Test-retest correlation is the statistical coefficient calculated between two sets of scores from a measure administered at different times. A test-retest correlation of $$+.80$$ or greater is generally considered to indicate good reliability for constructs expected to be stable. For example, the Rosenberg Self-Esteem Scale demonstrated a highly reliable test-retest correlation of $$+.95$$ when administered to students twice, a week apart.

Test-Retest Correlation

Assessing test-retest reliability involves measuring the same group of participants at two separate times and calculating the correlation between their scores. This statistical relationship is commonly visualized using a scatterplot. Since standard research designs often only test participants once, researchers might specifically design a two-phase study for a new measure to evaluate this reliability. Alternatively, data from a study designed to answer a different research question can be utilized if the measure was administered multiple times. For instance, if an instructor measures students' attitudes at the start and end of a semester to detect learning changes, the correlation between those two sets of scores can simultaneously be used to assess the measure's test-retest reliability, even if no overall attitude change occurred.

Assessing Test-Retest Reliability

A measurement tool can yield highly consistent results without actually measuring the intended construct. For instance, attempting to measure a person's self-esteem by using a ruler to measure their index finger length would demonstrate excellent test-retest reliability, as finger length is stable over time. However, this method would have absolutely no validity, as finger length has no conceptual relationship to self-esteem.

Example of Reliability Without Validity

Test-retest reliability is considered an appropriate standard of consistency for which type of psychological construct?

Test-retest reliability evaluates the extent to which a psychological measure yields consistent scores when administered to the same individuals across different times. This form of reliability is crucial when assessing constructs that are theoretically expected to remain stable, such as intelligence or self-esteem. However, it is not an appropriate standard for constructs that naturally vary over time, like mood or immediate stress levels.

Google

Psychologists consider three primary types of consistency when evaluating a measure: consistency over time (test-retest reliability), consistency across different items on a multiple-item measure (internal consistency), and consistency across different researchers or observers (inter-rater reliability).

Types of Reliability

After administering a psychological measure to a sample and collecting the resulting scores, researchers must thoroughly evaluate the tool for both reliability and validity. Even if a measurement tool has a strong track record in previous studies, it is essential to verify that it functioned effectively with the current participants and under the specific conditions of the new study. Regardless of expectations, this evaluation process generates new evidence regarding the measure's psychometric properties, which should be documented and added to the broader scientific literature.

Evaluating the Measure

Spielman, R. M., Jenkins, W. J., & Lovett, M. D. (2020, April 22). Psychology (2nd ed.). OpenStax. https://openstax.org/books/psychology-2e/pages/1-introduction

OpenStax Psychology (2nd ed.) Textbook

https://kpu.pressbooks.pub/psychmethods4e/

KPU Research Methods in Psychology - 4th American Edition

Inter-rater reliability represents the degree to which different observers or raters make consistent judgments when assessing behavior. It is critical when an assessment involves significant subjective judgment, demonstrating that the recorded behavior is independent of the specific person observing it. Researchers are expected to demonstrate the inter-rater reliability of their coding procedure by having multiple raters code the same behaviors independently and then showing that they are in close agreement.

Inter-rater Reliability

Test-Retest Reliability

Internal consistency is a measure of reliability that assesses how uniformly participants respond across the different items within a multiple-item measure. For instance, responses across items on the Rosenberg Self-Esteem Scale should reflect the underlying construct of self-esteem, much like outcomes in a simulated roulette game consistently reflect the underlying probabilities. It is a standard practice in psychological research to evaluate the internal consistency of any scale that uses multiple items to capture a single construct. Researchers typically determine this consistency by calculating specific statistical indices, most commonly a split-half correlation or Cronbach’s alpha ($$\alpha$$).

Internal Consistency

Match each type of measurement reliability with the aspect of consistency it evaluates.

Criterion validity involves evaluating how effectively a measure aligns with expectations by correlating it with other relevant variables. Researchers assess this by analyzing the correlations among multiple measures of the same psychological construct, or by verifying expected relationships with measures of conceptually distinct constructs. Additionally, a successful experimental manipulation inherently provides evidence for criterion validity. For example, if an experiment manipulates participants' moods by instructing them to think positive versus negative thoughts, and the subsequent mood measure reveals a distinct difference between the two groups, this simultaneously demonstrates that the manipulation worked and that the measurement tool was valid.

Criterion Validity

When newly collected data raises doubts about the reliability or validity of a psychological measure, researchers must investigate the underlying causes. This failure could stem from an inherent flaw in the measure, improper administration procedures, an inaccurate conceptual definition, or an ineffective experimental manipulation. For instance, if a mood measure indicates no difference between groups instructed to think positive versus negative thoughts, it might be because the participants failed to follow the cognitive instructions, or the thoughts simply did not influence their mood. In such cases, researchers must systematically revise the measure, refine the conceptual definition, or design a new manipulation.

Evaluating Measurement Failure

Even if a psychological measurement tool has been shown to be reliable and valid in previous studies, researchers must still evaluate its reliability and validity when used with a new sample of participants.

Learn Before

Related

Learn After