binsmooth (version 0.2.2)

rsubbins: Recursive subdivision PDF and CDF fitted to binned data

Description

Creates a PDF and CDF based on a set of binned data, using recursive subdivision on a step function.

Usage

rsubbins(bEdges, bCounts, m=NULL, eps1 = 0.25, eps2 = 0.75, depth = 3,
        tailShape = c("onebin", "pareto", "exponential"),
        nTail=16, numIterations=20, pIndex=1.160964, tbRatio=0.8)

Arguments

bEdges

A vector \(e_1, e_2, \ldots, e_n\) giving the right endpoints of each bin. The value in \(e_n\) is ignored and assumed to be Inf or NA, indicating that the top bin is unbounded. The edges determine \(n\) bins on the intervals \(e_{i-1} \le x \le e_i\), where \(e_0\) is assumed to be 0.

bCounts

A vector \(c_1, c_2, \ldots, c_n\) giving the counts for each bin (i.e., the number of data elements in each bin). Assumed to be nonnegative.

m

An estimate for the mean of the distribution. If no value is supplied, the mean will be estimated by (temporarily) setting \(e_n\) equal to \(2e_{n-1}\), and a warning message will be generated.

eps1

Parameter controlling how far the edges of the subdivided bins are shifted. Must be between 0 and 0.5.

eps2

Parameter controlling how wide the middle subdivsion of each bin should be. Must be between 0 and 1.

depth

Number of times to subdivide the bins.

tailShape

Must be one of "onebin", "pareto", or "exponential".

nTail

The number of bins to use to form the initial tail, before recursive subdivision. Ignored if tailShape equals "onebin".

numIterations

The number of iterations to optimize the tail to fit the mean. Ignored if tailShape equals "onebin".

pIndex

The Pareto index for the shape of the tail. Defaults to \(\ln(5)/\ln(4)\). Ignored unless tailShape equals "pareto".

tbRatio

The decay ratio for the tail bins. Ignored unless tailShape equals "exponential".

Value

Returns a list with the following components.

rsubPDF

A stepfun function giving the fitted PDF.

rsubCDF

A piecewise-linear approxfun function giving the CDF.

E

The right-hand endpoint of the support of the PDF.

shrinkFactor

If the supplied estimate for the mean is too small to be fitted with a step function, the bins edges will be scaled by shrinkFactor, which will be chosen less than (and close to) 1.

Details

First, a step function PDF is created, as described in stepbins. The bins of the resulting PDF are then recursively subdivided and shifted in a manner that preserves the area of the original bins, resulting in a step function with finer bins.

The methods stepbins and rsubbins are included in this package mainly for the purpose of comparison. For most use cases, splinebins will produce more accurate smoothing results.

References

Paul T. von Hippel, David J. Hunter, McKalie Drown. Better Estimates from Binned Income Data: Interpolated CDFs and Mean-Matching, Sociological Science, November 15, 2017. https://www.sociologicalscience.com/articles-v4-26-641/

See Also

stepbins

Examples

# NOT RUN {
# 2005 ACS data from Cook County, Illinois
binedges <- c(10000,15000,20000,25000,30000,35000,40000,45000,
              50000,60000,75000,100000,125000,150000,200000,NA)
bincounts <- c(157532,97369,102673,100888,90835,94191,87688,90481,
               79816,153581,195430,240948,155139,94527,92166,103217)
rsb <- rsubbins(binedges, bincounts, 76091, tailShape="pareto")

plot(rsb$rsubPDF, do.points=FALSE)
plot(rsb$rsubCDF, 0, rsb$E)

library(pracma)
integral(rsb$rsubPDF, 0, rsb$E)
integral(function(x){1-rsb$rsubCDF(x)}, 0, rsb$E) #mean is approximated
# }