Fitting Data to a Cauchy Distribution

Introduction

Elsewhere on the website, we describe two methods for fitting data to a Cauchy distribution: the method of moments and maximum likelihood estimation (MLE). We now describe another approach based on order statistics.

Weighted Order Statistics Estimate

As usual, our objective is to identify the Cauchy distribution parameters μ and σ that result in a good fit for the data. Let’s assume that the data x1, …, xn is ordered so that x1 ≤ ⋅⋅⋅ ≤ xn

For each i = 1, …, n, define

a_i

Next, define the following weights

Weights

The estimates for the parameters are

Parameter estimates

Worksheet Function

Real Statistics Function: The Real Statistics Resource Pack provides the following array function.

CAUCHY_FITX(R1, lab): returns a column array with the ordered statistics estimates of the mu and sigma parameters of the Cauchy distribution that best fits the data in the column array or range R1, along with the log-likelihood estimate based on these parameters. If lab = TRUE (default FALSE) then a column of labels is appended to the output.

Example

Example 1: Determine the parameters of a Cauchy distribution that best fit the data in range D2:H5 of Figure 1 using the various distribution fitting approaches described above.

 Data for Cauchy distribution

Figure 1 – Fitting data to a Cauchy distribution

We first note that the data in D2:H5 was obtained from a Cauchy distribution with mu = 2 and sigma = .5 by using the formula =T3_INV(RAND(),1,B2,B3) in all the cells of D2:H5. Thus, we expect that we will get a good fit for the data with a Cauchy distribution.

Figure 2 shows the results of fitting the data to a Cauchy distribution using the MLE, pseudo-method-of-moments, and ordered statistics approaches.

Estimating Cauchy distribution parameters

Figure 2 – Estimating Cauchy distribution parameters

The results are fairly similar, although we see that the ordered statistics approach yields estimates for the parameters that are closest to those in cells B2 and B3. Not surprisingly, though, the MLE approach yields the results with the largest log-likelihood value.

Examples Workbook

Click here to download the Excel workbook with the examples described on this webpage.

References

D’Agostino, R. B., Stephens, M. A. (1986) Goodness-of-fit techniques. Marcel Dekker, Inc.
https://www.hep.uniovi.es/sscruz/Goodness-of-Fit-Techniques.pdf

Chemoff, H., Gastwirth, J. L., Johns, M. V. (1967) Asymptotic distribution of linear combinations of functions of order statistics with applications to estimation. Ann. Math. Statist. 38, 52-73.
https://projecteuclid.org/journals/annals-of-mathematical-statistics/volume-38/issue-1/Asymptotic-Distribution-of-Linear-Combinations-of-Functions-of-Order-Statistics/10.1214/aoms/1177699058.full

Leave a Comment