Professional Documents
Culture Documents
Testing
http://ltj.sagepub.com/
Published by:
http://www.sagepublications.com
Additional services and information for Language Testing can be found at:
Email Alerts: http://ltj.sagepub.com/cgi/alerts
Subscriptions: http://ltj.sagepub.com/subscriptions
Reprints: http://www.sagepub.com/journalsReprints.nav
Permissions: http://www.sagepub.com/journalsPermissions.nav
Citations: http://ltj.sagepub.com/content/23/4/441.refs.html
This article provides renewed converging empirical evidence for the hypothesis
that asking test-takers to respond to text passages with multiple-choice questions induces response processes that are strikingly different from those that
respondents would draw on when reading in non-testing contexts. Moreover,
the article shows that the construct of reading comprehension is assessment
specific and is fundamentally determined through item design and text selection. The data come from qualitative analyses of 10 cognitive interviews conducted with non-native adult English readers who were given three passages
with several multiple-choice questions from the CanTEST, a large-scale
language test used for admission and placement purposes in Canada, in a
partially counter-balanced design. The analyses show that:
These findings support the development of response process models that are
specific to different item types, the design of further experimental studies of test
method effects on response processes, and the development of questionnaires
that profile response processes and strategies specific to different item types.
I Introduction
When people read in a non-testing context, they do not answer
multiple-choice (MC) questions in their heads. While this statement
Address for correspondence: Andr A. Rupp, Institut zur Qualittsentwicklung im Bildungswesen,
Humboldt Universitt zu Berlin, Unter den Linden 6 10099 Berlin, Germany; email:
Andre.Rupp@IQB.hu-berlin.de
Language Testing 2006 23 (4) 441474 10.1191/0265532206lt337oa
442
comprehension for mature readers, and different models of responding to MC reading comprehension questions for mature readers. The
subsequent section describes the participants, the methodology, and
the instruments that were used to collect and analyse the data in this
study. Following this, results from qualitative analyses of the interview transcripts are presented and are linked to relevant findings in
the reading assessment literature. The article then closes with a section that presents directions for future research and reflects on the
strengths and limitations of this study.
II Deconstructing reading comprehension
Without doubt, there is no such thing as the comprehension of a
text. As Kintsch (1998: 2) noted in his book entitled Comprehension:
a paradigm for cognition:
The terms understanding and comprehension are not scientific terms but are
commonsense expressions. As with other such expressions, their meaning is
fuzzy and imprecise . . . What seems most helpful here is to contrast
understanding with perception on the one hand and with problem solving on
the other.
Both of these quotes can be used to underline two key facts. First,
comprehension is more complex than mere perception even
though comprehension processes draw upon perception processes.
Second, and most relevant for the purpose of this article,
comprehension of texts in non-testing contexts is not simply the
result of a continual, conscious, and linear engagement in problemsolving activities, which, as will be shown in the article, contrasts
sharply with responding to MC questions about a passage in a testing context.
To provide a detailed theoretical context for the qualitative results
used to corroborate this contention later, the following three sections
of the article discuss a processing, reader-purpose, and genre
perspective on reading comprehension assessment (see Enright
et al., 2000).
444
446
448
Question-related strategies:
Answer the questions you know first
Use the process of elimination to make the best educated guess
Avoid choices that are too specific or too broad
Always look for choices that sound consistent with the main idea
Use prior knowledge to answer questions
Text-related strategies:
Read the first sentence of each paragraph for the main idea
Look for how the text is organized and ignore details
Try to predict where the authors points are leading
Get the gist of each paragraph
Pay special attention to the first part of the passage
Find short sentences within paragraphs
Preview key sentences
When reading, form ideas about the text
Relate what you read to what you already know
Look for context clues for the meaning of an unfamiliar word
General strategies:
Budget your time
Read the questions first before reading the text
Read the text first before reading the questions
Identify major reading question types
Look for key words
Remember that the questions follow the order of the passage
Dont try to read every word
Try to summarize after you read
Lurie et al.
(2005)
*
*
*
*
*
*
Goodman
et al. (2005)
Martinson
(2005)
*
*
Green and
Wolf (2000)
Source material
List of representative recommended strategies for reading comprehension sections of the GRE
Recommended strategy
Table 1
*
*
Goodman
et al. (2004)
Text-related strategies:
Read the first sentence of each
paragraph for the main idea
Look for how the text is organized
and ignore details
Try to predict where the authors
points are leading
General strategies:
Budget your time
Read the text first before reading
the questions
Read the questions first before
reading the text
Identify major reading
question types
Look for key words
Remember that the questions follow
the order of the passage
Dont try to read every word
Try to summarize after you read
Sullivan
et al. (2004)
Hinkel
(2004)
Lougheed
(2003)
Rymniak and
Shanks (2002)
Rogers
(2005)
Source material
Shmailo
(2002)
*
*
Gallagher
(2000)
List of representative recommended test-taking strategies for reading comprehension sections of the TOEFL
Recommended strategy
Table 2
Sullivan
et al. (2000)
450
Assessing reading with multiple-choice questions
Question-related strategies:
Answer the questions you know first
Use the process of elimination to
make the best educated guess
Avoid choices that are too specific
or too broad
Always look for choices that
sound consistent with the main idea
Use prior knowledge to
answer questions
452
social groups who share such knowledge of use in appropriate contexts (Grabe, 2000; 2002). Therefore, questions about a text from a
given genre A are likely to engage learners in very different response
processes than questions for a text from an alternative genre B due to
the inherent characteristics of the genres.
A recent study by Kobayashi (2002) investigated the impact of
different response formats and genre types on the performance of
Japanese adult ESL readers. The author found that text organization,
which is related to text type or genre, did not lead to strong performance differences for test formats that measured less integrative comprehension such as cloze tests or for learners of limited ESL
proficiency. On the contrary, stronger performance differences due to
organizational differences in texts were observed for testing formats
that measure more integrative forms of comprehension, especially
for learners with higher levels of ESL proficiency. This interaction
effect between learner competency and testing format highlights,
again, that the construct of reading comprehension that is assessed
and the processes that learners engage in will change as a result of
the testing format and text types used.
4 Levels of reading comprehension assessed with MC questions
The discussions so far have not addressed differences in item formats
and their influence on the response process. This article focuses
specifically on MC questions, but, even within an item format such
as MC, the quality and intensity of the reading comprehension
process that test-takers engage in can vary considerably across items.
Specifically, it appears reasonable to state that an analysis of the
structure and content of MC questions on any reading comprehension test will typically reveal that very different levels of reading
comprehension are assessed with different items. This becomes evident in typologies of such questions. A glance at the types of MC
questions published in testing preparation manuals for tests like the
TOEFL or the SAT reveal that they are likely to assess a mixture of
what could be termed local and global comprehension processes,
which will force readers to draw on component and integrative
processes to different degrees.
For example, using the TOEFL reading comprehension section as
a basis, Sheehan et al. (1999) differentiate between MC questions
that ask readers to identify the main idea of a passage, to directly
search a specific piece of information, to disambiguate vocabulary in
context, and to resolve anaphoric or pronominal references. As one
454
Therefore, apart from gaining a better understanding of how testtakers interact with the materials to successfully answer questions,
the goal of this study was to illustrate how reading comprehension is
a construct with many fine-grained differentiations and shades that
become operationalized differentially across passages and questions.
As stated earlier, a secondary goal of this study was to develop a
strategy inventory and a questionnaire that could be administered in
both small-scale and large-scale testing settings, which is available
from the first author on request and is currently being field tested.
III Method
The following sections describe the participants that were recruited
for this study, the instruments that were used to elicit data on
response processes and perceptions of texts and MC questions, and
the procedure for administering these instruments.
1 Participants
There were 10 participants in this study that were recruited from second language courses at a large Canadian university in Ontario; the
participants were selected because they had either recently taken the
CanTEST or were planning on taking it soon. Moreover, in dialogue
with the CanTEST administration office, the characteristics of our
participants matched those of typical test-takers in Ontario; Table 3
shows the characteristics of the participants.
The participants came from Argentina, Brazil, Canada, China,
Damaskus, Syria, and Sri Lanka; 3 participants were male and 7
were female. They had generally had taken several MC tests on other
27
34
19
33
19
20
18
26
25
10
Female
Female
Female
Female
Female
Male
Gender
Female
Brazil
Male
Argentina Female
Syria
Canada
China
Canada
China
China
Canada
Country
1.3
2.1
6.2
6.9
19.0
3.8
19.4
3.0
1.7
12.0
Time in
Canada
(in years)
21
Age
Description of participants
Participant
Table 3
18.1
23.3
13.0
20.3
15.0
25.0
14.1
19.2
17.0
15.0
Time in
school
(in years)
8.0
19.0
8.0
n/p
4.0
23.0
7.1
14.2
10.0
15.0
Time studying
English
(in years)
Bachelors
degree (i.p.)
Masters
degree
Masters
degree (i.p.)
Bachelors
degree (i.p.)
Masters
degree
Bachelors
degree (i.p.)
Bachelors
degree (i.p.)
High school
diploma
Masters
degree
Bachelors
degree (i.p.)
Education
Somewhat
Very much
Very much
Very much
Little
Very much
Much
Very much
Very much
Much
Experience
with MC
test
Experience
with MC
reading test
Little
Somewhat
Much
Somewhat
Much
Much
Somewhat Much
Comfort
level with
MC tests
456
Assessing reading with multiple-choice questions
458
section on each form of the test consists of a paragraph comprehension section and a cloze test, the latter not being of importance for
this study. The paragraph comprehension section consists of three
passages with about six items per passage, most of which are MC
questions with a few of them being open-ended questions. For the
purpose of this study, we utilized only the MC questions on passages
and classified the passages by content domain; we eventually chose
11 passages from 6 content domains. According to the schema presented in Enright et al. (2000: 30), the passages on the CanTEST are
all expository texts and the items required the test-takers to find
information, read for basic comprehension, and read to learn.
In addition to the CanTEST we also piloted a questionnaire that
profiles test-takers in MC reading comprehension contexts, a revised
version of which is to be administered with the CanTEST in the
future. The questionnaire contained many open-ended questions that
allowed us to fine-tune answer choices for the operational version,
which will be in a selected-response format due to time constraints
in the administration.
4 Procedure of administration
The participants were given three texts in different sequences. The
passages chosen as the first text were sampled from two domains
only whereas the other two passages were sampled from four
domains with the order of presentation partially counterbalanced.
Specifically, the first text was considered a preparation text where we
observed the participants while they were responding to the questions and recorded their behavior such as how long they spent on
each question or whether they seemed to focus on specific sections
of the text. For the second texts, participants were allowed to read the
text or the questions first and, after they responded to all questions,
we asked them, question by question, how they selected their answer,
how they would rate the difficulty of the question on a five-point
Likert scale from very easy to very difficult and what made the
questions difficult or easy; the median perceived difficulty ratings of
the items given by the participants are presented in Table 4.
As can be seen in Table 4, most items were rated by the participants as either very easy or easy even though an inspection of the
answers showed that the difficulty ratings did not correlate highly
with the answer scores. For the third text, we asked the participants
to think aloud while responding to each of the questions and asked
them similar questions for clarification.
Content
domain
Item 1
Item 2
Item 3
Item 4
Item 5
Item 6
Item 7
Median
per text
Environment
Statistics
Statistics
Food
Food
Technology
Technology
Language
Language
Other
academic
Other
academic
3.0
1.0
2.5
1.5
1.5
2.0
1.5
1.0
2.5
2.5
2.5
1.5
2.0
2.0
2.0
2.5
2.0
1.0
1.0
1.0
3.0
2.5
2.0
2.5
2.5
2.5
1.0
2.5
1.0
1.5
3.5
1.0
2.5
4.0
1.5
2.5
2.5
1.5
2.5
3.5
2.0
1.5
3.0
3.0
3.0
2.5
2.5
2.0
3.0
1.5
2.0
1.5
2.0
1.5
1.5
2.5
n/a
2.5
2.0
2.0
1.0
1.5
n/a
1.5
1.0
n/a
n/a
n/a
n/a
n/a
2.50
1.50
2.25
2.00
1.50
2.50
2.00
1.75
2.25
1.75
2.0
4.0
2.0
2.0
4.0
n/a
n/a
2.00
Notes : The scale ranges from 1 very easy to 5 very difficult; n/a not
applicable because the text had fewer items
460
4) Scan or read the questions first to assess the general topic of the
text implied in the questions as well as the question foci looking
for or underlining key words. Then answer the questions sequentially by reading the text in chunks.
In particular, participants experiences with MC tests played an
important role in developing their unconditional strategies. One participant, who had taken about 15 to 20 MC tests since he was 14
years old, expressed his unconditional strategy as follows:
So basically what I did is first I read the question pretty fast (to have a sense
of) what they (the questions) looking for, like, (and) what kind of answer
they looking for. So I just read (questions) fastly. And then I directly to the
answer sheet, I mean to the test (text) and so I just can (read it) until something match like some words match and then I try to find out I see some
similarities in the text I try to find out the right answer inside, so that is the
way I use . . . I mean, when the meaning of the question is clear, I go right
away I know it is somewhere here. Usually it is somewhere here I know the
main idea and the main idea should be somewhere here, and I can, like,
answer the question . . .
462
2) Scan or read the questions first before reading the text to assess
their difficulty.
If the questions are perceived to be easy, answer the first
question first and then proceed sequentially.
If the questions are perceived to be difficult, scan the text and
look for or underline key words first. Then answer the questions sequentially.
3) Scan or read the questions first before reading the text looking
for or underlining key words.
If the text is perceived to be short, read the entire text
looking for or underlining key words and answer the questions sequentially.
If the text is perceived to be long, immediately begin to
answer the questions sequentially.
4) Scan the text for length.
If the text is perceived to be long, start to answer questions
sequentially.
If the text is perceived to be short, read the text first looking
for or underlining key words. Then answer the questions
sequentially.
A few clear patterns emerge from the above strategies. First and foremost, for both unconditional and conditional strategies, the process
of responding to MC questions in our sample and context heavily
relied on key word matching, a process which test-takers often
facilitate by underlining or highlighting individual words or phrases
considered to be pertinent for understanding the text or answering
the questions. As one participant stated in response to how he located
relevant information for answering a question:
So they are talking about device here so I see the same word here, device,
so I stop over here and I know, like, the answers over here. So I just stop by
and try to like look around and I am pretty sure it is maybe one sentence
before or maybe one sentence after so that is how I proceed.
semantically or logically build on each other and induce such a linking. However, it is doubtful that those instances were frequent.
Second, the unconditional strategies were essentially variants of
the conditional strategies. The main deciding factor in choosing a
particular conditional strategy appeared to be the perceived difficulty
of the text or the questions, which are complex perceptions consisting of the perceived length of the text, the perceived familiarity with
the text type, the perceived familiarity with its topic, the perceived
familiarity with its vocabulary, and the perceived complexity of the
processes required to answer a question correctly.
Third, responding to MC questions was rarely a linear process in
which a text was read first to form a textbase with an integrated
situation model as suggested by the CI framework or to build a
coherent mental representation of nested structures as suggested by
the structure-building framework and then the set of MC questions
was answered. More commonly, texts were merely scanned for key
words and this took place either in chunks or after questions had
been scanned or had been read.
b Micro-level strategies: Our study showed that the response
process to individual items was a complex process with certain common features that could be abstracted.
464
One participant stated that she did not even attempt to eliminate
choices more carefully because:
the main idea was who can get this information I think when I read I got the
idea that, the information was already . . . what I remembered from the paragraph on, its like an answer while I was reading this was much more quickly
so I didnt need to eliminate, but there were three and like uh, no, uh, no, uh,
no. It was easy because in some way its like I knew what I was looking for.
as follows:
I picked C, because A, designing new homes for large families, youd need
demographics data to do that really not necessarily census data for new
homes. Theres always going to be expansion within a city now whether you are
going to build homes for large families, small families, all depending, would
you really rely on census data? Like federal government census data to do that?
No. Building components for computers, well, that just doesnt . . . why. . . I
dont know why wed use census data. Its about people not things its about
things that they need, but really how important is, like, a hard drive to somebody, you know, depending. Now finding a more favorable climatic zone, that
one was a good one, but no. Why would you? You dont need census data to figure out its cold there its warm here, you know, that just does not strike me.
Moving your business to a new area, yes. If you have a candy store, youre moving to a city with a lot of children, or if you have a toy store or something like
that or if youre moving to, like, a retirement community more like pharmacy
or, you know, chiropractors or something like that. See, that makes more sense
to me that youd want to know what kind of population you are going into.
466
On the other hand, items that asked for the main idea of a passage were usually perceived as easy, because test-takers could skip
local detailed information without necessarily losing key
information required to respond correctly (see Rupp et al., 2001).
One participant stated clearly her perception of easiness of the main
idea item:
I think its the main idea and its very declared in the first paragraph so thats
also the reason why I think this is actually very easy for me.
468
VI Conclusions
Our findings strongly suggest that the sequence and structure of MC
questions appear to provide important cues for test-takers that allow
or influence them to select response strategies, which may result in
response processes that deviate significantly from those predicted by
a model of reading comprehension in non-testing contexts. Despite
newer types of MC questions that focus more strongly on higher-level
reading comprehension, which can be found on the newer TOEFL or
SAT versions, for example, we hypothesize that test-takers frequently
segment a text into chunks that are aligned with individual questions
and focus predominantly on the microstructure representation of a
text base rather than the macrostructure of a situation model. As a
result, higher-order inferences that may lead to an integrated
macrostructure situation model in a non-testing context are often suppressed or are limited to grasping the main idea of a text.
It is indisputable that theoretical models of reading comprehension are often well suited to explain the cognitive processes of reading comprehension in non-testing contexts, but they neglect the
segmentation and localization functions of many types of MC questions. Put differently, it is certainly possible to link observations such
as responses to MC questions and verbal responses to interviewer
questions to individual components of theoretical models for higherorder reading comprehension such as the CI model, the structurebuilding framework, or a general information-processing model.
However, models for MC responding that postulate an integrated
reading comprehension process that is linearly followed by a decision-making process neglect testing format effects, which manifest
themselves in variation amongst response processes. That variation
is, in turn, grounded in the variation among test-takers, text types,
question types, and question formats.
470
472
474
Sarig, G. 1987: High-level reading in first and in the foreign language: some
comparative process data. In Devine, J. Carrell, P.L. and Eskey, D.E.
editors, Research in reading as a second language. Washington, DC:
Teachers of English to Speakers of Other Language 2, 10520.
Sheehan, K.M. and Ginther, A. 2001: What do passage-based MC verbal reasoning items really measure? An analysis of the cognitive skills underlying performance on the current TOEFL reading section. Paper presented
at the 2000 Annual Meeting of the National Council of Measurement in
Education.
Sheehan, K.M., Ginther, A. and Schedl, M. 1999: Development of a proficiency scale for the TOEFL reading comprehension section. TOEFL
Research Report. Princeton, NJ: Educational Testing Service.
Shmailo, L. 2002: Mastering the TOEFL CBT. New York: Kaplan.
Stanovich, K. 1980: Toward an interactive-compensatory model of individual
differences in the development of reading fluency. Reading Research
Quarterly 16, 3271.
1986: Matthew effects in reading: Some consequences of individual differences in the acquisition of literacy. Reading Research Quarterly 21,
360406.
Sullivan, P.N., Brenner, G.A. and Zhong, G.L.Q. 2004: Master the TOEFL
2005. Lawrenceville, NJ: Thomson/Peterson.
Sullivan, P.N., Zhong, G.L.Q. and Brenner, G.A. 2000 : Everything you need
to score high on the TOEFL. New York: Pearson Education.
Underwood, G. and Batt, V. 1996: Reading and understanding: an introduction to the psychology of reading. Cambridge, MA: Blackwell.
Van Dijk, T. 1985: Strategic discourse comprehension. In Ballmer, T., editor,
Linguistic dynamics: Discourses, procedures, and evolution. Berlin:
Walter de Gruyter, 3061.
Wenden, A. and Rubin, J. editors. 1987: Learner strategies in language learning. London: Prentice Hall International.