Treating weak reflexions in least-squares calculations

In photographic intensity measurements, the non-observability of a reflexion establishes a conditional probability distribution for the true value of F that is undefined below an observational threshold. By contrast, counter methods yield values of F²_o in which the errors can be expected to approach a normal distribution even for the weakest reflexions. In a least-squares refinement, preferably based on F², this normal distribution should not be distorted by the arbitrary exclusion, or other maltreatment, of 'unobserved' reflexions. Such selective tampering with weak intensities biases the input data and so risks systematic error in the refined parameters. The expected effect is illustrated by a numerical experiment.