01. SPSS Classroom Lectures| Basic Statistical Concepts (P1) | Reliability and Validity

Reliability and validity are the two core quality checks behind any questionnaire or measurement tool—reliability for consistency, validity for accuracy. A measurement is reliable when it produces the same numeric results repeatedly under similar conditions with the same subjects. Validity, by contrast, requires that the instrument actually measures the concept it claims to measure. The distinction matters because consistency alone can be misleading: an instrument can give stable results while still targeting the wrong construct.

A classic reliability-versus-validity example illustrates the risk. A wall clock that always shows 6:00 when someone enters the room is consistent, but it is not necessarily valid for telling the correct time. Likewise, a test that repeatedly returns the same score for a child’s recall of last day activities may be reliable, yet someone might incorrectly claim it measures the child’s IQ. Another analogy compares constructs: using a job satisfaction scale to measure job commitment may yield consistent scores, but those scores could reflect something else entirely—such as memory or another unrelated trait—rather than the intended concept.

Reliability can be assessed through the test–retest approach: administer the same instrument to the same people twice under similar conditions, then correlate the two sets of results. Higher correlation indicates greater consistency. In practice, test–retest is difficult because repeated participation is hard to arrange, and subjects may no longer respond neutrally after the first exposure. The example of repeatedly taking a GMAT test highlights how familiarity can introduce bias, undermining the attempt to measure pure consistency.

Because test–retest is often impractical, researchers use other reliability techniques. For categorical data, Cohen’s Kappa coefficient is commonly used. For internal consistency across items in a scale, Cronbach’s Alpha is widely applied. In modern survey research, construct reliability is frequently assessed using composite reliability, which fits naturally within confirmatory factor analysis.

Validity is assessed by checking how accurately the measurement aligns with the underlying trait it is meant to represent. For job satisfaction, the underlying trait might be reflected through multiple indicators—salary, environment, co-workers, and job security—so validity asks whether those indicators truly capture job satisfaction rather than something adjacent. Face validity is the first check: the instrument should appear, to experts and respondents, to measure the intended concept. Beyond that, predictive validity tests whether the measure forecasts related outcomes (e.g., GMAT performance predicting MBA performance). Content validity checks whether the instrument covers the full intended domain of the construct; measuring only reading skills when the construct includes reading, writing, and listening would miss major parts of the domain.

Construct validity, grounded in theory, looks for expected patterns of relationships among variables. It includes convergent validity—items intended to measure the same construct should correlate strongly—and discriminant validity—different constructs should relate differently, not collapse into one another. Statistical tools such as average variance extracted and the Fornell–Larcker criterion (or related methods) are used to support these claims. Reliability and validity together determine whether a measurement is both stable and meaningful for research conclusions.