Interactive module guide

Logistic Distribution Calculator & Guide

This logistic distribution calculator lets you calculate probabilities for the S-shaped CDF behind logistic regression, with slightly heavy tails.

Open calculator

Free · No sign-up · Calculations stay in your browser

The logistic distribution is the symmetric family whose cumulative curve is the exact logistic sigmoid 1/(1+e^{−(x−μ)/s}).

It looks like a normal distribution with slightly heavier tails (excess kurtosis 1.2), but its real importance is structural: modeling a binary outcome as a latent variable plus logistic noise crossing a threshold is precisely logistic regression, and the same curve underlies growth models, item response theory, and Elo-style rating systems.

This page provides the closed-form density, CDF, quantiles (the logit function), and exact moments.

Is this the right module for my question?

Use it when

Use it to understand logistic regression mechanically: the model’s probabilities are this CDF evaluated at a linear predictor, so log-odds are linear exactly because the latent noise is logistic.
Use it for growth-curve reasoning: adoption fractions and saturating processes follow the sigmoid, and quantile questions (“when does the process cross 90%?”) are one closed-form logit away.
Use it in paired-comparison and rating models: the probability that A beats B depends on the rating gap through this CDF, which is the mathematical core of Elo and Bradley–Terry systems.

Choose another method when

Avoid treating it as interchangeable with the normal in tail-sensitive work: its tails decay like e^{−x/s} rather than e^{−x²/2σ²}, so extreme events are noticeably more likely.
Avoid using it for heavy-tailed data in the power-law sense; its tails are exponential, far lighter than Pareto or Cauchy behavior.
Avoid parameter confusion when comparing with software: some libraries parameterize by the standard deviation rather than the scale s; the conversion is σ = sπ/√3 ≈ 1.814s.

Interactive tool

The calculator loads as you approach this section so the guide remains fast on mobile connections.

Example preview

Growth-curve crossing time

Inputs: X ~ Logistic(μ=0, s=1); find the 90th percentile.
Representative result: Q(0.9) = ln(9) ≈ 2.197 by the closed-form logit quantile.

Illustrative only. Load the interactive tool to enter your own values and review assumptions.

Watch the explanation

3:17 min

The player loads only after you press play. You can also watch on YouTube.

Read the complete transcript

The Logistic Distribution. Why can the same S curve describe adoption timing, rating differences, and binary regression probabilities? The logistic family uses finite location mu, positive scale s, and the entire real line. Its cumulative function is the exact sigmoid: one divided by one plus exponential negative standardized distance. Its density equals F times one minus F divided by s, producing a smooth symmetric peak. The quantile converts probability into log odds, adding s times log p over one minus p. Symmetry places mean, median, and mode at mu, where the cumulative probability is one half.

Variance is pi squared s squared over three, so standard deviation is one point eight one four s. Skewness is zero and excess kurtosis is exactly one point two for every positive scale. Probability beyond distance t on either side equals two divided by one plus exponential t over s. The interquartile range is two s log three, while median absolute deviation is s log three. At mu, the sigmoid is steepest, with derivative exactly one divided by four s. The identity F times one minus F equals s f links cumulative probability directly to density. With Y equal one when eta plus standard logistic noise is positive, probability equals sigmoid eta. That latent-error convention explains logistic regression, while a normal latent error instead produces probit regression. The same geometry supports adoption curves, item response models, and Bradley Terry or Elo comparisons.

Model technology-adoption thresholds with mu zero and s one, measuring years around the midpoint. The ninety percent crossing is log nine, approximately two point one nine seven years after midpoint. Probability below minus two is one over one plus e squared, approximately point one one nine two. Variance is pi squared over three, and standard deviation is approximately one point eight one three eight. Three standard deviations equal five point four four one four scale units from the center. There the logistic double tail is point zero zero eight six three, about three point two times normal.

Diagnose the family by plotting empirical log odds against x and checking for approximate linearity. Its tails are exponential, heavier than normal far out but much lighter than genuine power laws. An attractive sigmoid does not prove fit, and this explorer does not estimate regression coefficients. Reject nonpositive s, nonfinite mu, and probabilities outside the open unit interval before calculation. Log-logistic and generalized logistic models are different families, with different support, shape, and formulas. Report mu, s, event, population, result, endpoints, and how every normal comparison matched spread. Explore logistic probabilities, quantiles, sigmoid geometry, and tail comparisons free at Distri Scope dot com.

How to read the result

Read quantiles as log-odds: Q(0.9) − μ = s·ln 9 ≈ 2.197s says “90% of mass lies below 2.2 scale units,” and every additional factor of 10 in odds adds ln 10 ≈ 2.30 scale units.

Compared with a normal of equal variance, logistic tails are modestly heavier: |X−μ| exceeding 3 standard deviations has probability about 0.0090 against the normal’s 0.0027.

The central slope is 1/(4s): near μ, each unit of x moves the CDF by about 25%/s, the steepest part of the sigmoid.

Inputs and parameter meaning

Two parameters: the location μ (mean, median, and mode) and the scale s; the standard deviation is sπ/√3, about 81% larger than s.

The plotted range covers the central 99% of probability; both exponential tails extend beyond it.

Quantiles use the closed-form logit: Q(p) = μ + s·ln(p/(1−p)), so percentile questions never require iteration.

Formula and available parameters

Density or probability mass

f(x) = \frac{e^{-(x-\mu)/s}}{s\left(1 + e^{-(x-\mu)/s}\right)^2}

Cumulative distribution

F(x) = \frac{1}{1 + e^{-(x-\mu)/s}}

Mean

\mu

Variance

\sigma^2 = \frac{s^2\pi^2}{3}

Location (μ): interactive range -10 to 10; default 0.
Scale (s): interactive range 0.1 to 5; default 1.

What the method is doing

Density, CDF, and quantile function are all closed-form and mutually consistent; the density is computed in a form symmetric in ±(x−μ) for numerical stability far from the center.

Moments are exact: mean μ, variance s²π²/3, skewness 0, excess kurtosis exactly 6/5 at every parameter choice.

The identity F(x)(1−F(x)) = s·f(x) links the density to the CDF—the same identity that gives logistic regression its elegant score equations.

Worked example: adoption crossing ninety percent

A technology-adoption fraction over time follows a logistic curve centered at μ = 0 (the midpoint year) with scale s = 1 (in years). Find when adoption crosses 90%, the probability that a latent adopter’s threshold lies below −2 years, the standard deviation of the threshold distribution, and the 3σ tail compared with a normal.

1Crossing time: Q(0.9) = ln(0.9/0.1) = ln 9 ≈ 2.197 years after the midpoint—the closed-form logit answer.
2Early tail: F(−2) = 1/(1+e²) ≈ 0.1192, so about 12% of thresholds lie more than two years before the midpoint.
3Spread: the standard deviation is π/√3 ≈ 1.814 years, noticeably larger than the scale parameter 1.
4Tail comparison at 3σ (≈ 5.44 years): the logistic two-sided tail is ≈ 0.0090, roughly 3.3 times the normal’s 0.0027 at the same variance.

Interpretation

The logit quantile makes timeline questions trivial—every “when does adoption reach p?” is one logarithm. The tail comparison carries the practical caveat: laggards and ultra-early adopters are several times more common than a normal model would predict, which matters whenever planning targets the distribution’s extremes rather than its middle.

Common mistakes

Do not confuse the scale s with the standard deviation; reporting s where σ is expected understates spread by 45%.
Do not read logistic-regression coefficients as normal-latent effects: the latent noise is logistic, and coefficient scales differ from probit by roughly the factor 1.6–1.8.
Do not use the closed-form convenience as evidence of fit; convenient and correct are separate questions, and heavy-tailed data will still violate this model.

Limits and independent validation

The page covers the symmetric location-scale family; skewed generalizations (log-logistic, generalized logistic) are out of scope.

It describes the latent distribution, not fitted regression: estimating coefficients from data is a modeling task beyond this page.

Fixed kurtosis 1.2 means the family cannot adapt its tail weight—t distributions do that with a degree-of-freedom dial.

Before using the result

Verify one logit identity by hand: F(μ + s·ln 9) must be exactly 0.9; the point calculator confirms it.
Check the variance reading σ² = s²π²/3 against the summary panel.
On candidate data, plot empirical log-odds ln(F̂/(1−F̂)) against x; approximate linearity is the family’s signature.

normal distribution →Compare with the probit-latent alternative at matched variance.two proportion z test →Analyze the binary outcomes that logistic latent models generate.

← Laplace Distribution Rayleigh Distribution →

See every option in the Probability Distribution Explorer or review the DistriScope methodology. Educational information; last reviewed 2026-08-09. Verify consequential calculations independently.