Distribution Comparison Tool

DistriScope’s Family Comparator is built for questions that a single chart cannot answer. Several probability models can share a similar center while making very different claims about extreme observations, symmetry, discreteness, or the range of possible values. The comparator helps expose those differences by showing selected families together and allowing controlled parameter changes.

The page is especially useful during model education and early analytical planning. It can show why matching centers does not make two configurations equivalent, and how a single parameter change reshapes spread, skewness, and tail weight within one family. Comparison should begin with the mechanism that generated the data; the overlay is a reasoning aid, not an automatic recommendation.

A practical distribution comparison starts with a controlled question. If you want to study tail weight, align the candidates around a comparable center and spread, then keep the plotting range fixed while examining central and extreme regions. If you want to study support, include boundaries and impossible values rather than zooming only around the peak. For discrete and continuous families, remember that probability mass and density height have different meanings even when they share an axis. The distribution comparison tool makes these structural differences visible, but the interpretation must still name which parameterization and view were used.

Use comparisons to generate testable expectations for observed data. A heavy-tailed candidate should predict more extreme standardized observations; a skewed candidate should produce asymmetric quantiles; a bounded model should never assign probability outside its limits. Record those expectations before fitting so the visual exercise becomes a transparent model-checking plan. After data are collected, inspect empirical distributions, Q-Q behavior, information criteria, and uncertainty using the same sample. Close visual overlap between theoretical curves is not a reason to ignore independence, censoring, mixtures, measurement limits, or the mechanism that produced the observations.

Parameter matching should serve the comparison question. Equal means can isolate differences in spread or skew, while matching both mean and variance can expose tail and boundary behavior that moments do not capture. Some families cannot satisfy a requested match for every parameter combination, and different software may use rate versus scale or alternative shape conventions. Record the formulas or parameter labels shown by the tool so the comparison can be reproduced. When a graph changes, identify whether the cause is location, scale, shape, support, or axis range before drawing a substantive conclusion.

A clear comparison report should name every family, parameter value, support, and plotted function. Explain why the candidates were selected and which feature the shared view was designed to isolate. If axes were standardized or truncated, state that choice because it can change the apparent overlap. For decision problems, translate visible differences into the probability, quantile, or threshold that matters rather than ranking curves by aesthetics. Recheck that numerical probability independently before using it, and keep the original figure settings so another reader can repeat the experiment and challenge the interpretation.