Chi-Square Distribution ($\chi^2$)

The Chi-square distribution is a continuous probability distribution that is used in many hypothesis tests. The Chi-Square statistic always results in a positive value.

A Chi-Square variate (with $v$ degrees of freedom (df)) is the sum of $v$ independent, squared standard normal variates ($\sum\limits_{i=1}^v z_i^2$). It is denoted by $\chi^2_v$. The variance $s^2$ from a sample of normally distributed observations is distributed as $\chi^2$ with $v$ (the df) as a parameter referred to as df of the calculated variance. Symbolically,

$$\frac{v\cdot s^2}{\sigma^2} \sim \chi^2_v$$

Chi Square Distribution Table

The variance $s^2$ for $n$ observations from a $N(\mu, \sigma^2)$, the df is equal to $v=n-1$. The Chi-Square distribution is also used for the contingency (analysis of frequency) tables as an approximation to the distribution of complex statistics. All the families of Chi-Square distribution are specified by their degrees of freedom.

Chi-Square Family of Distributions

Chi-Square Distribution Case of the Gamma Distribution

The Chi-Square distribution is a particular case of the Gamma Distribution, the pdf is

$$P_{\chi^2}(x) = [2^{v/2}\Gamma(v/2)]^{-1} \chi^{(v-2)/2}e^{-x/2}, \quad x\ge 0$$

where $\Gamma(x)$ is the Gamma Distribution.

Normal Approximation to $\chi^2$

Method 1: The PDF and df of Chi-Square can be approximated by the normal distribution. For large $v$ df, the first two moments $z=\frac{(X-v)}{\sqrt{2v}}$, $X\sim \chi^2$.

Method 2: Fisher approximation (compensates the skewness of $X$)

$$\sqrt{2X} – \sqrt{2v-1} \sim N(0, 1)$$

Method 3: Approximation by Wilson and Hilferty is quite accurate. Defining $A=\frac{2}{9v}$, we have

$$\frac{\sqrt[3]{(X/v)}-1+A}{\sqrt{A}}\sim N(0, 1)$$

For the determination of percentage points

$$\chi^2_{v[P]}=v[z_P\sqrt{A}+1-A]^3$$

Generating Pseudo Random Variates

Following the schema allows the generation of random variates from $\chi^2_v$ distribution with $v>2$ df. It requires to generate serially random variates from the standard uniform $U(0,1)$ distribution.

Let $n=v$ degrees of freedom

\begin{align*}
C1 &= 1 + \sqrt{2/e} \approx 1.8577638850\\
C2 &= \sqrt{n/2}\\
C3 &= \frac{3n^2-2}{3n(n-2)}\\
C4 &= \frac{4}{n-2}\\
C5 &= n-2\\
\end{align*}

visit: https://gmstat.com for MCQ tests from various subjects

Leave a Comment

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.

Discover more from Statistics for Data Analyst

Subscribe now to keep reading and get access to the full archive.

Continue reading

Scroll to Top