Distribution comparison tool
Probability Distribution Family Comparator
A distribution family comparator places two or more probability models on a common visual scale so their support, center, dispersion, skewness, and tail behavior can be examined systematically.
What this tool helps you understand
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 help distinguish a bell-shaped continuous model from a count model, or show why matching means does not make two distributions equivalent. Comparison should begin with the mechanism that generated the data; the overlay is a reasoning aid, not an automatic recommendation.
Before you begin
Write down the statistical question, the unit of observation, and the quantity you want to estimate or explain before opening Distribution Family Comparison Tool. Confirm where the values came from, what units they use, and whether repeated observations are independent. Preserve the original inputs and record every parameter, transformation, and option used in the workspace. This creates a reproducible trail and makes it easier to compare the result with another package.
Treat the graph and numerical output as evidence within a model, not as a substitute for the study design. If a conclusion changes when a plausible parameter or assumption changes, report that sensitivity. Clear documentation is part of statistical accuracy because it allows another person to understand what was calculated, test the same conditions, and identify where an interpretation may need revision.
Core capabilities
Side-by-side family views
Inspect multiple distributions in one workflow so changes in shape and support are easier to recognize.
Controlled parameter comparison
Change one parameter at a time and observe whether location, spread, skew, or tail weight is affected.
Shared visual context
Use aligned plot ranges and clear legends to avoid misleading comparisons caused by unrelated axes.
Conceptual model screening
Identify obvious support or shape conflicts before moving to formal fitting and diagnostic checks.
A responsible workflow
- 1State the measurement type and its possible values, including whether it is discrete, continuous, bounded, or nonnegative.
- 2Select candidate families whose support is compatible with that measurement.
- 3Choose parameters that make a meaningful comparison, such as approximately matching centers or spreads.
- 4Inspect symmetry, modes, boundary behavior, and tails rather than comparing only the tallest point.
- 5Write down which observable patterns would distinguish the candidates in real data, then move to fitting or diagnostics.
Worked example: normal versus Student’s t
Compare a standard normal curve with a centered Student’s t distribution. At moderate degrees of freedom the curves can appear similar near zero, but the t distribution assigns more probability to the tails. Increase the degrees of freedom and the t curve approaches the normal curve. Keep the axis range fixed so the changing tail behavior remains visible.
Interpretation
A model with heavier tails treats extreme standardized observations as less surprising. That can affect interval estimates and tests, but it does not justify choosing the model without checking how the data were collected.
How to interpret the result
- Compare models on a common support. Overlaying a discrete probability mass function and a continuous density can be educational, but their vertical values have different meanings.
- Do not interpret density height as probability at an exact continuous value. Probability is represented by area over an interval.
- Matching the mean and variance is a useful controlled experiment, but distributions may still differ in skewness, kurtosis, modality, and tail risk.
- After a visual comparison, use the Data Fitting workspace for observed samples and report diagnostics rather than selecting the curve that merely looks familiar.
Key concepts behind the tool
Support before shape
Support is the first comparison because it rules out impossible candidates. A model for counts must respect integer values; a waiting-time model usually cannot generate negative values; a proportion may require bounded support. Shape comparison becomes meaningful only after this basic compatibility check.
Tail behavior
Two curves can match near the center and diverge sharply in their tails. That difference controls the modeled frequency of extreme observations and can dominate risk, reliability, and threshold calculations. Always compare tails on a range that makes their separation visible.
Standardized comparisons
Location and scale can obscure structural differences. Standardizing candidate distributions or matching selected moments helps isolate symmetry, skewness, and tail weight. It is a controlled learning device, not evidence that estimated real-world parameters are equal.
Limitations and verification
- An overlay can hide local differences when scales are poorly chosen; zoom and inspect tails separately.
- Visual similarity is not a goodness-of-fit test and does not account for parameter uncertainty.
- Some families use different parameter conventions across references and software.
- A candidate model must remain defensible from the sampling process and subject matter.
Related DistriScope resources
Educational information. Last reviewed 2026-07-30. Calculations should be independently verified for consequential decisions.