Reliability of Speaking Proficiency Tests

Published Date: 12 Feb 2018

Introduction

Testing, as a part of English teaching, is a very important procedure, not just because it can be a valuable source of information about the effectiveness of learning and teaching but also because it can improve teaching, and arouse the student's motivation to learn. Testing oral proficiency has become one of the most important issues in language testing since the role of speaking ability has become more central in language teaching with the advent of communicative language teaching (Nakamura, 1993). However, assessing speaking is challenging (Luoma, 2004). Validity and reliability, as fundamental concerns and essential measurement qualities of the speaking test (Bachman, 1990; Bachman & Palmer, 1996; Alderson et al, 1995), have aroused widespread attention. The validation of the speaking test is an important area of research in language testing.

Test of oral proficiency just started in China 15 years ago, and there are a few very dominant tests. An increasing number of Chinese linguists are putting their attention and efforts on analysis of their validity and reliability. Institutions began to introduce speaking tests into English exams in recent years with the widespread promotion of communicative language teaching (CLT). Publications that deal with speaking tests within institutions provide some qualitative assessments (Cai, 2002). But there is relatively little research literature relating to the reliability and validity of such measures within a university context. (Wen, 2001).

The College English Department at Dalian Nationalities University (DLNU) has been selected as one of thirty-one institutions of the College English Reform Demonstration Project in the Peoples' republic of China. In College English (CE) course of DLNU, the speaking test is one of the four subtests of the final examination of English assessment. The examination uses two different formats. One is a semi-direct speaking test, in which examinees talk to microphones connected to computers, and have their speeches recorded for the teachers to rate afterwards. The other is a face-to-face interview. This research in this paper aims to ascertain the degree of the reliability and validity of the speaking tests. By analyzing the results of the research, teachers will become more aware of the validity and reliability of oral assessments, including how to improve the reliability and validity of speaking tests. I, as a language teacher, will gain insight into the operation of language proficiency test, In order to better degree of reliability and validity of a particular test, I will also take other qualities of test usefulness into account when designing the language proficiency test., such as practicality and authenticity.

Research questions:

This study mainly addresses the questions of validity and reliability of the speaking test administered at DLNU. They are comprehensive concepts that involve analysis of test tasks, administration, rating criteria, examinee and tester's attitudes towards the test, the effect of the test on teaching and teacher or learner attitudes towards learning the tests (Luoma, 2004). Therefore, the purpose of this study is to answer the following research questions:

1. Is the speaking test administered at DLNU a valid and reliable test? This question can involve the following two sub-questions:

1) To what extent is the speaking test administered at DLNU reliable?

2) To what extent is the speaking test administered at DLNU valid?

2. In what aspects and to what extent may the validity and reliability of the speaking test administered at DLNU be improved?

Literature Review

This chapter presents a theoretical framework of speaking construct, ways of testing speaking, marking of speaking test and the reliability and validity of speaking test, also introduces the situation of speaking test in China.

Analyzing Speaking And Speaking Test

The Nature Of Speaking

Speaking, as a social and situation-based activity, is an integral part of people's daily lives (Luoma, 2004). Testing second language speaking is often claimed to be a much more difficult undertaking than testing other second language abilities, capacities or competencies, skills¼ˆUnderhill, 1987). Assessment is difficult not only because speaking is fleeting, temporal and ephemeral, but also because of the comprehensibility of pronunciation, the special nature of spoken grammar and spoken vocabulary, as well as the interactive and social features of speaking (Luoma, 2004), because of the “unpredictability and dynamic nature” of language itself (Brown, 2003). To have a clear understanding of what it means to be able to speak a language, we must understand that the nature and characteristics of the spoken language differ from those of the written form (Luoma, 2004; McCarthy & O'Keefe, 2004; Bygate, 2001) in its grammar, syntax, lexis and discourse patterns due to the nature of spoken language.

Spoken English involves reduced grammatical elements arranged into formulaic chunk expressions or utterances with less complex sentences than written texts. Spoken English breaks the standard word order because the omitted information can be restored from the instantaneous context (McCarthy & O'Keefe, 2004; Luoma, 2004; Bygate, 2001; Fulcher, 2003). Spoken English contains frequent use of the vernacular, interrogatives, tails, adjacency pairs, fillers and question tags which have been interpreted as dialogue facilitators (Luoma, 2004; Carter & McCarthy, 1995). The speech also contains a fair number of slips and errors such as mispronounced words, mixed sounds, and wrong words due to inattention, which is often pardoned and allowed by native speakers (Luoma, 2004). Conversations are also negotiable, unpredictable, and susceptible to social and situational context in which the talks happen (Luoma, 2004).

The Importance Of Speaking Test

Testing oral proficiency has become one of the most important issues in language testing since the role of speaking ability has become more central in language teaching with the advent of CLA (Nakamura, 1993). Of the four language skills (listening, speaking, reading, &writing), listening and reading occur in the receptive mode, while speaking and writing exist in the productive mode. Understanding and absorption of received information are foundational while expression and use of acquired information demonstrate an improvement and a more advanced test of knowledge. A lot of interests now in oral testing is partly because second language teaching is more than ever directed towards the speaking and listening skills¼ˆUnderhill, 1987). Language teachers are engaged in “teaching a language through speaking” (Hughes, 2002:7). On one hand, spoken language is the focus of classroom activity. There are often other aims which the teacher might have: for instance, helping the student gain awareness of practice in some aspect of linguistic knowledge (ibid). On the other hand, speaking test, as a device for assessing the learners' language proficiency also functions to motivate students and reinforce their learning of language. This represents what Bachman (1991) has called an “interface” between second language acquisition (SLA) and language testing research.

However, assessing speaking is challenging, “because there are many factors that influence our impression of how well someone can speak a language” (Luoma, 2004:1) as well as unpredictable or impromptu nature of the speaking interaction. The testing of speaking is difficult due to practical obstacles and theoretical challenges. Much attention has been given to how to perfect the assessment system of oral English and how to improve its validity and reliability. The communicative nature of the testing environment also remains to be considered (Hughes, 2002).

The Construct Of Speaking

Introduction To Communicative Language Ability (CLA)

A clear and explicit definition of language ability is essential to language test development and use (Bachman,1990). The theory on which a language test is based determines which kind of language ability the test can measure, This type of validity is called construct validity. According to Bachman (1990:84), CLA can be described as “consisting of both knowledge or competence and the capacity for implementing or executing that competence in appropriate, contextualized communicative language use”. CLA includes three components: language competence, strategic competence and pyschophysiological mechanisms. The following framework (figure 2.1) shows components of communicative language ability in communicative language use (Bachman,1990:85).

Knowledge Structures Language Competence

Knowledge of the world Knowledge Of Language

Strategic Competence

Psychophysiological Mechanisms

Context Of Situation

This framework has been widely accepted in the field of language testing. Bachman (1990:84) proposes that “language competence” essentially refers to a set of specific knowledge components that are utilized in communication via language. It comprises organizational and pragmatic competence. Two areas of organizational knowledge that Bachman (1990) distinguishes are grammatical knowledge and textual knowledge. Grammatical knowledge comprises vocabulary, syntax, phonology and graphology, and textual knowledge, comprises cohesion and rhetorical or conversational organization. Pragmatic competence shows how utterances or sentences and texts are related to the communicative goals of language users and to the features of the langue-use setting. It includes illocutionary acts¼Œor language functions, and sociolinguistic competence, or the knowledge of the sociolinguistic conventions that govern appropriate language use in a particular culture and in varying situations in that culture (Bachman, 1987).

Strategic competence refers to mastery of verbal and nonverbal strategies in facilitating communication and implementing the components of language competence. Strategic competence is demonstrated in contextualized communicative language use, such as socialcultural knowledge, real-world knowledge and mapping this onto the maximally efficient use of existing language abilities.

Psychophysiological competence refers to the visual and auditory skill used to gain access to the information in the administrator's instructions. Among other things, psychophysiological competence includes things like sound and light.

Fulcher's Construct Definition

To know what to assess in a speaking test is a prime concern. Fulcher (1997b) points out that the construct of speaking proficiency is incomplete. Nevertheless, there have been various attempts to reflect the underlying construct of speaking ability and to develop theoretical frameworks for defining the speaking construct. Fulcher's framework (figure 2.2) (Fulcher, 2003: 48) describes the speaking construct.

As Fulcher (2003) points out that there are many factors that could be included in the definition of the construct:

Phonology: the speaker must be able to articulate the words, have an understanding of the phonetic structure of the language at the level of the individual word, have an understanding of intonation, and create the physical sounds that carry meaning.

Fluency and accuracy: these concepts are associated with automaticity of performance and the impact on the ability of the listener to understand. Accuracy refers to the correct use of grammatical rules, structure and vocabulary in speech. Fluency has to do with the ‘normal' speed of delivery to mobilise one's language knowledge in the service of communication at relatively normal speed. The quality of speech needs to be judged in terms of the gravity of the errors made or the distance from the target forms or sounds.

Strategic competence: this is generally thought to refer to an ability to achieve one's communicative purpose through the deployment of a range of coping strategies. Strategic competence includes both achievement strategies and avoidance strategies. Achievement strategies contain overgeneralization/morphological creativity. Learners transfer knowledge of the language system onto lexical items that they do not know, for example, saying “buyed” instead of “bought”, Speakers also learn approximation: learners replace an unknown word with one that is more general or they use exemplification, paraphrasing (use a synonym for the word needed), word coinage (invent a new word for an unknown word), restructuring (use different words to communicate the same message), cooperative strategies (ask for help from the listener) , code switching (take a word or phrase from the common language with the listener in order to be understood) and non-linguistic strategies (use gestures or mime, or point to objects in the surroundings to help to communicate). Avoidance or reduction strategies consist of formal avoidance (avoiding using part of the language system) and functional avoidance (avoiding topical conversation). Strategic competence includes selecting communicative goals and planning and structuring oral production so as to fulfill them.

Textual knowledge: competent oral interaction involves some knowledge of how to manage and structure discourse, for example, through appropriate turn-taking, opening and closing strategies, maintaining coherence in one's contributions and employing appropriate interactional routines such as adjacency pairs.

Pragmatic and sociolinguistic knowledge: effective communication requires appropriateness and the knowledge of the rules of speaking. A range of speech acts, politeness and indirectness can be used to avoid causing offence.

Ways Of Testing Speaking

Clark (1979) puts forward a theoretical basis to discriminate three types of speaking tests: direct, semi-direct and indirect tests. Indirect tests belong to “procommunicative” era in language testing, in which the test takers are not actually required to speak. It has been regarded as having the least validity and reliability, while the other two formats are more widely used (O'Loughlin, 2001). In this section, the characteristics, advantages and disadvantages of the direct and semi-direct test are presented,

The Oral Proficiency Interview Format

One of the earliest and most popular direct speaking test formats, and one that continues to exert a strong influence, is the oral proficiency interview (OPI) –developed originally by the FSI (Foreign Service Institute) in the United States in the 1950s and later adopted by other government agencies. It is conducted with individual test-taker by a trained interviewer, who assesses the candidate using a global band scale (O'Loughlin, 2001). It typically begins with a warm-up discussion of a few easy questions, such as getting to know each other or talking about the day's events. Then the main interaction contains the pre-planned tasks, such as describing or comparing pictures, narrating from a picture series, talking about a pre-announced or examiner-selected topic, or possibly a role-play task or a reverse interview where the examinee asks question of the interviewer (Luoma. 2004). An important example of this type of test is the speaking component of the International English Language Testing System (IELTS), which is adopted in 105 different countries around the world each year.

The Advantage Of An Interview Format

The oral interview was recognized as the most commonly used speaking test format. Fulcher (2003) suggests that it is partly because the questions used can be standardized, making comparison between test takers easier than when other task types are used. Using this method, the instructor can get a sense of the oral communicative competence of students and can overcome weakness of written exams, because the interview, unlike written exams, “is flexible in that the questions can be adapted to each examinee's performance, and thus the testers have more controls over what happens in the interaction” (Luoma, 2004:35). It is also relatively easy to train raters and obtain high inter-rater reliability (Fulcher, 2003).

The Disadvantage Of An Interview Format

However, concern and skepticism exist about whether it is possible to test other competencies or knowledge because of the nature of the discourse that the interview produces (van Lier, 1989).

a. Issue of time

For the instructor, time management can be quite an issue. For instance, using a two-hour period for exams for 20 students means each student is allowed only six minutes for testing. This includes the time needed to enter the room and adjust to the setting. With such a time limit the student and instructor can hardly have any kind of normal real-world conversation.

b. Issue of asymmetrical relationship

The asymmetrical relationship between examiners and candidates elicits a form of inauthentic and limited socio-cultural contexts (van Lier, 1989; Savignon, 1985; Yoffe, 1997). Yoffe (1997) commented on ACTFL (American Council on the Teaching of Foreign Languages) OPI that the tester and the test-taker are “clearly not in equal positions” (Yofee, 1997).

The asymmetry is not specific to the OPI but is inherent in the notion of an 'interview' as an exchange wherein one person solicits information in order to arrive at a decision while the interlocutor produces what he or she perceives as most valued. The interviewee is, in most cases, acutely aware of the ramifications of the OPI rating and is, consequently, under a great deal of stress.

Van Lier (1989) also challenges the validity of OPI in terms of the asymmetry between them because “the candidate speaks as to a superior and is unwilling to take the initiative” (van Lier, 1989). Under the unequal relationship, the speech discourse, such as turn –taking, topic nomination and development, and repair strategies are all substantially different from normal conversational exchanges (see van Lier 1989).

c. Issue of interviewer variation

Given the fact that the interviewer has considerable power over the examinee in an interview, concerns have been aroused about the effect of the interlocutor (examiner) on the candidate's oral performance. Different interviewers vary in their approaches and attitudes toward the interview. Brown (2003) warns the danger of such variation to fairness. O'Sullivan (2000) conducts an empirical study that indicated learners perform better when interviewed by a woman, regardless of the sex of the learner. Underhill (1987:31) expresses his concern on the unscripted “flexibility… means that there will be a considerable divergence between what different learners say, which makes a test more difficult to assess with consistency and reliability.”

Testing Speaking In Pairs

There has been a shift toward a paired speakers format: two assessors examine two candidates at a time. One assessor interacts with the two candidates and rates them on a global scale, while the other does not take part in the interaction and just assesses--using an analytic scale. The paired oral test has been used as part of large-scale, international, standardized oral proficiency tests since the late 1980s (Ildikó, 2001). Key English Test (KET), Preliminary English Test (PET), First Certificate in English (FCE) and Certificate in Advanced English (CAE) make use of the paired format. In a typical test, the interaction begins with a warm-up, in which the examinees introduce themselves to the interlocutor, followed by two pair interaction task. The talk may involves comparing two photographs by each candidate at first, such as in Cambridge First Certificate (Luoma, 2004), then a two-way collaborative task between the two candidates based on more photographs, artwork or computer graphics, and ends up with a three-way discussion with the two examinees and the interlocutor about a general theme that is related to the earlier discussion.

The advantages of the paired interview format

Many researchers claim that the paired format is preferable to OPI. The reasons are:

a. The changed role of the interviewer frees up the instructors in order to pay closer attention to the production of each candidate than if they are participants themselves (Luoma, 2004).

b. The reduced asymmetry allows more varied interaction patterns, which elicits a broader sample of discourse and increased turn-takings than were possible in the highly asymmetrical traditional interview (Taylor, 2000).

c. The task type based on pair-work will generate a positive washback effect on classroom teaching and learning (Ildiko, 2001). In the case of the instructor following Communicative Language Teaching (CLT) methodology, where pair work may take up a significant portion of a class, it would be appropriate to incorporate similar activities in the exam. In that way the exam itself is much better integrated into the fabric of the course. Students can be tested for performance related to activities done in class. There may also be benefits in regards to student motivation. If students are aware that they will be tested on activities similar to the ones done in class, they may have more incentive to be attentive and use class time effectively.

The disadvantages of the paired interview format

There are, however, also concerns voiced regarding the paired format.

a. Mismatches between peer interactants

The most frequently raised criticisms against the paired speaking test relate to various forms of mismatches between peer interactants (Fulcher, 2003). Ildiko (2001) points out that when a candidate has to work with an incomprehensible or uncomprehending peer partner, it may negatively influence the candidate's performance. As a consequence, in such cases it is quite impossible to make a valid assessment of candidates' abilities.

b. Lack of familiarity between peer interactants

The extent to which this testing format actually reduces the level of anxiety of test-takers compared to other test formats remains doubtful (Fulcher, 2003). O'Sullivan (2002) suggests that the spontaneous support offered by a friend positively reduces anxiety and task performance under experimental conditions. However, the chances are quite high that the examinee will meet with strangers as his or her peer interactant. It is hard to imagine how these strangers can carry out some naturally flowing conversations. Estrangement, misinterpretation and even breakdown may occur during their talk.

c. Lack of control of the discussion

Problems are generated if the examiner loses control of the oral task (Luoma, 2004). When the instructions and task materials are not clear enough to facilitate the discussion, the examinees' conversation may go astray. Luoma (2004) points out that testers often feel uncertain about what amount of responsibility that they should give to the examinees. Furthermore, examinees do not know what kind of performance will earn them good results without the elicitation of the examiner. When one of the examinees has said too little, the examiner ought to monitor and jump in to give help when necessary.

Semi-Direct Speaking Tests

The term “semi-direct” is employed by Clark (1979:36) to describe those tests that are characterized “by means of tape recordings, printed test booklets, or other ‘non-human' elicitation procedures, rather than through face-to-face conversation with a live interlocutor.” Appearing during 1970s, and being an innovative adaptation of the traditional OPI, the semi-direct method normally follows the general structure of the OPI and makes an audio-recording of the test taker's performance which is later rated by one or more trained assessors (Malone, 2000). Examples of the semi-direct type used in the U.S.A. are the simulated oral proficiency interviews (SOPI) and the Test of Spoken English 2000 (TSE) (Ferguson, 2009). Examples in U.K. include the Test in English for Education Purpose (TEEP) and the Oxford-ARELS Examinations (O'Loughlin, 2001). Another mode of delivery is testing by telephone -- as in the PhonePass test (the test mainly consists of reading sentences aloud or repeating sentences), or even video-conferencing (Ferguson, 2009).

The Advantages Of The Semi-Direct Test Type

First, the semi-direct test is more cost efficient than direct tests, because many candidates can be tested simultaneously in large laboratories and administered by any teacher, language lab technician or aide in a language laboratory where the candidate hears taped questions and has their responses recorded (Malone, 2000).

Second, the mode of testing is quite flexible. It provides a practical solution in situations where it is not possible to deliver a direct test (O'Loughlin, 2001), and it can be adapted to the desired level of examinee proficiency and to specific examinee age groups, backgrounds, and professions (Malone, 2000).

Third, semi-direct testing represents an attempt to standardize the assessment of speaking while retaining the communicative basis of the OPI (Shohamy, 1994). It offers the same quality of interview to all examinees, and all examinees respond to the same questions so as to remove the effect that the human interlocutor will have on the candidate (Malone, 2000). The uniformity of the elicitation procedure greatly increases the reliability of the test.

Some empirical studies (Stansfield, 1991) show high correlations (0. 89- 0. 95) between the direct and semi-direct tests, indicating the two formats can measure the same language abilities and the SOPI can be the equivalent and surrogate of the OPI. However, there are also disadvantages.

The Disadvantages Of The Semi-Direct Test Type

First, the speaking task in semi-direct oral test is less realistic and more artificial than OPI (Clark, 1979; Underhill, 1987). Examinees use artificial language to “respond to tape-recorded questions -- situations the examinee is not likely to encounter in a real-life setting” (Clark, 1979:38). They may feel stressful while speaking to a microphone rather than to another person, especially if they are not accustomed to the laboratory setting (O'Loughlin, 2001).

Second, the communicative strategy and speech discourse elicited in these semi-direct SOPIs is quite different from that found in typical face-face interaction – being more formal, less conversation-like (Shohamy, 1994). Candidates tend to use written language in tape-mediated test, more of a report or narration; while, they focus more on interaction and on delivery of meanings in OPI.

Third, there are often technical problems that can result in poor quality recordings or even no recording in the SOPI format (Underhill, 1987).

In conclusion, one cannot assume any equivalence between a face-to face test and a semi-direct test (Shohamy, 1994). It may be that they are measuring different things, different constructs, so the mode of test delivery should be adopted on the basis of test purpose, accuracy requirement, practicability, and impartiality (Shohamy, 1994). Stansfield (1991) proposes the OPI is more applicable to the placement test and evaluation test of the curriculum, while SOPI is more appropriate for large-scale test with requirement of high reliability.

Marking Of Speaking Test

Marking and scoring is a challenge in assessing second language oral proficiency.. Since only a few elements of the speaking skill can be scored objectively, human judgments play major roles in assessment. How to establish the valid, reliable, effective marking criteria scales and high quality scoring instruments have always been central to the performance testing of speaking (Luoma, 2004). It is important to have clear, explicit criteria to describe the performance, as it is important for raters to understand and apply these criteria, making it possible to score them consistently and reliably. For these reasons, rating and rating scales have been a central focus of research in the testing of speaking (Ferguson, 2009).

Definition Of Rating Scales

A rating scale, also referred to as a “scoring rubric” or “proficiency scale” is defined by Davies et al as following (see Fulcher, 2003):

·consisting of a series of band or levels to which descriptions are attached

·providing an operational definition of the constructs to be measured in the test

·requiring training for its effective operation

Holistic And Analytic Rating Scales

There are different types of rating scales used for scoring speech samples. One of the traditional and commonly used distinctions is between holistic and analytic rating scales. Holistic rating scales also are referred to as global rating. With these scales, the rater attempts to match the speech sample with a particular band whose descriptors specify a range of defining characteristics of speech at that level. A single score is given to each speech sample either impressionistically or by being guided by a rating scale to encapsulate all the features of the sample (Bachman & Palmer, 1996).

Analytic rating scales: They consist of separate scales for different aspects of speaking ability (e.g. grammar / vocabulary; pronunciation, fluency, interactional management, etc). A score is given for each aspect (or dimension), and the resulting scores may be combined in a variety of ways to produce a composite single overall score. They include detailed guidance to raters, and rich information that they provide on specific strengths and weakness in examinee performance (Fulcher, 2003). Analytic scales are particularly useful for diagnostic purposes and for providing a profile of competence in the different aspects of speaking ability (Ferguson, 2009). The type of scale that is selected for a particular test of speaking will depend upon the purpose of the test

Validity And Reliability Of Speaking Test

Bachman And Palmers Theories On Test Usefulness

The primary purpose of a language test is to provide a measure that can be interpreted as an indicator of an individual's language ability (Bachman, 1990; Bachman and Palmer, 1996). Bachman and Palmer (1996) propose that test usefulness including six test qualities—reliability, construct validity, authenticity, interactiveness, impact (washback) and practicality. Their notion of usefulness can be expressed as in Figure2.3:

Usefulness=Reliability + Construct validity + Authenticity +

Interactiveness + Impact +Practicality

These qualities are the main criteria used to evaluate a test. “Two of the qualities --- reliability and validity --- are critical for tests and are sometimes referred to as essential measurement qualities” (Bachman & Palmer, 1996:19), because they are the “major justification for using test scores as a basis for making inferences or decisions” (ibid). The definitions of types of validity and reliability will be presented in this section.

Validity And Reliability

Defining Validity

The quotation from AERA (American Educational Research Association ) indicates:

“Validity is the most important consideration in test evaluation. The concept refers to the appropriateness, meaningfulness, and usefulness of the specific inferences made from test scores. Test validation is the process of accu­mulating evidence to support such inferences. A variety of inferences may be made from scores produced by a given test, and there are many ways of accumulating evidence to support any particular inference. Validity, however, is a unitary concept. Although evidence may be accumulated in many ways, validity always refers to the degree to which that evidence supports the inferences that are made from the score. The inferences regarding specific uses of a test are validated, not the test itself.”

(AERA et al., 1985: 9)

Messick stresses that “it is important to note that validity is a matter of degree, not all or none' (Messick, 1989:53). Validity lies in the test scores and inferences drawn from the scores. Validity is multifaceted and different types of evidence are needed to support any claims for the validity of scores on a test (Bachmann, 1990:89). The fact that there are many ways to establish the validity leads to the topic of types of validity.

Types Of Validity

It should be pointed out that different types of validity are distinguished by many writers based on different test purposes. The major validations mentioned here are from Alderson et al's (1995) framework.

Construct validity: it is seen as the most basic, fundamental type of validity. Bachman and Palmer (1996:21) state that: “Construct validity pertains to the meaningfulness and appropriateness of the interpretations that we make on the basis of test scores.” That is, to justify the interpretation of a test score, we need to provide evidence that the test score reflects the area(s) of language ability that we want to measure and little else. As we all know, construct validity is the specific definition of an ability that provides the basis for a given test or test task. The degree of construct validity is determined by the relationship between the purpose of a test and the theory on which the test is based. Alderson et al (1995) suggest ways to validate the construct validity: correlate each subtest with other subtests, with total test or with total minus self; multitrait-multimethod studies and factor analysis.

Internal Validity

Internal validity relates to studies of “the perceived content of the test and its perceived effect” (Alderson et al, 1995:171). There are two kinds of internal validity:

Face Validity: it can be defined as the extent to which the test appeals to test takers and test users (Bachman & Palmer, 1996). Regarded as the most superficial form of validity, it pertains to the public acceptability of a test. It is often determined impressionistically. Questionnaires to or interviews with candidates or administrator include questions like: does it look fair and appropriate to the test-takers and to the public? Does the test appear to measure what it claims to measure? Do the test tasks look something like what you might do in a real world setting? (Henning, 1987; Ferguson, 2009).

Content Validity: it is “concerned with whether or not the content of the test is sufficiently representative and comprehensive for the test to be a valid measure of what it is supposed to measure.”(Henning, 1987:94). It involves only the test, and not the performance of test takers. Its validation should be based on the analysis of the language being tested and on the particular course objectives. Alderson et al (1995) suggest methods to validate content validity: compare test content with specifications/syllabus, questionnaires to and interviews with “experts” such as teachers, subject specialists, applied linguists, expert judges rate test items and texts according to precise list of criteria, etc..

External Validity

External validity relates to studies “comparing students' test scores to measures of their ability gleaned from outside the test” (Alderson et al, 1995:171). It comprises two types:

Concurrent validity: It in essence involves the comparison of the test scores with some other measures for the same candidates taken at roughly the same time as the test (Alderson et al, 1995:177). This produces a correlation coefficient which suggests the extent t which the tests are measuring the same thing. Ways to validate concurrent validity include: correlate students' test scores with students scores on other tests; students' test scores with other measures of ability such as students' or teachers' ratings (ibid).

Predictive Validity: It shows that the test can predict how successful the learners will be at using the language in the future (Underhill, 1987). Ways to validate predictive validity include: correlate students' test scores with their scores on tests taken some time later; correlate students' test scores with other measures of ability taken some time later, such as subject teachers' assessments, language teachers' assessment; and also students' scores with success in study or at work (Alderson et al, 1995).

Reliability

Defining Reliability

Reliability is defined as consistency and stability of measurement (Bachman & Palmer, 1996). “A reliable test score will be consistent across different characteristics of the testing situation.” (Bachman & Palmer, 1996:19). It shows that the accuracy and consistency of the measurement are reflected in obtaining the similar results over repeated tests involving the same subjects.

Types Of Reliability

Luoma (2004) presents three types of reliability particularly relevant for speaking test.

Intra-rater reliability or internal consistency: it means that raters agree with themselves, over a period of few days, about the ratings that they give. In other words, if a person rates a test one day then rates the test the same on another day, the test is said to have high intra-rater reliability.

Inter-rater reliability: it means that different raters rate performances similarly. They do not necessarily need to agree completely. However, well-defined criteria help raters agree, and frequent disagreements may indicate either that some raters are not able to apply the criteria consistently or that the criteria need to be defined better.

Parallel form reliability: examinees are asked to take two or more of the different forms of tests, and their scores are analyzed for consistency. If the scores are not consistent, the forms cannot be considered parallel--- assuming of course that the raters are internally consistent.

Relationship Between Validity And Reliability

Bachman (1990: 161) points out: “The concerns of reliability and validity can thus be seen as leading to two complementary objectives in designing and developing tests: (1) to minimize the effects of measurement error, and (2) to maximize the effects if the language abilities we want to measure.”

As the two prime concerns of the qualities in the test, validity and reliability are interrelated concepts. Reliability is a prerequisite to validity in performance assessment in the sense that no test can achieve its intended purpose if the test results are unreliable. Similarly, inferences cannot be drawn from an invalid test regardless of its reliability.

Speaking Test In China

In this section, the importance of English language and development of speaking test in China are discussed with a view to offering a better understanding of the background of the empirical research.

The Importance Of English Language In China

In China, English is one of the three core compulsory subjects along with mathematics and Chinese that are tested for students wishing to enter schools of higher level (Cheng, 2008). As an obligatory subject for all majors in Chinese universities and colleges, students are required to pass the CET (College English Test) Band IV to obtain their bachelor's degrees. Apart from the academic requirements for English, better English skills and abilities are also preferential accesses to employments, selection and promotion.

Development Of Speaking Tests In China

Despite all the efforts of the Chinese Educational Ministry to promote English proficiency nationwide, the situation of English education cannot satisfy the needs of the social development (Cai, 2002). Students' pragmatic competence is weak compared to other language competences, especially their oral abilities (ibid). Many empirical studies have been conducted showing that spoken ability is the weakest of the four basic skills (Wen, 2001). The weakness of spoken language learning has been a leading problem for English education for a long time.

The development of speaking test in China started in the 1990s, for the language testing had focused on testing of reading and writing abilities. The Cambridge Business English Certificate (BEC) was first introduced to China in 1993 with an oral component. The speaking sub-test in the Test for English Majors (TEM) began in 1994. The National Matriculation English Test -- Oral Subtest (NMETOS) was formally introduced in 1995 in China (see Cheng, 2008). The CET-- Spoken English Test (CET-SET) started in 1999. The Public English Testing System (PETS) with a speaking component has begun to be promoted since 1999.

The reasons the development of the speaking test started late in China are various. The following are major reasons. On one hand, objectively there are many difficulties involved in the construction and administration of the speaking assessment due to the discrepancies among colleges and universities, regions and areas in terms of teaching resources, students' level of English upon entering college, and the social needs they face. Unified effective assessment instruments are lacking and the measurement tools are hard to grasp and implement. Large students population, limited testers and time make it impracticable to administer the speaking test.

On the other hand, subjectively, communicative assessment has received little attention. Hughes (1989) states that there is a great discrepancy between the predominance of the communicative approach and the accurate measurement of communication ability. With the widespread promotion of communicative language teaching (CLT) in EFL countries replacing the traditional grammar-centered, text-centered, and teacher-centered methods, English teachers have been trying to carry out CLT in their classrooms. However, communicative speaking assessment has not been seriously practiced in a way that reflects authentic interaction in test task design. So the matters concerning validity and reliability have been given little attention in oral assessments in China.

Induction To CET-SET

CET is a large-scale standardized test administered nationwide by the National College English Testing Committee on behalf of the Higher Education Department of Chinese Ministry of Education (CME) (Cheng, 2008). The main aim is to measure English proficiency of college and university undergraduate students in accordance with the National College English Teaching Syllabus (CME, 1999). As a norm-referenced test, nowadays, CET “has nearly become the unifying criterion in judging the English level of non-English major students of university in educational field and even in the whole society” (Zhu, 2004). In most colleges and universities, the certificate of CET-4 is one of the requirements to obtain a Bachelor's degree. As we can see, the CET has exerted a huge amount of influence on English language teaching and learning at the tertiary level in China due to its high stake (Cheng, 2008).

CET came into being in 1987, but its component Spoken English Test (SET) started in 1999. It is available to students who have passed the CET-4 with a score of 80% or above or the CET-6 with a score of 75% or above. Every speaking test is administered by an interviewer, who talks to the candidates and controls the direction and topic of the conversation, also rates their performance, along with an assessor, who listens to a learner speaking and only makes an evaluative judgement on what he or she hears. The candidates are comprised by 3 or 4 learners. Topics cover various fields based on examinees' outlooks to the life and the world, such as, ideal jobs, campus life, holiday, festival activities, TV programmes, education in China, environment and human, modern life style, social communication, etc. The rating scale is listed in Appendix 1. The test procedure consists of three parts as it is shown in the following table 2.3.

Table 2.3 CET-SET procedure (National College English Syllabus for Non-English Majors, 1999).

Background Of Testing At DLNU

Introduction To Dlnu And College English (Ce) Teaching

Dalian Nationalities University (DLNU) offers engineering and applied sciences as major disciplines. As its name suggests, 65% students are from 55 different minority ethnic groups. The College English Department (CED), which offers courses to non-English major students, has been selected as one of 31 institutions of College English Reform Demonstration Project due to its unremitting efforts and practice on the reform of CE teaching. It makes full use of modern computer and network technologies and has constructed a new model of CE teaching, to help develop students' autonomic learning ability. In April 2004, its “National New Model of CE Teaching Research” entered the list of "Tenth Five-Year Plan" Research Projects of National Education Science Organization.

College English Course And Syllabus

CE teaching and learning is an integral part of higher education in China. “College English Curriculum Requirements (For Trial Implementation)” (CECR) (CME, 2004) provides colleges and universities with the guidelines for non-English major students. DLNU, along with many other colleges and universities, takes the CECR as their CE teaching syllabus. According to CECR (CME, 2004), College English has knowledge and practical skills of the English language as its main components along with learning strategies and intercultural communication; it takes theories of foreign language teaching as its guide and incorporates different teaching models and approaches.

Teaching Requirement

English proficiency requirements are divided into three levels, namely, basic requirements, intermediate requirements, and higher requirements. The basic requirement is the minimum level that all the undergraduates of non-English majors must attain before graduation. The intermediate and higher requirements are respectively set for those who, having laid a good foundation of English, can afford time to learn more of the language. Institutions of higher learning should set their own objectives in the light of their specific circumstances, strive to create favorable conditions, and encourage students to adjust their objectives in line with their own performance and try to meet the intermediate or higher requirements (check the CECR (trial) in Appendix 2 for detailed description of three levels of requirements) .

Speaking Tests At DLNU

In College English course of DLNU, the speaking test is one of the four subtests of the final examination of English assessment. It has been adopted and administered since 2003, when CED started their practice on CE teaching reform. Almost all the freshmen are required to take the speaking test as part of their English final examinations at the end of first and second semesters (a small proportion of students from special ethnic groups take Russian or Japanese as their foreign language learning). Their speaking test scores take up 10 percent of the final score of English subject, which will be one of the criteria appraised for selection of scholarship awards and recorded into students' archives. 90% of the score cover these sub-tests--- 60% of score coming from computer-based test of listening and reading, 10% from writing score and 20% from teacher evaluations. Teacher evaluations are based on students' attendance, daily performance in class and after-class assignments.

Relationship Between Teaching And Testing

Due to the promotion of CLT, communicative language abilities have been stressed and developed in class. CE teaching in DLNU adopted the model of “2+2+2”, which means students take 2 periods of CE classes at regular classroom with a teacher's presence, 2 periods at computer labs with a teacher's presence, and another 2 periods at computer labs with teacher's absence. Students use two different course books in CE classes --- Reading & Writing, and Viewing, Listening & Speaking textbooks. They are two series of “New Horizontal College English” battery textbooks published by Foreign Language Teaching and Research Press (FLTRPP).

In Reading & Writing class, teachers impart language knowledge partially by speaking activities, such as, discussions, presentations, or debates revolving around the topics of the text. Topical knowledge, grammatical and textual knowledge are exercised through these activities. In Viewing, Listening & Speaking class, colloquial language takes up a high proportion. By watching and simulating sample conversations in the video, learners' pragmatic knowledge is enhanced. Dialogues between learners and between the teacher and the learner are transferred through the computer.

All the topics in the speaking test are chosen from the ones that have been practiced in class.

Purpose

Because the CET-SET is not available to most of college students as mentioned above, they may take less notice of the important role that speaking plays in foreign language acquisition. These two types of speaking tests, especially the computer-aided test, make it practicable to have more than 3,000 students' oral abilities tested within one week. Setting up the speaking test on a school scale can bring many benefits to language learners. It can be facilitated to adopt different effective test items to cater to the teaching syllabus, and it can function as a stimulus to promote students' practice of speaking English and check the teaching and learning outcome. Its purpose has been stated clearly in CECR (Trial):

to develop students' ability to use English in an all-round way, especially in listening and speaking, so that in their future work and social interactions they will be able to exchange information effectively through both spoken and written channels, and at the same time they will be able to enhance their ability to study independently and improve their cultural quality so as to meet the needs of China's social development and international exchanges. (CME, 2004)

Format

Two different formats are used at DLNU. One is semi-direct speaking test, in which examinees talk into the microphones connected to computers, and have their speeches recorded for the teachers to rate afterwards. 94% of students (approximately 2,900) participate in this form. The other is face-to-face interview, in which one interviewer talks to one examinee each time. Students from Department of Chemical Engineering and Technology (Dep. of CEAT) are interviewed by this form. Because this faculty in DLNU has been awarded the “Sate Key Discipline”, students have higher English requirements in NMET (minimum l10 out of 150 points) and the faculty is well-staffed and well-equipped. Students' CE class are offered on small class scale (20-30 students), while those of other departments are usually on large class scale (50-70 students). Besides CE class, students are offered classes of English Video by foreign teachers, Translation, Extensive Reading, etc. Thus, we can see students' English classes are taught and tested intensively.

Construction Of The Test

The speaking test follows the Basic Requirement of CECR for speaking:

Students should be able to communicate in English in the course of learning, to conduct discussions on a given theme, and to talk about everyday topics with people from English-speaking countries. They should be able to give, after some preparation, short talks on familiar topics with clear articulation and basically correct pronunciation and intonation. They are expected to be able to use basic conversational strategies in dialogue.

(CECR) (CME, 2004)

The structures of computer-aided and interview oral proficiency test are shown in Table 3.1 and 3.2, and the detailed test contents are attached in the Appendix 3 and 4:

Table3.1 The structure and description of the computer-aided speaking test

In the computer-aided speaking test, testers' recorded speeches forwarded to teachers who do not teach this class to mark. The interviewers conduct both interviewing and marking.

Research Methodology

Subjects

The subjects of this research included 24 testers and 225 test takers, who were involved in the speaking tests administered from June 16 to June 20, 2009 at CED of School of Foreign Language and Culture in DLNU.

Testers

Note: In the above table, T=Tester; F=Female; M=Male; A. P. =Associate Professor; L=Lecturer; Year= Years they have taught English

Note: In the above table, T=Tester; F=Female; M=Male; A. P. =Associate Professor; L=Lecturer;

Test Takers

The test-taker subjects are freshmen, who have studied two semesters at DLNU. 112 students from Class 1- 4 of Grade 2008, majoring in Chemical Engineering and Technology (CEAT), received face-to-face interviews as their speaking test, and 113 students from Class 1-4 of Grade 2008, majoring in International Economy and Trade (IET) received the computer-aided speaking test. These two departments have the same entrance requirement scores (110 out of 150 points at NMET), higher than those of other departments. Both departments offered extra English classes apart from CE. For example, the Department of IET offered Business English class. Besides, some of their professional courses were taught in English. Those students in face-to-face testing groups and those in the computer-aided groups are presumably comparable in English levels.

Instruments

In this study, three kinds of instruments are adopted in quadrangulation to collect various forms of data: testing materials, questionnaire, telephone interview and test records are adopted.

Testing Materials

The necessary testing materials were collected, including College English Curriculum Requirements (trial), test guidelines and specifications, test paper, scoring criteria, test takers' scores and auditory records of the test.

Questionnaire

Questionnaire survey is important as a method to understand test-taker preferences and opinions (Fulcher, 2003; Alderson et al, 1995). The questionnaire in this empirical study aims at gathering the opinions of subjects, as well as comments and suggestions. It is judged and corrected by the expert in the field of Language Testing before being sent to the subjects. Questionnaires are classified into four versions: questionnaires to test takers of two different formats and questionnaires to testers of two different formats. Each questionnaire, along with the background information about the subject, can be divided into four parts (see the Appendix 5-8): Part I, subjects' opinions about various aspects of validity and reliability; Part II, opinions about aspects of the current speaking test which should be improved and corresponding suggestions¼›Part III, opinions about the preferred test format; Part IV, the effect on English teaching and learning.

To ensure the questionnaire was designed scientifically, one of the four versions was selected to test its validity and reliability. Factor analysis and internal consistency analysis were assessed using SPSS. The results are shown in Table 4.2:

In this table, the Kaiser-Meyer-Olkin Measure of Sampling Adequacy. (KMO) value of .705, exceeded the recommended value of .6 and the Barlett's Test of Sphericity reached statistical significance, supporting the factorability of the correlation matrix. Furthermore, the questionnaire has good internal consistency, as determined by Cronbach alpha coefficient reported of .747. Cronbach's alpha indicates a high degree of reliability.

Telephone Interview

Interviews with testers through telephone are complementary and supportive of the other resources. Questions to the testers cover the evaluation of usefulness of the test in the light of Bachman and Palmer's (1996) checklist for logical evaluation of test usefulness and their own perspectives on the reliability and validity of the tests, test operation and improvement, and the effects the test creates on teaching and learning.

Data Collection And Analysis

The test materials, students' tests scores, and audio records were collected from CED at DLNU. Then 235 copies of students' questionnaires, along with 24 copies of testers' questionnaires were collected. Each data collected was analyzed qualitatively or quantitatively to answer the research questions thoroughly.

Data were then entered into Excel and Statistical Package for Social Science 16.0 (SPSS 16.0) for statistical analyses. Subjects' answers on questionnaires were processed by SPSS to make basic descriptive analysis and frequency analysis, and then correlation analysis was conducted to provide more information about the validity and reliability of the test investigation.

Result And Discussion

This chapter presents results of the data analyses including both qualitative data and quantitative data. Alderson et al (1995) emphasize that it is best to validate a test in as many ways as possible. Therefore, research data from various sources are provided as far as possible.

Theoretical Evaluation

Results And Discussion From Testing Materials

Evaluation Of Test Content

Alderson et al suggest (1995:173) a common way to validate the content validity of a test is to analyze its content and compare that content with “its specification, a formal teaching syllabus or curriculum”. Luoma (2004) also suggests validating the test by relate the test task to the test purpose and test construct.

According to the CECR (CME, 2004), the construct is defined as the ability to “communicate in English in the course of learning, to conduct discussions on a given theme, to talk about everyday topics with people from English-speaking countries”, to “give, after some preparation, short talks on familiar topics with clear articulation and basically correct pronunciation and intonation”, and to “use basic conversational strategies in dialogue.” Thus, this construct involves the communicative competences in the light of the theory of CLA. However, the “sub-skills to be measured” section, grammatical accuracy, pronunciation, intonation, use of fairly accurate vocabulary and sentence pattern, etc., have been given much weight. In the interview format of test, communicative competence is stressed and tested, but without much consideration. Therefore, the computer-aided test has not attached enough importance to communicative competence in its oral test.

The Curriculum Requirements have defined the topics as “everyday topics” and “familiar topics.” The topics tested at DLNU cover quite a wide range, such as pets, music, shopping, public speaking, safety, hobbies, crimes, values on money, drunk driving, movie, examination, hair colouring, family relations, love and marriage, and so on. These are very familiar topics for students and occur in their daily lives. These topics are compatible with the teaching syllabus. Thus, the content validity is considered to be fairly high. However, one aspect that can be improved is that academic study is not included as a topic in the test. Public speaking can be regarded as a skill for presentation, but academic topic and situation should be more introduced into test content as a primary aspect of examinees' daily life.

Evaluation On Scoring Criteria

The rating criteria also need to be evaluated to check if it is coherent with the test purpose and the construct (ibid). The rating scale is defined as “the speech is complete and coherent in answering question, rich in content, with correct pronunciation and fluency and almost no grammatical errors.” The task demands and performance qualities are seen in terms of pronunciation, fluency, grammar, and coherence, so they are assessed in terms of linguistic criteria instead of communicative criteria. They are not quite coherent with the communicative purpose as the construct defines. Furthermore, the criteria are not defined concretely and precisely enough to make them easy to use.

Luoma (2004) recommends that the test administration and scoring processes can be evaluated in terms of their consistency and their coherence with the construct definition. In additions to these, validation of the test includes examinee attitudes to and experiences with the test, the washback effect or the effect of the test on teaching and the teacher or learner attitudes towards learning and the test (ibid), which will be evaluated and discussed in the following sections through various instruments.

Results From Telephone Interview

Bachman and Palmer (1996:150) propose a checklist for logical evaluation of the usefulness of a given test. The questions listed in Table 5.1 are ones to indicate the degree to which the reliability and the validity have been satisfied in the speaking tests at DLNU. Answers are elicited through telephone interview.

These questions offer a very detailed elicitation for an in-depth investigation from a theoretical perspective. With the exceptions of questions 5, 8, 9, and 10, answers to these questions along with the corresponding explanations, demonstrate a high and positive result of the logical evaluation of the reliability and validity of the speaking test. From these explanations, we can safely draw a conclusion that at the theoretical level the speaking test at DLNU has a high degree of reliability and validity.

Empirical Evaluation

Every facet of validity and reliability, and the impact of the speaking test are included in the questionnaires.

Results From Questionnaires Of Students Of Interview Speaking Test

A Strongly disagree,

B Disagree,

C Agree,

D Completely agree.

In this table, the numbers of objective questions are listed in the first left column; where the data under the Column A, B, C and D are the percentage of frequency, which shows the number of repeated choices. The “mean” column shows the average choice number for each question, and the column of “Std. Deviation” is a measure of the variability or dispersion of the data set, or a probability distribution. A low standard deviation indicates that the data points tend to be very close to the mean, whereas high standard deviation indicates that the data are spread over a large range of values.

In the following part, the result of each question is analyzed one by one with the statistical data.

Question 1. The test scores accurately estimate candidates' oral proficiency.

This question was used to investigate the face validity of the test. The data in this table shows that 82.1 % of students think the interview test can fairly and accurately test their oral proficiency. Only 16.1% students responded negatively. The standard deviation for responses to this question is the lowest of all, only 0.456, which indicates that test-takers have little variation in their responses to this question. In students' eyes, the test is quite valid.

Question 2. The interviewer can keep the friendly attitude all the time.

As one facet of the reliability of the test, this question aims at investigating if the attitude of the interviewer has an effect on the examinees' performance. The cumulative percent of positive response from students amounts to 97.3%, and the mean score of 3.54 is the highest of all the questions. It indicates that the effect of that the variation of the interviewers has on test-takers is quite minor. Thus, in this aspect, the reliability of the test score is high.

Question 3. The topic answering in the first part of the speaking test is the most capable of testing the candidate's oral proficiency.

Question 4. The impromptu question and answer in the second part of the speaking test is the most capable of testing the candidate's oral proficiency.

Question 3 and 4 investigate the face validity of the test from th