ror-image-like method, with some mathematical manipulations based upon published normative data (derived from normal persons). The fundamental flaw in this reasoning-which no amount of mathematical tweaking can offset-is that these lower values are derived from symptomatic patients, and while some of them may be "normal" in the sense of disease-free, others may not. A short distal motor latency does not assure that the nerve is diseasefree. It is simply incorrect to use these values, representing a mixture of true normal values and false-negative disease values, as the basis for diagnostic judgment.
To put it another way, the shape of the left side of the frequency distribution curve in these symptomatic patients is likely not identical to that in true normals. This shape discrepancy will be reduplicated on the right side of the curve in Yaar's procedure, and give rise to erroneous criteria of abnormality. Moreover, the central tendency of the curve is likely shifted to the right.
These influences may be seen in Yaar's results. HIs proposed upper limit of "normative" median motor distal latency is 4.66 ms (mean + 2 SD), which is considerably longer than most upper limits derived from studies of true normal subjects, e.g., Melvin et al. 2 Unfortunately, Yaar does not present a systematic comparison of his results with those from true normals.
A lesser point: Yaar goes on to state that his procedure " . . . fails when the latencies have other than symmetric frequency distribution." We 1 and others have previously pointed out that not a few measurements in electrodiagnostic medicine probably exhibit skewed distributions in the normal population, notably some-such as distal latencies-which are biophysically constrained in one direction. We concur with Yaar that " . . . most published normative data approached the data as if they are normally distributed [Gaussian] anyway, whether the author declares so in the text or not." Because of this, we have recommended that a test for Gaussian fit be implemented (even something as elementary as visual inspection of a frequency distribution histogram) before Gaussian statistics are applied to the normative data.
Yaar closes his article by listing the methodological and fiscal obstacles to gathering quality true normative data, and offers his new method as a cheap and easy alternative. A number of aphorisms are relevant to this proposal: one that comes immediately to mind is you get what you pay for. We should beware of dubious shortcuts and insist on quality data for electrodiagnostic medical decision-making.