4
$\begingroup$

The empirical cumulative distribution function of a random variable, given observations $x_\left( k \right) > x_\left( k-1 \right)$, $k \in \mathbb N$, $k \le n$, is defined as $F_{emp}(x_\left( k \right) > X \ge x_\left( k-1 \right)) = \frac k {n+1}$ and $F_{emp}(X \ge x_\left(n\right))=1$.

Why? As long as we're interpolating, wouldn't it make sense to use some interpolation method with less error? A simple nearest neighbour or piecewise average interpolant would be an improvement, and a cubic interpolant would get us a differentiable empirical density function, too.

The above definition won't even give you the piecewise infimum of the cdf, because the variable is random. It certainly approaches the true function as $n\to\infty$, but then so would any other interpolant. Surely at least linear interpolants were considered.

$\endgroup$
4
$\begingroup$

The EDF is the CDF of the population constituted by the data themselves. This is exactly what you need to describe and analyze any resampling process from the dataset, including nonparametric bootstrapping, jackknifing, cross-validation, etc. Not only that, it's perfectly general: any kind of interpolation would be invalid for discrete distributions.

$\endgroup$
  • $\begingroup$ For discrete distributions, there is no point in defining the ecdf at all in $\mathbb R$. It makes no sense to ask "what proportion of families have at most 1.5 children", for example. As for sampling from the sample - I have never thought of that as a particularly good idea to begin with, either. $\endgroup$ – sesqu Nov 25 '10 at 15:02
  • 5
    $\begingroup$ @sesqu Hmm... You ought to rethink that last statement, because it seems you have rejected out of hand some significant advances in statistical concepts and methodology. The first statement seems to misunderstand what the ecdf is; it perhaps is related to an unusual conception of what a distribution function is. Are you using a standard definition such as that given at mathworld.wolfram.com/DistributionFunction.html ? $\endgroup$ – whuber Nov 26 '10 at 16:01
4
$\begingroup$

The empirical CDF is just one estimator for the CDF. It's consistent, converges pretty quickly in general, and is dead simple to understand. If you want something fancier you could certainly get a kernel density estimate for the PDF and integrate it to get another estimate for the CDF, which would do some kind of interpolation as you suggest. But if it ain't broke....

$\endgroup$
  • 2
    $\begingroup$ Indeed, in the case of a random sample, the empirical CDF converges uniformly. Massart's refinement of the Dvoretzky-Kiefer-Wolfowitz inequality shows that this occurs exponentially fast in probability. $\endgroup$ – cardinal May 13 '11 at 23:26
  • 1
    $\begingroup$ @cardinal--oh, of course. I was thinking only of continuous CDFs. It wouldn't make much sense to interpolate discrete distributions. $\endgroup$ – whuber May 14 '11 at 1:30
  • 1
    $\begingroup$ @whuber: I suspect that for continuous cdfs it must be the case. I'll see if I can come up with a complete and simple argument tomorrow. Intuitively, I would expect that the left-continuous with right-hand limits modification of the empirical cdf could also be shown to satisfy the DKW inequality in the continuous case. That result would then seem to imply your statement. $\endgroup$ – cardinal May 14 '11 at 2:15
  • 3
    $\begingroup$ I wonder then if there's an interpolated estimator with better finite sample properties, besides trivial cases like $N=2$ or something. Though I imagine it'd require extra assumptions about the true distribution. The efficiency of the ecdf is quite remarkable, really. $\endgroup$ – JMS May 14 '11 at 2:40
  • 2
    $\begingroup$ @JMS, I would guess that even restricting to continuous distributions may not yield a uniformly better interpolated estimator. I say this because we can approximate a discrete distribution by a continuous one to arbitrary precision. (Even with continuously differentiable $F$ we may not get such a statement.) It's a good question. $\endgroup$ – cardinal May 14 '11 at 18:42
0
$\begingroup$

I can't answer this question in full generality, but I think I can state one circumstance where it certainly is not useful: The Anderson-Darling test: \begin{align*} A^2/n &:= \int_{-\infty}^{\infty} \frac{(F_{n}(x) -F(x))^2}{F(x)(1-F(x))} \, \mathrm{d}F(x) \\ &= \int_{-\infty}^{x_0} \frac{F(x)}{1-F(x)} \, \mathrm{d}F(x) + \int_{x_{n-1}}^{\infty} \frac{1-F(x)}{F(x)} \, \mathrm{d}F(x) + \sum_{i=0}^{n-2} \int_{x_i}^{x_{i+1}} \frac{(F_n(x) - F(x))^2}{F(x)(1-F(x))} \mathrm{d}F(x) \end{align*}

Here, $F$ is the cumulative distribution function of the normal distribution, namely, $$ F(x) := \frac{1}{2}\left[1 + \mathrm{erf}\left(\frac{x}{\sqrt{2}}\right) \right] $$ and $F_n$ is the empirical cumulative distribution function $$ F_n(x) := \frac{1}{n} \sum_{i=0}^{n-1} \mathbb{1}_{x_i \le x} $$ (We will abuse notation a bit and let $F_{n}$ denote the linearly interpolated version of $F_n$ as well.)

I repeatedly generated $n$ $~N(0,1)$ random numbers, sorted them, and then considered $F_n$ first as a step function, and then as a sequence of linear interpolants. Each interior integral was computed via Gaussian quadrature of ridiculously high degree, and the tails via exp-sinh.

Did the empirical distribution fit the cumulative distribution better with linear interpolation than step interpolation? No, in fact they are indistinguishable as $n\to \infty$ and one is not uniformly better than the other for small $n$:

enter image description here

Code to reproduce:

#include <iostream>
#include <random>
#include <utility>
#include <boost/math/distributions/anderson_darling.hpp>
#include <quicksvg/scatter_plot.hpp>

template<class Real>
std::pair<Real, Real> step_vs_linear(size_t n)
{
    std::random_device rd;
    Real mu = 0;
    Real sd = 1;
    std::normal_distribution<Real> dis(mu, sd);

    std::vector<Real> v(n);
    for (size_t i = 0; i < n; ++i) {
        v[i] = dis(rd);
    }
    std::sort(v.begin(), v.end());
    Real Asq = boost::math::anderson_darling_normality_step(v, mu, sd);
    Real step = Asq;
    //std::cout << "n = " << n << "\n";
    //std::cout << "Step: A^2 = " << Asq << ", Asq/n = " << Asq/n << "\n";

    Asq = boost::math::anderson_darling_normality_linear(v, mu, sd);
    Real line = Asq;
    //std::cout << "Line: A^2 = " << Asq << ", Asq/n = " << Asq/n << "\n";

    return std::pair<Real, Real>(step, line);
}

int main(int argc, char** argv)
{
    using std::log;
    using std::pow;
    using std::floor;
    size_t samples = 10000;
    std::vector<std::pair<double, double>> linear_Asq(samples);
    std::vector<std::pair<double, double>> step_Asq(samples);
    std::default_random_engine generator;
    std::uniform_real_distribution<double> distribution(3, 18);

#pragma omp parallel for
    for(size_t sample = 0; sample < samples; ++sample) {
        size_t n = floor(pow(2, distribution(generator)));
        auto [step , line] = step_vs_linear<double>(n);
        step_Asq[sample] =  std::make_pair<double, double>(std::log2(double(n)), std::log(step/n));
        linear_Asq[sample] = std::make_pair<double, double>(std::log2(double(n)), std::log(line/n));
        if (sample % 10 == 0) {
            std::cout << "Sample " << sample << "/" << samples << "\n";
        }
    }

    std::string title = "Linear (blue) vs step (orange) Anderson-Darling test";
    std::string filename = "ad.svg";
    std::string x_label = "log2(n)";
    std::string y_label = "ln(A^2/n)";
    auto scat = quicksvg::scatter_plot<double>(title, filename, x_label, y_label);
    scat.add_dataset(linear_Asq, false, "steelblue");
    scat.add_dataset(step_Asq, false, "orange");
    scat.write_all();
}

Anderson-Darling tests:

#ifndef BOOST_MATH_DISTRIBUTIONS_ANDERSON_DARLING_HPP
#define BOOST_MATH_DISTRIBUTIONS_ANDERSON_DARLING_HPP

#include <cmath>
#include <algorithm>
#include <boost/math/distributions/normal.hpp>
#include <boost/math/quadrature/exp_sinh.hpp>
#include <boost/math/quadrature/gauss_kronrod.hpp>

namespace boost { namespace math {

template<class RandomAccessContainer>
auto anderson_darling_normality_step(RandomAccessContainer const & v, typename RandomAccessContainer::value_type mu = 0, typename RandomAccessContainer::value_type sd = 1)
{
    using Real = typename RandomAccessContainer::value_type;
    using std::log;
    using std::pow;
    if (!std::is_sorted(v.begin(), v.end())) {
        throw std::domain_error("The input vector must be sorted in non-decreasing order v[0] <= v[1] <= ... <= v[n-1].");
    }

    auto normal = boost::math::normal_distribution(mu, sd);

    auto left_integrand = [&normal](Real x)->Real {
        Real Fx = boost::math::cdf(normal, x);
        Real dmu = boost::math::pdf(normal, x);
        return Fx*dmu/(1-Fx);
    };
    auto es = boost::math::quadrature::exp_sinh<Real>();
    Real left_tail = es.integrate(left_integrand, -std::numeric_limits<Real>::infinity(), v[0]);

    auto right_integrand = [&normal](Real x)->Real {
        Real Fx = boost::math::cdf(normal, x);
        Real dmu = boost::math::pdf(normal, x);
        return (1-Fx)*dmu/Fx;
    };
    Real right_tail = es.integrate(right_integrand, v[v.size()-1], std::numeric_limits<Real>::infinity());


    auto integrator = boost::math::quadrature::gauss<Real, 30>();
    Real integrals = 0;
    int64_t N = v.size();
    for (int64_t i = 0; i < N - 1; ++i) {
        auto integrand = [&normal, &i, &N](Real x)->Real {
            Real Fx = boost::math::cdf(normal, x);
            Real Fn = (i+1)/Real(N);
            Real dmu = boost::math::pdf(normal, x);
            return (Fn - Fx)*(Fn-Fx)*dmu/(Fx*(1-Fx));
        };
        auto term = integrator.integrate(integrand, v[i], v[i+1]);
        integrals += term;
    }


    return v.size()*(left_tail + right_tail + integrals);
}


template<class RandomAccessContainer>
auto anderson_darling_normality_linear(RandomAccessContainer const & v, typename RandomAccessContainer::value_type mu = 0, typename RandomAccessContainer::value_type sd = 1)
{
    using Real = typename RandomAccessContainer::value_type;
    using std::log;
    using std::pow;
    if (!std::is_sorted(v.begin(), v.end())) {
        throw std::domain_error("The input vector must be sorted in non-decreasing order v[0] <= v[1] <= ... <= v[n-1].");
    }

    auto normal = boost::math::normal_distribution(mu, sd);

    auto left_integrand = [&normal](Real x)->Real {
        Real Fx = boost::math::cdf(normal, x);
        Real dmu = boost::math::pdf(normal, x);
        return Fx*dmu/(1-Fx);
    };
    auto es = boost::math::quadrature::exp_sinh<Real>();
    Real left_tail = es.integrate(left_integrand, -std::numeric_limits<Real>::infinity(), v[0]);

    auto right_integrand = [&normal](Real x)->Real {
        Real Fx = boost::math::cdf(normal, x);
        Real dmu = boost::math::pdf(normal, x);
        return (1-Fx)*dmu/Fx;
    };
    Real right_tail = es.integrate(right_integrand, v[v.size()-1], std::numeric_limits<Real>::infinity());


    auto integrator = boost::math::quadrature::gauss<Real, 30>();
    Real integrals = 0;
    int64_t N = v.size();
    for (int64_t i = 0; i < N - 1; ++i) {
        auto integrand = [&](Real x)->Real {
            Real Fx = boost::math::cdf(normal, x);
            Real dmu = boost::math::pdf(normal, x);
            Real y0 = (i+1)/Real(N);
            Real y1 = (i+2)/Real(N);
            Real Fn = y0 + (y1-y0)*(x-v[i])/(v[i+1]-v[i]);
            return (Fn - Fx)*(Fn-Fx)*dmu/(Fx*(1-Fx));
        };
        auto term = integrator.integrate(integrand, v[i], v[i+1]);
        integrals += term;
    }


    return v.size()*(left_tail + right_tail + integrals);
}

}}
#endif
$\endgroup$
  • $\begingroup$ I find this graphic impossible to interpret due to the overplotting of the orange and blue points and the lack of any description of what the axes even mean. $\endgroup$ – whuber Sep 26 '19 at 16:59
  • $\begingroup$ The x-axis is the logarithm of the number of samples. The y-axis is $\ln(A^2/n)$. If linear interpolation was distinctly better than step interpolation, then you'd expect that the blue points would in general lie below the orange points. (Sadly, the overplotting of the orange and blue points is difficult to avoid with SVG.) $\endgroup$ – user14717 Sep 26 '19 at 17:06
  • $\begingroup$ @whuber: Should be much better now, with code to reproduce the graphic posted. Originally, I missed the measure $dF(x)$. $\endgroup$ – user14717 Sep 26 '19 at 20:08
  • $\begingroup$ If, as you write, you are extrapolating out into the tails, then all bets are off: the results will depend on the extrapolation and that's not relevant to this thread. However, I still cannot see any systematic difference between the orange and blue dots in the graphic, even at its full original scale. $\endgroup$ – whuber Sep 26 '19 at 22:13
  • 1
    $\begingroup$ @whuber: No problem; my communication skills always need work. (BTW I've been reading your stuff on this site; all excellent.) $\endgroup$ – user14717 Sep 27 '19 at 18:37

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.