Exponentially Modified Gaussian Distribution¶
The distribution of an independent Gaussian value plus a positive exponential value, yielding a precise right-skewed convolution family.
Core Idea¶
The exponentially modified Gaussian distribution (EMG or ex-Gaussian) is the law of \(Z=X+Y\) for independent \(X\sim N(\mu,\sigma^2)\) and \(Y\sim\operatorname{Exp}(\lambda)\), with \(\sigma,\lambda>0\). Its density is the convolution of those two component densities. Using \(\tau=1/\lambda\), the mean is \(\mu+\tau\) and variance is \(\sigma^2+\tau^2\); the added positive exponential creates a right tail. It is a defined distribution, not a label for every skewed histogram.[ref-767419c358ee][ref-7f6005f275cf]
Scope of Application¶
Original chromatography research uses EMG functions to characterize asymmetric elution peaks. A later capillary liquid-chromatography study fits fatty-acid profiles and compares fitted symmetric and tail contributions to peak moments. Original Stroop response-time research fits the same ex-Gaussian form and finds distributional differences that mean RT alone obscured. The common model does not imply the same physical causes in both settings.[ref-6e5180d72369][ref-5947a9468e76][^ref-7bada9991d27]
In those measured chromatographic profiles, the empirical second moment did not equal the sum of fitted Gaussian and exponential variance contributions. The ideal independent-sum variance equation remains mathematically valid; the empirical mismatch warns against identifying the fitted components with independent physical causes.[^ref-5947a9468e76]
The frozen “Gaussian minus exponential distribution” redirect remains unresolved, not an alias: \(X-Y\) has a left rather than right exponential tail. EMG and ex-Gaussian are accepted naming surfaces for the right-tailed sum.[^ref-767419c358ee]
Clarity¶
This entry uses \(\lambda\) for rate and \(\tau=1/\lambda\) for mean/scale. SciPy instead uses \(K=\tau/\sigma\), loc \(=\mu\), and Scale \(=\sigma\). A larger \(\lambda\) shortens the tail; a larger \(\tau\) lengthens it. Although \(Y\) is nonnegative, \(X\) is unbounded, so \(Z\) has full-real support for positive \(\sigma\). Only the \(\sigma\to0\) limit becomes a shifted exponential bounded below by \(\mu\).[^ref-767419c358ee]
Manages Complexity¶
Three parameters condense a full right-skewed density into location, symmetric spread and one-sided tail scale. That can reveal differences hidden by a single mean and supply a common language for comparing chromatographic peak shapes or reaction-time conditions. But a good fit does not uniquely identify an adsorption mechanism or an attention process. Original model-comparison work cautions that ex-Gaussian parameters do not map one-to-one onto cognitive processes.[ref-5947a9468e76][ref-7bada9991d27][^ref-53bebe662eaa]
Abstract Reasoning¶
Independence gives \(f_Z(z)=\int_0^\infty f_X(z-y)f_Y(y)\,dy\) and lets means and variances add. The exponential's third cumulant gives standardized skewness \(2\tau^3/(\sigma^2+\tau^2)^{3/2}\). As \(\tau\to0\), the law tends to the Gaussian; as \(\sigma\to0\), it tends to \(\mu+\operatorname{Exp}(\lambda)\). Valid long observations matter for fitting \(\tau\), while uncritical outlier trimming can bias the tail; independently invalid artifacts still require treatment.[ref-767419c358ee][ref-7f6005f275cf]
Knowledge Transfer¶
The same mathematical roles map to a chromatographic zone and a Stroop response-time distribution: Gaussian location/spread, exponential right-tail scale, independence as a modeling assumption, and the convolution readout. What does not transfer is a causal claim. A chromatography tail and a slow response may be fitted by one family without sharing a generator. The broader “symmetric variation plus one-sided delay” skeleton is a possible future prime; the exact EMG remains a typed probability distribution.[ref-5947a9468e76][ref-7bada9991d27][^ref-53bebe662eaa]
[^ref-767419c358ee]: SciPy developers, scipy.stats.exponnorm, Notes on density, independent sum, support and parameter conversion.
[^ref-7f6005f275cf]: R. Ulrich and J. Miller, “Effects of Truncation on Reaction Time Analysis”, Journal of Experimental Psychology: General 123 (1994), Appendix moments and truncation analysis; PDF access intermittent.
[^ref-6e5180d72369]: E. Grushka, “Characterization of Exponentially Modified Gaussian Peaks in Chromatography”, Analytical Chemistry 44 (1972), 1733–1738; publisher record/first-page access only.
[^ref-5947a9468e76]: “Additivity of Statistical Moments in the Exponentially Modified Gaussian Model of Chromatography”, Analytica Chimica Acta 478 (2003), 99–110, publisher abstract and indexed excerpts.
[^ref-7bada9991d27]: A. Heathcote, S. J. Popiel and D. J. K. Mewhort, “Analysis of Response Time Distributions: An Example Using the Stroop Task”, Psychological Bulletin 109 (1991), 340–347, original abstract.
[^ref-53bebe662eaa]: D. Matzke and E.-J. Wagenmakers, “Psychological Interpretation of the Ex-Gaussian and Shifted Wald Parameters”, Psychonomic Bulletin & Review 16 (2009), original abstract.
Relationships to Other Abstractions¶
Current abstraction Exponentially Modified Gaussian Distribution Domain-specific
Parents (2) — more general patterns this builds on
-
Exponentially Modified Gaussian Distribution is a kind of Probability Distribution Domain-specific
This independent normal-plus-exponential law is one parametric probability-distribution family.
-
Exponentially Modified Gaussian Distribution presupposes Convolution Prime
The density of the sum is defined as the convolution of its independent Gaussian and exponential component densities.
Hierarchy paths (6) — routes to 3 parentless roots
- Exponentially Modified Gaussian Distribution → Probability Distribution → Random Variable → Function (Mapping)
- Exponentially Modified Gaussian Distribution → Convolution → Function (Mapping)
- Exponentially Modified Gaussian Distribution → Probability Distribution → Probability → Measure → Set and Membership
- Exponentially Modified Gaussian Distribution → Probability Distribution → Probability → Measure → Aggregation → Micro Macro Linkage
- Exponentially Modified Gaussian Distribution → Probability Distribution → Random Variable → Probability → Measure → Set and Membership
- Exponentially Modified Gaussian Distribution → Probability Distribution → Random Variable → Probability → Measure → Aggregation → Micro Macro Linkage
Neighborhood in Abstraction Space¶
Exponentially Modified Gaussian Distribution sits in a sparse region of the domain-specific corpus (67th percentile for distinctiveness): few abstractions share its structure, so a faithful description tends to retrieve it precisely.
Family — Foundations of Probability & Inference (29 abstractions)
Nearest neighbors
- Cramér's Theorem (Large Deviations) — 0.85
- Covariance Matrix — 0.85
- Probability Distribution — 0.84
- Cochran's Theorem — 0.84
- Random Variable — 0.84
Computed from structural-signature embeddings · 2026-10-08