TL;DR
This paper introduces a SAS/IML macro that efficiently computes and visualizes probability values for Pearson distributions, a versatile family of distributions useful for modeling data with unknown characteristics. This tool aims to simplify statistical analysis by providing an accessible way to obtain these values, overcoming the limitations of older, less user-friendly methods.
What Problem Does This Paper Address?
Many statistical analyses rely on the assumption of normally distributed data. However, real-world data often deviates from normality, making this assumption problematic. Pearson distributions offer a powerful framework to approximate a wide range of non-normal distributions using just the first four moments of the data (mean, variance, skewness, and kurtosis). While tables and older software exist for calculating percentage points of these distributions, they often require cumbersome interpolation methods, making them impractical for routine statistical analysis, especially in hypothesis testing where precise probability values (p-values) are crucial. This paper addresses the need for a more user-friendly and computationally efficient tool to work with Pearson distributions.
Methods
The core of this work is a SAS/IML macro program. The macro takes the second, third, and fourth moments of a dataset as input, along with a specific percentage point (x₀). It first calculates the B-coefficients (skewness and kurtosis) and then uses a κ-criterion to determine which of the seven Pearson distribution types (plus the normal distribution) best fits the data. Based on the identified distribution type, the macro calculates the specific parameters of its density function. Finally, it employs the SAS subroutine QUAD for numerical integration to compute the probability value corresponding to the input percentage point. For visualization, an accompanying small macro can generate plots of the density function with the calculated probability value indicated.
Datasets
The paper does not explicitly mention training datasets. For evaluation, the authors compare the parameters computed by their SAS/IML macro against values found in Elderton and Johnson (1969)'s tables. They also evaluate the computed probability values against percentage points from Pearson and Hartley (1972)'s Table 32.
Key Insights and Ideas
- Pearson distributions are a flexible tool: They can approximate a wide variety of data shapes, making them invaluable when the underlying distribution is unknown.
- Traditional methods are cumbersome: Relying on printed tables and manual interpolation for Pearson distribution values is inefficient and prone to error, hindering practical statistical analysis.
- Automation is key: A programmatic approach, like the SAS/IML macro, significantly streamlines the process of obtaining and visualizing these probability values.
- Visualizing probability values is important: The ability to graph the density function and highlight the probability value provides a crucial visual aid for understanding the statistical implications.
Key Results
The authors demonstrate that their SAS/IML macro accurately computes the parameters for various Pearson distribution types, with absolute differences from Elderton and Johnson (1969) being very small (mostly less than 0.001). Furthermore, the computed probability values for common percentage points (2.5% and 97.5%) are highly precise, with absolute differences less than 0.0001 compared to values from Pearson and Hartley (1972). The macro also offers an optional graphing feature to visualize these results.
Limitations and Future Work
The paper notes that the first moment (mean) is not an explicit input for the macro, as it's implicitly handled in the calculation of higher-order moments. A limitation mentioned is that Elderton and Johnson (1969) do not provide tables for all Pearson distribution types (specifically III and V). For future work, the authors suggest developing similar programs in other widely used statistical languages like R or Stata to broaden accessibility.
Summary updated