10.3 Sampling Techniques, Bias, & Making Population Inferences

10.3 Sampling Techniques, Bias, & Making Population Inferences

Statistical inference is the mathematical process of drawing valid conclusions and making predictions about a large population based on observations collected from a representative sample. On the TABE 13&14 Mathematics assessment, inferential statistics questions evaluate your understanding of population parameters versus sample statistics, the identification of sampling bias, proportional scaling algorithms, margin of error principles, and comparative population analysis.

---

Populations vs. Samples: Parameters vs. Statistics

In statistical research, gathering data from every single member of a large population (a census) is often practically impossible due to budget constraints, time limitations, or destructive testing procedures. Instead, researchers collect data from a sample.

Inferential Statistics Workflow:
┌────────────────────────────────────────────────────────┐
│                     POPULATION (N)                     │
│               True Parameters: μ, σ, p                 │
│                                                        │
│    [ Sampling Design ]        [ Statistical Inference ]│
│            │                             ▲             │
│            ▼                             │             │
│                     SAMPLE (n)                         │
│              Sample Statistics: x̄, s, p̂                │
└────────────────────────────────────────────────────────┘

[!TIP] Memory Mnemonic:

• Population corresponds to Parameter (both start with P).
• Sample corresponds to Statistic (both start with S).

---

Representative (Probability) Sampling Methods

To make valid inferences, a sample must be representative—it must accurately reflect the diversity and characteristics of the overall population without systematic favoritism.

1. Simple Random Sampling (SRS)

Every individual in the population and every possible group of size $n$ has an equal probability of being selected. Methods include drawing names from a thoroughly mixed drum, using a table of random digits, or running a computer random number generator ($RNG$).

2. Systematic Sampling

Elements are selected at regular numerical intervals from an ordered list or continuous production stream after a randomly selected starting point between $1$ and $k$: $k = \frac{\text{Population Size } (N)}{\text{Sample Size } (n)}$ Example: Inspecting every $25\text{th}$ automobile engine coming off an assembly line.

3. Stratified Random Sampling

The population is first divided into distinct, non-overlapping homogeneous sub-groups called strata based on a shared demographic or operational characteristic (such as age group, shift, department, or grade level). A simple random sample is then drawn from each stratum in proportion to its size in the population. Advantage: Guarantees that key minority subgroups are proportionally represented in the final sample.

4. Cluster Sampling

The population is naturally divided into diverse, heterogeneous mini-groups called clusters (often based on geographic proximity, city blocks, or school classrooms). Several entire clusters are chosen at random, and every single individual within the selected clusters is surveyed. Difference vs. Stratified: Stratified samples some individuals from all strata; Cluster sampling samples all individuals from some clusters.

Sampling Methods Summary Table

---

Sampling Bias & Sources of Survey Error

Sampling bias occurs when the design of a study systematically favors certain outcomes or systematically excludes segments of the population. Data collected from a biased sample cannot be used to make valid statistical generalizations.

Common Types of Bias on TABE

1. Convenience Sampling (Selection Bias): Choosing individuals who are easiest to reach (e.g., a surveyor standing outside a specialty organic grocery store asking shoppers about average national food spending).
2. Voluntary Response Bias: Allowing individuals to self-select into the sample (e.g., online website polls, call-in radio hotlines, social media comment threads). This systematically over-represents individuals with extreme negative or passionate viewpoints.
3. Undercoverage Bias: Occurs when the sampling frame (the list from which the sample is drawn) inadvertently leaves out entire segments of the population (e.g., conducting a telephone survey using only landlines excludes mobile-only households).
4. Nonresponse Bias: Occurs when a large percentage of selected individuals refuse or fail to respond to the survey, and non-respondents hold systematically different attitudes than respondents.
5. Response Bias & Leading Questions: Poorly worded, loaded, or emotionally charged question phrasing that pushes respondents toward a particular answer (e.g., "Given the dangerous traffic hazards on Main Street, do you agree the speed limit should be reduced?").

---

Making Inferences & Estimating Population Parameters

When a sample is unbiased and representative, the sample proportion ($\hat{p}$) or sample mean ($\bar{x}$) serves as a reliable point estimate for the unknown population parameter.

Proportional Scaling Formulas

$\mathbf{\hat{p} = \frac{x}{n} = \frac{\text{Count with Target Characteristic in Sample}}{\text{Total Sample Size}}}$ $\mathbf{\text{Estimated Population Total} = \hat{p} \times N = \left(\frac{x}{n}\right) \times N}$

Worked Example 1: Capture-Recapture Wildlife Biology

Problem: Wildlife biologists tag and release $250$ largemouth bass into a lake. Two weeks later, they catch a random sample of $400$ bass and find that $20$ of them have tags. What is the estimated total bass population in the lake?

1. Set up the proportion: $\frac{\text{Tagged in Sample}}{\text{Total Sample}} = \frac{\text{Tagged in Population}}{\text{Total Population } (N)}$ $\frac{20}{400} = \frac{250}{N}$
2. Simplify the sample proportion: $\frac{20}{400} = 0.05$ ($5%$ of the population is tagged).
3. Solve for $N$: $0.05 \times N = 250 \implies N = \frac{250}{0.05} = \mathbf{5,000\text{ bass}}$

Worked Example 2: Industrial Defect Estimation

Problem: In a random sample of $600$ medical syringes inspected from a daily manufacturing run of $75,000$ syringes, $9$ syringes are found to have hairline cracks. Estimate the total number of defective syringes in the daily run.

1. Sample proportion defective: $\hat{p} = \frac{9}{600} = 0.015 = 1.5%$.
2. Multiply by population size $N = 75,000$: $\text{Estimated Defects} = 0.015 \times 75,000 = \mathbf{1,125\text{ defective syringes}}$

---

Margin of Error & Sample Size Principles

Because a sample only represents a portion of the population, sample statistics naturally vary from sample to sample—a phenomenon known as sampling variability.

The Margin of Error (MOE)

The margin of error establishes a confidence interval around the point estimate within which the true population parameter is expected to fall:

$\text{Confidence Interval} = \hat{p} \pm \text{MOE}$

Example: A poll reports that $54%$ of voters support a bond measure with a margin of error of $\pm 3%$. The plausible range for true voter support is $54% - 3%$ to $54% + 3% \implies [51%, 57%]$.

The Inverse Square-Root Sample Size Rule

As the sample size ($n$) increases, the standard error shrinks and the margin of error decreases according to an inverse square-root relationship: $\text{MOE} \propto \frac{1}{\sqrt{n}}$

• Quadrupling the sample size ($4n$) cuts the margin of error in half ($\frac{1}{\sqrt{4}} = \frac{1}{2}$).
• Increasing sample size increases precision and narrowness of the estimate, but cannot fix or correct for systematic sampling bias!

---

Comparative Population Analysis: Centers & Spread

TABE Level A questions frequently present comparative dot plots, histograms, or box plots representing two distinct populations (e.g., Plant A vs. Plant B output) and ask you to evaluate differences in center (mean/median) relative to spread (variability / IQR / MAD).

Comparative Dot Plot (Daily Response Times in Minutes):

Team Alpha (Lower Spread, Low Mean):
  •  •  •  •  •  •  •  •  •  •
  +-----+-----+-----+-----+-----+-----
  2     4     6     8     10    12    (Minutes)  --> Mean = 6.0 min, Range = 4 min

Team Beta (Higher Spread, Higher Mean):
  •     •     •     •     •     •     •     •     •     •
  +-----+-----+-----+-----+-----+-----
  2     4     6     8     10    12    (Minutes)  --> Mean = 9.5 min, Range = 10 min

Analytical Rules for Comparing Populations

1. Compare Centers: If Team Alpha's mean response time is $6.0\text{ min}$ and Team Beta's is $9.5\text{ min}$, Team Alpha is significantly faster on average.
2. Compare Variability: Team Alpha's compact cluster indicates highly consistent, predictable performance, whereas Team Beta's wide dispersion indicates high inconsistency.