7.3 Simple, Compound, and Theoretical vs. Experimental Probability
Key Takeaways
Probability ranges from 0 to 1 inclusive. The favorable-outcomes ratio requires a nonempty finite sample space with equally likely outcomes.
For at least one success, subtract the probability of zero successes from 1.
Multiply fixed branch counts to count ordered outcomes. Check how they are selected before assigning equal probabilities.
Multiply probabilities for independent events. For a general intersection, condition on an event with positive probability. Subtract overlaps when finding a union.
Compare the observed relative frequency with the proposed model. With independent trials and a fixed probability, the chance of being within any fixed margin of that probability approaches 1.
7.3 Simple, Compound, and Theoretical vs. Experimental Probability
Quick Answer: A probability from 0 to 1 describes how likely an event is. Divide favorable outcomes by all outcomes for a finite model with equally likely outcomes. If outcomes have different probabilities, add the probabilities of the favorable outcomes. Use the complement for “at least one” and the conditional multiplication rule for “both.” Experimental probability is an observed relative frequency that can be compared with a model.
Foundations of Probability and the Probability Scale
Probability measures the likelihood of an event, which is a set of outcomes. The set of all possible outcomes of an experiment is called the sample space (S).
The Classical Probability Formula
For a finite, nonempty sample space with equally likely outcomes,
The Standard Probability Scale
Every probability is between 0 and 1 inclusive. The endpoint labels below apply to a finite model whose listed individual outcomes all have positive probability.
0.0 (0%) 0.25 (25%) 0.50 (50%) 0.75 (75%) 1.0 (100%)
|-------------------------|----------------------|----------------------|-------------------------|
Impossible Unlikely Equally Likely Likely Certain
(Cannot occur) as Not Occurring (Must occur)
- P(E) = 0 (0%): An impossible event in a finite model has probability 0 (e.g., rolling an 8 on a standard single six-sided die).
- P(E) = 0.5 (50%): Equally likely as not. Equivalent to an unbiased coin flip landing on heads.
- P(E) = 1 (100%): A certain event in a finite model has probability 1 (e.g., rolling a number strictly less than 7 on a standard six-sided die).
Equivalent Mathematical Forms
Probabilities can be written as fractions, decimals and percentages. For example, . With unequal probabilities, use their weights instead. For a spinner whose red sector covers half the circle and whose blue and green sectors each cover a quarter, a model with equally likely directions assigns red probability . Counting only the three color names and dividing by three would lose that information.
Complementary Events and the "At Least One" Strategy
The complement of event A (denoted A', A^c, or not A) consists of all outcomes in the sample space that are not contained in A. Because an event either happens or does not happen, the sum of an event and its complement is always 1 (100%):
P(A) + P(A') = 1 ===> P(A') = 1 - P(A)
The "At Least One" Problem-Solving Shortcut
Word problems asking for the probability that an event occurs "at least once" across multiple trials can be extremely tedious to calculate directly (because you would need to calculate and add the probabilities of exactly 1 success, exactly 2 successes, exactly 3 successes, etc.).
The complement of "at least one success" is "zero successes" (none). Therefore:
Example: A prize spinner lands on red with probability 0.10 on each spin. Assume the spins are independent. Over three spins, what is the probability of at least one result that is not red?
- The complement is three red results, with probability .
- The probability of at least one nonred result is , or 99.9%.
Sample Spaces and the Fundamental Counting Principle
A list, table or tree can help identify the outcomes before finding the desired probability.
The Fundamental Counting Principle (Multiplication Counting Rule)
If stage 1 has n1 choices, each of those has n2 next choices, and each partial sequence at each stage has the same number of next choices for that stage, multiply the stage counts to find the total number of ordered outcomes:
Example 1: Labels With Repeated Digits
A set of inventory labels uses four positions, each filled by a digit from 0 through 9. Leading zeros and repeated digits are allowed. N = 10 × 10 × 10 × 10 = 10^4 = 10,000 possible labels
Example 2: Labels Without Repeated Digits
Keep the four positions and allow a leading zero. Each digit can now appear only once. N = 10 × 9 × 8 × 7 = 5,040 possible labels
Permutations vs. Combinations Overview
- Permutations: Arrangements where order matters (e.g., job assignments like President/Vice-President, inventory labels, finishing order in a race). For an ordered choice of r different items from n distinct items, with 0 ≤ r ≤ n, P(n, r) = n! / (n - r)!.
- Combinations: Selections where order does not matter (e.g., selecting a committee of 3 members from 10 applicants, choosing a sample of 4 parts from a bin). For an unordered choice of r different items from n distinct items, with 0 ≤ r ≤ n, C(n, r) = n! / (r!(n - r)!).
If a choice leaves different numbers of available next steps, add the counts from those branches. For example, a meal offers either a sandwich with one of three fillings or a salad with one of two dressings. There are options. Even when you can count the options, check how an outcome is selected before assuming the options are equally likely.
Compound Events: Independent vs. Dependent Probability
A compound event involves the occurrence of two or more simple events combined with the logical operators AND (intersection) or OR (union).
1. Independent Events (Multiplication Rule)
Events A and B are independent when their joint probability is the product of the two probabilities. When A has positive probability, this means that the probability of B given A equals its unconditional probability. Independent random draws with replacement and independent tosses of a coin are common models.
Example: An urn contains 6 blue marbles and 4 green marbles (10 total). Choose a marble uniformly at random, record its color and return it. Choose the second marble uniformly at random from the full urn, independently of the first draw. What is the probability of drawing two blue marbles? P(Blue_1 and Blue_2) = (6 / 10) × (6 / 10) = 36 / 100 = 9 / 25 = 0.36
2. The General Multiplication Rule
When the probability of A is positive, the probability of both A and B is the probability of A times the probability of B given A. In a dependent case this conditional probability can differ from the unconditional probability of B. If A has probability zero, the joint probability is zero and conditioning on A with this formula is undefined.
Here P(B|A) denotes the conditional probability of B within the outcomes in A.
Example: Using the same urn (6 blue, 4 green; 10 total), two marbles are drawn without replacement, choosing uniformly from the marbles available on each draw. What is the probability that both are blue?
- First draw: P(Blue_1) = 6 / 10.
- Second draw: With one blue marble removed, 5 blue marbles remain out of 9 total marbles: P(Blue_2 | Blue_1) = 5 / 9.
- Compound probability: P(both blue) = (6 / 10) × (5 / 9) = 30 / 90 = 1 / 3 ≈ 0.333
3. The Addition Rule: Mutually Exclusive vs. Overlapping Events
When finding the probability that event A OR event B occurs:
- Mutually Exclusive (Disjoint) Events: Events that cannot happen at the same time (P(A and B) = 0). Example: Drawing one card uniformly from a standard 52-card deck, a king or queen has probability 4/52 + 4/52 = 8/52 = 2/13.
- Non-Mutually Exclusive (Overlapping) Events: Events that can occur simultaneously. You must subtract the joint probability to avoid double-counting the overlap: Example: Drawing one card uniformly from a standard 52-card deck, a king or heart has probability 4/52 + 13/52 - 1/52 = 16/52 = 4/13.
Models and Observed Frequencies
A probability model assigns probabilities to outcomes. The probabilities are nonnegative and sum to 1. In a uniform model each outcome receives the same share. A model learned from trials can assign different shares, using the observed frequency of each outcome divided by the number of trials.
Suppose a spinner has three equal-angle sectors and the model treats each color as equally likely. Among 60 spins, the model gives an expected count of for each color. If the observed red count is 24, the relative frequency is . A difference from one third is possible by chance. If large, well-controlled sets of spins consistently differ in a similar way, check the spinner, the spin method and how boundary cases were recorded. An empirical model from a separate 100 spins with 52 red, 29 blue and 19 green would assign 0.52, 0.29 and 0.19. Evaluate that model against fresh observations when the conditions remain the same.
The Law of Large Numbers
For independent repeated trials with the same success probability, the chance that the observed proportion is within a fixed distance of that probability approaches 1 as the number of trials grows. For instance, 7 heads in 10 independent fair tosses is possible. This is a long-run result. A particular larger run can still land farther from one half than a smaller run. Changing test conditions can invalidate the assumption of a common probability.
Simulating a Chance Process
To model independent trials with a 20% chance of a match each time, select a digit uniformly from 0 through 9 independently for each trial. Use 0 and 1 for a match and the other digits for no match. In a possible 10-trial run with digits 8, 1, 0, 7, 2, 4, 1, 6, 3 and 9, there are three matches, so the simulated relative frequency is 30%. The preset probability is 20%. Repeating the entire simulation can show how the estimates vary.
Population Predictions and Quality Control Applications
A random sample from a batch can support an estimate for the batch. The observed sample fraction is an estimate of the full batch fraction. The projection can differ from the unknown actual total.
Project a Count From a Sample Rate
Step-by-Step Quality Control Walkthrough
Suppose a random sample of 500 brackets from a batch of 80,000 contains 15 defective brackets.
- Calculate Sample Defect Rate (Experimental Probability): Observed fraction = 15 / 500 = 0.03 = 3.0%
- Project Defective Units Across Full Production Run: Projected count = 80,000 × 0.03 = 2,400 brackets
- Estimate Financial Scrap Impact: If each scrapped bracket incurs a manufacturing loss of $4.25: Total Scrap Loss = 2,400 × $4.25 = $10,200
A maintenance technician has a storage bin containing 15 high-precision pressure sensors: 10 sensors are fully calibrated and 5 sensors are uncalibrated. The technician draws 2 sensors in sequence without replacement, with each remaining sensor equally likely at each draw. What is the probability that both selected sensors are fully calibrated, and what is the probability that at least one of the selected sensors is uncalibrated?
P(both calibrated) = 4/9; P(at least one uncalibrated) = 5/9
P(both calibrated) = 1/3; P(at least one uncalibrated) = 2/3
P(both calibrated) = 3/7; P(at least one uncalibrated) = 4/7
P(both calibrated) = 9/21; P(at least one uncalibrated) = 11/21
A library generates inventory labels consisting of 2 uppercase letters chosen from {A, B, C, D, E, F} followed by 3 numerical digits chosen from {1, 2, 3, 4, 5, 6, 7, 8}. Letters cannot be repeated within any label. How many unique labels can be created if digits may be repeated, and how many unique labels can be created if digits cannot be repeated either?
With digit repetition = 18,432 labels; without digit repetition = 12,096 labels
With digit repetition = 15,360 labels; without digit repetition = 10,080 labels
With digit repetition = 12,288 labels; without digit repetition = 8,064 labels
With digit repetition = 20,480 labels; without digit repetition = 10,080 labels
A quality assurance inspection team tests a random sample of 600 precision electronic micro-switches from an automated production run of 150,000 units. In the sample, exactly 18 micro-switches fail dimensional tolerances. Based on this experimental sample, what is the defect rate (relative frequency), how many defective micro-switches are predicted across the entire production run of 150,000 units, and what is the projected scrap loss if each defective micro-switch costs the manufacturer $3.50?
Defect rate = 0.025 (2.5%); Predicted defects = 3,750; Scrap loss = $13,125
Defect rate = 0.030 (3.0%); Predicted defects = 4,200; Scrap loss = $14,700
Defect rate = 0.030 (3.0%); Predicted defects = 4,500; Scrap loss = $15,750
Defect rate = 0.035 (3.5%); Predicted defects = 5,250; Scrap loss = $18,375
A six-sided die may be weighted. In 120 independent rolls under the same conditions, it lands on 6 a total of 32 times. Which comparison between the observed frequency and a uniform model is justified?
The observed share is about 26.7% versus about 16.7% in a uniform model. The difference calls for checking the model and collecting more data.
The observed share is about 26.7%. This one run proves the true probability of a six is exactly 26.7%.
Both shares are about 16.7% because the die has six numbered faces.
The uniform model requires the number of sixes to be exactly 20 in every run of 120 rolls.
Sections you finish are checked off in the contents.