Issues of Reliability and Validity
By Al Clayton
Vol. 11, No. 4, 1989, p. 11
Too often, standardized exams are not accurate even within the narrow range of things they are able to measure. The reasons for this problem can be found in the somewhat technical areas of test reliability and validity.
If we could administer a test to a person, then have that person completely forget what is on the test, and finally re-administer the test, would the two test scores be the same? If the answer is yes, the test is reliable. Since people do not forget the test they take, test-makers use a variety of statistical methods to determine reliability and report the results on a scale from zero (no reliability, the score was a random-chance accident) to one (perfect reliability.)
Reliability over 0.9 is considered quite good. Most major achievement tests (such and the California, Iowa, Stanford, and Metropolitan) reach this level.
Sub-sections of these same tests ( e.g., a subsection on calculating percents in an arithmetic test) provide the more detailed information that could be useful in teaching. Sub-test scores, however, are often far less reliable than the whole exam. Many achievement tests report such sub-scores, with the cautions against making decisions based on unreliable sub-test results buried deep inside complex manuals.
Tests for young children are particularly unreliable, with typical ratings only from 0.5 to 0.7 This is largely due to the rapid developmental changes children go through. Young children are also less able to subordinate their immediate thoughts and emotions to a task such as taking a test. The lack of reliability in the tests of young children contributes to the dangers of using such tests for placing them in special programs.
Decisions about students should never be made solely or primarily on the basis of test scores. Since reliability is never perfect, such decisions will be wrong a certain percentage of the time. The lower the reliability, the higher the “measurement error” and the more likely a mistake will be made. Despite this fact, and even though test manufacturers admit tests should not be used as sole determinants, they often are used for deciding placement, grade retention and graduation.
Validity encompasses a set of concepts designed to answer the question, “Are we measuring what we think we are measuring?” A test is only valid in relation to the use made of it. A test that does not measure what it claims to measure is not only invalid, it is dangerous.
A number of aspects of validity should be considered in assessing a test’s utility, but test-makers rarely look at more than the most simple. Partly this is because doing a comprehensive validation study can be expensive and time-consuming. More fundamentally, examining the deeper issues of validity can call into question the entire test itself.
Construct validity examines how well a test actually measures the underlying theoretical construct it claims to measure. For example, does the test accurately measure “academic potential” or “competence” or “reading”? To answer these questions requires a theoretical grasp of the construct to be measured (e.g., “reading”) as well as knowledge of how the tests scores will be used.
Consider, for example, a “spelling” test in which a student is expected to find the correctly spelled word among four or five choices or decide if an underlined word is already spelled correctly or incorrectly. This is actually a test of spelling recognition. Test manufacturers treat the two, spelling and spelling recognition, as essentially the same tling, and the test will often be used as an indicator of the ability to spell.
Often, however, tests are a bad substitute for the real thing. For example, “reading tests” do not measure reading, they measure some reading skills. Because reading is more than the sum of a set of separate skills, reading tests are based on a faulty understanding of reading and learning to read. Similarly, a test that is used to make statements about “school achievement” may really be a test of another construct such as “verbal ability.” Many tests are used as if they possessed wide-ranging validity when there is little evidence supporting such assumptions.
These are the tests that are used to determine the educational fate of children and the content of school programs. Making educational decisions based on instruments that fail to measure what they claim to measure is a recipe for disaster for many children, most of ten those from low-income and minority-group backgrounds.