Theoretical Distributions - ICAI Foundation Statistics

Fundamentals of Theoretical Distributions

  • Theoretical probability distributions exist both in mathematical theory and in real-world phenomena, classifying into discrete or continuous forms.
  • A parameter is a defining numerical characteristic of a population or a probability distribution (e.g., the population mean).
  • A trial is an action or experiment intended to produce an outcome that is neither certain nor impossible.
  • Bernoulli trials consist of independent trials where each individual trial results in only one of two mutually exclusive outcomes (success or failure).
  • The cumulative distribution function of a random variable XX is defined as F(x)=P(X≠x)F(x) = P(X \ne x) or F(x)=P(X≤x)F(x) = P(X \le x).

Binomial Distribution

  • A bi-parametric discrete probability distribution specified by parameters nn (number of trials) and pp (probability of success).
  • The binomial random variable xx takes discrete whole number values in the range 0≤x≤n0 \le x \le n.
  • The probability mass function is given by:   f(x)=nCxpxqn−xf(x) = {^n C_x} p^x q^{n - x}
  • Mean and variance are given by:   Mean=np\text{Mean} = n pVariance=np(1−p)\text{Variance} = n p (1 - p)
  • The mean of a binomial distribution is always strictly greater than its variance.
  • The variance attains its maximum possible value of n4\frac{n}{4} when p=0.5p = 0.5.
  • The distribution is completely symmetrical when p=0.5p = 0.5; in this case, the mean and mode are equal.
  • Depending on the parameter values, the distribution may have either one mode (unimodal) or two modes (bimodal).
  • Fitting a binomial distribution to empirical data is commonly performed using the method of moments.
  • Real-world applications include modeling discrete win/loss outcomes, such as the results of One Day International (ODI) cricket matches.

Poisson Distribution

  • A uni-parametric discrete distribution defined solely by the parameter m=npm = n p.
  • Key underlying assumptions include:
    • The probability of an occurrence in a small time interval is constant.
    • The probability of multiple occurrences in an infinitesimal time interval is negligible.
    • Occurrences in non-overlapping intervals are mutually independent.
  • A fundamental property of the Poisson distribution is that its mean and variance are equal:   Mean=Variance=m\text{Mean} = \text{Variance} = m
  • The distribution is always positively skewed and can be unimodal or bimodal.
  • Acts as a limiting form of the binomial distribution under the conditions that n→∞n \rightarrow \infty and p→0p \rightarrow 0 such that npn p remains finite.
  • Fitting a Poisson distribution to an observed frequency distribution requires equating parameter mm directly to the sample mean of the distribution.
  • Appropriate for modeling rare event frequencies, such as the number of misprints per page in a thick book.

Normal Distribution

  • The principal continuous probability distribution, defined across the domain −∞<x<∞-\infty < x < \infty with probability density function:   f(x)=1σ2πe−12(x−μσ)2f(x) = \frac{1}{\sigma \sqrt{2 \pi}} e^{-\frac{1}{2} \left(\frac{x - \mu}{\sigma}\right)^2}
  • The normal probability curve is bell-shaped and symmetric about the mean μ\mu, with the total area under the curve equaling 11.
  • Because of bilateral symmetry, the area from −∞-\infty to μ\mu is 0.500.50, and the area from μ\mu to ∞\infty is 0.500.50.
  • The mean, median, and mode are always identical.
  • Points of inflexion for a general normal curve lie at μ−σ\mu - \sigma and μ+σ\mu + \sigma; for a standard normal distribution, they lie at −1-1 and 11.
  • The area under the curve covered by (μ−3σ,μ+3σ)(\mu - 3 \sigma, \mu + 3 \sigma) accounts for all but 0.27%0.27\% of the distribution (99.73%99.73\% total coverage).
  • For a standard normal distribution, the mean deviation about the median equals 0.800.80.
  • The quartile deviation (QDQD) is approximately 0.675σ0.675 \sigma (e.g., for σ=4\sigma = 4, QD=2.70QD = 2.70).
  • The notation Φ(a)\Phi(a) designates the cumulative standard normal area evaluated from −∞-\infty to aa.
  • Additive/reproductive property: If XX and YY are two independent normally distributed random variables, their sum (X+Y)(X + Y) is also normally distributed.
  • Well suited for modeling continuous socioeconomic and physical measurements, such as the wages of factory workers.