Which reliability coefficient is appropriate for ratio/interval data?

Which reliability coefficient is appropriate for ratio/interval data?

• A. Pearson's r
• B. Intraclass Correlation Coefficient (ICC)
• C. Standard error of measurement (SEM)
• D. Spearman's Rho

Answer: (B)

The main idea is reliability of measurements on continuous data. When you want to know how consistently measurements agree across raters or across occasions for data that have a true ratio/interval scale, the intraclass correlation coefficient is the go-to statistic. It doesn’t just look at whether scores move together (that’s what a Pearson or Spearman correlation does); it assesses whether the actual scores are similar enough across measurements to be considered reliable by accounting for variability between subjects, between raters, and within subjects. Pearson's r measures the strength of a linear relationship between two variables, not the agreement between repeated measurements, so it isn’t a reliability coefficient. Spearman's rho is similar but based on ranks, suitable for ordinal data or nonparametric assumptions, not ideal for assessing agreement on interval/ratio scores. The standard error of measurement tells you how precise a single score is but doesn’t quantify reliability across raters or occasions. ICC captures the degree of agreement or consistency of measurements on a ratio/interval scale and can be tailored with different models depending on whether raters are fixed or random and whether you’re interested in single measurements or averages.

The main idea is reliability of measurements on continuous data. When you want to know how consistently measurements agree across raters or across occasions for data that have a true ratio/interval scale, the intraclass correlation coefficient is the go-to statistic. It doesn’t just look at whether scores move together (that’s what a Pearson or Spearman correlation does); it assesses whether the actual scores are similar enough across measurements to be considered reliable by accounting for variability between subjects, between raters, and within subjects.

Pearson's r measures the strength of a linear relationship between two variables, not the agreement between repeated measurements, so it isn’t a reliability coefficient. Spearman's rho is similar but based on ranks, suitable for ordinal data or nonparametric assumptions, not ideal for assessing agreement on interval/ratio scores. The standard error of measurement tells you how precise a single score is but doesn’t quantify reliability across raters or occasions.

ICC captures the degree of agreement or consistency of measurements on a ratio/interval scale and can be tailored with different models depending on whether raters are fixed or random and whether you’re interested in single measurements or averages.