12.3 Random Sampling, Bias, Simulations, Expected Value & Population Predictions

12.3 Random Sampling, Bias, Simulations, Expected Value & Population Predictions

Statistical literacy in middle-school mathematics extends beyond calculating basic summary statistics on small, isolated datasets. It requires understanding statistical inference—the inductive process of drawing defensible conclusions about a large, unobserved population based on the empirical observations of a representative sample. Educator candidates must possess both the mathematical rigor to design simulations and compute expected values, and the pedagogical insight to help students identify sampling bias, evaluate media claims, and utilize proportional reasoning to formulate population predictions.

---

The Architecture of Statistical Inference: Parameters vs. Statistics

The central purpose of inferential statistics is bridging the divide between two distinct mathematical entities:

Because conducting a complete census of an entire population is typically prohibitively expensive, time-consuming, or physically impossible, researchers collect a sample to calculate statistics (like $\bar{x}$ and $\hat{p}$) that serve as point estimates for the true population parameters ($\mu$ and $p$).

---

Representative Probability Sampling Methodologies

For sample statistics to serve as valid estimates of population parameters, the sample must be representative—it must mirror the demographic and characteristic distributions of the parent population. Non-zero, known probability selection mechanisms are required to eliminate conscious or unconscious human selection bias.

1. Simple Random Sampling (SRS)

A Simple Random Sample (SRS) of size $n$ is chosen in such a way that every individual in the population has an equal chance of being selected, and every conceivable combination of $n$ individuals has an equal probability of constituting the sample.

• Implementation: Numbering every individual in the sampling frame from $1$ to $N$, and selecting $n$ numbers using a pseudorandom number generator (PRNG) or a table of random digits without replacement.
• Strength: Completely unbiased in expectation ($E(\bar{x}) = \mu$).
• Limitation: Can be logistically difficult or geographically expensive for dispersed populations.

2. Stratified Random Sampling

In Stratified Random Sampling, the population is partitioned into non-overlapping, homogeneous subgroups called strata based on an important shared characteristic (e.g., grade level, socioeconomic status, geographic district). An independent SRS is then drawn from within each stratum, typically proportional to each stratum's share of the total population.

• Implementation: If a school has 600 students (300 in Grade 6, 200 in Grade 7, 100 in Grade 8), a proportional stratified sample of size 60 draws 30 sixth-graders, 20 seventh-graders, and 10 eighth-graders via independent random lotteries within each grade.
• Strength: Guarantees representation of key minority subpopulations and reduces sampling variability ($s$) across strata.

3. Systematic Sampling

In Systematic Sampling, elements are selected at regular, fixed numerical intervals throughout an ordered list of the population. The sampling interval $k$ is calculated as $k \approx \frac{N}{n}$. A random integer $r$ between $1$ and $k$ is selected as the starting point, and every $k$-th individual thereafter is included: $r, r + k, r + 2k, r + 3k, \dots$

• Strength: Simple, rapid execution in field settings (e.g., quality control inspection on an assembly line).
• Hazard (Periodicity Bias): If the ordered list contains a hidden cyclical pattern that matches the interval $k$ (e.g., inspecting every 8th item when a manufacturing machine has 8 rotating heads), severe bias is introduced.

4. Cluster Sampling

In Cluster Sampling, the population is naturally organized into diverse, heterogeneous mini-groups called clusters (e.g., homeroom classrooms, city blocks, schools within a region). A random sample of entire clusters is selected, and a complete census is conducted on every individual within the chosen clusters.

• Comparison to Stratified: A vital middle-school distinction: Stratified sampling takes some individuals from all groups (strata are internally homogeneous), whereas Cluster sampling takes all individuals from some groups (clusters should ideally be internally heterogeneous, mini-mirrors of the population).

---

Non-Probability Sampling Designs & Systemic Biases

When samples are collected without objective randomization, or when survey procedures are flawed, systemic bias occurs. Bias is the systematic tendency of an estimator to overestimate or underestimate a population parameter.

Flawed Sampling Procedures

1. Convenience Sampling: Selecting individuals who are easiest or most convenient to reach (e.g., surveying students sitting near the teacher's desk, or interviewing shoppers at a mall entrance). Convenience samples consistently fail to represent the broader population.
2. Voluntary Response Sampling: Allowing individuals to self-select into the sample (e.g., call-in radio polls, social media surveys, voluntary online feedback forms). This methodology severely over-represents individuals with strong, polarized, or negative opinions, completely distorting true population sentiment.

Non-Sampling Errors and Biases

Even when an SRS is selected, procedural errors can corrupt inferential validity:

• Undercoverage: Occurs when certain segments of the population are systematically excluded from the sampling frame (the list from which the sample is drawn). Example: Using a residential landline telephone directory excludes mobile-only households, under-representing younger adults.
• Non-Response Bias: Occurs when individuals selected for the sample fail or refuse to respond, and the non-respondents differ systematically in their views or characteristics from those who do participate.
• Response Bias & Wording Bias: Occurs when the behavior of the interviewer, lack of anonymity, or loaded, leading question phrasing influences participant answers. For example, asking "Do you support wasting taxpayer money on redundant programs?" produces radically different results than asking "Do you support funding municipal administrative reviews?"

Critical Candidate Takeaway: Increasing the sample size ($n$) does NOT eliminate bias. A sample of 1,000,000 voluntary online respondents is just as biased as a sample of 100 voluntary respondents; it simply produces an extraordinarily precise estimate of a totally unrepresentative subpopulation!

---

Designing and Conducting Probability Simulations

In middle-school classrooms, theoretical probabilities for complex processes can be difficult to calculate analytically. A simulation is an experimental model that replicates the probabilistic features of a real-world scenario using physical manipulatives or digital technology.

Step-by-Step Simulation Framework

1. Identify the Real-World Probability Model: Clearly state the theoretical probabilities of the component outcomes (e.g., a basketball player makes free throws with $70\%$ probability; each shot has $P(\text{Hit}) = 0.70$ and $P(\text{Miss}) = 0.30$).
2. Select an Appropriate Simulation Tool:
   • Fair coins: For binary events with $P = 0.50$ (e.g., guessing on a true/false question).
   • Standard six-sided dice: For events involving sixths ($P = 1/6, 2/6, 3/6$).
   • Spinners: Continuous central angles divided proportionally to model fractional probabilities.
   • Random Number Generators / Random Digit Tables: Digits 0 through 9 provide maximum flexibility for decimal and percentage probabilities.
3. Establish an Explicit Digit Assignment:
   • Digits must accurately reflect the theoretical probabilities.
   • For the $70\%$ free-throw shooter: Assign digits $0, 1, 2, 3, 4, 5, 6$ to represent a "Made Shot" (7 out of 10 digits = $70\%$), and digits $7, 8, 9$ to represent a "Missed Shot" (3 out of 10 digits = $30\%$).
4. Define a Single Trial: State explicitly how many digits are read or what stopping condition ends the trial (e.g., "Read 5 consecutive random digits to simulate a sequence of 5 free-throw attempts").
5. Execute Repeated Trials ($N \ge 30-100$): Run multiple independent trials and record the observed result of each trial (e.g., recording whether the player made at least 4 out of 5 shots).
6. Calculate Empirical Relative Frequency: Divide the total number of successful trials by the total number of simulated trials.

---

Expected Value of Discrete Random Variables

The expected value ($E(X)$ or $\mu_X$) of a discrete random variable is the theoretical long-run weighted average outcome of an experiment repeated independently over many trials. It represents the central balance point of the probability distribution.

Formal Mathematical Formula

If a discrete random variable $X$ takes on distinct numerical values $x_1, x_2, x_3, \dots, x_k$ with corresponding probabilities $P(X = x_1), P(X = x_2), \dots, P(X = x_k)$:

$$E(X) = \mu_X = \sum_{i=1}^{k} x_i \times P(X = x_i) = x_1 P(x_1) + x_2 P(x_2) + \dots + x_k P(x_k)$$

where $\sum_{i=1}^k P(x_i) = 1$.

Concept of a "Fair Game"

In financial mathematics, games of chance, and insurance modeling, decisions depend on net expected return:

$$\text{Net Payoff} = \text{Gross Winnings} - \text{Cost to Play}$$ $$E(\text{Net}) = E(\text{Gross}) - \text{Cost}$$

The Fair Game Criterion: A game of chance is mathematically fair if and only if the expected net value equals zero:

$$E(\text{Net Return}) = 0$$

• If $E(\text{Net}) < 0$: The game is mathematically unfavorable to the player (the standard model for state lotteries, raffles, and casino games).
• If $E(\text{Net}) > 0$: The game is favorable to the player.

Important Conceptual Distinction: The expected value does not have to equal any possible individual outcome of the experiment. For example, the expected value of rolling a fair six-sided die is:

$$E(X) = 1\left(\frac{1}{6}\right) + 2\left(\frac{1}{6}\right) + 3\left(\frac{1}{6}\right) + 4\left(\frac{1}{6}\right) + 5\left(\frac{1}{6}\right) + 6\left(\frac{1}{6}\right) = \frac{21}{6} = 3.5$$

No single roll can ever produce 3.5; it represents the mathematical mean outcome over thousands of rolls.

---

Making Population Inferences and Proportional Predictions

A primary application of probability and statistics in the Grade 4–8 curriculum is using sample proportions to make quantitative predictions about large populations.

Proportional Estimation Model

If a representative random sample of size $n$ contains $x$ individuals exhibiting a specific characteristic, the sample proportion is:

$$\hat{p} = \frac{x}{n}$$

Assuming the sample is representative, this sample proportion serves as the point estimate for the population proportion ($p \approx \hat{p}$). To predict the total number of individuals in a population of size $N$ that possess this characteristic, set up a direct proportion:

$$\frac{x}{n} = \frac{\text{Predicted Total}}{N} \implies \text{Predicted Total} = N \times \hat{p} = N \times \left(\frac{x}{n}\right)$$

The Capture-Recapture Method (Lincoln-Petersen Index)

In environmental science and biological field research, ecologists cannot count every animal in an open habitat. Instead, they use the capture-recapture method, which directly applies proportional reasoning:

1. Capture an initial sample of $M$ animals, mark them with harmless tags, and release them back into the population.
2. Allow sufficient time for the marked animals to mix randomly and uniformly throughout the total population $N$.
3. Capture a second independent sample of $C$ total animals, and count how many marked animals are recaptured ($R$).
4. Equate the proportion of marked animals in the second sample to the proportion of marked animals in the total population: $$\frac{R}{C} \approx \frac{M}{N} \implies N \approx \frac{M \times C}{R}$$

Sample Size, Variability & Margin of Error

Students must understand that point estimates are subject to sampling variability—different random samples from the same population yield slightly different statistics. The precision of an estimate depends on the sample size $n$:

• The standard error of a sample proportion is inversely proportional to the square root of $n$ ($SE \propto \frac{1}{\sqrt{n}}$).
• To cut the margin of error in half, the sample size must be quadrupled ($4n$).
• Counter-intuitively, the size of the population $N$ has virtually no impact on the margin of error, provided the population is substantially larger than the sample ($N \ge 10n$ or $20n$). A random sample of 1,000 voters yields the same margin of error whether drawn from the city of Austin or from the entire United States.

---

Sampling Methods Matrix

---

Step-by-Step Worked Mathematical Examples

Worked Example 1: Expected Value of a School Raffle Ticket

Problem: A middle school student council sells 500 raffle tickets at $5.00 each to raise money for a library renovation. The raffle prizes are structured as follows:

• 1 Grand Prize: $500 cash gift card
• 2 Second Prizes: $100 cash gift card each
• 5 Third Prizes: $20 gift certificate each
• Remaining tickets: No prize ($0)

1. Part A: What is the expected gross payout of a single raffle ticket?
2. Part B: What is the expected net value of purchasing one ticket, and is this raffle a fair game?

Step-by-Step Solution:

• Part A: Calculate Expected Gross Payout: Construct the discrete probability distribution for gross winnings $X$:

  • $P(X = 500) = \frac{1}{500}$
  • $P(X = 100) = \frac{2}{500}$
  • $P(X = 20) = \frac{5}{500}$
  • $P(X = 0) = \frac{500 - (1 + 2 + 5)}{500} = \frac{492}{500}$

  Apply the expected value formula:

  $$E(X) = \sum x_i P(x_i) = 500\left(\frac{1}{500}\right) + 100\left(\frac{2}{500}\right) + 20\left(\frac{5}{500}\right) + 0\left(\frac{492}{500}\right)$$ $$E(X) = \frac{500}{500} + \frac{200}{500} + \frac{100}{500} + 0 = \frac{800}{500} = \text{\textdollar}1.60$$

  The expected gross payout is $1.60 per ticket.

• Part B: Calculate Expected Net Value and Evaluate Fairness: The net return equals the expected gross payout minus the ticket purchase cost ($5.00):

  $$E(\text{Net}) = E(X) - \text{Cost} = 1.60 - 5.00 = -\text{\textdollar}3.40$$

  Because $E(\text{Net}) = -\text{\textdollar}3.40 \ne 0$, the raffle is mathematically unfair to the purchaser. On average, a ticket buyer loses $3.40 per ticket purchased, which is exactly how the student council raises funds (retaining an expected profit of $3.40 per ticket sold).

---

Worked Example 2: Wildlife Population Estimation via Capture-Recapture

Problem: A Texas state park biologist tags 150 largemouth bass in a reservoir and releases them. A month later, the biologist nets 120 bass and finds that 18 of them have tags. What is the estimated total bass population in the reservoir?

Step-by-Step Solution:

• Step 1: Identify the given quantities:
  • Initial tagged fish $M = 150$
  • Second capture sample size $C = 120$
  • Recaptured tagged fish $R = 18$
  • Unknown total population $N$
• Step 2: Set up the proportional estimation equation: $$\frac{R}{C} = \frac{M}{N} \implies \frac{18}{120} = \frac{150}{N}$$
• Step 3: Solve for $N$ algebraically:
  • Simplify the sample proportion: $\frac{18}{120} = \frac{3}{20} = 0.15$
  • Substitute and solve: $$0.15 N = 150 \implies N = \frac{150}{0.15} = 1,000$$
  • Equivalently, cross-multiply: $18N = 150 \times 120 = 18,000 \implies N = \frac{18,000}{18} = 1,000$
• Conclusion: The estimated population in the reservoir is 1,000 bass.

---

Diagnostic Misconceptions & Pedagogical Strategies

1. The "Sample Must Be a Large Percentage of the Population" Fallacy: Middle-school students intuitively assume that a sample of 500 people from a city of 1,000,000 cannot possibly be accurate because it represents "only 0.05% of the people." Pedagogical strategy: Use the soup ladle analogy: "When a chef tastes a pot of soup to check seasoning, does the chef need to eat half the pot? No, a single well-stirred spoonful is enough, whether the soup is in a 2-quart saucepan or a 50-gallon vat." The key is random mixing (stirring), not the fraction of the pot consumed.
2. Conflating Stratified Sampling with Cluster Sampling: Students routinely mix up these two probability designs. Pedagogical strategy: Anchor the phrase: "Stratified takes some from all; Cluster takes all from some." Draw diagrams showing 4 distinct classrooms: Stratified picks 5 random students from each of the 4 rooms; Cluster rolls a 4-sided die, picks Classroom 2, and tests every student in that room.
3. Expecting the Expected Value to be an Achievable Single Outcome: Students frequently assume that an expected value must be an integer or a value listed on the prize board. Pedagogical strategy: Emphasize the word average. If a family has an average of 2.4 children, no single family has 0.4 of a child; it is an arithmetic mean across the entire population.