2 Measurement of Variables: Scaling, Reliability and Validity
3 Major Sources of Errors in Measurement Since 100% control for precise and unambiguous measurement of variables is unattainable, error does occur. Much potential error is SYSTEMATIC (results from a bias) while the remainder is RANDOM (occurs erratically).Sources of Measurement DifferencesRespondentsSituational factorsMeasurer or researcherData collection instruments
4 Validity and Reliability Accuracy of measurement. The degree to which an instrument measures that which is supposed to be measured.Validity Coefficient: An estimate of the validity of a measure in the form of a correlation coefficient.Reliability:Consistency of measurement. The degree to which an instrument measures the same way each time it is used under the same conditions with the same subjects.Reliability Coefficient: An estimate of the reliability of a measure usually in the form of a correlation coefficient.From Sproull (1995):Validity is an extremely important and perhaps the most important, aspect of a measure. If an instrument does not measure accurately what it is supposed to measure, there is no reasons to use it, even if it does measure consistently.Of validity and reliability, validity is far more important than reliability for the reason given in #1.It is questionable practice to use measures for which: (1) there is no evidence of validity or (2) there is evidence of validity but the validity estimates are low.Validity is specific. An instrument or measure can be valid for a specific criterion but not for other criteria, for a specific group of people but not other groups.There are several types of validity and reliability estimates. Because of the confusion in using different names and meanings for the various types of estimates, standard labels and standard definitions for the various types of both validity and reliability estimates have been determined.A single instrument may have several types of validity and reliability estimates.Each type of validity and reliability estimate has a different purpose.Validity and reliability are always estimated, not proven.The statistical symbol for a validity coefficient is rxy indicating correlation of two different measures. The statistical symbol for a reliability coefficient is rxx indicating two measures of the same variables.A typical validity coefficient would be approximately .45 or higher. Higher would be better, However, validity coefficients rarely exceed .60 and many are in the range of .30 to .40.A typical reliability coefficient for a a researcher designed instrument is approximately .70 or higher. For an instrument designed by a testing service, one would expect .90 or higher.If the researcher designs an instrument, validity and reliability estimates should be made by the researcher. If an instrument is purchased, the company from which it is purchased should provide validity and reliability information which the researcher examines prior to purchase.
5 Possible Conditions of Validity and Reliability When examining an instrument for validity and reliability remember that three types of conditions may exist. An instrument might show evidence of being:Both valid and reliable orReliable but not valid orNeither reliable nor validNOTE: An instrument which is valid will also have some degree of reliability.
6 About Reliability and Validity Coefficients Validity and reliability are estimated by using correlation coefficients. These statistics estimate the degree of validity or reliability.Thus, it is not a question of an instrument having or not having validity or reliability;It is a question of to what degree is an instrument valid for a specific purpose and to what degree does the instrument evidence specific types of reliability.Reliability estimates are done after validity is assessed.We will discuss the notions of internal and external validity in the context of experimental designs.
7 Types of Validity Content Validity The representativeness of the content of the instrument to the objectives of using the instrument.Usual Process:1. Examine objectives;2. Compare objectives to content of instrument.
8 Types of Validity (cont’d) Criterion-Related ValidityPredictive: The degree to which a measure predicts a second future measure.Usual Process:1. Assess validation sample on predictor;2. Assess validation sample on criterion at later time;3. Correlate scoresConcurrent: The degree to which a measure correlates with another measure of the same variable which has already been validated.1. Assess validation sample on new measure;2. Assess validation on already validated measure of same variable at about the same time;3. Correlate scores.
9 Types of Validity (cont’d) Construct ValidityThe degree to which a measure relates to expectations formed from theory for hypothetical constructsUsual Process1. Assess validation sample on major variable2. Assess validation sample on several hypothetically related variables3. Analyze to see of major variable differentiates Ss on the related variables
10 Types of Reliability (Consistency) Estimates V.IMP: A MEASURE CAN BE RELIABLE BUT TOTALLY LACK VALIDITY.StabilityTest-retestEquivalenceParallel formsInternal ConsistencySplit-halfKR20 (Kuder-Richardson)Coefficient (Cronbach’s) alphaInterater reliability
11 Types of Reliability (Consistency) Estimates: STABILITY Test-Retest ReliabilityUsed to assess the stability of a measure over time.Usually indicated by a correlation coefficient.Number of forms (of instrument): 1Number of administrations: 2Usual Process:Administer the instrument to the reliability sample at Time 1.Wait a period of time (e.g., 2-4 weeks)Administer copies of the same instrument to the same sample at Time 2.Correlate the scores from Time 2 and Time 1.
12 Types of Reliability (Consistency) Estimates: EQUIVALENCE Equivalence Forms Reliability (Also known as Parallel Forms or Alternate Forms Reliability).Used to assess the equivalence of two forms of the same instrument.Usually indicated by a correlation coefficient.Number of forms (of instrument): 2Number of administrations: 2Usual Process:Administer Form A of the instrument to the reliability sampleBreak the sample for a short rest period (10-20 minutes)Administer Form B of the instrument to the same reliability sampleCorrelate the scores from Form A and Form BNeeded when two or more versions (Forms) of the instrument will be used.
13 Types of Reliability (Consistency) Estimates: INTERNAL CONSISTENCY Split-Half ReliabilityUsed to assess the internal consistency or equivalence of two halves of an instrument.Usually indicated by a correlation coefficient plus Spearman-Brown Prophecy Formula.Number of forms (of instrument): 1Number of administrations: 1Usual Process:Obtain or generate an instrument in which the two halves were formulated to measure the same variable.Administer the instruments to the reliability sample.Correlate the summed scores from the first half (often the odd numbered items) with the summed scores from the second half (often the even numbered items).Computer the Spearman-Brown Prophecy Formula to correct for splitting one instrument into halves.
14 Types of Reliability (Consistency) Estimates: INTERNAL CONSISTENCY KR20 (Kuder-Richardson Reliability)Used to assess the internal consistency of items on an instrument when responses are dichotomous.Usually indicated by the correlation generated using the KR-20 formula (There are other forms of this formula. This is used when there are two responses: correct or incorrect).Number of forms (of instrument): 1Number of administrations: 1Usual Process:Generate or select an instrument.Administer the instrument to the reliability sample.Compute the variance (x)2 of the scores.Computer the proportion of correct and incorrect responses to each item.Compute the KR-20 formula.
15 Types of Reliability (Consistency) Estimates: INTERNAL CONSISTENCY Coefficient Alpha (Cronbach’s Alpha)Used to assess internal consistency of items on an instrument when responses are nondichotomous.Usually indicated by the coefficient generated using Cronbach’s formula (more generic version of the KR-20 formula)Number of forms (of instrument): 1Number of administrations: 1Usual Process: Same as previous slide.
16 Types of Reliability (Consistency) Estimates: INTERSUBJECTIVE INTERRATER RELIABILITYUsed to assess the degree to which two or more judges (raters) rate the same variables in the same way.Usually needed when two or more judges (raters) will be used in a research study.Usual Process:Select or generate an instrumentRandomly select a number of objects or events to be ratedTrain the ratersHave each rater rate each object or event independentlyCorrelate the scores of the two raters.
17 Practicality of Measurement Practicality has been defined in terms of the following three characteristics: Economy, Convenience and InterpretabilityEconomySome trade-off between ideal needs and budgetInstrument length (limiting factor: cost)Choice of data collection methodNeed for Fast and economical scoringThe scientific requirements of a project call for the measurement processes to be reliable and valid, while the operational requirements call for it to be practical.
18 Practicality of Measurement (cont’d) ConvenienceA measuring device passes the convenience test if it is easy to administer.Detailed and clear instructions with examples, if needed.Pay close attention to design and layoutAvoid crowding of material, carryover of items from one page to anotherInterpretabililtyRelevant when persons rather than test designers interpret results. In that case, test designers should include:A statement of the functions the test was designed to measure and the procedures by which it was developedDetailed instructions for administering and scoringEvidence of reliability etc.