Sampling Distributions, Standard Error, and Z-Scores

Foundations of Inferential Statistics and Sample Location

• Goal of Inferential Statistics:

  • Determines how a sample group performs overall in comparison to a known baseline or population parameter.
  • Evaluates whether an intervention or experimental treatment (e.g., antidepressant medication vs. a sugar pill/placebo, or a newly piloted academic curriculum vs. existing standard curricula) produces a true shift in performance or score location.
  • Focuses on sample location relative to the original population: assessing whether sample statistics fall well above, well below, or near the population mean.

• Individual Scores vs. Aggregate Sample Means:

  • Probability of individual scores focuses on single raw scores within a distribution (e.g., calculating the probability of an individual scoring higher than $115$ or lower than $85$ on a standard IQ test).
  • In psychological and scientific research, studies rarely evaluate isolated individual scores; typical sample sizes comprise large groups (e.g., $n = 300$ or more participants).
  • A single participant's performance is insufficient to evaluate treatment efficacy because individual extreme scores or outliers do not indicate a systematic group shift.
  • Researchers evaluate the sample mean ($M$) as the representative central value of a group to infer population parameters and evaluate treatment effects across the entire sample.

Sampling Error and Sampling Distributions

• Sampling Error:

  • Definition: The natural, statistical disparity between a sample statistic (such as the sample mean, $M$) and the corresponding population parameter (the population mean, $\mu$).
  • Cause: Occurs inherently due to random selection during sampling. Repeated random draws from the exact same population inevitably yield slightly different sample statistics.
  • Concrete Example: Sampling U Albany college students at the Campus Center to record overall GPA will yield sample means that are generally close to the actual student body population mean, but two separate random samples drawn sequentially will rarely be identical.

• Sampling Distribution of the Mean:

  • Definition: A probability distribution created by drawing an infinite or large number of random samples of a specific, fixed size ($n$) from a population, calculating the sample statistic (sample mean, $M$) for each sample, and plotting all collected sample means on a frequency distribution curve.
  • Critical Requirement: The sample size ($n$) must remain strictly constant across every drawn sample added to the distribution.
  • Central Clustering: Because sample means drawn at random are most likely to be close to the true population parameter, sample means cluster around the actual population mean ($\mu$).

Discrete Population Numerical Demonstration

• Population Parameters:

  • Consider a finite target population containing exactly $N = 4$ individuals.
  • Variable measured: Age in years.
  • Raw population scores: $24$, $26$, $28$, $30$
  • Individual selection probability: $P = 0.25$ for each person in a single random draw.
  • Population Mean ($\mu$):         $\mu = \frac{24 + 26 + 28 + 30}{4} = 27$
  • Population Standard Deviation ($\sigma$):         $\sigma \approx 2.236$

• Sampling Procedure ($n = 2$ with replacement):

  • Drawing samples of size $n = 2$ with replacement yields $4 \times 4 = 16$ total possible sample combinations.
  • Full enumeration of potential sample pairs and their corresponding sample means ($M$):
    • Sample $(24, 24) \rightarrow M = 24$
    • Sample $(24, 26)$ or $(26, 24) \rightarrow M = 25$
    • Sample $(24, 28)$, $(28, 24)$, or $(26, 26) \rightarrow M = 26$
    • Sample $(24, 30)$, $(30, 24)$, $(26, 28)$, or $(28, 26) \rightarrow M = 27$
    • Sample $(26, 30)$, $(30, 26)$, or $(28, 28) \rightarrow M = 28$
    • Sample $(28, 30)$ or $(30, 28) \rightarrow M = 29$
    • Sample $(30, 30) \rightarrow M = 30$

• Sampling Distribution Statistics ($n = 2$):

  • Frequency of occurrence peaks at $M = 27$ ($4$ out of $16$ outcomes, probability $0.25$).
  • Mean of the Sampling Distribution ($\mu_M$):         $\mu_M = 27$         The mean of all sample means is identical to the original population mean ($\mu = 27$).
  • Standard Deviation of the Sampling Distribution (Standard Error, $\sigma_M$):         $\sigma_M \approx 1.581$         Notice that $\sigma_M = 1.581$ is smaller than the population standard deviation ($\sigma = 2.236$).

Standard Error and Properties of Sampling Distributions

• Standard Error ($\sigma_M$):

  • Definition: The standard deviation of a sampling distribution of sample means. It quantifies the average distance or deviation that sample means fall from the true population mean.
  • Purpose: Provides a direct standard measurement of the magnitude of sampling error present in the data.
  • Calculation Formula:$\sigma_M = \frac{\sigma}{\sqrt{n}}$
    • $\sigma$ = Population standard deviation
    • $n$ = Sample size of each sample
  • Verification with Discrete Demonstration Data:$\sigma = 2.236$$n = 2$$\sigma_M = \frac{2.236}{\sqrt{2}} = \frac{2.236}{1.414} \approx 1.581$

• Influence of Sample Size ($n$):

  • As sample size ($n$) increases, standard error ($\sigma_M$) decreases.
  • Larger sample sizes reduce sampling error, producing sample means that cluster more tightly around the true population mean.
  • Methodological Limitations: While maximised sample sizes are theoretically ideal, real-world research constraints include financial limits (compensating participants), structural bounds, and finite participant pools (e.g., student subject pools in Psychology 101 courses).

• The Central Limit Theorem:

  • Condition 1: If the original underlying population distribution is normally distributed, the resulting sampling distribution of sample means will ALWAYS be normally distributed, regardless of sample size ($n$).
  • Condition 2: If the original population distribution is non-normal (e.g., uniform, skewed, or flat distributions), the sampling distribution of sample means will increasingly approximate a normal bell-shaped curve as sample size ($n$) increases.
  • Threshold Rule: A sample size of $n \ge 30$ is generally recognized as sufficient to invoke the Central Limit Theorem and guarantee a normally distributed sampling distribution.

Interactive Visualisations and Simulations

• IQ Simulation Demonstration ($\mu = 100$, $\sigma = 50$):

  • Drawing iterative samples of size $n = 10$ generates sample means clustering near $100$ (e.g., sample means recorded at $99.03$ and $93.22$).
  • Increasing sample size to $n = 30$ produces a tight normal bell-shaped distribution centered directly at $\mu = 100$
  • Empirical Standard Error computed via formula for $n = 30$ with standard deviation of $\sigma = 15$:         $\sigma_M = \frac{15}{\sqrt{30}} \approx 2.74$         (Observed simulation standard error converged dynamically to $2.54$ through $2.74$ as sample iterations accumulated).

• Uniform Parent Distribution Simulation:

  • Parent population set to a flat uniform distribution with population mean $\mu = 3.5$
  • Drawing samples of size $n = 30$ systematically creates a bell-shaped normal sampling distribution with an empirical mean of $M = 3.52$, verifying the Central Limit Theorem.

Standardizing Sample Means: Z-Scores for Sampling Distributions

• Formula for the Z-score of a Sample Mean:$Z = \frac{M - \mu}{\sigma_M} = \frac{M - \mu}{\frac{\sigma}{\sqrt{n}}}$

  • $M$ = Observed sample mean
  • $\mu$ = Hypothesized or known population mean
  • $\sigma_M$ = Standard error of the mean

• Demonstration Calculation:

  • Given parameters: Observed sample mean $M = 6$, sample size $n = 36$, population mean $\mu = 8$, population standard deviation $\sigma = 6$
  • Step 1: Compute Standard Error ($\sigma_M$):$\sigma_M = \frac{\sigma}{\sqrt{n}} = \frac{6}{\sqrt{36}} = \frac{6}{6} = 1$
  • Step 2: Compute Z-score ($Z$):$Z = \frac{M - \mu}{\sigma_M} = \frac{6 - 8}{1} = -2$
  • Step 3: Determine Probability / Proportion in Tail:         Consulting the standard normal Z-table under Column C (tail area) for $Z = -2$ yields:         $P(Z < -2) = 0.02275$         Expressed as a percentage, there is a $2.275\%$ probability of obtaining a sample mean equal to or more extreme than $6$ by random chance.

Interactive Class Exercises & Questions

• Group 1 Problem Walkthrough:

  • Given Parameters: Population mean $\mu = 15$, population standard deviation $\sigma = 6$, sample size $n = 36$, target sample mean $M = 17$
  • Standard Error Calculation:$\sigma_M = \frac{6}{\sqrt{36}} = \frac{6}{6} = 1$
  • Z-score Calculation:$Z = \frac{17 - 15}{1} = 2$
  • Tail Probability Determination:         Consulting Column C for $Z = 2$ gives a tail proportion of $0.02275$ ($2.275\%$ or rounded to $0.0228$ / $2.28\%$).

• Group 2 Problem Walkthrough:

  • Given Parameters: Population mean $\mu = 20$, population standard deviation $\sigma = 10$, sample size $n = 25$, target sample mean $M = 17$
  • Standard Error Calculation:$\sigma_M = \frac{10}{\sqrt{25}} = \frac{10}{5} = 2$
  • Z-score Calculation:$Z = \frac{17 - 20}{2} = \frac{-3}{2} = -1.5$
  • Tail Probability Determination:         Consulting Column C (tail area below negative Z) for $Z = -1.5$ gives a proportion of $0.0668$ ($6.68\%$ or $6.681\%$).

• Questions & Discussion on Table Lookup Mechanics:

  • Question: How do you know which table column to consult when evaluating probabilities?
  • Answer:
    • Negative Z-Scores: Use the negative Z-score table. Column C gives the area in the lower tail (probability of obtaining a score below that negative Z-score).
    • Positive Z-Scores: Use the positive Z-score table. Column C gives the area in the upper tail (probability of obtaining a score above that positive Z-score).
    • Column B: Provides the area falling between the mean ($Z = 0$) and the calculated Z-score.
    • Third Table (Range): Used when finding probabilities bounded between two distinct Z-scores.

Introduction to Hypothesis Testing and Statistical Significance

• Alpha Level and Significance Thresholds:

  • In statistical and psychological research, the universal standard cutoff point for statistical significance is $5\%$ ($\alpha = 0.05$).
  • If the probability of obtaining a given sample statistic purely by random chance is $5\%$ or lower ($P \le 0.05$), the result is declared statistically significant.

• Two-Tailed vs. One-Tailed Critical Boundaries:

  • Two-Tailed (Non-Directional) Test: Evaluates overall change without predicting direction (higher or lower). The $5\%$ alpha region is split evenly between both tails ($2.5\%$ or $0.025$ per tail).
  • Critical Cutoff Z-score: $Z = \pm 1.96$. Any Z-score exceeding $1.96$ or falling below $-1.96$ is statistically significant.
  • One-Tailed (Directional) Test: Used when there is an explicit directional prediction (e.g., predicting exclusively that a curriculum increases scores). The entire $5\%$ ($0.05$) probability region is concentrated into a single tail.

• Comprehensive Experimental Evaluation Example (IQ Curriculum Study):

  • Known Population Parameters: $\mu = 100$, $\sigma = 15$
  • Sample Size: $n = 35$ (sufficient for Central Limit Theorem)
  • Observed Sample Mean: $M = 127$
  • Step 1: Compute Standard Error ($\sigma_M$):$\sigma_M = \frac{15}{\sqrt{35}} \approx \frac{15}{5.916} \approx 2.535$
  • Step 2: Compute Z-score ($Z$):$Z = \frac{127 - 100}{2.535} = \frac{27}{2.535} \approx 10.65$
  • Conclusion: The observed Z-score of $Z = 10.65$ dramatically exceeds the critical threshold of $1.96$. The probability of obtaining this sample mean by random chance is far below $0.05$. Therefore, the sample mean significantly differs from the population mean, proving that the novel curriculum produced a statistically significant increase in IQ scores.

Course Administration

• Exit Ticket:
  • Date: Active for the 16th.
  • Access Password: distribution
• Upcoming Examinations:
  • First Exam Date: Scheduled for next Friday.