1/125
Looks like no tags are added yet.
Statistics
The development and application of methods to collect, analyse, and interpret data. Helps us make informed decisions despite uncertainty and variation.
Population
The collection of ALL objects of interest (the big picture, all possible).
Sample
A selection from the population (the pieces actually measured).
Parameter
A descriptive measure of the POPULATION, e.g. mean μ, standard deviation σ, variance σ².
Statistic
A descriptive measure of the SAMPLE, e.g. mean x-bar, standard deviation s, variance s². (Think: parameter, but for the sample.)
Random variable (r.v.)
A characteristic that changes from object to object in the population (what we go out and measure).
Discrete r.v.
Set of possible values is finite, or infinite but countable (can be listed in a sequence). Finite number of decimal places. Example: number of classes.
Continuous r.v.
Possible values are infinite and impossible to list (infinitely many decimals possible). Judge by what could theoretically happen, not by what got measured. Example: height, temperature.
Three main topics of Stat 255
Descriptive stats (Sets 1-2), Probability (Sets 3-16 and 18), Inferential stats (Sets 17-27).
Descriptive statistics
Ways to describe a data set: charts/graphs, mean, median, standard deviation, variance.
Probability (course topic)
How likely events are to occur. Look at the population to see how likely events will be in the sample.
Inferential statistics
Use the sample to make generalizations about the population.
Notation: population mean vs sample mean
Population mean = μ (parameter). Sample mean = x-bar (statistic).
Example: poll Stat 255 students for height to learn about adults in Canada. Identify population, sample, parameter, statistic, r.v.
Population: adults who live in Canada. Sample: Stat 255 students who answered the poll. Parameter: average height of adults in Canada (μ). Statistic: average height of the students who answered (x-bar). R.v.: height.
Example: 200 random UVic students, record number of classes. Identify population and r.v.
Population: UVic students registered this semester. R.v.: number of classes registered in. Parameter μ = average for all current UVic students; statistic x-bar = average for the 200 measured.
Discrete or continuous: number of accidents at an intersection in a year
Discrete
Discrete or continuous: price of a cup of coffee in dollars
Discrete (finite number of decimal places)
Discrete or continuous: top running speed of a greyhound in km/h
Continuous
Discrete or continuous: lowest temperature in the month
Continuous
Discrete or continuous: number of courses a student is registered in
Discrete
Numerical data
Real numbers that measure something (length, height, weight, temperature, midterm scores).
Categorical data
Categories (eye colour, year of school, quality of customer service).
Ordinal data
Categorical data whose categories CAN be ordered (bad/fair/good/excellent; letter grades F to A).
Nominal data
Categorical data whose categories are NOT ordered (eye colour, major).
Is age categorical or numerical?
Usually numerical.
Why is the "average eye colour" coded 1=blue, 2=brown, 3=green meaningless?
The numbers are just placeholders for categories, not measured values. Always check whether recorded numbers are real measurements. Changing the codes would change the average.
How is categorical data usually analysed?
With tables (counts and relative frequencies) and charts (bar chart, pie chart), not means.
Relative frequency
Count in a category divided by the total (proportion of the data in that category). Example: 5 of 15 = 0.3333 = 33.33%.
Course rounding rule
Give answers with at least 3 decimals of accuracy and do not round until the very last step.
Histogram
Bar-chart-like graph for NUMERICAL data. You choose the range of values in each bar; bar height = frequency. Total number of students = sum of bar heights.
Sample mean (x-bar)
(x1 + x2 + … + xn) / n = (1/n) × sum of xi. Same thing as the average.
Sample median (x-tilde)
The middle observation AFTER sorting. If n is even, average the two middle values. 50% of data above, 50% below.
Mode
The most common value. May be several modes if tied.
Mean, median, mode for 0, 0, 2, 3, 6, 7, 10, 11, 20
Mean = 59/9 = 6.556, median = 6, mode = 0.
Mean, median, mode for 1, 1, 2, 8, 15, 15, 25, 100
Mean = 167/8 = 20.875, median = (8+15)/2 = 11.5, modes = 1 and 15.
What happens to mean and median if the outlier 100 is removed from 1, 1, 2, 8, 15, 15, 25, 100?
Mean drops a lot (20.875 to 67/7 = 9.571). Median changes very little (11.5 to 8). The mean is sensitive to outliers; the median is not.
Does x-bar or x-tilde have to be an observed value?
No. Neither has to appear in the sample, so do not round final answers to make them "possible".
Which measures of centre work for categorical data?
Only the mode. Mean and median cannot be used for categorical data.
Mean on a histogram
The "balance point" of the picture.
Symmetric vs skewed histogram
Symmetric: mean sits at the middle. Skewed: a long tail on one side, mean is pulled toward the tail.
Measures of location
Mean, median, mode (tell us where the "center" of the data is).
Measures of variability
Sample variance and sample standard deviation (tell us how spread out the data is).
Sample variance formula
s² = sum of (xi - x-bar)² / (n - 1).
Sample standard deviation
s = square root of the sample variance.
Shortcut formula for sample variance
s² = 1/(n-1) × [ sum of xi² - (1/n)(sum of xi)² ].
Why divide by n - 1 in sample variance?
Course formula for a SAMPLE uses n - 1 (not n). Use the calculator stat mode: s is the "sample standard deviation" key, then square it for s².
Sample 1: 10,20,49,50,51,80,90 vs Sample 2: 10,48,49,50,51,52,90. Both have mean = median = 50. Which is more spread out and by how much?
Sample 1. s1 = 28.873 (s² = 833.667) vs s2 = 23.130 (s² = 535). Bigger variance / standard deviation = more spread out.
Units of mean, variance, and standard deviation (observations in km)
Mean: km. Variance: km². Standard deviation: km (same units as the original data).
Sample space (S)
A list of all possible outcomes of a random experiment. Outcomes must be exhaustive and mutually exclusive.
Exhaustive (sample space)
ALL possible outcomes are in the list.
Mutually exclusive (sample space outcomes)
No overlap: no two outcomes in the list can happen at the same time.
Simple event
A single outcome from S.
Event
A set of one or more simple events. Example: born in first 3 months = {Jan, Feb, Mar}.
Three ways to find a probability
Subjective (personal opinion), relative frequency, classical method. (Subjective is not used in this course.)
Relative frequency probability
P(A) = (number of times A occurred) / (number of times the experiment was run). Estimated from many observations.
Classical probability
P(A) = n(A) / n(S) = (number of ways A can occur) / (number of outcomes in S). Use when outcomes are equally likely.
Probability of an event in terms of simple events
The sum of the probabilities of its simple events.
Tree diagram
Diagram that lists all possible outcomes in the sample space, one branch per choice at each stage.
Sample space for flipping a fair coin 3 times
8 outcomes: HHH, HHT, HTH, HTT, THH, THT, TTH, TTT. Each has probability 1/8.
Pea plants: round R is dominant over wrinkled r. Cross Rr × Rr. P(round)?
3/4 = 0.75 (outcomes RR, Rr, rR are round, rr is wrinkled).
Rabbits: EeFf × Eeff. P(firm ears AND long fur)?
3/8. Draw ONE tree with 8 equally likely branches; 3 give firm ears (E_) and long fur (ff).
Heterozygous
Carries two different alleles for a trait (e.g. Rr).
Allele assumption in this course
Each allele a parent carries has an equal (50%) chance of being passed on, so you can count tree branches.
Why can't you multiply 3/4 × 1/2 for the rabbit ears and fur traits?
Multiplying assumes the traits are independent, and we are not allowed to assume that. Use ONE tree with all characteristics on it. Multiplying separate trees loses marks.
Axioms of probability
1) P(S) = 1. 2) P(A) ≥ 0 for any event A. 3) For mutually exclusive events, P(A1 or A2 or …) = P(A1) + P(A2) + …
Range of a probability
Always between 0 and 1. No negative probabilities.
Empty (impossible) event
Symbol ∅, with P(∅) = 0.
Complement of A
A' ("not A") = all outcomes in S not in A. P(A') = 1 - P(A), usually used as P(A) = 1 - P(A').
Union (A ∪ B)
"A or B": outcomes in A, in B, or in both (at least one occurs).
Intersection (A ∩ B)
"A and B": outcomes in both A and B (the overlap).
Mutually exclusive (disjoint) events
A ∩ B = ∅, so P(A and B) = 0 (no overlap, cannot happen together).
General addition rule (two events)
P(A or B) = P(A) + P(B) - P(A and B). Same rule works with counts n( ).
General addition rule (three events)
P(A or B or C) = P(A)+P(B)+P(C) - P(A and B) - P(A and C) - P(B and C) + P(A and B and C).
How to fill in a 3-circle Venn diagram
Start in the middle (triple overlap) and work outward using subtraction.
Addition rule when A and B are mutually exclusive
P(A or B) = P(A) + P(B) since P(A and B) = 0.
Rabbit colony (160): n(G)=104, n(E)=105, n(F)=126, n(F∩G)=90, n(E∩F)=80, n(G∪E)=149, none=5. Find P(G∩E∩F).
n(G∩E) = 104+105-149 = 60. Then 155 = 104+105+126-60-90-80+n(triple), so n(triple) = 50. P = 50/160 = 0.3125.
Same rabbit colony: P(grey only)?
n(grey only) = 4, so P = 4/160 = 0.025.
Michelle's work habits: E1 has prob p, E2 = 8p, E3 = 3p, E4 = 2p, E5 = 6p, mutually exclusive and exhaustive. P(not E1)?
20p = 1 so p = 1/20. P(E1') = 1 - 1/20 = 19/20.
Conditional probability
P(A|B) = P(A and B) / P(B): probability of A given that B is known to have occurred.
Mouse table: P(D∩G)=0.03, P(D'∩G)=0.12, P(D∩G')=0.17, P(D'∩G')=0.68. Find P(D|G) and P(D|G').
P(D|G) = 0.03/0.15 = 0.20. P(D|G') = 0.17/0.85 = 0.20. Same, so having the marker does not change the chance of disease.
Relative risk
P(D|E) / P(D|E'), where D = disease and E = has the risk factor. It is a ratio of probabilities, NOT a probability (can be below, equal to, or above 1).
Interpreting relative risk
RR < 1: disease less likely with the risk factor. RR = 1: no difference. RR > 1: disease more likely with the risk factor.
Aspirin data: aspirin 139 heart attacks / 10,898 none; placebo 239 / 10,795. Relative risk of heart attack with aspirin?
P(D|aspirin) = 139/11037, P(D|placebo) = 239/11034. RR = 0.584. Heart attack is less likely with aspirin; the chance is greater when taking the placebo.
Independent events (definition)
P(A|B) = P(A) (or P(B|A) = P(B)). Equivalent: P(A and B) = P(A) × P(B).
Mutually exclusive events (definition)
Cannot occur at the same time: P(A and B) = 0.
Can two events with positive probability be both mutually exclusive and independent?
No. If mutually exclusive, P(A and B) = 0, which is not P(A)P(B) when both are positive. Mutually exclusive events are NOT independent (knowing one occurred tells you the other did not).
Rule for several independent events
P(E1 and E2 and … and En) = P(E1) × P(E2) × … × P(En).
Does P(E1 and … and En) = product of probabilities prove independence for n ≥ 3?
No. The product rule can hold without the events being independent. You would have to check all pairs, triples, etc.
Mouse example: P(D)=0.20, P(G)=0.15, P(D∩G)=0.03. Independent?
Yes, 0.20 × 0.15 = 0.03. D and G' are also independent, since independence carries to complements.
Are D and D' independent?
No. P(D and D') = 0, but P(D)P(D') = 0.20 × 0.80 = 0.16.
Dog allergies (7% of population), 3 random strangers. P(exactly one allergic)?
3 × (0.07)(0.93)(0.93) = 0.1816 (three branches, each with one allergic person).
Same: P(none allergic) and P(at least one)?
None: 0.93³ = 0.8044. At least one: 1 - 0.93³ = 0.1956.
P("At least one") trick
P(at least one) = 1 - P(none). Much faster than adding all the branches.
Multiplication rule
P(A and B) = P(A) × P(B|A) = P(B) × P(A|B). If independent this reduces to P(A) × P(B).
Law of total probability on a tree
P(H) = sum over branches of P(branch) × P(H | branch). Add the products of the tree paths that end in H.
Balding men: 28% balding; heart attack 18% if balding, 11% if not. P(heart attack)?
0.28(0.18) + 0.72(0.11) = 0.0504 + 0.0792 = 0.1296.
Bayes' theorem (tree-diagram form)
P(A|B) = P(A and B) / P(B), where the numerator is one tree path and the denominator is the sum of all tree paths that end in B.
TV plants: A 50% (1% defective), B 30% (2%), C 20% (2%). P(plant C | defective)?
P(D) = 0.5(0.01) + 0.3(0.02) + 0.2(0.02) = 0.015. P(C|D) = 0.004/0.015 = 0.267.
False positive rate
P(+ | D')
False negative rate
P(- | D)