Notes on Distributions, Means, and Center of Gravity (Transcribed Lecture)

Factorials and quick computation

  • Four factorial divided by three factorial:

    • 4!=4×3×2×14! = 4 \times 3 \times 2 \times 1
    • 3!=3×2×13! = 3 \times 2 \times 1
    • Quotient can be computed by direct calculation or by cancellation:
    • Direct: 4!3!=246=4\frac{4!}{3!} = \frac{24}{6} = 4
    • Cancellation: common factors cancel to leave only the 4 that the 3! didn’t have.
  • This cancellation theme appears often: many expressions share factors that simplify.

  • Two constant parameters in the example context

    • There are two quantities that behave as constants (parameters) across problems:
    • the sample size n (in the context, this tends to be fixed per problem);
    • the probability of success pi (or π).
    • What changes when you vary the problem is the value of x (the number of successes), which affects the factorials and exponents below.
    • Hence, n and pi are kind of fixed parameters, while x changes with the scenario.
  • Exam framing and formulas

    • An exam may provide the formula, but you still need to know how to apply it and where to put the values.
    • Reading and applying the formula matter even if the algebra is supplied.
    • The math can be straightforward, but the formulas can be tricky to apply in practice.
  • Example: probability that four students all pass

    • If you pick four students and the probability a single student passes is π=0.782 ,\pi = 0.782\,, then the probability all four pass is:
    • P(X=4)=(44)π4(1−π)4−4=π4=(0.782)4≈0.374.P(X=4) = {4 \choose 4} \pi^4 (1-\pi)^{4-4} = \pi^4 = (0.782)^4 \approx 0.374.
    • This illustrates using the binomial probability mass function (PMF) in the simple case where all trials succeed.
    • General takeaway: for binomial with parameters (n, π), P(X=x)=(nx)πx(1−π)n−x.P(X=x) = {n \choose x} \pi^x (1-\pi)^{n-x}. In the specific all-success case, this reduces to P(X=n)=πn.P(X=n) = \pi^n.
    • If you instead ask for the probability that at least one fails (not all pass) for n trials, you can use the complement: P(X≠n)=1−P(X=n)=1−πn.P(X \ne n) = 1 - P(X=n) = 1 - \pi^n.
  • Shortcut remarks for binomial probabilities

    • The trick “at least one” is often easiest as the complement of “all pass.”
    • Likewise, for none passing, you would consider the complement of all passing for a different framing.
  • Quick recap on constants and reading the formula

    • You’ll often see formulas that require you to plug in n, π (or p), and x. The efficiency comes from recognizing when you can cancel, use complements, or reduce to a straightforward power.

Poisson distribution and its relation to Binomial

  • Poisson mass function (PMF)

    • The Poisson PMF is:
    • P(X=x)=e−λλxx!,x=0,1,2,…P(X=x) = e^{-\lambda} \frac{\lambda^x}{x!}, \quad x = 0,1,2,\dots
    • Here, λ\lambda is the average rate (mean number of events in a fixed interval).
  • Relationship to Binomial

    • If the size n is large and the probability of success p is small, the binomial distribution with parameters (n, p) can be well approximated by a Poisson distribution with λ=np\lambda = n p.
    • This approximation is common when the number of trials is large and successes are rare per trial.
  • Example: births per day

    • Given total births in a year: 6,013 births per year, with roughly the same rate each day (365 days/year):
    • λ=6013365≈16.474.\lambda = \frac{6013}{365} \approx 16.474.
    • P(X=15) for this Poisson model:
    • P(X=15)=e−λλ1515!≈0.0957.P(X=15) = e^{-\lambda} \frac{\lambda^{15}}{15!} \approx 0.0957.
    • P(X \ge 5) (days with at least five births)
    • Instead of summing from 5 to infinity, use the complement of P(X ≤ 4):
      • P(X≥5)=1−∑x=04e−λλxx!.P(X \ge 5) = 1 - \sum_{x=0}^{4} e^{-\lambda} \frac{\lambda^x}{x!}.
    • In the example, this yields a very large probability (near 1). The instructor computed a value around 0.9997, indicating that on most days there are at least five births.
    • Practical note: Discrete distributions are often visualized as histograms; probabilities for specific x are obtained from the PMF table or calculator.
  • Using technology for Poisson/binomial checks

    • Tools like Minitab can compute Poisson probabilities, including P(X=x) and tail probabilities, and can produce graphs for probability distributions.
    • You can enter the mean (λ) and target x values to retrieve P(X=x) or P(X ≥ x).
    • For binomial and Poisson problems, a calculator or software can speed up computation and help verify work.
  • When to use Poisson in practice

    • Use Poisson when:
    • n is large and p is small, with λ = n p not too large.
    • The problem context involves counting the number of events in a fixed interval of time or space.
    • If λ is too large or the binomial parameters do not meet the approximation criteria, fall back to exact binomial calculation.
  • Numerical example: validating a Poisson calculation

    • With λ ≈ 16.474 and x = 15:
    • P(X=15)=e−16.47416.4741515!≈0.096.P(X=15) = e^{-16.474} \frac{16.474^{15}}{15!} \approx 0.096.
    • For at least five births per day, compute either directly or via complement; the calculator example yielded approximately 0.9997 (very high probability).
  • Note on notation and interpretation

    • λ is the average (mean) rate, sometimes called the mean number of occurrences in the given period.
    • λ is sometimes estimated from data as the total number of events divided by the number of periods observed.

Normal distribution and z-scores (including critical values)

  • Normal distribution basics (conceptual, not all formulas shown here)

    • The normal curve is symmetric and characterized by its mean μ and standard deviation σ.
    • A common task is to find values corresponding to a certain tail probability, e.g., 5% in the right tail.
  • Critical z-value for 95% in the right tail

    • The value where the area to the right is 0.05 is denoted as z_{0.05}:
    • The instructor identifies z_{0.05} ≈ 1.645.
    • This is a standard critical value used in confidence intervals and hypothesis testing.
    • Notation reminder: z_{0.05} denotes the z-score with 5% area to its right (not to its left).
  • Converting to a data value (the actual height example)

    • If a normal distribution has mean μ and standard deviation σ, a value c that is μ + z σ corresponds to a percentile determined by z.
    • The problem example discussed a pediatric height where a height taller than 95% of children corresponds to c = μ + z σ with z = z_{0.05} = 1.645; equivalently, the 95th percentile is at that z-score.
    • To compute a specific height: c = μ + z σ, then interpret the percentile in the context of the distribution.
  • Visualization and labeling

    • When presenting a normal curve, include labels for the mean μ and inflection points (where the curve changes concavity).
    • For a percentile description, indicate the corresponding z-score and the resulting data value c.
  • Quick analogy for probability interpretation

    • If you want the height taller than 95% of patients, you look far to the right on the curve; the area to the right of c is 0.05.
  • Practical note

    • In many classroom contexts, you’ll be given μ and σ and asked to compute percentile thresholds; the z-score table or software can provide the needed z-values (e.g., z_{0.05} = 1.645).

Weibull distribution and area between two values

  • Weibull distribution basics

    • A two-parameter Weibull distribution with shape α > 0 and scale β > 0 has the CDF:
    • F(t)=1−exp⁡(−(tβ)α),t≥0.F(t) = 1 - \exp\left(-\left(\frac{t}{\beta}\right)^{\alpha}\right), \quad t \ge 0.
    • The parameters are α (shape) and β (scale).
  • Proportion below a threshold (CDF form)

    • The probability that X is below t is given by the CDF above: P(X≤t)=F(t)=1−e−(t/β)α.P(X \le t) = F(t) = 1 - e^{-(t/\beta)^{\alpha}}.
  • Proportion between two values

    • To find the probability that X lies between a and b (a < b), you can compute the CDF at b and at a and subtract, or use the exponential form directly:
    • Using CDF difference:
      • P(a<X≤b)=F(b)−F(a)=[1−e−(b/β)α]−[1−e−(a/β)α]=e−(a/β)α−e−(b/β)α.P(a < X \le b) = F(b) - F(a) = \left[1 - e^{-(b/\beta)^{\alpha}}\right] - \left[1 - e^{-(a/\beta)^{\alpha}}\right] = e^{-(a/\beta)^{\alpha}} - e^{-(b/\beta)^{\alpha}}.
    • This is a convenient way to avoid subtracting two non-simplifying numbers if you only need the difference.
  • Practical note

    • Weibull distributions appear in reliability analysis and life data modeling; the two-parameter form is flexible for various hazard rate shapes.

Center of gravity: means, medians, and trimming

  • Notation for a data sample

    • Data consists of n values: x<em>1,x</em>2,…,x<em>n.x<em>1, x</em>2, \dots, x<em>n. The ith data value is often written as xi.
    • The original sample size is n; x_i denote individual observations.
    • The sum is represented with the summation symbol: ∑<em>i=1nx</em>i.\sum<em>{i=1}^{n} x</em>i.
  • Sample mean (center of gravity)

    • Defined as:
    • Xˉ=1n∑<em>i=1nx</em>i.\bar{X} = \frac{1}{n} \sum<em>{i=1}^{n} x</em>i.
    • Also written as the average value and described as the center of gravity of the data.
    • Pros/cons: Very algorithmic and easy to compute; always computed the same way; sensitive to outliers.
  • Sample median (chronological center)

    • The median is the middle value when data are ordered from smallest to largest.
    • If n is odd:
    • The median is the value at position n+12\frac{n+1}{2} in the ordered list: i.e., the middle value.
    • If n is even:
    • The median is the average of the two central values at positions n2\frac{n}{2} and n2+1\frac{n}{2} + 1 in the ordered list:
      • median=x<em>(n/2)+x</em>(n/2+1)2,\text{median} = \frac{x<em>{(n/2)} + x</em>{(n/2+1)}}{2}, where x(k)x_{(k)} denotes the kth order statistic.
    • Important: When even, there are two central values; when odd, there is a single central value.
    • Median versus mean
    • Mean reflects the sum/center of gravity and can be skewed by extreme values.
    • Median reflects the central order position and is more robust to outliers.
  • Trimming (robust to outliers)

    • Trimming removes a fraction of extreme values from both ends before computing the mean.
    • If r is a fraction between 0 and 1 (e.g., 0.1 for 10%), then the largest 100r% and smallest 100r% are removed.
    • The trimmed mean is then computed on the remaining data.
    • Range of trimming: typically not more than 50% trimmed (i.e., r ≤ 0.5).
    • Example with 11 values and 10% trimming:
    • 10% of 11 is 1.1; rounded down, trim 1 value from the top and 1 from the bottom after ordering.
    • Consequences: If the trimmed mean is close to the original mean, the data are not heavily skewed; large differences hint at outliers affecting the mean.
  • Practical guidance on choosing mean vs median

    • Use the mean for symmetric, unimodal data without strong outliers.
    • Use the median for skewed data or when outliers are present.
  • Quick summary of ordering and centering methods

    • Mean is algorithmic: add up numbers and divide by n; always the same procedure.
    • Median is logical: must order data first and then apply the odd/even rule; choice of method depends on parity and whether trimming is used.

Practical exam strategies and notes on computation tools

  • Formula sheets and exam strategy

    • Formula sheets may be provided, but you must know how to apply and interpret results.
    • Being comfortable with reading and applying the formulas is as important as memorizing them.
  • When to use shortcuts and not rely on heavy computation

    • In binomial problems, use complements (e.g., none pass, at least one not pass) to reduce computation when possible.
    • For Poisson, use λ = n p or λ as given; when x lies far from the mean, consider tail probabilities and complementary forms to avoid summing many terms.
  • Graphs and interpretation

    • Discrete distributions (binomial, Poisson) are inherently histogram-like; probabilities are not smooth curves but bars.
    • Normal distributions appear as smooth bell curves; labeling mean and inflection points helps interpret the curve.
  • Using technology for checking work

    • Minitab and similar tools can perform Poisson/binomial calculations and generate distribution plots.
    • You can input λ or mean, and then ask for P(X=x) or tail probabilities and view the results.
  • Quick example recap for exam readiness

    • Factorial/cancellation trick: use cancellation to simplify factorial expressions (e.g., 4!3!=4\frac{4!}{3!} = 4).
    • Binomial problems: use the PMF P(X=x)=(nx)πx(1−π)n−xP(X=x) = {n \choose x} \pi^x (1-\pi)^{n-x}; when x=n, this reduces to P(X=n)=πnP(X=n) = \pi^n.
    • Complementary probabilities: use P(X≥a)=1−P(X<a)=1−∑x=0a−1P(X=x)P(X \ge a) = 1 - P(X < a) = 1 - \sum_{x=0}^{a-1} P(X=x) when convenient.
    • Poisson: you can approximate Binomial when appropriate; compute P(X=x) with P(X=x)=e−λλxx!P(X=x) = e^{-\lambda} \frac{\lambda^x}{x!}; for tail probabilities, use complementary sums or a calculator.
    • Normal percentiles: use z-scores such as z0.05=1.645z_{0.05} = 1.645 to locate 95th percentile on the right; convert to data value with c=μ+zσ.c = \mu + z\sigma.
    • Weibull: remember the CDF F(t)=1−e−(t/β)αF(t) = 1 - e^{-(t/\beta)^\alpha} and area between two values via P(a < X \le b) = e^{-(a/\beta)^\alpha} - e^{-(b/\beta)^\alpha}.$n
  • Additional note on distributions and sections

    • The lecturer references sections 1.1–1.3 (front half) and 1.4–1.6 (back half); formula sheets and practice will be provided for binomial, Poisson, Weibull, and other distributions.
    • The discussion also touched on log-normal and normal distributions and the idea that different distributions require different mathematical treatments and tools.
  • Center of gravity recap (closing)

    • The data’s center can be measured by the mean (sensitive to outliers) or median (robust to outliers).
    • Trimming provides a middle-ground approach by discarding a portion of extreme values before computing a mean.
    • Ultimately, the choice between mean, median, and trimmed mean depends on data shape and outlier presence.

Quick worked-context examples (from the lecture)

  • Example 1: factorial cancellation in a quotient
    • Show the cancellation: \frac{4!}{3!} = \frac{4 \cdot 3 \cdot 2 \cdot 1}{3 \cdot 2 \cdot 1} = 4.
  • Example 2: binomial “all pass” probability
    • With n = 4, π = 0.782: P(X=4) = {4 \choose 4} (0.782)^4 (1-0.782)^{0} = (0.782)^4 \approx 0.374.
    • Complement: P( ext{not all pass}) = 1 - (0.782)^4 \approx 0.626.
  • Example 3: Poisson probability P(X=15)
    • With λ = 16.474: P(X=15) = e^{-16.474} \frac{16.474^{15}}{15!} \approx 0.0957.
  • Example 4: Poisson tail P(X ≥ 5)
    • Use sample mean λ ≈ 16.474: P(X \ge 5) = 1 - \sum_{x=0}^{4} e^{-λ} \frac{λ^x}{x!} \approx 0.9997.
  • Example 5: normal percentile to height
    • Concept: to get the 95th percentile height, use c = μ + z{0.05} σ with z{0.05} ≈ 1.645.
  • Example 6: Weibull two-parameter area between values
    • CDF: F(t) = 1 - e^{-(t/β)^α}.
    • Between a and b: P(a < X ≤ b) = e^{-(a/β)^α} - e^{-(b/β)^α}.$$