Introduction

Terminology:
● Population: Entire data set
● Sample: Subset of population
● Sample Space: all possible outcomes of data set
● Discrete: Fixed number of options
● Continuous: Infinite number of options
● Random/Stochastic variable: assigned number to identify outcome
● Distributions
○ Symmetric
○ Uniform
○ Bimodal
○ Skewed
○ J-Shaped
● Stochastic process
○ Random process
Describing data set:
● Central tendency
○ Mean
○ Median
○ Mode
● Dispersion
○ Standard deviation (root mean square)

○
Normal distribution:
● Also called gaussian distribution or bell curve
● Relates mean to standard deviation

●
Probability axioms:
● Probability: Likelihood that event will happen
● Axiom 1: probability is between 0 and 1
● Axiom 2: P=1 means event must happen
● Axiom 3: sum of all probabilities equals 1
Probability rules:
● Mutually exclusive
○ Events cannot occur at same time
○ RULE: 𝑃(𝐴∪𝐵) = 𝑃(𝐴) + 𝑃(𝐵)
○ eg. Flipping coin and getting heads or tails
● Mutually inclusive
○ Two event may or may not occur together
○ RULE: 𝑃(𝐴∪𝐵) = 𝑃(𝐴) + 𝑃(𝐵) − 𝑃(𝐴∩𝐵)
■ So we don’t double count overlap
● Independent events
○ Outcome of A doesn’t influence B
○ RULE: 𝑃(𝐴∩𝐵) = 𝑃(𝐴) × 𝑃(𝐵)
○ eg. Flipping two coins separately
● Dependent events:
○ Also called conditional probabilities
○ Probability of A given B happening
○ Denoted: 𝑃(𝐴|𝐵)
○ RULE: 𝑃(𝐴∩𝐵) = 𝑃(𝐵) 𝑃(𝐴|𝐵)
Probability distribution:
● Probability mass functions (PMF)
○ For discrete random variables
○ Mean: µ = Σ𝑥𝑖 𝑃(𝑥𝑖)
2 2
○ Variance: σ = Σ(𝑥𝑖 − µ) 𝑃(𝑥𝑖)
● Probability density function (PDF)
○ Since infinite number of outcomes, probability of given outcome is 0
■ ∴intervals must be used
○ Use integral instead of sums
○ 𝑃(𝑎 ≤ 𝑥 ≤ 𝑏) = ∫𝑓(𝑥) 𝑑𝑥
○ Mean: µ = ∫𝑥 𝑓(𝑥) 𝑑𝑥
2
○ Variance: σ = ∫(𝑥 − µ) 𝑓(𝑥) 𝑑𝑥
● Cumulative distribution function (CDF)
○ Use probability function to determine probability of event in certain range
■ Adjust bonds of integration

Discrete probability distributions

● Types
 ○ Binomial distribution
○ Poisson distribution
Choose notation:
● Permutations
○ Denoted: nPk
○ Order matters
○ Number of unique ordered sets
● Combinations
○ Denoted: nCk
○ Order independent
○ Number of unique non-ordered sets
Binomial distribution:
● Used for only 2 possible outcomes
○ Eg. pass or fail
○ Probability of certain outcome is process is repeated x amount of times

● where p is probability of desired event, k is #of desired events, n is

total number of events
Poisson distribution:
● Probability of given events in fixed interval of time or space
● Used for
○ Constant mean rate
○ Events independent of time since last event
○ Time or spatially discrete problems
● Assumptions:
○ Probability of event is independent of time interval (given same length of time)
○ Probability of event is independent of other events
𝑘 −λ
λ𝑒
● 𝑃(𝑘) = 𝑘!
where k is number of time desired event occurs, λ is probability of event in
given time interval of interest

Continuous probability distributions

Types:
● Normal/gaussian distribution
● Standard log-normal distribution
● Exponential distribution
Normal/gaussian distribution:
● Symmetric and negative
● Models random error

●
 ○ But hard to solve so use change of variables
(𝑥−µ)
○ 𝑧= σ
■ How many standard deviations away from mean
○ Set standard integral:

■
■ Set z1=0 to make it a single variable function
■ Use tables
● Using matlab to find probability
○ p=normcdf(z) where z is change of variable OR
■ Form -∞ to point (NOT 0)
○ p=normpdf(x, μ, σ)
■ Form -∞ to point (NOT 0)
Standard lognormal distribution:
● Strictly positive and occasionally very large
○ eg. Lifetime of equipment
● Logarithm that is normally distributed
○ Take ln of variable apply standard normal distribution
Exponential distribution
● Likelihood of event increase or decreases exponentially with time
−λ𝑥
● PDF function: 𝑓(𝑥, λ) = λ𝑒 where λ is rate parameter
1
● Mean (μ) = standard deviation (σ) = λ
−λ𝑥1 −λ𝑥2
● CDF: 𝑃(𝑥1 ≤ 𝑥 ≤ 𝑥2) = 𝑒 −𝑒

Test for normality:

● Test how normal an experimental distribution is
● 2 measures
○ Skewness: skewness of data
■ Denoted S
■ 𝑆 = 0 is perfectly symmetric
■ |𝑆| > 1 ⇒ very skewed
○ Kurtosis: sharpness of data (also indication of tails/extremes)
■ Denoted K
■ 𝐾 = 3 is perfect normal distribution
■ note: Some programs subtract 3 automatically
■ |𝐾 − 3| > 3 ⇒ very sharp/blunt
 ● Standardized moment chart to describe data relative to normal distribution

Population and samples

Distribution parameter estimation:
● Impractical to measure all population
○ ∴usually measure only sample
● How do we determine if our sample accurately represents the mean
● Central limit theorem
○ Distribution of sample means is normally distributed
■ For n>30 (#of data points per sample)
■ Regardless of population distribution
○ ↗ #of a data points (n) per sample →
○ ↘ standard deviation of, mean of samples
○ Properties
■ If population is normally distributed → samples mean ( ) is normally
distributed
■ If #of data points per sample (n)>30 → samples mean ( ) is normally
distributed

■ If n>30 →
Interval estimation of mean:
● Determining error in our sample mean
●
○ δ is confidence interval
○ Standard in 95% confidence interval (ASME standard)
● Confidence level (C)
○ Probability that population mean (μ) lies within confidence interval
○ 𝐶 = 1 − α where α is level of significance
○ C is % chance event will happen
○ α is % chance event will not happen
 ● Assume standard deviation of sample is equal to standard deviation of population
○ 𝑆= σ
𝑧α/2𝑆
● δ= where α is significance
𝑛
○ To find zα/2 reverse the z process using tables
○ Using matlab
■ z = norminv(p)
■ Linkes probability to z value
○ z1 is at norminv(α/2) AND z2 is at norminv(C+α/2)
● One-sided intervals
○ Only interested in upper of lower limit

■ Upper:

■ Lower:
○ DON’T divide α by 2 since all area (probability) is on one side
Student’s t-distribution:
● Use when n<30
● Same procedure as normal distribution BUT
○ Use t instead of z
○ Matlab:
α
■ tinv(p, nu) where 𝑝𝑢𝑝𝑝𝑒𝑟 = 𝐶 + 2
■ ν (nu) is degree of freedom
● As ν→∞, distribution approaches normal distribution
● As ν→∞, distribution flattens and widens
Estimation of population variance:
2
● Use chi-squared (χ ) distribution
● Use matlab
○ chi2inv(p, nu)
2
● χ is only positive therefore bounds are:
α α
○ 𝑝1 = 2
𝑝2 = 𝐶 + 2

Correlation
Linear correlation:
● Linear correlation coefficient (rx, y)
○ 𝑟 = 1 → strong positive correlation
○ 𝑟 =− 1 → strong negative correlation
○ 𝑟 =± 0. 1 → no correlation
● Only provides data on correlation
○ NO slope
○ NO non-linear correlation
 ● Matlab:
○ corr(x, y) where x and y are arrays of values
● Significance of linear correlation coefficient
○ ↗ data points → ↗ significance
○ Table gives minimum correlation coefficient needed to accept correlation
○ Depends on
■ #of data points sampled
■ Significance level wanted (α) (%that correlation is due to pure chance)

Correlation and causation:

● Correlation NOT causation