Comprehensive Study Guide on Sampling and Sampling Distributions

Learning Objectives & Fundamental Concepts

• Chapter Learning Objectives:

  • LO 1: Select a simple random sample from a finite population.

  • LO 2: Identify when a sample is being drawn from a finite or an infinite population and explain how to construct a corresponding frame.

  • LO 3: Use sample data to calculate the point estimate of the population mean ($\mu$).

  • LO 4: Use sample data to calculate the point estimate of the population standard deviation ($\sigma$).

  • LO 5: Use sample data to calculate the point estimate of the population proportion ($p$).

  • LO 6: Calculate $\sigma_{\bar{x}}$, the standard deviation of the sample mean (standard error of the mean), for a sample of size $n$ taken from a population with a known mean $\mu$ and standard deviation $\sigma$, and apply the finite population correction factor when appropriate.

  • LO 7: Describe and explain the distribution form and parameters of the sampling distribution of the sample mean ($\bar{x}$) for a sample of size $n$ taken from a population with a known mean $\mu$ and standard deviation $\sigma$.

  • LO 8: Use the sampling distribution of the sample mean to calculate the probability that the sample mean will fall between two specified values.

  • LO 9: Calculate $\sigma_{\bar{p}}$, the standard error of the proportion, for a sample of size $n$ taken from a population with a known proportion $p$.

  • LO 10: Describe and explain the distribution form and parameters of the sampling distribution of the sample proportion ($\bar{p}$) for a sample of size $n$ taken from a population with a known proportion $p$.

  • LO 11: Use the sampling distribution of the sample proportion to calculate the probability that the sample proportion will fall between two specified values.

• Fundamental Statistical Definitions:

  • Element: The individual entity on which data are collected.

  • Population: The complete collection of all elements of interest in a study.

  • Sample: A subset of elements selected from the population.

  • Sampled Population: The actual population from which the sample is drawn.

  • Frame: An explicit operational list of all elements from which the sample will be selected.

  • Target Population: The population about which statistical inferences and conclusions are intended to be made.

  • Parameter: A numerical characteristic or summary measure of a population (such as $\mu$, $\sigma$, or $p$).

  • Sample Statistic: A numerical characteristic or summary measure calculated from a sample (such as $\bar{x}$, $s$, or $\bar{p}$).

  • Primary Objective of Sampling: Collecting sample data allows researchers to perform statistical inference and answer specific research questions about a population without incurring the excessive time and financial costs of conducting a complete census.

Introductory Case Studies & The Sampling Problem

• Case 1: Texas US Senate Candidate Preference:

  • Context: Political party leaders in Texas sought to estimate the proportion of registered voters favoring a specific candidate for election to the US Senate.

  • Sampling Data: A sample of $n = 400$ registered voters in Texas was selected, of which $160$ voters expressed a preference for the candidate.

  • Point Estimate: The estimate of the population proportion favoring the candidate is calculated as:$\bar{p} = \frac{160}{400} = 0.40$

  • Population Type: Because the number of registered voters in Texas is countable, this represents sampling from a finite population with a defined frame (the voter registration list).

• Case 2: Tire Mileage Longevity Testing:

  • Context: A tire manufacturer developed a new tire design intended to provide improved mileage over the company's current line. Testing was conducted to estimate the mean useful life.

  • Sampling Data: A sample of $n = 120$ newly designed tires was produced and subjected to testing, yielding a sample mean mileage of:$\bar{x} = 36,500\text{ miles}$

  • Point Estimate: The estimate of the population mean useful life for all new tires was established at $36,500\text{ miles}$.

  • Population Type: The sample population is conceptual, consisting of all potential tires that could be manufactured by the ongoing production process at that specific point in time. Because no exhaustive list of elements can be compiled, it represents an infinite population where constructing a frame is impossible.

• The Reality of Sampling Error:

  • Sample statistics represent estimates rather than exact values of population parameters.

  • In the voter study, exactly $40\%$ ($0.40$) of the entire electorate is not expected to support the candidate; likewise, the true mean mileage of the tire population is not expected to equal exactly $36,500\text{ miles}$.

  • Discrepancies occur naturally because a sample contains only a fraction of the population, leading to unavoidable sampling error.

Electronics Associates, Incorporated (EAI) Problem

• Organizational Context & Parameters:

  • The director of personnel at Electronics Associates, Incorporated (EAI) was tasked with developing a comprehensive employee profile across the organization's workforce.

  • Population Size: $N = 2,500$ employees.

  • Data File: Stored in the electronic file designated EAI.

  • Population Parameters for Annual Salary:

    • Population Mean: $\mu = \$71,800$

    • Population Standard Deviation: $\sigma = \$4,000$

  • Population Parameter for Management Training Completion:

    • A total of $1,500$ of the $2,500$ employees completed the company's management training program.

    • Population Proportion:$p = \frac{1,500}{2,500} = 0.60$

• Managerial Dilemma & Sample Profile:

  • If personnel records are not readily accessible in a centralized system, collecting data on all $2,500$ employees is cost-prohibitive and time-consuming.

  • The director evaluated taking a simple random sample of $n = 30$ employees to estimate the parameters $\mu$, $\sigma$, and $p$.

Selecting a Simple Random Sample from a Finite Population

• Definition of Simple Random Sample (Finite):

  • A simple random sample of size $n$ from a finite population of size $N$ is selected such that each possible sample of size $n$ has the exact same probability of being chosen.

  • Probability sampling guarantees that statistical inferences drawn from the sample are mathematically valid.

• Two-Step Selection Procedure Using Random Numbers:

  • Step 1: Assign a distinct random number between $0$ and $1$ to each of the $N$ elements in the population frame.

  • Step 2: Select the $n$ elements corresponding to the $n$ smallest (or lowest) random numbers generated.

  • Because every subset of size $n$ has an identical chance of receiving the $n$ smallest random values, the selection satisfies the mathematical condition of simple random sampling.

• Implementation in Microsoft Excel:

  • Generate random numbers between $0$ and $1$ using the formula =RAND().

  • The mathematical algorithm of =RAND() ensures that every value in the interval $[0, 1]$ has an equal probability of occurrence.

  • Sorting Procedure:

    • Select a cell within the random number range.

    • Navigate to the Home tab on the ribbon.

    • In the Editing group, click Sort & Filter.

    • Select Sort Smallest to Largest.

    • The records at the top of the sorted table (rows $2$ through $n + 1$) constitute the simple random sample.

• Application Example: 2019 National Baseball League:

  • Population: $N = 15$ teams (Table 7.1).

  • Objective: Select $n = 5$ teams for in-depth interviews concerning minor league franchise management.

  • Data Generation: The formula =RAND() was entered into cells B2:B16.

    • Arizona was assigned $0.850862$.

    • Atlanta was assigned $0.706245$.

    • Washington was assigned $0.066942$.

  • Sorting Results: Washington had the lowest generated number ($0.066942$), followed by Miami, San Francisco, Saint Louis, and New York. These five teams formed the sample.

• Application Example: EAI Workforce Sample:

  • Population: $N = 2,500$ employees.

  • Objective: Select $n = 30$ employees.

  • Procedure: In file EAI, column D was assigned the label Random Numbers in cell D1, and =RAND() was entered into cells D2:D2501.

    • Employee 1 received random number $0.820711$.

    • Employee 2 received random number $0.292880$.

  • Sorting Results: The sorted range placed the $30$ lowest numbers in rows $2$ to $31$. Employee $648$ held the smallest random number and was selected first; Employee $13$ (originally in row $14$) became the $20\text{th}$ observation in the sample.

Selecting a Random Sample from an Infinite Population

• The Nature of Infinite Populations:

  • Arises when populations are infinite or elements are generated by an ongoing process without limits on total output.

  • Because compiling an exhaustive frame is impossible, simple random sampling cannot be executed.

• Conditions for a Random Sample from an Infinite Population:

  • Condition 1: Each element selected must originate from the exact same population.

  • Condition 2: Each element must be selected completely independently of all other elements.

• Quality Control Case Study (Breakfast Cereal Packaging):

  • Process: A production line fills boxes of breakfast cereal targeting a mean weight of $24\text{ ounces}$.

  • Sampling Protocol: A quality control inspector takes periodic samples of $n = 12$ boxes to detect machine underfilling or overfilling.

  • Satisfying Condition 1: All $12$ boxes must be gathered at approximately the same point in time to guarantee that machine settings or production conditions have not drifted.

  • Satisfying Condition 2: The machinery must be engineered such that the filling weight of each individual box occurs independently of adjacent boxes.

• Customer Arrival Case Study (Fast-Food Restaurant):

  • Context: Surveying arriving fast-food restaurant customers to develop a visitor profile.

  • Satisfying Condition 1: The surveyor must only select individuals who actually enter and make a purchase; surveying an individual who enters solely to use the restroom violates the same-population criterion.

  • Satisfying Condition 2 (Avoiding Selection Bias):

    • Interviewer bias occurs if an employee subjectively selects customers within a preferred age demographic.

    • Cluster/dependency bias occurs if an interviewer samples groups of friends or families arriving together, as their characteristics are correlated.

    • McDonald's Implementation: McDonald's established an objective protocol based on promotional discount coupons. Whenever an arriving customer presented a discount coupon, the next customer served was invited to complete the profile survey. Because coupon presentation was independent and random across arriving patrons, independent selection was maintained.

• Ongoing Industrial and Operational Examples:

  • Parts manufactured along an automated assembly line.

  • Repetitive experimental trials conducted inside a scientific laboratory.

  • Financial transactions processed at commercial bank branches.

  • Inbound customer telephone calls arriving at technical support centers.

  • Foot traffic moving through retail store checkout lanes.

Point Estimation Procedures & Population Correspondence

• Sample Statistics and Point Estimators:

  • Point Estimator of $\mu$: The sample mean ($\bar{x}$).

  • Point Estimator of $\sigma$: The sample standard deviation ($s$).

  • Point Estimator of $p$: The sample proportion ($\bar{p}$).

  • The numerical result computed from a specific sample is known as the point estimate.

• Mathematical Formulas for Sample Statistics:

  • Sample Mean:$\bar{x} = \frac{\sum x_i}{n}$

  • Sample Standard Deviation:$s = \sqrt{\frac{\sum (x_i - \bar{x})^2}{n - 1}}$

  • Sample Proportion:$\bar{p} = \frac{x}{n}$(where $x$ is the number of elements possessing the characteristic of interest).

• Computation from EAI Sample ($n = 30$):

  • Salary Sum: $\sum x_i = 2,154,420$

  • Sum of Squared Deviations: $\sum (x_i - \bar{x})^2 = 325,009,260$

  • Number Completed Training: $x = 19$

  • Point Estimate of Mean Salary:$\bar{x} = \frac{2,154,420}{30} = \$71,814$

  • Point Estimate of Standard Deviation:$s = \sqrt{\frac{325,009,260}{29}} = \$3,348$

  • Point Estimate of Completed Training Proportion:$\bar{p} = \frac{19}{30} \approx 0.63$

• Comparison of EAI Population Parameters and Sample Point Estimates:

  • Mean Annual Salary: Population Parameter $\mu = \$71,800$ vs. Point Estimate $\bar{x} = \$71,814$.

  • Standard Deviation of Salary: Population Parameter $\sigma = \$4,000$ vs. Point Estimate $s = \$3,348$.

  • Management Training Proportion: Population Parameter $p = 0.60$ vs. Point Estimate $\bar{p} = 0.63$.

• Practical Alignment of Sampled and Target Populations:

  • In the EAI case, the sampled population and target population are identical (all $2,500$ employees).

  • Amusement Park Scenario: An amusement park seeks to study customer age and duration of park visits across a typical summer. If data are gathered on a day when the park is reserved exclusively for the employees and families of a single corporation, the sampled population diverges sharply from the target population, invalidating point estimates.

Sampling Distribution of the Sample Mean ($\bar{x}$)

• The Sample Mean as a Random Variable:

  • Because different random samples yield distinct data subsets, $\bar{x}$ varies from sample to sample.

  • A second sample of $n = 30$ EAI employees produced $\bar{x} = \$72,670$ and $\bar{p} = 0.70$.

  • In a simulation of $500$ distinct simple random samples of size $n = 30$ from EAI, the relative frequency histogram formed an approximately symmetric, bell-shaped distribution centered near $\mu = \$71,800$.

  • Sampling Distribution: The probability distribution of all possible values of a sample statistic.

• Expected Value of $\bar{x}$:

  • The mean of the sampling distribution of $\bar{x}$ is its expected value, denoted $E(\bar{x})$:$E(\bar{x}) = \mu$

  • Unbiased Estimator: Because $E(\bar{x}) = \mu$, $\bar{x}$ is an unbiased estimator of $\mu$. For EAI, $E(\bar{x}) = \$71,800$.

• Standard Deviation of $\bar{x}$ (Standard Error of the Mean):

  • Notation:

    • $\sigma_{\bar{x}}$ = Standard deviation of the sample mean (standard error of the mean).

    • $\sigma$ = Standard deviation of the population.

    • $n$ = Sample size.

    • $N$ = Population size.

  • Finite Population Formula:$\sigma_{\bar{x}} = \sqrt{\frac{N - n}{N - 1}}\left(\frac{\sigma}{\sqrt{n}}\right)$

  • Infinite Population Formula:$\sigma_{\bar{x}} = \frac{\sigma}{\sqrt{n}}$

  • Finite Population Correction Factor (FPCF):$\sqrt{\frac{N - n}{N - 1}}$

  • Rule of Thumb for FPCF:

    • When $\frac{n}{N} > 0.05$, the finite population formula must be applied.

    • When $\frac{n}{N} \le 0.05$, the FPCF is close to $1.0$ and can be ignored, allowing use of the infinite formula $\sigma_{\bar{x}} = \frac{\sigma}{\sqrt{n}}$.

  • EAI Standard Error Calculation:

    • $\frac{n}{N} = \frac{30}{2,500} = 0.012$

    • Because $0.012 \le 0.05$, the FPCF is omitted:$\sigma_{\bar{x}} = \frac{4,000}{\sqrt{30}} \approx 730.30$

• Form of the Sampling Distribution:

  • Case 1 (Normally Distributed Population): When the underlying population follows a normal distribution, the sampling distribution of $\bar{x}$ is normally distributed for any sample size $n$.

  • Case 2 (Non-Normal Population & The Central Limit Theorem):

    • Central Limit Theorem (CLT): In selecting random samples of size $n$ from any population, the sampling distribution of $\bar{x}$ can be approximated by a normal distribution as the sample size becomes large.

    • Illustrations across non-normal distributions (Figure 7.6):

      • Population I: Continuous uniform distribution.

      • Population II: "Rabbit ear" symmetric distribution with concentrated probability mass in both tails.

      • Population III: Exponential distribution with pronounced right skew.

    • Behavior at Sample Sizes: At $n = 2$, shapes reflect population irregularities; at $n = 5$, uniform and rabbit-ear shapes begin resembling normal curves; at $n = 30$, all three sampling distributions appear approximately normal.

    • Sample Size Guidelines: For most populations, $n \ge 30$ ensures a sound normal approximation. If the population exhibits extreme skewness or heavy outliers, $n \ge 50$ may be required.

Practical Probability Calculations for the Sample Mean

• EAI Sampling Problem ($n = 30$ within $\pm \$500$ of $\mu$):

  • $\mu = \$71,800$, $\sigma_{\bar{x}} = 730.30$, Target Range: $[\$71,300, \$72,300]$.

  • Z-score Computations:

    • Upper endpoint ($72,300$):$z = \frac{72,300 - 71,800}{730.30} = \frac{500}{730.30} \approx +0.68$

    • Lower endpoint ($71,300$):$z = \frac{71,300 - 71,800}{730.30} = \frac{-500}{730.30} \approx -0.68$

  • Standard Normal Table Method:

    • Cumulative probability at $z = +0.68$: $0.7517$

    • Cumulative probability at $z = -0.68$: $0.2483$

    • Probability calculation:$P(71,300 \le \bar{x} \le 72,300) = 0.7517 - 0.2483 = 0.5034$

  • Excel Function Method (NORM.DIST):

    • Formula: =NORM.DIST(x, mean, standard_dev, cumulative)

    • Upper endpoint: =NORM.DIST(72300, 71800, 730.30, TRUE) yields $0.7532$

    • Lower endpoint: =NORM.DIST(71300, 71800, 730.30, TRUE) yields $0.2468$

    • Probability calculation:$P(71,300 \le \bar{x} \le 72,300) = 0.7532 - 0.2468 = 0.5064$

    • The Excel result ($0.5064$) is more accurate due to avoiding the rounding of $z$ to two decimal places.

  • Managerial Interpretation: A sample of $n = 30$ yields a $50.64\%$ probability of obtaining a sample mean within $\$500$ of the population mean, with a $1 - 0.5064 = 0.4936$ ($49.36\%$) probability that the sampling error exceeds $\$500$.

Impact of Sample Size on the Sampling Distribution of the Mean

• Effects of Increasing Sample Size from $n = 30$ to $n = 100$:

  • The expected value remains unaffected: $E(\bar{x}) = \mu = \$71,800$.

  • The standard error decreases inversely with $\sqrt{n}$:$\sigma_{\bar{x}} = \frac{\sigma}{\sqrt{n}} = \frac{4,000}{\sqrt{100}} = \frac{4,000}{10} = 400$

• Probability Evaluation for $n = 100$ within $\pm \$500$ of $\mu$:

  • Target Range: $[\$71,300, \$72,300]$.

  • Excel function =NORM.DIST(72300, 71800, 400, TRUE) yields $0.8944$.

  • Excel function =NORM.DIST(71300, 71800, 400, TRUE) yields $0.1056$.

  • Net probability:$P(71,300 \le \bar{x} \le 72,300) = 0.8944 - 0.1056 = 0.7888$

  • Conclusion: Expanding the sample size from $30$ to $100$ contracts the dispersion of the sampling distribution, raising the probability of keeping sampling error within $\$500$ from $0.5064$ to $0.7888$.

Sampling Distribution of the Sample Proportion ($\bar{p}$)

• The Sample Proportion as a Random Variable:

  • $\bar{p} = \frac{x}{n}$, where $x$ represents the count of elements possessing the specified attribute.

  • In repeated random sampling, $\bar{p}$ varies and has an expected value, standard deviation, and probability distribution.

• Expected Value of $\bar{p}$:

  • The mean of all possible sample proportions equals the population proportion:$E(\bar{p}) = p$

  • Because $E(\bar{p}) = p$, $\bar{p}$ is an unbiased estimator of $p$. For EAI, $E(\bar{p}) = 0.60$.

• Standard Deviation of $\bar{p}$ (Standard Error of the Proportion):

  • Finite Population Formula:$\sigma_{\bar{p}} = \sqrt{\frac{N - n}{N - 1}}\sqrt{\frac{p(1 - p)}{n}}$

  • Infinite / Large Population Formula ($\frac{n}{N} \le 0.05$):$\sigma_{\bar{p}} = \sqrt{\frac{p(1 - p)}{n}}$

  • Calculation for EAI ($n = 30$, $p = 0.60$):$\sigma_{\bar{p}} = \sqrt{\frac{0.60(1 - 0.60)}{30}} = \sqrt{\frac{0.60(0.40)}{30}} = \sqrt{\frac{0.24}{30}} = \sqrt{0.008} \approx 0.0894$

• Form of the Sampling Distribution of $\bar{p}$:

  • Because $x$ is a binomial random variable and $n$ is constant, the exact distribution of $\bar{p}$ is discrete.

  • Normal Approximation Criteria: The binomial distribution and the sampling distribution of $\bar{p}$ can be approximated by a normal distribution whenever:$np \ge 5\quad \text{and}\quad n(1 - p) \ge 5$

  • Verification for EAI ($n = 30$, $p = 0.60$):

    • $np = 30(0.60) = 18 \ge 5$

    • $n(1 - p) = 30(0.40) = 12 \ge 5$

    • Both criteria are satisfied; the normal distribution is appropriate.

Practical Probability Calculations for the Sample Proportion

• EAI Sampling Problem ($n = 30$ within $\pm 0.05$ of $p$):

  • Target Range: $[0.55, 0.65]$, with $p = 0.60$ and $\sigma_{\bar{p}} = 0.0894$.

  • Upper endpoint: =NORM.DIST(0.65, 0.60, 0.0894, TRUE) yields $0.7120$

  • Lower endpoint: =NORM.DIST(0.55, 0.60, 0.0894, TRUE) yields $0.2880$

  • Probability calculation:$P(0.55 \le \bar{p} \le 0.65) = 0.7120 - 0.2880 = 0.4240$

• Increasing Sample Size to $n = 100$:

  • Standard Error:$\sigma_{\bar{p}} = \sqrt{\frac{0.60(0.40)}{100}} = \sqrt{\frac{0.24}{100}} = \sqrt{0.0024} \approx 0.0490$

  • Upper endpoint: =NORM.DIST(0.65, 0.60, 0.0490, TRUE) yields $0.8462$

  • Lower endpoint: =NORM.DIST(0.55, 0.60, 0.0490, TRUE) yields $0.1538$

  • Probability calculation:$P(0.55 \le \bar{p} \le 0.65) = 0.8462 - 0.1538 = 0.6924$

  • Comparative Assessment: Expanding the sample size from $30$ to $100$ lowers the standard error of the proportion from $0.0894$ to $0.0490$, boosting the probability of landing within $0.05$ of $p$ by $0.6924 - 0.4240 = 0.2684$ ($26.84\%$).

Alternative Probability and Non-Probability Sampling Methods

• Probability Sampling Methods (Known Selection Probabilities):

  • Stratified Random Sampling:

    • Population is partitioned into non-overlapping groups called strata ($H$ strata), where every element belongs to exactly one stratum.

    • Strata are formed using variables such as department, geographic location, age, or industry type.

    • Elements within each stratum should be as homogeneous as possible (alike within, low internal variance).

    • A separate simple random sample is drawn from each stratum, and results are combined.

    • Yields higher statistical precision than simple random sampling with a smaller total sample size.

  • Cluster Sampling:

    • Population is divided into separate groups called clusters, with each element belonging to one cluster.

    • Elements within clusters should be heterogeneous (not alike within, serving as small-scale replicas of the total population).

    • A simple random sample of entire clusters is selected, and every element within the chosen clusters is observed.

    • Area Sampling: City blocks or defined geographical sectors serve as clusters.

    • Requires a larger total sample size to achieve comparable precision to simple random sampling, but offers major cost and travel efficiencies.

  • Systematic Sampling:

    • From a population list of $N$ elements, sample size $n$ is obtained by sampling $1$ element for every $k = \frac{N}{n}$ elements.

    • Example: If $N = 5,000$ and $n = 50$, then $k = \frac{5,000}{50} = 100$. One element is randomly chosen from the first $100$ elements, and every $100\text{th}$ element thereafter is selected.

    • Assumes properties of a simple random sample when the population frame is in random order.

• Non-Probability Sampling Methods (Unknown Selection Probabilities):

  • Convenience Sampling:

    • Elements are included based on ease of accessibility without known selection probabilities.

    • Examples: University professors recruiting student volunteers; agricultural inspectors haphazardly selecting oranges from crates; consumer research volunteer panels; wildlife captures.

    • Precludes valid statistical inferences or probability statements about sampling error.

  • Judgment Sampling:

    • Elements are chosen based on the discretion of an expert believed to identify representative items.

    • Example: A political journalist interviewing two or three specific senators deemed representative of the entire Senate.

    • Quality depends entirely on personal judgment, making statistical evaluation impossible.

Sampling Error versus Non-Sampling Error

• Sampling Error:

  • The variation or difference between a sample statistic and its corresponding population parameter resulting purely from random chance in sample selection.

  • Reflected mathematically by standard errors $\sigma_{\bar{x}}$ and $\sigma_{\bar{p}}$.

  • Unavoidable when sampling instead of conducting a census, but decreases toward zero as sample size $n$ grows.

• Non-Sampling Error:

  • Deviations and biases introduced into study findings by factors unrelated to random sampling mechanics.

  • Does not diminish as sample size increases; massive samples remain vulnerable to non-sampling error.

  • Forms of Non-Sampling Error:

    • Coverage Error: Mismatch between the operational frame/sampled population and the target population research objective (e.g., website penningtondailytimes.com sampling current users when the business question concerns non-users or rival readers).

    • Nonresponse Error: Certain segments systematically fail to respond (e.g., a website survey conducted via pop-up windows underrepresents visitors with pop-up blockers and overrepresents those without blockers).

    • Measurement Error: Flawed instruments or data collection errors, such as:

      • Ambiguous questions (e.g., asking if articles are "compelling and accurate" confuses respondents who find them compelling but inaccurate).

      • Leading questions (e.g., stating "Many readers find stories compelling… do you?" induces positive response bias).

      • Respondent recall failure or deliberate untruthfulness.

• Recommended Procedures to Minimize Non-Sampling Error:

  • Carefully define the target population prior to sampling and draw a probability sample directly from it.

  • Design data collection protocols rigorously and train all data collectors.

  • Pretest instruments and procedures to correct ambiguities prior to implementation.

  • Employ stratified sampling when key qualitative population categories are known.

  • Employ cluster sampling when heterogeneous subgroups are physically identifiable.

  • Employ systematic sampling when quantitative ordering exists across the population.

Big Data Architecture and Analytics

• Data Generation Metrics:

  • Approximately $2.5\text{ quintillion bytes}$ ($2.5 \times 10^{18}\text{ bytes}$) of data are generated worldwide each day.

  • Historical Escalation:

    • 1992: $100\text{ gigabytes}$ generated worldwide per day.

    • 1997: $100\text{ gigabytes}$ generated worldwide per hour.

    • 2002: $100\text{ gigabytes}$ generated worldwide per second.

  • Per-Minute Global Activity:

    • $216,000$ Instagram posts.

    • $204,000,000$ emails sent.

    • $12\text{ hours}$ of video uploaded to YouTube.

    • $277,000$ tweets posted on Twitter.

  • Scale Progression: Terminology expanded from petabytes to yottabytes (Table 7.6).

• The Four V's of Big Data:

  • Volume, Velocity, Variety, and Veracity.

• Structural Dimensions of Big Data:

  • Tall Data: Data sets characterized by an extreme number of observations ($n$) relative to variables, causing traditional hypothesis tests and standard errors to become negligible (e.g., consumer goods sentiment analysis across millions or billions of social media comments).

  • Wide Data: Data sets characterized by an extreme number of variables relative to observations (e.g., facial recognition algorithms analyzing millions of pixel attributes per high-resolution image across few image subjects).

Implications of Big Data for Sampling Error (Pennington Daily Times Case)

• Baseline Parameters for penningtondailytimes.com (PDT):

  • Primary revenue derived from digital advertising.

  • Standard deviation of customer visit duration: $\sigma = 20\text{ seconds}$.

  • Historical ad click-through proportion: $p = 0.51$.

• Diminishing Standard Errors Across Sample Sizes:

  • Sample Mean Visit Time Standard Error ($\sigma_{\bar{x}} = \frac{20}{\sqrt{n}}$, Table 7.7):

    • At $n = 10$: $\sigma_{\bar{x}} = 6.32456\text{ seconds}$

    • At $n = 100,000$: $\sigma_{\bar{x}} = 0.06325\text{ seconds}$

    • At $n = 1,000,000,000$: $\sigma_{\bar{x}} = 0.0063\text{ seconds}$

  • Sample Click Proportion Standard Error ($\sigma_{\bar{p}} = \sqrt{\frac{0.51(0.49)}{n}}$, Table 7.8):

    • At $n = 10$: $\sigma_{\bar{p}} = 0.15811$

    • At $n = 100,000$: $\sigma_{\bar{p}} = 0.00158$

    • At $n = 1,000,000,000$: $\sigma_{\bar{p}} = 0.00002$

  • At $n = 1,000,000,000$, standard errors are exactly $\frac{1}{10,000}$ of their magnitude at $n = 10$.

• Large Sample Statistical Anomaly Analysis:

  • Assume prior year population mean visit duration was $\mu = 84\text{ seconds}$.

  • PDT gathers a massive sample of $n = 1,000,000$ website visitors.

  • Standard error:$\sigma_{\bar{x}} = \frac{20}{\sqrt{1,000,000}} = \frac{20}{1,000} = 0.02\text{ seconds}$

  • Probability that $\bar{x}$ falls within $84 \pm 0.15\text{ seconds}$ ($[83.85, 84.15]$):

    • At $83.85$: $z = \frac{83.85 - 84}{0.02} = -7.5$

    • At $84.15$: $z = \frac{84.15 - 84}{0.02} = +7.5$

    • Because $P(z \le -7.5) \approx 0.0000$ and $P(z \le 7.5) \approx 1.0000$, the probability is $1.0000$.

    • Excel formula =NORM.DIST(84.15, 84, 0.02, TRUE) - NORM.DIST(83.85, 84, 0.02, TRUE) equals $1.0000$.

  • Observed Sample Mean Result: The new sample reveals $\bar{x} = 88\text{ seconds}$.

  • Three Plausible Explanations for the Discrepancy:

    1. Sampling Error: Discarded as an explanation because $n = 1,000,000$ produces a near-zero standard error ($0.02$), making random variation of this magnitude virtually impossible.

    2. Non-Sampling Error: Potential data contamination, bot traffic, or measurement flaws that must be audited.

    3. Change in Population Mean: If non-sampling error is ruled out, the population mean itself has shifted from $84\text{ seconds}$ to a higher value, indicating altered visitor engagement.

Dialogue and Peer Exchange

• Student Exchange During Practice Quiz Review:

  • Prompt/Question: "You ready for your practice quiz?"

  • Reaction/Response: "Oh my god. 17, 18, one It's not actually gonna be two, three, four."

  • Prompt/Question: "Do you even know what you're learning in stats right now?"

  • Reaction/Response: "Me neither. I just been cheating it. With some Exercises, applications, practice quiz…"