Chapter 3: Understanding Test Quality-Concepts of Reliability and Validity
are two technical properties of a test that indicate the quality and usefulness of the test. These are the two most important features of a test. You should examine these features when evaluating the suitability of the test for your use. This chapter provides a simplified explanation of these two complex ideas. These explanations will help you to understand reliability and validity information reported in test manuals and reviews and use that information to evaluate the suitability of a test for your use.
- What makes a good test?
- Test reliability
- Interpretation of reliability information from test manuals and reviews
- Types of reliability estimates
- Standard error of measurement
- Test validity
- Methods for conducting validation studies
- Using validity evidence from outside studies
- How to interpret validity information from test manuals and independent reviews.
Principles of Assessment Discussed
Use only reliable assessment instruments and procedures.
Use only assessment procedures and instruments that have been demonstrated to be valid for the specific purpose for which they are being used.
Use assessment tools that are appropriate for the target population.
What makes a good test?
An employment test is considered "good" if the following can be said about it:
- The test measures what it claims to measure consistently or reliably. This means that if a person were to take the test again, the person would get a similar test score.
- The test measures what it claims to measure. For example, a test of mental ability does in fact measure mental ability, and not some other characteristic.
- The test is job-relevant. In other words, the test measures one or more characteristics that are important to the job.
- By using the test, more effective employment decisions can be made about individuals. For example, an arithmetic test may help you to select qualified workers for a job that requires knowledge of arithmetic operations.
The degree to which a test has these qualities is indicated by two technical properties: reliability
refers to how dependably or consistently a test measures a characteristic. If a person takes the test again, will he or she get a similar test score, or a much different score? A test that yields similar scores for a person who repeats the test is said to measure a characteristic reliably.
How do we account for an individual who does not get exactly the same test score every time he or she takes the test? Some possible reasons are the following:
- Test taker's temporary psychological or physical state. Test performance can be influenced by a person's psychological or physical state at the time of testing. For example, differing levels of anxiety, fatigue, or motivation may affect the applicant's test results.
- Environmental factors. Differences in the testing environment, such as room temperature, lighting, noise, or even the test administrator, can influence an individual's test performance.
- Test form. Many tests have more than one version or form. Items differ on each form, but each form is supposed to measure the same thing. Different forms of a test are known as parallel forms or alternate forms. These forms are designed to have similar measurement characteristics, but they contain different items. Because the forms are not exactly the same, a test taker might do better on one form than on another.
- Multiple raters. In certain tests, scoring is determined by a rater's judgments of the test taker's performance or responses. Differences in training, experience, and frame of reference among raters can produce different test scores for the test taker.
Principle of Assessment: Use only reliable assessment instruments and procedures. In other words, use only assessment tools that provide dependable and consistent information.
These factors are sources of chance or random measurement error in the assessment process. If there were no random errors of measurement, the individual would get the same test score, the individual's "true" score, each time. The degree to which test scores are unaffected by measurement errors is an indication of the reliability of the test
Reliable assessment tools produce dependable, repeatable, and consistent information about people. In order to meaningfully interpret test scores and make useful employment or career-related decisions, you need reliable tools. This brings us to the next principle of assessment.
Interpretation of reliability information from test manuals and reviews
Test manuals and independent review of tests provide information on test reliability. The following discussion will help you interpret the reliability information about any test.
The reliability of a test is indicated by the reliability coefficient
. It is denoted by the letter "r," and is expressed as a number ranging between 0 and 1.00, with r = 0 indicating no reliability, and r = 1.00 indicating perfect reliability. Do not expect to find a test with perfect reliability. Generally, you will see the reliability of a test as a decimal, for example, r = .80 or r = .93. The larger the reliability coefficient, the more repeatable or reliable the test scores. Table 1 serves as a general guideline for interpreting test reliability. However, do not
select or reject a test solely based on the size of its reliability coefficient. To evaluate a test's reliability, you should consider the type of test, the type of reliability estimate reported, and the context in which the test will be used.
Table 1. General Guidelines for
|Reliability coefficient value||Interpretation|
|.90 and up||excellent|
|.80 - .89||good|
|.70 - .79||adequate|
|below .70||may have limited applicability|
Types of reliability estimates
There are several types of reliability estimates, each influenced by different sources of measurement error. Test developers have the responsibility of reporting the reliability estimates that are relevant for a particular test. Before deciding to use a test, read the test manual and any independent reviews to determine if its reliability is acceptable. The acceptable level of reliability will differ depending on the type of test and the reliability estimate used.
The discussion in Table 2 should help you develop some familiarity with the different kinds of reliability estimates reported in test manuals and reviews.
Table 2. Types of Reliability Estimates
- Test-retest reliability indicates the repeatability of test scores with the passage of time. This estimate also reflects the stability of the characteristic or construct being measured by the test.
Some constructs are more stable than others. For example, an individual's reading ability is more stable over a particular period of time than that individual's anxiety level. Therefore, you would expect a higher test-retest reliability coefficient on a reading test than you would on a test that measures anxiety. For constructs that are expected to vary over time, an acceptable test-retest reliability coefficient may be lower than is suggested in Table 1.
- Alternate or parallel form reliability indicates how consistent test scores are likely to be if a person takes two or more forms of a test.
A high parallel form reliability coefficient indicates that the different forms of the test are very similar which means that it makes virtually no difference which version of the test a person takes. On the other hand, a low parallel form reliability coefficient suggests that the different forms are probably not comparable; they may be measuring different things and therefore cannot be used interchangeably.
- Inter-rater reliability indicates how consistent test scores are likely to be if the test is scored by two or more raters.
On some tests, raters evaluate responses to questions and determine the score. Differences in judgments among raters are likely to produce variations in test scores. A high inter-rater reliability coefficient indicates that the judgment process is stable and the resulting scores are reliable.
Inter-rater reliability coefficients are typically lower than other types of reliability estimates. However, it is possible to obtain higher levels of inter-rater reliabilities if raters are appropriately trained.
- Internal consistency reliability indicates the extent to which items on a test measure the same thing.
A high internal consistency reliability coefficient for a test indicates that the items on the test are very similar to each other in content (homogeneous). It is important to note that the length of a test can affect internal consistency reliability. For example, a very lengthy test can spuriously inflate the reliability coefficient.
Tests that measure multiple characteristics are usually divided into distinct components. Manuals for such tests typically report a separate internal consistency reliability coefficient for each component in addition to one for the whole test.
Test manuals and reviews report several kinds of internal consistency reliability estimates. Each type of estimate is appropriate under certain circumstances. The test manual should explain why a particular estimate is reported.
Standard error of measurement
Test manuals report a statistic called the standard error of measurement (SEM)
. It gives the margin of error that you should expect in an individual test score because of imperfect reliability of the test. The SEM represents the degree of confidence that a person's "true" score lies within a particular range of scores. For example, an SEM of "2" indicates that a test taker's "true" score probably lies within 2 points in either direction of the score he or she receives on the test. This means that if an individual receives a 91 on the test, there is a good chance that the person's "true" score lies somewhere between 89 and 93.
The SEM is a useful measure of the accuracy of individual test scores. The smaller the SEM, the more accurate the measurements.
When evaluating the reliability coefficients of a test, it is important to review the explanations provided in the manual for the following:
- Types of reliability used. The manual should indicate why a certain type of reliability coefficient was reported. The manual should also discuss sources of random measurement error that are relevant for the test.
- How reliability studies were conducted. The manual should indicate the conditions under which the data were obtained, such as the length of time that passed between administrations of a test in a test-retest reliability study. In general, reliabilities tend to drop as the time between test administrations increases.
- The characteristics of the sample group. The manual should indicate the important characteristics of the group used in gathering reliability information, such as education level, occupation, etc. This will allow you to compare the characteristics of the people you want to test with the sample group. If they are sufficiently similar, then the reported reliability estimates will probably hold true for your population as well.
For more information on reliability, consult the APA Standards, the SIOP Principles, or any major textbook on psychometrics or employment testing. Appendix A lists some possible sources.
Validity is the most important issue in selecting a test. Validity refers to what characteristic
the test measures and how well
the test measures that characteristic.
- Validity tells you if the characteristic being measured by a test is related to job qualifications and requirements.
- Validity gives meaning to the test scores. Validity evidence indicates that there is linkage between test performance and job performance. It can tell you what you may conclude or predict about someone from his or her score on the test. If a test has been demonstrated to be a valid predictor of performance on a specific job, you can conclude that persons scoring high on the test are more likely to perform well on the job than persons who score low on the test, all else being equal.
- Validity also describes the degree to which you can make specific conclusions or predictions about people based on their test scores. In other words, it indicates the usefulness of the test.
Principle of Assessment: Use only assessment procedures and instruments that have been demonstrated to be valid for the specific purpose for which they are being used.
It is important to understand the differences between reliability
. Validity will tell you how good a test is for a particular situation; reliability will tell you how trustworthy a score on that test will be. You cannot draw valid conclusions from a test score unless you are sure that the test is reliable. Even when a test is reliable, it may not be valid. You should be careful that any test you select is both reliable and valid for your situation.
A test's validity is established in reference to a specific purpose; the test may not be valid for different purposes. For example, the test you use to make valid predictions about someone's technical proficiency on the job may not be valid for predicting his or her leadership skills or absenteeism rate. This leads to the next principle of assessment.
Similarly, a test's validity is established in reference to specific groups. These groups are called the reference groups. The test may not be valid for different groups. For example, a test designed to predict the performance of managers in situations requiring problem solving may not allow you to make valid or meaningful predictions about the performance of clerical employees. If, for example, the kind of problem-solving ability required for the two positions is different, or the reading level of the test is not suitable for clerical applicants, the test results may be valid for managers, but not for clerical employees.
Test developers have the responsibility of describing the reference groups used to develop the test. The manual should describe the groups for whom the test is valid, and the interpretation of scores for individuals belonging to each of these groups. You must determine if the test can be used appropriately with the particular type of people you want to test. This group of people is called your target population or target group.
Principle of Assessment: Use assessment tools that are appropriate for the target population.
Your target group and the reference group do not
have to match on all factors; they must be sufficiently similar so that the test will yield meaningful scores for your group. For example, a writing ability test developed for use with college seniors may be appropriate for measuring the writing ability of white-collar professionals or managers, even though these groups do not have identical characteristics. In determining the appropriateness of a test for your target groups, consider factors such as occupation, reading level, cultural differences, and language barriers.
Recall that the Uniform Guidelines
require assessment tools to have adequate supporting evidence for the conclusions you reach with them in the event adverse impact occurs. A valid personnel tool is one that measures an important characteristic of the job you are interested in. Use of valid tools will, on average, enable you to make better employment-related decisions. Both from business-efficiency and legal viewpoints, it is essential to only use tests that are valid for your intended use.
In order to be certain an employment test is useful and valid, evidence must be collected relating the test to a job. The process of establishing the job relatedness of a test is called validation
Methods for conducting validation studies
The Uniform Guidelines
discuss the following three methods of conducting validation studies. The Guidelines
describe conditions under which each type of validation strategy is appropriate. They do not express a preference for any one strategy to demonstrate the job-relatedness of a test.
- Criterion-related validation requires demonstration of a correlation or other statistical relationship between test performance and job performance. In other words, individuals who score high on the test tend to perform better on the job than those who score low on the test. If the criterion is obtained at the same time the test is given, it is called concurrent validity; if the criterion is obtained at a later time, it is called predictive validity.
- Content-related validation requires a demonstration that the content of the test represents important job-related behaviors. In other words, test items should be relevant to and measure directly important requirements and qualifications for the job.
- Construct-related validation requires a demonstration that the test measures the construct or characteristic it claims to measure, and that this characteristic is important to successful performance on the job.
The three methods of validity-criterion-related, content, and construct-should be used to provide validation support depending on the situation. These three general methods often overlap, and, depending on the situation, one or more may be appropriate. French (1990) offers situational examples of when each method of validity may be applied.
First, as an example of criterion-related validity, take the position of millwright. Employees' scores (predictors) on a test designed to measure mechanical skill could be correlated with their performance in servicing machines (criterion) in the mill. If the correlation is high, it can be said that the test has a high degree of validation support, and its use as a selection tool would be appropriate.
Second, the content validation method may be used when you want to determine if there is a relationship between behaviors measured by a test and behaviors involved in the job. For example, a typing test would be high validation support for a secretarial position, assuming much typing is required each day. If, however, the job required only minimal typing, then the same test would have little content validity. Content validity does not apply to tests measuring learning ability or general problem-solving skills (French, 1990).
Finally, the third method is construct validity. This method often pertains to tests that may measure abstract traits of an applicant. For example, construct validity may be used when a bank desires to test its applicants for "numerical aptitude." In this case, an aptitude is not an observable behavior, but a concept created to explain possible future behaviors. To demonstrate that the test possesses construct validation support, ". . . the bank would need to show (1) that the test did indeed measure the desired trait and (2) that this trait corresponded to success on the job" (French, 1990, p. 260).
Professionally developed tests should come with reports on validity evidence, including detailed explanations of how validation studies were conducted. If you develop your own tests or procedures, you will need to conduct your own validation studies. As the test user, you have the ultimate responsibility for making sure that validity evidence exists for the conclusions you reach using the tests. This applies to all tests and procedures you use, whether they have been bought off-the-shelf, developed externally, or developed in-house.
Validity evidence is especially critical for tests that have adverse impact. When a test has adverse impact, the Uniform Guidelines require that validity evidence for that specific employment decision be provided.
The particular job for which a test is selected should be very similar to the job for which the test was originally developed. Determining the degree of similarity will require a job analysis. Job analysis is a systematic process used to identify the tasks, duties, responsibilities and working conditions associated with a job and the knowledge, skills, abilities, and other characteristics required to perform that job.
Job analysis information may be gathered by direct observation of people currently in the job, interviews with experienced supervisors and job incumbents, questionnaires, personnel and equipment records, and work manuals. In order to meet the requirements of the Uniform Guidelines, it is advisable that the job analysis be conducted by a qualified professional, for example, an industrial and organizational psychologist or other professional well trained in job analysis techniques. Job analysis information is central in deciding what to test for and which tests to use.
Using validity evidence from outside studies
Conducting your own validation study is expensive, and, in many cases, you may not have enough employees in a relevant job category to make it feasible to conduct a study. Therefore, you may find it advantageous to use professionally developed assessment tools and procedures for which documentation on validity already exists. However, care must be taken to make sure that validity evidence obtained for an "outside" test study can be suitably "transported" to your particular situation.
The Uniform Guidelines, the Standards, and the SIOP Principles state that evidence of transportability is required. Consider the following when using outside tests:
- Validity evidence. The validation procedures used in the studies must be consistent with accepted standards.
- Job similarity. A job analysis should be performed to verify that your job and the original job are substantially similar in terms of ability requirements and work behavior.
- Fairness evidence. Reports of test fairness from outside studies must be considered for each protected group that is part of your labor market. Where this information is not available for an otherwise qualified test, an internal study of test fairness should be conducted, if feasible.
- Other significant variables. These include the type of performance measures and standards used, the essential work activities performed, the similarity of your target group to the reference samples, as well as all other situational factors that might affect the applicability of the outside test for your use.
To ensure that the outside test you purchase or obtain meets professional and legal standards, you should consult with testing professionals. See Chapter 5
for information on locating consultants.
How to interpret validity information from test manuals and independent reviews
To determine if a particular test is valid for your intended use, consult the test manual and available independent reviews. (Chapter 5
offers sources for test reviews.) The information below can help you interpret the validity evidence reported in these publications.
In evaluating validity information, it is important to determine whether the test can be used in the specific way you intended, and whether your target group is similar to the test reference group.
Test manuals and reviews should describe
- Available validation evidence supporting use of the test for specific purposes. The manual should include a thorough description of the procedures used in the validation studies and the results of those studies.
- The possible valid uses of the test. The purposes for which the test can legitimately be used should be described, as well as the performance criteria that can validly be predicted.
- The sample group(s) on which the test was developed. For example, was the test developed on a sample of high school graduates, managers, or clerical workers? What was the racial, ethnic, age, and gender mix of the sample?
- The group(s) for which the test may be used.
The criterion-related validity
of a test is measured by the validity coefficient
. It is reported as a number between 0 and 1.00 that indicates the magnitude of the relationship, "r," between the test and a measure of job performance (criterion). The larger the validity coefficient, the more confidence you can have in predictions made from the test scores. However, a single test can never fully predict job performance because success on the job depends on so many varied factors. Therefore, validity coefficients, unlike reliability coefficients, rarely exceed r = .40.
Table 3. General Guidelines for Interpreting Validity Coefficients
|Validity coefficient value||Interpretation|
|above .35||very beneficial|
|.21 - .35||likely to be useful|
|.11 - .20||depends on circumstances|
|below .11||unlikely to be useful|
As a general rule, the higher the validity coefficient the more beneficial it is to use the test. Validity coefficients of r =.21 to r =.35 are typical for a single test. Validities for selection systems that use multiple tests will probably be higher because you are using different tools to measure/predict different aspects of performance, where a single test is more likely to measure or predict fewer aspects of total performance. Table 3 serves as a general guideline for interpreting test validity for a single test. Evaluating test validity is a sophisticated task, and you might require the services of a testing expert. In addition to the magnitude of the validity coefficient, you should also consider at a minimum the following factors:
- level of adverse impact associated with your assessment tool
- selection ratio (number of applicants versus the number of openings)
- cost of a hiring error
- cost of the selection tool
- probability of hiring qualified applicant based on chance alone.
Here are three scenarios illustrating why you should consider these factors, individually and in combination with one another, when evaluating validity coefficients:
You are in the process of hiring applicants where you have a high selection ratio and are filling positions that do not require a great deal of skill. In this situation, you might be willing to accept a selection tool that has validity considered "likely to be useful" or even "depends on circumstances" because you need to fill the positions, you do not have many applicants to choose from, and the level of skill required is not that high.
Now, let's change the situation.
You are recruiting for jobs that require a high level of accuracy, and a mistake made by a worker could be dangerous and costly. With these additional factors, a slightly lower validity coefficient would probably not be acceptable to you because hiring an unqualified worker would be too much of a risk. In this case you would probably want to use a selection tool that reported validities considered to be "very beneficial" because a hiring error would be too costly to your company.
Here is another scenario that shows why you need to consider multiple factors when evaluating the validity of assessment tools.
A company you are working for is considering using a very costly selection system that results in fairly high levels of adverse impact. You decide to implement the selection tool because the assessment tools you found with lower adverse impact had substantially lower validity, were just as costly, and making mistakes in hiring decisions would be too much of a risk for your company. Your company decided to implement the assessment given the difficulty in hiring for the particular positions, the "very beneficial" validity of the assessment and your failed attempts to find alternative instruments with less adverse impact. However, your company will continue efforts to find ways of reducing the adverse impact of the system.
Again, these examples demonstrate the complexity of evaluating the validity of assessments. Multiple factors need to be considered in most situations. You might want to seek the assistance of a testing expert (for example, an industrial/organizational psychologist) to evaluate the appropriateness of particular assessments for your employment situation.
When properly applied, the use of valid and reliable assessment instruments will help you make better decisions. Additionally, by using a variety of assessment tools as part of an assessment program, you can more fully assess the skills and capabilities of people, while reducing the effects of errors associated with any one tool on your decision making.
A document by the:
U.S. Department of Labor
Employment and Training Administration