Inferential Statistics, Sampling Distributions, and Standard Error

Foundations of Descriptive and Inferential Statistics

• Distinction Between Sample and Population Data:

  • Calculations such as $z$-scores in descriptive contexts evaluate data relative to the specific mean of the observed sample.
  • A population is a theoretical group containing every individual or unit that fits the research definition (e.g., every individual living in Guam constitutes the Guam population).
  • Global population parameters (such as the true population mean $\mu$) are usually unknown because gathering data from an entire population is practically impossible.
  • Samples consist of the specific subsets of individuals from whom data is directly collected.

• Descriptive Statistics vs. Inferential Statistics:

  • Descriptive Statistics: Focuses exclusively on summarizing, organizing, and describing the specific sample obtained using metrics such as sample mean ($\bar{x}$), sample standard deviation ($s$), or sample $z$-scores. Descriptive statistics make no broader claims beyond the immediate sample (e.g., reporting whether a sample prefers McDonald's over Wendy's without inferring preferences for the entire population).
  • Inferential Statistics: The process of making formal inferences, generalizations, and conclusions about an unobserved theoretical population based on calculations derived from a smaller representative sample.
  • Because observing an entire population is nearly impossible, inferential statistics collects sample data and applies statistical tools to infer population parameters.

• Role of Estimation:

  • Statistical inference is intrinsically tied to estimation, which is the procedure used to approximate true population parameters.
  • When population parameters are not provided directly (unlike introductory worksheets where population data is explicitly given), researchers must estimate parameters by shifting the unit and layer of analysis from individual values to group/sample values.

Sampling Distribution of Means

• Shift in Analysis Layer:

  • Standard distributions analyze individual data points (e.g., a sample of 20 individuals yields a distribution consisting of 20 individual scores).
  • To estimate theoretical population parameters (such as mean $\mu$ and standard deviation $\sigma$), analysis shifts from a distribution of individual scores to a distribution of sample means.

• Definition and Structure of Sampling Distribution of Means:

  • A sampling distribution of means is a theoretical distribution where each data point represents the mean of a distinct sample drawn from the population (e.g., Value 1 = Mean of Sample 1, Value 2 = Mean of Sample 2, Value 3 = Mean of Sample 3).
  • It represents a distribution of "groups of people" rather than individual values (e.g., one university class represents a single sample mean within the distribution, and another class represents another sample mean).

• Evolution of Research Questions across Layers:

  • Individual Layer Question: "How different is an individual (e.g., Ryan) from the rest of the class?"
  • Group/Sample Layer Question: "How different is an entire class (sample mean) from the broader theoretical population?"

• Theoretical Properties of the Sampling Distribution of Means:

  • The mean of the sampling distribution of means is theoretically equal to the true population mean ($\mu$).
  • If every possible sample of a given size were collected, their respective means computed, and the average of all those sample means calculated, that average would equal the exact population mean.
  • Because collecting every possible sample from a population is physically impossible, this concept serves as a theoretical foundation enabling inference from a single observed sample.

Calculating Sample Standard Deviation and Degrees of Freedom

• Parameters Required to Construct Distributions:

  • Constructing any statistical distribution requires two foundational parameters: the mean and the standard deviation.

• Monthly Ross Visits Case Study:

  • Survey Question: "In a month, how many times do you visit Ross (regardless of branch)?"
  • Sample Size ($n$): $n = 19$ participants.
  • Sample Values Recorded: $7$, $3$, $7$, $4$, $3$, $8$, $2$, $10$, $3$, $3$, $10$, $0$.
  • Calculated Sample Mean ($\bar{x}$): $\bar{x} = 4.3$ visits per month (approximately once per week).
  • Calculated Sample Standard Deviation ($s$ or $SD_{ws}$): $s = 3.6$.

• Formulas for Standard Deviation:

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

• Concept of Degrees of Freedom ($df$):

  • Formula:         $df = n - 1$
  • Degrees of Freedom represents the allowance or flexibility for sample values to vary when estimating unknown population parameters.
  • It reflects the specific number of observations in a sample that are free to vary when computing a statistic given fixed parameters.
  • Higher degrees of freedom indicate a greater number of possible numerical combinations within the sample model.

• Degrees of Freedom as a Statistical Correction:

  • Because complete population data is unknown, dividing by $n - 1$ instead of $n$ serves as a mathematical correction or statistical penalty for incomplete population knowledge.
  • Dividing by a smaller denominator ($n - 1$ instead of $n$) yields a larger standard deviation value.
  • In statistical analysis software like Jamovi, using the sample standard deviation formula ($n - 1$) produces a larger standard deviation, which consequently yields lower calculated $z$-scores compared to standard population calculations.

Concrete Examples and Mathematical Proofs of Degrees of Freedom

• Example 1: Fixed Sample Mean of 4 ($n = 5$):

  • Target Sample Mean: $\bar{x} = 4$
  • Sample Size: $n = 5$
  • Required Total Sum for Mean of 4:         $\text{Total Sum} = 4 \times 5 = 20$
  • First 4 Recorded Scores:
    • Rafi: $1$
    • Amkylia: $2$
    • Adriana: $5$
    • Erin: $0$
  • Sum of First 4 Scores:         $1 + 2 + 5 + 0 = 8$
  • Calculation for 5th Score:         $\text{5th Score} = 20 - 8 = 12$
  • Mathematical Principle: The first 4 participants ($n - 1 = 4$) are free to vary to any numerical value (e.g., $5, 6, 7, 3$). However, the 5th participant's value is completely fixed at $12$ to maintain the fixed mean of $4$.

• Example 2: Fixed Sleep Duration Mean of 7 Hours ($n = 5$):

  • Target Average Sleep Duration: $\bar{x} = 7\, \text{hours}$
  • Sample Size: $n = 5$
  • Required Total Sum of Hours:         $\text{Total Hours} = 7 \times 5 = 35\, \text{hours}$
  • Recorded Responses for First 4 Participants:
    • Christina: $5\, \text{hours}$
    • Faith: $4\, \text{hours}$
    • Participant 3: $6\, \text{hours}$
    • Participant 4: $4\, \text{hours}$
  • Sum of First 4 Participants' Sleep Hours:         $5 + 4 + 6 + 4 = 19\, \text{hours}$
  • Calculation for 5th Participant's Sleep Hours:         $\text{5th Score} = 35 - 19 = 16\, \text{hours}$
  • Mathematical Principle: In a sample size of $n$, exactly $n - 1$ scores possess the freedom to vary. The final $1$ score is constrained by the sample mean. For a sample size of $n = 30$, exactly $29$ scores are free to change while $1$ score remains fixed.

Standard Error of the Mean (SEM)

• Definition of Standard Error:

  • The Standard Error of the Mean ($SE$ or $SEM$) measures the estimated variability or standard deviation of the sampling distribution of sample means.
  • It quantifies expected sampling variability across sample means drawn from a population.

• Inapplicability of Individual Standard Deviation to Group Distributions:

  • Individual standard deviation ($s$) evaluates dispersion among individual values, which is theoretically higher than dispersion between aggregated group means.
  • Aggregating individual observations into groups creates greater homogeneity across group means than across individuals.
  • Because sampling distributions evaluate group sample means rather than individual scores, sample standard deviation ($s$) cannot be directly applied without transformation.

• Formula for Standard Error of the Mean ($SEM$):     $SE = \frac{s}{\sqrt{n}}$     where $s$ is the sample standard deviation and $n$ is the sample size.

• Standard Error Calculation for Ross Visits Data:

  • Sample Standard Deviation ($s$): $3.6$
  • Sample Size ($n$): $19$
  • Standard Error Equation:         $SE = \frac{3.6}{\sqrt{19}}$

• Units of Distance in Sampling Distributions:

  • When constructing a sampling distribution of means, distance from the mean is measured in units of Standard Error ($SE$) rather than standard deviations ($s$).
  • A sample mean with a $z$-score of $3$ is exactly $3$ standard errors away from the population mean ($3\,SE$).