What is the Erlang Distribution

Historical Context and Origin

The Erlang distribution was first developed by Danish mathematician and engineer Agner Krarup Erlang (1878–1929) in the early 20th century. Erlang was pioneering the field of telephone traffic engineering, aiming to understand and optimize the efficiency of telephone exchanges. His primary goal was to model and predict the number of telephone calls an operator or a switching station would receive simultaneously, thereby establishing reliable staffing levels and necessary equipment capacity to minimize congestion and dropped calls.

Erlang’s work laid the foundation for modern queuing theory, which is now indispensable across numerous industries, including telecommunications, logistics, and healthcare. The distribution itself arises naturally from considering a series of independent, identically distributed exponentially distributed events. When modeling the time until k such events occur, the resulting distribution of the total time is the Erlang distribution with shape parameter k.

The distribution’s historical roots firmly establish its utility in fields that require the precise modeling of arrival rates and service times. Beyond its initial use in traffic modeling, its clean mathematical properties and practical interpretability have cemented its status as a cornerstone in statistical modeling wherever sequential, random processes are involved.

Mathematical Definition and Parameters

The Erlang distribution is a continuous probability distribution defined by its probability density function (PDF). Understanding this function is crucial for applying the distribution correctly in analytical models. The PDF requires two key parameters: the shape parameter ($k$) and the rate parameter ($lambda$ or its reciprocal, $mu$).

The probability density function using the scale parameter ($mu$) is given by the formula:

f(x; k, μ) = x^k-1e^-x/μ / μ^k(k-1)!

Here is a breakdown of the variables and parameters used in the equation:

• x: Represents the continuous random variable (the time elapsed).
• k: The Shape Parameter. This must be a positive integer ($k ge 1$). It signifies the number of independent sequential phases or events being summed.
• μ: The Scale Parameter. This must be a positive real number ($mu > 0$). It dictates the scaling of the time axis.

It is common in literature to see the distribution defined using the rate parameter ($lambda$) instead of the scale parameter ($mu$). The two are reciprocals of each other: $mu = 1/lambda$. When using the rate parameter, $lambda$ represents the mean rate of occurrence of the underlying events (e.g., the average number of events per unit time). Both formulations are mathematically equivalent, but the context often dictates which parameter is more intuitive for a specific application.

Key Characteristics and Statistical Moments

Analyzing the statistical moments of the Erlang distribution provides valuable insights into the expected behavior and variability of the time being modeled. These moments—the mean, mode, variance, skewness, and kurtosis—are all functions of the shape parameter ($k$) and the rate parameter ($lambda$).

The fundamental properties of the distribution when expressed in terms of the shape parameter $k$ and the rate parameter $lambda$ are:

• Mean (Expected Value): The average time until the $k$-th event occurs is given by the ratio of the shape parameter to the rate parameter. Mean = k/λ
• Variance: Measures the spread or dispersion of the distribution around the mean. Variance = k/λ²
• Mode: Represents the most probable time value. For $k > 1$, the mode is positive. Mode = (k-1)/λ
• Skewness: Quantifies the asymmetry of the distribution. As $k$ increases, the distribution becomes less skewed and more symmetrical. Skewness = 2/√k
• Kurtosis (Excess): Measures the “tailedness” of the distribution relative to a normal distribution. Kurtosis = 6/k

It is important to observe how these moments change with $k$. As the shape parameter increases, the variance decreases relative to the mean, and the skewness decreases significantly. This convergence towards symmetry reflects the underlying principle that the sum of many independent, identically distributed random variables (the $k$ exponential phases) approaches a Normal distribution, as predicted by the Central Limit Theorem.

Relationship to Other Distributions

The Erlang distribution is tightly connected to several other fundamental probability distributions, often serving as a specialized case or a foundation for more general models. These relationships highlight its place within the broader framework of probability theory.

The most significant relationship is with the Gamma distribution. The Erlang distribution is specifically a special case of the Gamma distribution where the shape parameter ($k$) is constrained to be a positive integer. This integer constraint is key to the Erlang distribution’s interpretation in terms of stages or phases (e.g., waiting for the 1st, 2nd, or 3rd event).

Furthermore, the Erlang distribution exhibits strong ties to the Exponential distribution:

• When the shape parameter, $k$, is set equal to 1, the Erlang distribution simplifies directly to the Exponential distribution. This reflects the fact that both distributions model the time until the first event in a Poisson process when $k=1$.
• Conversely, the Erlang distribution models the sum of $k$ independent, identically distributed random variables, each following an Exponential distribution with the same rate parameter $lambda$.

Finally, a relationship exists with the Chi-Squared distribution. If the scale parameter ($mu$) of the Erlang distribution is set to 2 (meaning the rate parameter $lambda = 1/2$), the resulting distribution is equivalent to a Chi-Squared distribution with $2k$ degrees of freedom. This connection is highly relevant in statistical inference and estimation methods.

Visualizing Parameter Impact

The visual shape of the Erlang distribution’s probability density function is highly sensitive to changes in its shape ($k$) and scale ($mu$) parameters. Visualizing this variability helps in understanding how different assumptions about the number of stages or the rate of occurrence affect the resulting time distribution.

The following plot shows the shape of the Erlang distribution when it takes on different parameters, illustrating the profound effect parameter selection has on the curve:

It’s interesting to see just how much the shape of the distribution changes depending on the values used for the shape and scale parameters. When the shape parameter $k$ is small (e.g., $k=1$), the distribution is highly right-skewed. As $k$ increases, the peak shifts to the right and the curve becomes increasingly concentrated around the mean, demonstrating a practical convergence toward the symmetry of the Normal distribution. This flexibility makes the Erlang model suitable for modeling a wide range of waiting times that involve multiple sequential steps.

Real-World Applications in Queueing Systems

One of the most profound applications of the Erlang distribution is within queuing theory, which provides the mathematical framework necessary to analyze systems characterized by random arrivals and service constraints. The distribution is crucial for modeling service times and inter-arrival times, offering crucial insights for operational management across industries.

1. Call Centers and Traffic Engineering

The Erlang distribution is a cornerstone in modeling the time intervals between incoming calls at a call center, as well as the expected duration of service. Formulas derived from the Erlang distribution (such as Erlang C) allow managers to determine the required staffing capacity during different times of the day. This precision ensures that the organization can handle the incoming calls in a timely fashion, maintaining high service quality, without incurring unnecessary losses by staffing too many employees during lulls.

2. Retail and Logistics Settings

In retail environments, the Erlang distribution is used for modeling the frequency of interpurchase times by consumers. This provides retailers and other businesses with an idea of how often a given consumer is expected to purchase a product or service from them. This predictive capacity is highly valuable for strategic inventory control, enabling businesses to optimize stock levels, and for staffing points of service, such as checkout counters, based on predicted customer flow.

Applications in Reliability and Biology

The Erlang distribution’s concept of sequential stages makes it highly applicable outside of customer service queues, particularly in analyzing complex biological and engineering processes.

3. Medical and Biological Settings

The Erlang distribution is widely used to model cell cycle time distribution, which is critical in oncology and developmental biology. Since the cell cycle involves a sequence of discrete, time-consuming phases (G1, S, G2, M), the Erlang distribution, with its integer shape parameter $k$ corresponding to the number of phases, provides a highly accurate model for the total time required to complete the cycle. This has a variety of different applications in medical settings, including predicting cell growth rates under varying conditions.

4. Reliability Engineering

In quality control and reliability analysis, the distribution models the time until failure for complex systems where failure requires the completion of several distinct, sequential wear-out or degradation stages. This allows engineers to calculate the mean time between failures (MTBF) and design robust maintenance schedules, drastically improving the lifespan and operational reliability of industrial machinery and complex hardware systems.

Summary and Conclusion

The Erlang distribution stands as a powerful and indispensable tool in applied probability and statistics, bridging the theoretical gap between the simple Exponential model and the more complex Gamma distribution. Originating from the practical necessities of telephone traffic engineering, its utility has expanded vastly into modern fields requiring the analysis of sequential, time-dependent processes.

Its strength lies in its ability to model the waiting time for the $k$-th event in a Poisson process, offering a realistic framework for systems involving multiple, discrete stages. By effectively manipulating the integer shape parameter $k$ and the rate parameter $lambda$, analysts can accurately describe phenomena ranging from call center congestion to product reliability and biological growth rates, ensuring optimized operational efficiency across numerous industries.