Test Administration

TEST ADMINISTRATION FACTORS AFFECTING TEST ADMINISTRATION
When we talk about reliability, we are interested in random sources of error.  Observed Score = True Score + Error
When tests are actually administered, however, there are other sources of error aside from random error. 2 FACTORS AFFECTING TEST ADMINISTRATION
Rapport
Ethnicity
Language
Training of Test Administrators
Expectancy Effects
Use of Reinforcers
Computer-Assisted Testing
Subject Variables 3 1 RAPPORT
Importance  Feldman of establishing rapport & Sullivan (1960) - WISC
Enhanced rapport vs. neutral rapport younger children (through grade 3) did not benefit from enhanced rapport
Older children (grades 5-9) produced higher IQ scores under enhanced rapport  Enhanced  Neutral Rapport = mean IQ of 122 Rapport = mean IQ of 109 4 RAPPORT Children score lower on IQ test when the administrator made disapproving comments (“I thought you could do better) then when administrators made neutral or positive comments (Witmer, Bernstein and Dunham, 1971)
Children unfamiliar with the administrator did significantly worse on a reading test compared to children familiar with the administrator (DeRosa and Patalano, 1991)
5 RAPPORT
Importance  Fuchs of establishing rapport & Fuchs (1986) - meta-analysis  22 studies involving 1489 children IQ point increase when the examiner was familiar with the test taker, in general  7.6 IQ point increase when familiarity and lower SES cooccurred 4 6 2 RAPPORT
Importance  Self-report of establishing rapport? vs. interview of attitudinal surveys  People disclose MORE on self-report than they do to interviewers  People disclose MORE to computers than they do to human interviewers
Conclusions:  Rapport is important in situations that are not viewed as “personal” or those typically subject to social desirability. 7 ETHNICITY
Should children of one ethnicity be tested only by test administrators of the same ethnicity?
Majority of studies have found nonsignificant effects for cross-ethnic administration of most intelligence tests.
The only significant findings have been when paraprofessionals have administered the tests.
Why no differences?  Standardized procedures 8 LANGUAGE How valid are tests given in English to bilingual or Limited-English Proficient (LEP) individuals?
What about translating tests?
Language
Standard of practice: administer a test in the most proficient language.  BUT - what about the normative sample?  How comparable are the scores from these individuals?  Can IRT help? 
Interpreters: another potential source of bias 9 3 TRAINING
Administration and scoring errors are a large source of bias.  Typical graduate training: 2-4 administrations of a test (in class)  importance of fieldwork placements of testing practice obtained in fieldwork placements  majority
Error rates on WAIS administrations decrease after 10 administrations(!) 10 EXPECTANCY EFFECTS
Also known as: Rosenthal effects  Robert Rosenthal, Harvard University perform in a manner consistent with experimenter’s (test administrator’s) expectations  Subjects  works with humans, works with rats  Effects not limited to experiments, also occurs on standardized tests  students asked to score ambiguous responses will give more points to people they like, or think are bright.  People find what they expect 11 EXPECTANCY EFFECTS
Expectancy and test administration  Rosenthal - expectancy effects are triggered by nonverbal cues, and the experimenter/ administrator may not even be aware  Expectancy effects have small and varied influence on test outcomes; careful study is required 12 4 USE OF REINFORCEMENT
No clear and consistent difference between studies using reinforcement showing positive or negative effects.
Individual studies, however, are compelling  Terrell, Taylor, & Terrell (1978) found a 17.6 point increase in IQ scores when African-American children were given culturally appropriate feedback by African-American test administrators. 13 USE OF REINFORCEMENT
General guidelines:  Check with the testing manual first  Generally OK to reward EFFORT, not answers. 14 COMPUTER-ASSISTED TEST ADMINISTRATION
Advantages  The obvious connection to Item Response Theory and the ability to tailor tests to a persons ability  Highly Standardized  Precision of Timing  Lessened Dependence on Human Testers  Pacing (no need to rush respondents)  Control of Bias (from the test administrator, etc.) 15 5 COMPUTER-ASSISTED TEST ADMINISTRATION
Computer adaptive versions of tests have shown no large differences between computer assisted and paper-and-pencil versions
Computer versions can be more accurate and take less time (e.g. IRT and CAT)
Some people enjoy the computer format and even prefer it 16 COMPUTER-ASSISTED TEST ADMINISTRATION
One study (Locke and Gilbert, 1995) showed that when respondents are asked about sensitive material (e.g. MMPI, drinking, etc.) they were more honest when the tests were administered via computer vs. questionnaire or interview.
CAT has been applied to the MMPI, personnel selection and cognitive tests among others 17 COMPUTER-ASSISTED TEST ADMINISTRATION
The big concern with computer aided testing is that it will lead to the computer generated reports landing in the wrong (inexperienced) hands and misinterpreted 18 6 SUBJECT VARIABLES
The state of the subject can also be a source of error when administering a test  Illness  Insomnia  Test-anxiety  Drugs (prescription and recreational)  Hormones (e.g. menstruation) – variations in perceptual motor coordination varied with cycle (better away from menses; effects reverse for other tasks 19 FACTORS AFFECTING BEHAVIORAL ASSESSMENT
Issues that arise when people (judges) act as the testing instrument
Human judges are subject to problems that add to the error when assessing respondents  Reactivity  Drift  Expectancies (same as with test administration)  Deception 20 REACTIVITY
The reliability of behavioral assessments is usually assessed using inter-rater reliability (consistency among raters) or by having supervisor make periodic checks
Reliability tends to be highest when the judges know they are being evaluated either by supervisor or against one another
This increase is called REACTIVITY 21 7 REACTIVITY
One study (Reid, 1970) showed that the reliability of observers rating decreased 25% when they were told that they would not be compared to a standard
Many studies report inter-rater reliability, but care should be exercised because these numbers typically are calculated during training and drop during the administration 22 DRIFT
Evaluators (i.e. judges, observers) are typically trained and given a strict set of rules in which to follow when evaluating targets
After some time the evaluators may not be following as strict a set of guidelines as they were trained to follow; this is called DRIFT
To avoid this, evaluators should be periodically retrained on the assessment criteria 23 DECEPTION
How do you assess someone who does not want to be “accurately” assessed?
People in general are awefull at detecting lying  Secret service agents only scored above chance (Ekman and O’Sullivan, 1991)
Lie detection as an industry  Lie detectors (even though questionable at best) tests of honesty and integrity (questionable validity)  Commercial 24 8