Research Design Part 2 Variability, Validity, Reliability
Selecting an Instrument Reliability & validity of the instrument Are the subject characteristics the same Is the instrument the best one? Why? Purchase cost? Availability? Copyright? Simplicity of administering & scoring
Should you use an existing instrument, modify an instrument, develop a new instrument?
Existing, Modifying, New Instruments Pros Quick to use Have established R & V Can build on knowledge base est. with instrument Cons May not “fit” research question exactly May require training for administration, scoring, or analysis May incur cost to purchase, score, or analyze…MBTI May be too long for the purposes at hand, take too much time to complete
Existing, Modifying, New Instruments Pros Can be modified to better suit research question Most of the work of creating the tool has been completed May be able to compare some results with previous results Cons Changing a known quantity into something unknown Previous reliability and validity indicators may no longer apply
Existing, Modifying, New Instruments Pros Can develop instrument to fit specific need Instrument itself may make a significant contribution to the field of research Cons Requires time, effort, resources, expertise Requires knowledge of scale development procedures Runs risk that instrument will not be reliable or valid for purpose at hand
Modifying or Developing a New Inst. Determine what behaviors/traits to measure Intrinsic vs. extrinsic rewards Review the literature to determine how traits are measured Herzberg; rec literature Consult experts in the field to review the instrument Pilot test Clarity, ambiguity, time of completion, directions Revise Check for a distribution of scores Is there a problem with question or lack of variance on the item
Design Validity
4 types Statistical conclusion validity Construct validity Internal validity External validity
Statistical Conclusion Validity Accurate determination of whether a relationship exists Inflated error rates from multiple tests Extraneous variance Low statistical power
Construct Validity Degree to which a test/ measurement measures a hypothetical construct Threats Using one method to measure the construct Inadequate explanation of a construct Measuring just one construct & making inferences Using 1 item to measure personality
External Validity Generalizability of the results Population external validity Characteristics & results can only be generalized to those with similar characteristics Demographics Psych experiments with college students Ecological external validity Conditions of the research are generalizable to similar characteristics Physical surroundings Time of day
Internal Validity Internal validity is strongest when the study’s design (subjects, instruments/measurements, and procedures) effectively controls possible sources of error so that those sources are not reasonably related to the study’s results.
Internal Validity History Extraneous incidents/events that occur during the research to effect results Only impacts studies across time Attendance at football games/coaching change Selection If there are systematic differences in groups of subjects Gender Compare GRE scores & grad school performance between sequences Occurs when random sampling isn’t used
Internal Validity Statistical regression If doing pre-test/post-test those scoring extremely high or low on first test will often “regress to the mean” on the second test Scoring based more on luck than actual performance The regression effect causes the change & not the treatment Don’t group the high/low scores for the post-test
Internal Validity Pre-testing Pre-test can increase/decrease motivation Gives subjects opportunities to practice Practice can be a positive so they get a true score
Internal Validity Instrumentation Changes in calibration of the exam GRE – new & old Changes in observer scoring Fatigue/ boredom American Idol Attrition/Mortality Subjects drop out/lost Low scorers on GRE drop out of grad school school
Internal Validity Experimenter effect Presence, demeanor of researcher impacts +/- Course instructor is PI Course evals Subject effect Subjects’ behaviors change because they are subjects Subjects want to present themselves in the best light Hawthorn effect
Practice problems