density-estim: Density Estimators

In retimes: Reaction Time Analysis

Description

Functions to identify the peak of a probability density distribution using a Gaussian kernel estimator (kernestim), a fixed-width histogram estimator (histestim), and a distance density estimator (distestim).

Usage

kernestim(x, smoothing = NULL)
histestim(x, smoothing = NULL)
distestim(x, smoothing = NULL)

Arguments

x	Vector of data.
smoothing	Smothing parameter, also named bandwidth. If the argument is defined as NULL, the smoothing parameter will be calculated from the data (see details).

Details

The functions kernestim and histestim implement two well-known methods to estimate the density of an empirical distribution (see Van Zandt, 2000, for their application in RT analysis). The functions can be used to find the value corresponding to the peak of an empirical distribution.

The function kernestim center a Gaussian density over each observation and identifies the value with the greater density.

The function histestim divides the data in bins, starting from the lower to the higher value of data. The function searches the bin with the higher data frequency. The peak of the distribution is identified calculating the mean of the data into the bin.

The function distestim is an experimental method which mixes the two previous techniques. Around each data point (pivot), an interval [x_{i}-h/2, x_{i}+h/2] is built, where h is the smoothing parameter. The function searches the interval (bin) with the higher data frequency. The output value is the weighted average of the values into the selected bin, in which each observation is weighted on the basis of the distance from the pivot. If bins with equal densities are found, the bin presenting the smallest deviance from the pivot is chosen.

For the Gaussian kernel estimator, the smoothing parameter is calculated using the Silverman's method (Silverman, 1986). Differently, using histogram and distance estimators, the smoothing parameter is calculated as: (Q_{0.975}-Q_{0.025}) / √{n}, where Q_{p} are the quantiles for α = 0.05 and n is the sample dimension.

Value

The value at which the peak of density is located.

Author(s)

Davide Massidda [e-mail]

References

Silverman, B. W. (1986). Density estimation for statistics and data analysis. London: Chapman & Hall.

Parzen, E. (1962). On estimation of a probability density function and mode. Annals of Mathematical Statistics, 33(3), 1065-1076.

Van Zandt, T. (2000). How to fit a response time distribution. Psychonomic Bulletin & Review, 7(3), 424-465.

Examples

x <- rexgauss(1000, mu=500, sigma=100, tau=250)
k <- kernestim(x); k
h <- histestim(x); h
d <- distestim(x); d
plot(density(x))
segments(k,0,k,1,col="red")
segments(h,0,h,1,col="blue")
segments(d,0,d,1,col="green")

Example output

Reaction Time Analysis (version 0.1-2)
[1] 606.6124
[1] 595.3623
[1] 595.375

retimes documentation built on May 2, 2019, 12:34 p.m.