Distributions Relating to Inference Based on the Normal Distribution
Normal reference distributions in inference
Normal-based inference expresses a result on a standardized scale: subtract the model center and divide by the relevant standard deviation or standard error. The resulting z value says how many reference-scale units the observation lies above or below the center. Its sign gives direction; its magnitude determines which tail areas are small.
A normal reference distribution is a model, not a decorative curve. Before using it, identify what quantity is being standardized and whether the conditions supporting the stated reference distribution are plausible. The same raw difference can have a different standardized meaning when its uncertainty differs.
Tail probabilities answer conditional questions: assuming the stated model and its center are correct, how often would results at least this far into the tail occur? A small tail probability is not the probability that the model is false, and it does not measure effect size or practical importance.
Under a standard normal reference model, which z values mark the central 95% approximately?
A calculation gives a two-sided normal-reference tail probability of 0.004 for a test statistic. Which interpretation is justified?