Invalid measurement validity.

Concurrence looks at how a measure relates to a gold standard. If, as frequently occurs, there isn’t one, the next best is the correlation with other established measures. This assumes that the latter are themselves valid, which may not be true, and risks stifling innovation, a source of concern if a paper is criticized on this ground. More straightforwardly, concurrence also means that a test should be consistent with previous approaches to the function measured and should be internally consistent. This might be relevant to current attempts to redefine cerebral palsy. Predictive validity concerns a different aspect: correlating the scale with a future assessment, such as the value of the admission Glasgow Coma Score in predicting the outcome from traumatic brain injury. As it is not particularly well correlated, other measurements are needed if intervention studies are being planned. Accuracy and reliability are associated concepts. A test with high concurrent validity on correlation measures may still be inaccurate. If different categories are being defined, are they unambiguous and clearly separated from each other? If the results are part of a continuum, do they reflect that? The difference between centiles and standard deviations affects widely used instruments, such as growth charts. The mean is the 50th centile, 1 standard deviation below it approximates to the 33rd centile, 2 to the 2nd centile, etc. It is not immediately obvious that in statistical terms the difference between the 50th and the 33rd centile is the same as between the 33rd and the 2nd. Reliability, as in intraand interobserver variation or test–retest reproducibility, is essential for validity, but a reliable test may not be valid if it does not measure what is intended. The final aspect, which relates to the study as a whole rather than measurement itself, concerns internal and external validity. Internal validity means that all confounding factors have been controlled for. External validity indicates how well a finding can be generalized to other groups of people (population validity) or situations (ecological validity). Essentially, measurement validity is an assessment of how good the test is at doing what it is intended to do. Many of the tests we use in clinical life and in research are not ideal in these terms, particularly some of those hallowed by time. If a researcher is developing a new test, such as a new quality of life instrument, all these issues need to be considered. It is upsetting to have to send back a piece of work that has clearly taken many hours in terms of preparation, data collection, analysis, and writing up, all because of a problem in the fundamental validity of the tests used.