Ecological validity of web tests

Construct validity is the argument that a score reflects a named idea — working memory, not “being clever in general.” It is built from theory, patterns of correlations, and failed rival explanations. A 20-trial number-Stroop lite can feel like control. Feeling is not a validity coefficient. This site never reports a licensed IQ. 練習であり臨床IQではない。

Criterion validity asks whether a score lines up with something you care about that is not the test itself — a later grade, a supervisor rating, or a Wechsler index. Concurrent means measured at the same time; predictive means later. A visual 2-back lite has no published criterion table. It is not an employment screen and not an IQ. 練習であり臨床IQではない。

Content validity is whether the item set covers the skill or knowledge you claimed. A spelling test with only the letter Q is a coverage failure. Shape-span lite samples a tiny visuospatial sequence. It does not cover “all spatial intelligence.” Experts, blueprints and reviews build content arguments. A weekend hackathon does not. 練習であり臨床IQではない。

A 95% confidence interval is a procedure: if you repeated the study under the same model, 95% of such intervals would cover the parameter. It is not the probability that this one interval magically contains the truth after you saw it. Licensed IQ manuals turn a standard error into a range around a scaled score. Digit-span lite does not. Treat any integer from this site as a toy count. 練習であり臨床IQではない。

Item response theory (IRT) models each item: harder items need more of the trait for a 50% chance of success; discriminating items separate nearby people. Computerized adaptive tests pick the next item from that model. Pattern-matrices lite is a fixed original set. It does not estimate a theta with published item parameters. It is not an IQ. 練習であり臨床IQではない。