If a candidate is unable
to produce coherent sentences that convey the intended meaning in English, then the other dimensions
of content and rhetoric may be of limited value. Conversely, if a candidate is strong in the mechanical
skills of written language, then s/he has a foundation upon which to learn higher order comprehension
and rhetorical skills.
The content aspect, or comprehensiveness of the information given in a candidates response, is
associated with the Reading Comprehension subscore. This accounts for 20% of the Overall score. It is
not only a measure of how well the candidate is able to understand textual input, but also how well the

2013 Pearson Education, Inc. or its affiliate(s).

Page 16 of 30

candidate then demonstrates understanding by responding to it. Thus, this is not a measure of pure
comprehension in the cognitive sense, but rather of comprehension and usage.
Finally, the rhetoric aspect is informed by the Organization and Voice & Tone subscores. This aspect
also accounts for 20% of the Overall score. Producing accurate lexical and structural content is
important, but effective communication depends on producing clear, succinct writing which allows for
ease of reading and gives a positive impression to the reader.
In the Versant Writing Test scoring logic, the linguistic, content, and rhetoric aspects are weighted 6020-20 respectively to reflect their importance for successful written communication.

5.2 Score Use

Once a candidate has completed a test, the Versant testing system analyzes the written performances
and posts the scores at www.VersantTest.com. Test administrators and score users can then view and
print out the test results from a password-protected section of the website called ScoreKeeper.
Score users of the Versant Writing Test may be business organizations, educational and government
institutions. Business organizations may use Versant Writing Test scores as part of the screening, hiring,
selection, language monitoring or promotion process. Within a pedagogical research setting, Versant
Writing Test scores may be used to evaluate the level of written English skills of individuals entering
into, progressing through, and leaving English language courses.
The Versant Writing Test score scale covers a wide range of abilities in written English communication.
In most cases, score users must decide what Versant Writing Test score is considered a minimum
requirement in their context (i.e., a cut score). Score users may wish to base their selection of an
appropriate cut score on their own localized research. Pearson can provide a Benchmarking Kit and
further assistance in establishing cut scores.

6. Data Resources for Scoring Development

6.1 Data Collection
Both native speakers and non-native speakers of English were recruited as participants from August
2009 through November 2009 to take a prototype data-collection version of the Versant Writing Test.
The purposes of this field testing were 1) to validate operation of the test items with both native and
non-native speakers, 2) to calibrate the difficulty of each item based on a large sample of candidates at
various levels and from various first language backgrounds, and 3) to collect sufficient written English
samples to develop automatic scoring models for written English. The description of participants is
presented in Table 4.

2013 Pearson Education, Inc. or its affiliate(s).

Page 17 of 30

Table 4. Description of participants in the field testing whose responses were used to
develop automated scoring models (n=1,768).

Native
73

Non-Native

31% : 63%
Unknown = 6%
20 - 73
mean = 35.6

44% : 49%
Unknown = 7%
19 - 67
mean = 28.0

English (U.S., U.K.,

and Australia)

Angami, Arabic, Armenian, Assamese, Bengali,

Bhojpuri, Cantonese, Catalan, Cebuano, Chinese,
Czech, Dutch, Farsi, Filipino, Fookien, French,
Garhwali, German, Gujarati, Haryanvi, Hindi,
Italian, Japanese, Kalenjin, Kannada, Korean,
Kumani, Lotha, Marathi, Maithali, Malayalam,
Manipuri, Mao, Marathi, Nepali, Oriya,
Portuguese, Punjabi, Rajasthani, Rongmei,
Russian, Serbian, Spanish, Swedish, Tagalog,
Taiwanese, Tamil, Telugu, Thai, Turkish, Urdu,
Vietnamese, Visayan, Waray-waray, Yoruba

Number of
Participants
Male : Female
Age Range

Languages

1695

6.1.1 Native Speakers

A total of 73 educated adult native English speakers were recruited. Most were from the U.S. with a few
from the U.K. and Australia. Most of them took the test multiple times producing a total of 706
completed tests. Each test was comprised of a unique set of items, so items did not overlap between
the tests. The mean age of the native speaker sample was 35.6 and the male:female ratio was 31:63.
While the Versant Writing Test is specifically designed for non-native candidates, responses from native
speakers were used to validate the appropriateness of the test items and their performance was also
used to evaluate the scoring models.
6.1.2 Non-Native Speakers
A total of 1695 non-native candidates were recruited from various countries representing both
university students and working professionals.
There were a total of 46 countries represented in the field test, but the majority of the data were
collected in Argentina, China, Germany, India, Italy, Japan, Korean, Philippines, Spain, and Taiwan. A
total of 55 different languages were reported. The male:female ratio was 44:49 with 7% of the
candidates being unreported. The mean candidate age was 28.

6.2 Expert Human Rating

During the field test of the Versant Writing Test, more than 50,000 responses were collected from
native speakers and English learners. Subsets of the response data were presented to trained raters for
developing the automatic scoring models.

2013 Pearson Education, Inc. or its affiliate(s).

Page 18 of 30

Selected item responses to Passage Reconstruction and Email Writing from a subset of candidates were
presented to twenty-one educated native English speakers to be judged for content accuracy and
vocabulary usage. Before the native speakers began rating responses, they were trained to evaluate
responses according to analytical and holistic rating criteria. All raters held a masters degree in either
linguistics or TESOL.
The raters logged in to a web-based rating system and evaluated the written responses to Passage
Reconstruction and Email Writing items for such traits as vocabulary, grammar, organization, voice and
tone, email conventions, and task completion. Rating stopped when each item had been judged by at
least two raters.

7. Validation
7.1 Validity Study Design
A series of validity analyses were conducted to examine three aspects of the Versant Writing Test
scores:
Structural Validity
1. Reliability: whether or not the Versant Writing Test is structurally reliable and assigns scores
consistently,
2. Dimensionality: whether or not the five different subscores of the Versant Writing Test are
sufficiently distinct,
3. Accuracy: whether or not the automatically scored Versant Writing Test scores are
comparable to the scores that human listeners and raters would assign,
4. Differentiation among known populations: whether or not Versant Writing Test scores reflect
expected differences and similarities among known populations (e.g., natives vs. English
learners).
Concurrent Validity
Relation to scores of tests or frameworks with related constructs: how closely do Versant Writing Test
scores predict the reliable information in scores of a well-established English test for a workplace
context (e.g., TOEIC); and how do Versant Writing Test scores correspond to the six levels of the
Common European Framework of Reference (CEFR).
7.1.1 Validation Sample
A total of 124 subjects were recruited for a series of validation analyses. These validation subjects were
recruited separately from the field test candidates. Care was taken to ensure that the training dataset
and validation dataset did not overlap for independent validation analyses. This means that the written
performance samples provided by the validation candidates were excluded from the datasets used for
training the scoring models.
Validation subjects were recruited from a variety of countries, first language backgrounds, and
proficiency levels and were representative of the candidate population using the Versant Writing Test.
A total of five native speakers were included in the validation dataset. Table 5 below summarizes the
demographic information of the validation participants.

2013 Pearson Education, Inc. or its affiliate(s).

Page 19 of 30

Table 5. Description of Participants Used to Validate the Scoring Models and Estimate Test Reliability (n=124).

Number of
Participants

124 (including 5 native speakers)

Male : Female ratio

44% : 56%

Age Range

19 - 66
mean = 30.4

Languages

Arabic, Chinese, English, Filipino, French,

German, Hindi, Italian, Japanese, Korean,
Malayalam, Russian, Spanish, Tagalog,
Tamil, Telugu, Visayan

7.2 Structural Validity

To understand the consistency and accuracy of the Versant Writing Test Overall scores and the
distinctness of the subscores, the following were examined: descriptive statistics of the validation
sample, the standard error of measurement of the Versant Writing Test Overall score; the reliability of
the Versant Writing Test (split-half reliability); the correlations between the Versant Writing Test
Overall score and its subscores, and between pairs of subscores; comparison of machine-generated
Versant Writing Test scores with listener-judged scores of the same Versant Writing Tests. These
qualities of consistency and accuracy of the test scores are the foundation of any valid test.
7.2.1 Descriptive Statistics
The mean Overall score of the validation sample was 51.74 with a standard deviation of 15.27 (on a
scale of 20-80). Table 6 summarizes some descriptive statistics for the validation sample.
Table 6. Descriptive Statistics for the Validation Dataset (n=124).

Measure

Statistic

Mean

51.74

Standard Error

1.37

Median

51.55

Standard Deviation

15.27

Sample Variance

233.07

Kurtosis

-0.44

Skewness

0.06

2013 Pearson Education, Inc. or its affiliate(s).

Page 20 of 30

7.2.2 Standard Error of Measurement

The Standard Error of Measurement (SEM) provides an estimate of the amount of error, due to
unreliability, in an individuals observed test score and shows how far it is worth taking the reported
score at face value (Luoma, 2003: 183). The SEM of the Versant Writing Test Overall score is 2.2.
7.2.3 Test Reliability
Score reliabilities were estimated by the split-half method. Split-half reliability was calculated for the
Overall score and all subscores. The split-half method divides a test into two halves and the scores
from these two halves are correlated. Then, the correlation coefficient is corrected for full-test
reliability using the Spearman-Brown Prophecy Formula. The split-half reliabilities were calculated for
both the listener-judged scores and the machine-generated scores. The reliability coefficients are
summarized in Table 7.
The values in Table 7 suggest that there is sufficient information in a Versant Writing Test item response
set to extract reliable information, and that the effect on reliability of using the Versant automated
system, as opposed to careful human rating, is quite small. The high reliability3 is a good indication that
the computerized assessment will be consistent for the same candidate assuming there are no changes in
the candidates language proficiency level.
Table 7. Split-half Reliabilities of Versant Writing Test Human Scoring versus Machine Scoring (n=124).

Split-half Reliability
for Human Scores

Split-half Reliability
for Machine Scores

Overall

0.93

0.98

Grammar

0.97

0.98

Vocabulary

0.89

0.91

Organization

0.77

0.87

Voice & Tone

0.79

0.90

Reading Comprehension

0.92

0.93

Score

The reliability for the Organization and Voice & Tone subscores is lower than the reliability of the other
subscores because these subscores are estimated solely from Email Writing, of which only two items
are presented in the test. However, the agreement between two raters for these subscores was
sufficiently high: inter-rater reliability for Organization was 0.90 and inter-rater reliability for Voice &
Tone was 0.93 at the item level (corrected for under-estimation).
7.2.4 Dimensionality: Correlations among Subscores
Ideally, each subscore on a test provides unique information about a specific dimension of the
candidates ability. For language tests, the expectation is that there will be a certain level of covariance
between subscores given the nature of language learning. This is due to the fact that when language
3

The possible reliability coefficient range is 0 to 1. The closer the coefficient is to 1.0, the greater the reliability is.

2013 Pearson Education, Inc. or its affiliate(s).

Page 21 of 30

learning takes place, the candidates skills tend to improve across multiple dimensions. However, if all
the subscores were to correlate perfectly with one another, then the subscores might not be measuring
different aspects of facility with the language.
Table 8 presents the correlations among the Versant Writing Test subscores and the Overall score for
the same validation sample of 124 candidates, which includes five native English speakers.
Table 8. Inter-correlation between Subscores on the Versant Writing Test (n=124).

Grammar
Vocabulary
Organization
Voice & Tone
Reading
Comp.

Grammar
0.81
0.77
0.79

Vocab

Organization

Voice &
Tone

0.81
0.83

0.98

0.91

0.88

0.87

0.89

Reading
Comp.

Overall
0.96
0.96
0.89
0.91
0.96

As expected, test subscores correlate with each other to some extent by virtue of presumed general
covariance within the candidate population between different component elements of written language
skills. The Organization and Voice & Tone subscores correlate highly with one another since they are
both representing the rhetoric aspect of written language from the same set of items. However, the
correlations between the remaining subscores are below unity (i.e., below 1.0), which indicates that the
different scores measure different aspects of the test construct.
7.2.5 Machine Accuracy
An analysis for internal quality of the test involved comparing scores from the Versant Writing Test,
which uses automated language processing technologies, versus careful human judgments from expert
raters.
Table 9 presents Pearson Product-Moment correlations between machine scores and human scores,
when both methods are applied to the same performances on the same Versant Writing Test responses.
The candidate sample is the same set of 124 validation candidates that was used in the reliability and
subscore analyses. The human scores in Table 9 were calculated from a single human judgment, which
means that the correlation coefficients are conservative (higher coefficients can be obtained with
multiple human ratings).
Table 9. Correlation Coefficients between Human and Machine Scoring of
Versant Writing Test Responses (n = 124).

Score Type

Correlation

Overall

0.98

Grammar

0.99

Vocabulary

0.98

Organization

0.90

Voice & Tone

0.91

Reading Comprehension

0.96

2013 Pearson Education, Inc. or its affiliate(s).

Page 22 of 30

The correlations presented in Table 9 suggest that scoring a Versant Writing Test by machine will yield
scores that closely correspond with human ratings. Among the subscores, the human-machine relation
is closer for the linguistic (Grammar and Vocabulary) and content (Reading Comprehension) aspects of
written language than for the rhetoric aspect (Organization and Voice & Tone), but the relation is close
for all five subscores. At the Overall score level, Versant Writing Test machine-generated scores are
virtually indistinguishable from scoring that is done by multiple independent human judgments.
7.2.6 Differentiation among Known Populations
The next validity analysis examined whether or not the Versant Writing Test scores reflect expected
differences between native English speakers and English language learners. Overall scores from a total of
400 tests completed by the native speakers and 1709 tests completed by the learners representing a
range of native languages were compared. Figure 2 presents cumulative distributions of Overall scores
for the native and non-native speakers. Note that the range of scores displayed in this figure is from 10
through 90, whereas Versant Writing Test scores are reported on a scale from 20 to 80. Scores
outside the 20 to 80 range are deemed to have saturated the intended measurement range of the test
and are therefore reported as 20 or 80.

Percent Cumulative Density [%]

The results show that native speakers of English consistently obtain high scores on the Versant Writing
Test. Fewer than 5% of the native sample scored below 70, which was mainly due to performance in
Email Writing (i.e. rhetorical written skills rather than language skills). Learners of English as a second
or foreign language, on the other hand, are distributed over a wide range of scores. Note also that only
10% of the non-natives scored above 70. In sum, the Overall scores show effective separation between
native and non-native candidates.

Versant Writing Test Score

Figure 2. Cumulative density functions of Versant Writing Test Overall scores for the native and non-native
groups (native n=400 and non-native n=1709).

2013 Pearson Education, Inc. or its affiliate(s).

Page 23 of 30

7.3 Concurrent Validity

One important goal of the validity studies is to understand how the Versant Writing Test relates to
other measures of English proficiency. Since the Versant Writing Test has an emphasis on workplace
English, it would be most sensible to explore a relationship with another well-known workplace English
test. For this reason, a study was undertaken to compare the automatically derived Versant Writing
Test Overall scores with the Test of English for International Communication (TOEIC). In addition,
another study was undertaken to identify the correspondence between the scores on the Versant
Writing Test and the six levels of the Common European Framework of Reference (CEFR).
7.3.1 Versant Writing Test and TOEIC
The TOEIC Listening and Reading test was used as a concurrent validation. The TOEIC Listening and
Reading test is claimed to measure a non-native speakers listening and reading skills in English as these
skills are used in the workplace. The test was developed about 30 years ago as a measure of receptive
language skill and has been widely accepted and used worldwide. (Liao, Qu, & Morgan, 2010). The
Listening and Reading subscores are both reported in the range of 5 to 495 for a total score between 10
and 990.
Method
The study was conducted between November 2009 and February 2010. The participants were 28
Japanese and 27 South Koreans who represented a mix of full-time students and working professionals.
Of the 55 participants, 26 were male and 29 female with a mean age of 24. The participants were
recruited by agents in Japan and Korea acting on Pearsons behalf (a university professor and two
business professionals).
The participants took both the Versant Writing Test and TOEIC with a gap between sittings of no less
than 30 days. All participants were asked to take a shorter version of the Versant Writing Test as a
demo test so their resulting performance would more closely relate to their proficiency levels, rather
than reflect their unfamiliarity with the Versant Writing Test. They took the Versant Writing Test
individually at their home, school, or workplace. The TOEIC tests were administered during the official
test administrations. No institutional TOEIC tests were used.
Results
The inter-correlation matrix between the subscores of each test is given in Table 10. The values across
all subscores are at or above r=0.68. Not surprisingly, the highest correlation coefficients (0.96 and
0.91) exist between subscores (or modules) and the overall scores for the same test. This is true for
both Versant Writing Test and TOEIC.

2013 Pearson Education, Inc. or its affiliate(s).

Page 24 of 30

Table 10. Pearson Correlation Coefficients for Versant Writing Test and TOEIC (n=55).

TOEIC Reading
TOEIC Listening
TOEIC Total
Versant Writing Test

TOEIC
Reading
0.84
0.96
0.70

TOEIC
Listening

TOEIC
Total

0.96
0.68

0.72

Though the sample size is small, this matrix shows an expected pattern of relationships among the
subscores of the tests, bearing in mind that they all relate to English language ability but assess different
dimensions of that ability.
The Versant Writing Test and TOEIC Total correlated moderately at r=0.72, as shown in Figure 3,
indicating that there is general English ability as a covariance, but that these tests measure different
aspects of language performance. The Versant Writing Test correlated higher with TOEIC Reading
(r=0.70) than with TOEIC Listening (r=0.68), which is expected because more content is presented
through reading than listening in the Versant Writing Test.

TOEIC Total Score

Versant Writing Test Score vs. TOEIC Total Score

Versant Writing Test Score

Figure 3. Scatterplot showing the relationship between the Versant Writing Test and TOEIC (n=55).

7.3.2 Versant Writing Test and CEFR Level Estimates

In order to identify the correspondence between scores on the Versant Writing Test and CEFR, a
standard-setting procedure was conducted following the guidelines of the Manual for Relating Language
Examinations to the Common European Framework of Reference (Council of Europe, 2001). The goal was to
identify minimum scores (cut scores) on the Versant Writing Test scale that maps to the A1 through C2

2013 Pearson Education, Inc. or its affiliate(s).

Page 25 of 30

proficiency levels of the CEFR. A secondary goal of the study was to empirically demonstrate that two
item types found on the Versant Writing Test, Passage Reconstruction and Email Writing, can be reliably
evaluated by English language testing experts.
Method
A set of analytic descriptors containing six levels was developed from the CEFR scales, corresponding to
CEFR levels A1, A2, B1, B2, C1, and C2. Six English language testing experts were recruited as expert
judges. They were instructed to utilize the CEFR descriptors to grade holistically, and choose the CEFR
level that best fit each response. A response set of written samples was created using the following
procedure: 240 candidates who took a field test version of the Versant Writing Test were selected via
stratified random sampling. This sampling technique was used to assure that the response set contained
written samples from a wide variety of language backgrounds and equally distributed proficiency levels,
approximately 40 per CEFR level. The candidates came from China, Costa Rica, France, Germany, India,
Iran, Japan, Korea, Mexico, the Netherlands, Russia, Spain, Taiwan, Thailand, and the United States.
Eleven of the candidates were excluded from analysis either before or after the rating process due to
incomplete data (most or all responses were blank), leaving 229 individual candidates in the response
set. Each candidate contributed a total of five written responses from two tasks: three Passage
Reconstruction responses and two Email Writing responses. The response set therefore consisted of
1145 written samples: 687 Passage Reconstruction responses and 458 Email Writing responses.
Results
Raters demonstrated a high level of consistency with one another in their assigned scores (r=0.98). This
high level of inter-rater reliability demonstrates that candidates can be consistently classified into CEFR
levels based on performances elicited by these tasks. The CEFR ratings from the six raters and the
Versant Writing Test scores for each candidate were entered into a Rasch model to produce an ability
estimate for each candidate on a common logit scale. Initial CEFR boundaries were then estimated from
Rasch ability estimates, as shown in Table 11.
Table 11. CEFR Score Boundaries as Logits from a Rasch Model.

Facetstep

CEFR Level

1
2
3
4
5
6

A1
A2
B1
B2
C1
C2

Expectation Measure at
CEFR Boundary (Logits)
-4.43
-2.45
-0.68
0.88
2.39
4.22

Candidates Versant Writing Test scores were then lined up next to their CEFR-based ability estimates
to establish the score boundaries. When comparing the aggregated expert judgments with the Versant
Writing Test scores to establish a CEFR Level, 68% of candidates are correctly classified and 99% of
candidates are classified correctly or one step away. Table 12 below provides the final mapping between
the two scales.

Table 12. Mapping of CEFR Levels with Versant Writing Test Scores.

2013 Pearson Education, Inc. or its affiliate(s).

Page 26 of 30

CEFR Level

Versant Writing Test

Score Range

A1

20-29

A2

30-43

B1

44-53

B2

54-66

C1

67-76

C2

77-80

Figure 4 plots the relation between each candidates Versant Writing Test score (shown on the x-axis)
and their CEFR ability estimate in logits as estimated from the judgments of the six panelists (shown on
the y-axis). The figure also shows the original Rasch-based CEFR boundaries (horizontal dotted lines)
and the slightly adjusted boundaries (vertical dotted lines).

CEFR Estimate (Logits)

Versant Writing Test Score vs. CEFR Estimate

Versant Writing Test Score

Figure 4. Scatterplot showing Rasch-based CEFR ability estimates as derived from human
judgments and Versant Writing Test scores.

The Pearson correlation coefficients for Versant Writing Test scores and CEFR estimates is 0.95,
revealing that the Versant Writing Test instrument yields test scores which are highly consistent with
judges evaluation of written performance using the CEFR scales.
The raters CEFR ratings were based on two tasks (Email Writing and Passage Reconstruction) which
elicit linguistic, content and rhetorical skills. However, it is important to note that the Versant Writing
Test Overall score is derived not only from performance on these two tasks, but also on Sentence
2013 Pearson Education, Inc. or its affiliate(s).

Page 27 of 30

Completion and Dictation which assess linguistic skills more reliably. Therefore, some error in CEFR
classification is to be expected when individuals have substantially different linguistic skills than content
and rhetorical skills.

8. Conclusion
This report has provided details of the test development process and validity evidence for the Versant
Writing Test. The information is provided for test users to make an informed interpretive judgment as
to whether test scores would be valid for their purposes. The test development process is documented
and adheres to sound theoretical principles and test development ethics from the field of applied
linguistics and language testing:
the items were written to specifications and were subjected to a rigorous procedure of
qualitative review and psychometric analysis before being deployed to the item pool;
the content was selected from both pedagogic and authentic material;
the test has a well-defined construct that is represented in the cognitive demands of the tasks;
the scores, item weights and scoring logic are explained;
the items were widely field tested on a representative sample of candidates.
This report provides empirical evidence demonstrating that Versant Writing Test scores are structurally
reliable indications of candidate ability in written English and are suitable for high-stakes decision-making.

9. About the Company

Pearson: Pearsons Knowledge Technologies group and Ordinate Corporation, the creator of the
Versant tests, were combined in January, 2008. The Versant tests are the first to leverage a completely
automated method for assessing spoken language.
Ordinate Testing Technology: The Versant automated testing system was developed to apply
advanced speech recognition techniques and data collection to the evaluation of language skills. The
system includes automatic telephone and computer reply procedures, dedicated speech recognizers,
speech analyzers, databanks for digital storage of speech samples, and score report generators linked to
the Internet. The Versant Writing Test is the result of years of research in statistical modeling,
linguistics, testing theory and speech recognition. The Versant patented technologies are applied to
Pearsons own language tests such as the Versant series and also to customized tests. Sample projects
include assessment of spoken English, assessment of spoken aviation English, childrens reading
assessment, adult literacy assessment, and collections and human rating of spoken language samples.
Pearsons Policy: Pearson is committed to the best practices in the development, use, and
administration of language tests. Each Pearson employee strives to achieve the highest standards in test
publishing and test practice. As applicable, Pearson follows the guidelines propounded in the Standards
for Educational and Psychological Testing, and the Code of Professional Responsibilities in Educational
Measurement. A copy of the Standards for Educational and Psychological Testing is available to every
employee for reference.
Research at Pearson: In close cooperation with international experts, Pearson conducts ongoing
research aimed at gathering substantial evidence for the validity, reliability, and practicality of its current
products and investigating new applications for Ordinate technology. Research results are published in
international journals and made available through the Versant website (www.VersantTest.com).

2013 Pearson Education, Inc. or its affiliate(s).

Page 28 of 30

10. References
Alderson, J. C. (2000). Assessing reading. Cambridge: Cambridge University Press.
Bachman, L.F. (1990). Fundamental considerations in language testing. Oxford: Oxford University Press.
Bachman, L.F. & Palmer, A.S. (1996). Language testing in practice. Oxford: Oxford University Press.
Buck, G. (2001). Assessing listening. Cambridge: Cambridge University Press.
Council of Europe (2001). Common European Framework of Reference for Languages: Learning,
teaching, assessment. Cambridge: Cambridge University Press.
Godfrey, J.J. & Holliman, E. (1997). Switchboard-1 Release 2. LDC Catalog No.: LCD97S62.
http://www.ldc.upenn.edu.
Gough, P.B., Ehri, L.C., and Treiman, R. (1992). Reading acquisition. Hillsdale, NJ: Erlbaum.
Grabe, W., and Kaplan, R.C. (1996). Theory and practice of writing. New York: Longman.
Liao, C-w., Qu, Y., and Morgan, R. (2010). The Relationship of Test Scores Measured by the TOEIC
Listening and Reading Test and TOEIC Speaking and Writing Tests (TC-10-13). Retrieved from
Educational Testing Service website: http://www.ets.org/research/policy_research_reports/tc10-13
Luoma. (2003). Assessing speaking. Cambridge: Cambridge University Press.
McLaughlin, G.H. (1969). SMOG grading: A new readability formula. Journal of Reading, 12(8), 639646.
Oakeshott-Taylor, J. (1977). Information redundancy, and listening comprehension. In R. Dirven
(ed.), Hrverstndnis im Fremdsprachenunterrict. Listening comprehension in foreign language
teaching. Kronberg/Ts.: Scriptor.
Oller, J. W. (1971). Dictation as a device for testing foreign language proficiency. English Language
Teaching, 25(3),254-259.
Perry, J. (2001). Reference and reflexivity. Stanford, CA: CSLI Publications.
Segalowitz, N., Poulsen, C., and Komoda, M. (1991). Lower level components or reading skill in
higher level bilinguals: Implications for reading instruction. In J.H. Hulstijn and J.F. Matter (eds.),
Reading in two languages, AILA Review, Vol. 8,. Amsterdam: Free University Press, 15-30.
Sigott, G. (2004). Towards identifying the C-test construct. New York: Peter Lang.
Storey, P. (1997). Examining the test-taking process: a cognitive perspective on the discourse cloze
test. Language Testing, 14(2), 214-231.

2013 Pearson Education, Inc. or its affiliate(s).

Page 29 of 30

4040 Campbell Avenue

Suite 200
Menlo Park, California 94025
USA

4940 Pearl East Circle

Suite 200
Boulder, Colorado 80301
USA

Version 0313b

2013 Pearson Education, Inc. or its affiliate(s).

Page 30 of 30