Which statistic is used to determine whether two scores are significantly different in banding procedures?

Which statistic is used to determine whether two scores are significantly different in banding procedures?

• A. Standard Error of Difference (SED)
• B. Standard Error of Measurement (SEM)
• C. Cronbach's alpha
• D. Pearson r

Answer: (A)

The main idea is that banding decisions rely on whether the difference between two observed scores is larger than what would be expected from measurement error. The standard error of difference does exactly that: it combines the imprecision of each score (their standard errors of measurement) to establish a threshold for how big a difference could occur by chance. You estimate the standard error of difference from the standard errors of measurement for the two scores, often using a formula like SED = sqrt(SEM1^2 + SEM2^2). Then you compare the actual observed difference to a critical value (commonly 1.96 times the SED for a 95% confidence level). If the observed gap surpasses this threshold, you conclude the scores differ beyond what measurement error would produce, and they would fall into different band levels. The other statistics don’t fit this purpose. The standard error of measurement describes how much a single score might vary if it were repeated, not the variability of the difference between two scores. Cronbach’s alpha measures internal consistency reliability, not differential significance between two scores. Pearson r assesses the strength of a linear relationship between two variables, not whether their scores differ significantly.

The main idea is that banding decisions rely on whether the difference between two observed scores is larger than what would be expected from measurement error. The standard error of difference does exactly that: it combines the imprecision of each score (their standard errors of measurement) to establish a threshold for how big a difference could occur by chance.

You estimate the standard error of difference from the standard errors of measurement for the two scores, often using a formula like SED = sqrt(SEM1^2 + SEM2^2). Then you compare the actual observed difference to a critical value (commonly 1.96 times the SED for a 95% confidence level). If the observed gap surpasses this threshold, you conclude the scores differ beyond what measurement error would produce, and they would fall into different band levels.

The other statistics don’t fit this purpose. The standard error of measurement describes how much a single score might vary if it were repeated, not the variability of the difference between two scores. Cronbach’s alpha measures internal consistency reliability, not differential significance between two scores. Pearson r assesses the strength of a linear relationship between two variables, not whether their scores differ significantly.