LESSON 25 - ALPHA COEFFICIENT RELIABILITY: DETERMINING ALPHA COEFFICIENT RELIABILITY USING SPSS

Internal consistency reliability hinges on whether items in a single instrument move together. For quantitative measures administered once, researchers typically rely on alpha coefficient methods—most notably Cronbach’s alpha for Likert-type (agree/disagree) or other “no right/wrong” item formats—and use cutoffs to judge whether the scale is dependable. A reliability coefficient of 0.7 or higher is generally considered acceptable, with interpretation for Cronbach’s alpha ranging from weak to strong: values above 0.9 indicate high internal consistency, 0.7–0.9 indicates acceptable-to-good consistency, 0.6–0.7 is acceptable but weaker, 0.5–0.6 is weak, and below 0.5 suggests no meaningful internal consistency.

The lesson also lays out practical ways to raise alpha reliability. Increasing the sample size, adding more items to the instrument, and piloting the instrument can improve the coefficient—but piloting is framed as an improvement step rather than a reliability test in itself. Reliability is treated as distinct from validity: piloting may help reliability, yet it does not establish validity.

To demonstrate the process, the class walks through running Cronbach’s alpha in SPSS. After entering scale items in the Data Editor, the workflow is: Analyze → Scale → Reliability Analysis, then selecting “Alpha” and choosing the appropriate statistic calculation (correlation-based). The output produces three key tables. One table reports the number of items used; another reports Cronbach’s alpha value; and a third provides the inter-item correlation matrix, showing how each item correlates with every other item. In the worked example, 18 items were used and Cronbach’s alpha came out to 0.827, which falls in the “high/strong” reliability range and supports confidence in using the questionnaire with respondents.

Beyond Cronbach’s alpha, the lesson introduces a second alpha coefficient family member: the Kuder–Richardson 20 and 21 tests (KR-20/21). These are positioned as more appropriate for knowledge or achievement questions where responses can be scored as correct versus incorrect. KR-20 is highlighted as the most frequently applied homogeneity index and is described as relying on the ratio of correct and incorrect answers across items. A key constraint is emphasized: KR-20/21 is valid when items are effectively divided into two categories (e.g., correct/incorrect) and when items carry equal weight. If items have different weights, the method should not be used.

Overall, the session connects theory to practice: internal consistency reliability is assessed through alpha-based methods in SPSS for single-administration instruments, with Cronbach’s alpha suited to Likert-style scales and KR-20/21 suited to dichotomously scored knowledge tests. The next step in the broader course is shifting from reliability to validity for qualitative instruments.