Validitas dan Reliabilitas

01Core Definitions and Conceptual Framework

Validity refers to the degree to which a measurement instrument truly measures the construct it was designed to assess. A depression scale that actually measures anxiety instead of depression has poor validity, regardless of how reproducible its scores are. Validity is threatened by systematic error (bias), which pushes all measurements consistently in one direction away from the true value. The relationship is inversely proportional: as systematic error increases, validity decreases.

Reliability refers to the consistency and reproducibility of a measurement instrument when applied repeatedly under the same conditions. If a blood pressure cuff yields readings of 130, 145, 118, and 155 mmHg for the same patient within minutes, the instrument has poor reliability. Reliability is threatened by random error, which causes unpredictable fluctuations in measurements that scatter values around a central point without a consistent direction.

The most heavily tested conceptual framework is the target analogy. Imagine a dart board:

• High validity + High reliability: All darts land in the bullseye (accurate and precise).

• Low validity + High reliability: All darts cluster tightly together, but far from the bullseye (precise but inaccurate, representing systematic bias).

• High validity + Low reliability: Darts are scattered, but their average position is the bullseye (accurate on average but imprecise).

• Low validity + Low reliability: Darts are scattered all over, nowhere near the bullseye (neither accurate nor precise).

The fundamental rule that exam writers exploit: reliability is a necessary but not sufficient condition for validity. A measurement can be perfectly reproducible (reliable) while consistently measuring the wrong thing (invalid). However, if a measurement yields wildly inconsistent results (unreliable), it cannot possibly be measuring the intended construct accurately (valid). In formula terms, the maximum possible validity of a test is mathematically limited by its reliability.

02Types of Validity

Face validity is the weakest form because it relies only on subjective appearance. A questionnaire can "look right" without actually measuring the intended construct. Content validity is stronger because it involves systematic review by domain experts. However, neither involves statistical testing.

Criterion validity is where exam writers frequently set traps. The key discriminator between concurrent and predictive validity is timing: concurrent validity compares the test against a gold standard measured at the same time point, while predictive validity assesses whether the test can forecast a future outcome or criterion. The nature of the comparison (same-time benchmark vs. future outcome) is what distinguishes them, not the quality of the criterion itself.

Construct validity is regarded as the broadest, most overarching form. Good construct validity requires both convergent and discriminant evidence. A valid depression measure must correlate with other depression instruments (convergent) AND must fail to correlate with unrelated constructs like physical fitness (discriminant). If only one of these is demonstrated, construct validity is incomplete.

Internal vs. External validity is a classic trade-off tested on exams. Strict inclusion criteria (such as enrolling only males aged 25 to 35 with no comorbidities) increase internal validity by reducing confounders but decrease external validity by limiting generalizability. Conversely, broad inclusion criteria improve generalizability but open the door to confounding.

03Types of Reliability

Reliability is assessed in four primary ways. The exam tests your ability to match the correct reliability type to the correct scenario and the correct statistical measure.

Test-retest reliability measures stability over time. The critical design decision is the retest interval: if the interval is too short, participants remember their previous answers and the reliability estimate is artificially inflated (recall bias). If the interval is too long, the underlying trait may have genuinely changed, and the reliability estimate is artificially deflated. The ideal interval depends on the construct being measured.

Inter-rater reliability is assessed whenever human judgment is part of the measurement process. Cohen's Kappa is the standard for two raters with categorical outcomes because it corrects for agreement that would occur purely by chance. A simple percent agreement is misleading because two raters could agree frequently on common findings just by chance alone. Kappa removes this chance agreement from the calculation, providing a more honest estimate.

Internal consistency is the most commonly reported reliability statistic in health research. Cronbach's Alpha represents the average of all possible split-half correlations within the instrument. The important caveat: Cronbach's Alpha is sensitive to the number of items. Instruments with many items (100+) will produce artificially high alpha values even if some items are redundant or poorly written. This is why split-half reliability is preferred for long instruments.

In Indonesian health research practice, validity is tested using the Pearson Product Moment Correlation method, where each item's r-hitung (calculated r) is compared against the r-tabel at significance level $\alpha =0.05\backslash alpha\; =\; 0.05$. An item is considered valid if r-hitung > r-tabel. Reliability is tested using Cronbach's Alpha, with a threshold of $\alpha \ge 0.60\backslash alpha\; \backslash geq\; 0.60$ considered acceptable. This standard is widely referenced across Indonesian university-based health research and is consistent with the approach used in Kemenkes-affiliated studies.

04Statistical Measures and Interpretation Thresholds

A commonly tested subtlety: Cronbach's Alpha is only a lower-bound estimate of true reliability. It tells you the minimum reliability of your instrument. It does not confirm that the instrument measures a single construct (unidimensionality). A high alpha can occur even when items measure multiple unrelated dimensions, as long as there are many items.

Cohen's Kappa interpretation scale (Landis and Koch, 1977) is the most frequently referenced:

05The Relationship Between Validity and Reliability

This is the section that test-writers draw from most frequently because it tests whether students understand the conceptual hierarchy rather than just memorizing definitions.

The core rules:

1. Reliability is necessary but not sufficient for validity. You cannot have a valid test without reliability, but reliability alone does not guarantee validity.

2. Random error reduces reliability. Increasing sample size or repeating measurements can reduce the effect of random error.

3. Systematic error reduces validity. Increasing sample size does NOT fix systematic error. Only recalibration or methodological redesign can address it.

4. Precision $\approx \backslash approx$ Reliability. This mapping appears on exams. Precision reflects reproducibility.

5. Accuracy $\approx \backslash approx$ Validity. Accuracy reflects closeness to the true value.

06Differential Concepts and Distractors

07Traps and High-Yield Pearls

The most common way students get questions on validity and reliability wrong involves the direction of the necessary relationship. Students frequently select answer choices implying that a valid instrument must also be reliable (correct) but then fail to recognize that a reliable instrument is not necessarily valid (the reverse is false). The vignette might describe a screening tool that produces highly reproducible scores (high reliability) yet misclassifies patients because it is measuring the wrong construct entirely (low validity). The "gotcha" is recognizing that consistency alone does not prove correctness.

A second high-yield trap involves confusing the timing element in criterion validity. Concurrent and predictive validity are both subtypes of criterion validity, but the distinguishing feature is exclusively about when the gold standard comparison occurs, not about the nature of the gold standard itself. If a new test is compared against an established measure at the same time, it is concurrent. If the new test is used to predict an outcome that has not yet occurred, it is predictive.

A third trap involves systematic error and sample size. Students are taught that increasing sample size improves precision by reducing random error, which is correct. They then incorrectly generalize this to systematic error, assuming that a larger sample will also improve accuracy. It will not. Systematic error (bias) affects every measurement in the same direction regardless of sample size. Only recalibration, blinding, improved methodology, or instrument redesign corrects systematic error.

For the Indonesian exam context, remember that the standard validity test uses the Pearson Product Moment correlation with comparison against the r-tabel at $\alpha =0.05\backslash alpha\; =\; 0.05$, and the standard reliability threshold uses Cronbach's Alpha $\ge 0.60\backslash geq\; 0.60$. These thresholds are lower than international research standards ($\ge 0.70\backslash geq\; 0.70$), reflecting pragmatic norms in Indonesian health sciences research. If a vignette describes an Indonesian health survey and asks whether the instrument is "reliable," look for the Cronbach's Alpha value and compare it against 0.60.

The core competency being tested across all questions on this topic: Can you distinguish between what an instrument measures (validity) and how consistently it measures it (reliability), identify the correct type of each, match it to the appropriate statistical test, and recognize when systematic versus random error is the underlying problem?