test development. why develop a new test? u meet the needs of a special group of test takers u...

52
Test Development

Upload: aron-carr

Post on 22-Dec-2015

214 views

Category:

Documents


1 download

TRANSCRIPT

Test Development

Why Develop a New Test?

meet the needs of a special group of test takers

sample behaviours from a newly defined test domain

improve the accuracy of test scores for their intended purpose

Tests need to be revised

First Four Steps

Defining the test universe, audience, and purpose

Developing a test plan Composing the test items Writing the administration instructions

Continued Steps of Test ConstructionDiagram of Test Construction (p. 234)

Constructing Scales Piloting the Test Standardizing the Test Collecting Norms Validation & Reliability Studies Manual Writing Test Revision

Defining the Test Universe, Audience, & Purpose

Defining the test universe. – prepare a working definition of the

construct– locate studies that explain the construct– locate current measures of the construct

Defining the Test Universe, Audience, & Purpose

Defining the target audience. – make a list of characteristics of persons

who will take the test--particularly those characteristics that will affect how test takers will respond to the test questions (e.g., reading level, disabilities, honesty, language)

Defining the Test Universe, Audience, & Purpose

Defining the purpose.– includes not only what the test will

measure, but also how scores will be used– e.g., will scores be used to compare test

takers (normative approach) or to indicate achievement (criterion approach)?

– e.g., will scores be used to test a theory or to provide information about an individual?

Developing a Test Plan

A test plan includes a definition of the construct, the content to be measured (test domain), the format for the questions, and how the test will be administered and scored

Defining the Construct Define construct after reviewing

literature about the construct and any available measures

Operationalize in terms of observable and measurable behaviours

Provides boundaries for the test domain (what should and shouldn’t be included)

Specify approximate number of items needed

Choosing the Test Format Test format refers to the type of

questions the test will contain (usually one format per test for ease of test takers and scoring)

Test formats have two elements:– stimulus (e.g., a question or phrase)– mechanism for response (e.g., multiple

choice, true-false). May be objective or subjective test format

Composing the Test Items

test items are the stimuli presented to the test taker (may or may not take the form of questions)

the form chosen depends on decisions made in the test plan (e.g., purpose, audience, method of administration, scoring)

Test Types Structured Response

– Multiple Choice– True False, Forced Choice– Likert Scales

Free Response– Essay, Short Answer– Interview Questions– Fill in the Blank– Projective Techniques

Multiple Choice Multiple choice most common in educational

testing (and also some personality and employment testing)– consists of a stem and a number of responses--

should only be one right answer– the wrong answers are called distractors because

they may appear correct--should be realistic enough to appeal to uninformed test taker

– easy scoring but downside is that test takers can get some correct by guessing

Multiple Choice Pros

more answer options (4-5) reduce the chance of guessing that an item is correct

many items can aid in student comparison and reduce ambiguity, increase reliability

Cons measures narrow facets of performance reading time increased with more answers transparent clues (e.g., verb tenses or letter uses “a”

or “an”) may encourage guessing difficult to write four or five reasonable choices takes more time to write questions

True/False

True/False is also used in educational testing and some personality testing– in educational testing the test taker can

again gain some advantage by guessing

True/False (cont.) Ideally a true/false question should be

constructed so that an incorrect response indicates something about the student's misunderstanding of the learning objective.

This may be a difficult task, especially when constructing a true statement.

Forced Choice Items

Forced-Choice is similar to multiple-choice but is used in personality and attitude tests (e.g., MBTI)– test taker must choose between unrelated but

equally acceptable responses

Forced Choice Items(cont.)

Example

Place an “X” in the space to the left of the work that of the word in each pair that best describes your personality.

1. ____ Sunny 2. ____ Outgoing

____ Friendly ____ Loyal

Likert Scales

Likert scales are usually reliable and highly popular (e.g., personality and attitude tests)– item is presented with an array of response

options (e.g., 1 to 5 or 1 to 7 scale), usually on an agree/disagree or approve/disapprove continuum

Test Types

Structured Response– Advantages

Great Breadth Quick Scoring

– Disadvantages Limited Depth Difficult to assess higher levels of skills Guessing/Memorization vs. Knowledge

Subjective Items subjective items are less easily scored

but provide the test taker with fewer cues and open wider areas for response--often used in education– essay questions - responses can vary in

breadth and depth and scorer must determine to what extent the response is correct (often by examining match with predetermined correct response)

Essay Questions Provide a freedom on response that

facilitates assessing higher cognitive behaviors (e.g., analysis and evaluation)

Allows respondent to focus on what they have learned and does not limit them to specific questions

Interview Questions– interview questions are often used in organizational

settings--interviewer decides what is a good or poor answer

test plan should be based on knowledge, skills, abilities and other characteristics required to perform the job

Information can be obtained from a job description, job analysis, current job incumbent

Projective Techniques

Projective techniques are often employed in clinical settings– uses a highly ambiguous stimulus to elicit

an unstructured response (i.e., the test taker “projects” his or her perception and perspective onto a neutral stimulus)

– variety of stimuli (e.g., pictures, words) and responses may be verbal or drawing pictures

Sentence Completion Sentence-Completion format presents an

incomplete sentence that the test taker completes (e.g., “I feel happiest when …)

subjective tests are at risk for judgment error and inter-rater reliability is therefore of particular importance--scoring keys and training important

Test Types

Subjective Items – Advantages

Can Test Higher Cognitive skills Encourages organize/develop thoughts

– Disadvantages Difficult to Grade Judgement error (e.g., interrater reliability) Requires Advance - Objective Scoring Key

Writing Good Items Basis building block of test construction Little attention given to writing items an art that requires originality, creativity, combined with

knowledge of test domain and good item writing practices not all items will perform as expected--may be too easy or

difficult, may be misinterpreted, etc. rule of thumb to write at least twice as many items as you

expect to use Broad vs. Narrow items

Writing Good Items (cont.)

Suggestions:– identify item topics by consulting test plan

(increases content validity)– ensure that each item presents a central

idea or problem– write items drawn only from testing

universe– write each item in clear and direct manner

Writing Good Items (cont..)

Suggestions:– use vocabulary and language appropriate for

the target audience (e.g., age, culture)– take into account sexist or racist language

(e.g., mailman, fireman)– make all items independent (e.g.,one question

per question)– ask an expert to review items to reduce

ambiguity and inaccuracy

Writing Administration Instructions

specify the testing environment to decrease variation or error in test scores

should address:– group or individual administration– requirements for location (e.g., quiet)– required equipment– time limits or approximate completion time– script for administrator and answers to questions

test takers may ask

Specifying Administration and Scoring Methods

determine such things as how test will be administered (e.g., orally, written, computer--individually or in groups)

method of scoring, but also whether scored by hand by test administrator, or accompanied by scoring software, or sent to test publisher for scoring

Scoring MethodsCumulative model: most common

– assumes that the more a test taker responds in a particular fashion the more he/she has of the attribute being measured (e.g., more “correct” answers, or endorses higher numbers on a Likert scale)

– correct responses or responses on Likert scale are summed

– yields interval data that can be interpreted with reference to norms

Scoring Methods (cont.)

Categorical model: place test takers in a group

– e.g., a particular pattern of responses may suggest diagnosis of a certain psychological disorder

– typically yields nominal data because it places test takers in categories

Scoring Methods (cont…) Ipsative model: test takers scores are not compared to that of

other test takers but rather compare the scores on various scales WITHIN the test taker (Which scores are high & low)

– e.g., a test taker may complete a measure of interpersonal problems of various types and the test administrator may want to determine which of the types the test taker feels is most problematic for him or her

Cumulative model may be combined with categorical or ipsative model

Response Bias In preparing an item review, each question can

be evaluated from two perspectives: Is the item fair? Is the item biased?

Tests are subject to error and one form comes from the test takers

Response Sets/Styles Are patterns of responding that result in misleading

information and limit the accuracy and usefulness of the test scores

Reasons for misleading information

1. Information requested is too personal

2. Distort their responses

3. Answer items carelessly

4. May feel coerced into completing the test

Response Style

– People always agree (acquiescence) or disagree (criticalness) with statements without attending to the actual content

– Usually, when items are ambiguous

Solution: use both positively- and negatively-keyed items

Social Desirability

Some test takers choose socially acceptable answers or present themselves in a favourable light

People often do not attend as much to the trait being measured as to the social acceptability of the statement

This represents unwanted variance

Social Desirability (cont.)

Example items:

– Friends would call me spontaneous.

– People I know can count on me to finish what I start.

– I would rather work in a group than by myself.

– I often get stressed-out in many situations.

Faking

Faking -- some test takers may respond in a particular way to cause a desired outcome

– may “fake good” (e.g., in employment settings) to create a favourable impression

– may “fake bad” (e.g., in clinical or forensic settings) as a cry for help or to appear mentally disturbed

– may use some subtle questions that are difficult to fake because they aren’t clearly face valid

“Faking Bad”– People try to look worse than they really are

Common problem in clinical settings

– Reasons: Cry for help Want to plea insanity in court Want to avoid draft into military Want to show psychological damage

– Most people who fake bad overdo it

Impression Management

– Mitigating IM:

Use positive and negative impression scales (endorsed by 10% of the population)

Use lie scales to “flag” those who score high (e.g., “I get angry sometime”).

Inconsistency scales (e.g., two different responses to two similar questions)

(Use multiple assessment methods (other than self-report)

Random Responding

Random responding may occur when test takers are unwilling or unable to respond accurately.

– likely to occur when test taker lacks the skills (e.g., reading), does not want to be evaluated, or lacks attention to the task

– try to detect by embedding a scale that tends to yield clear results from vast majority such that a different result suggests the test taker wasn’t cooperating

Random Responding– Detection:

Duplicate items:

“I love my mother.”

“I hate my mother.”

Infrequency scales:

“I’ve never had hair on my head.”

“I have not seen a car in 10 years.”

Random Responding– May occur for several reasons:

People are not motivated to participate Reading or language difficulties Do not understand instructions / item content Too confused or disturbed to respond

appropriately

Piloting and Revising Tests can’t assume the test will perform as

expected pilot test scientifically investigates the

test’s reliability and validity administer test to sample from target

audience analyze data and revise test to fix any

problems uncovered--many aspects to consider

Setting Up the Pilot Test

test situation should match actual circumstances in which test will be used (e.g., in sample characteristics, setting)

developers must follow the American Psychological Association’s codes of ethics (e.g., strict rules of confidentiality and publish only aggregate results)

Conducting the Pilot Test

depth and breadth depends on the size and complexity of the target audience

adhere strictly to test procedures outlined in test administration instructions

generally require large sample may ask participants about the testing

experience

Analyzing the Results

can gather both quantitative and qualitative information

use quantitative information for such things as item characteristics, internal consistency, convergent and discriminate validity, and in some instances predictive validity

Revising the Test Choosing the final items requires

weighing each item’s content validity, item difficulty and discrimination, inter-item correlation, and bias

when new items need to be added or items need to be revised, the items must again be pilot tested to ensure that the changes produced the desired results

Validation and Cross-Validation Validation is the process of obtaining

evidence that the test effectively measures what it is supposed to measure (i.e., reliability and validity)

first part of establishing content validity is carried out as the test is developed--that it measures the constructs (construct validity) and predicts an outside criterion is determined in subsequent data collection

Validation and Cross-Validation when the final revision of a test yields

scores with sufficient evidence of reliability and validity, test developers then conduct cross-validation--a final round of test administration to another sample

because of chance factors the reliability and validity coefficients will likely be smaller in the new sample--referred to as shrinkage