Comprehensive Study Guide on Variables, Samples, Populations, Correlation, and Variance

Foundations of Statistical Variables and Constants

• Definition of Variable: A variable is anything that varies—defined mathematically and conceptually as any characteristic, attribute, or entity that possesses more than $1$ value.

  • Examples of human attributes that serve as variables include political affiliation, intelligence quotient (I.Q.), degree of introversion, preference for television shows (such as liking the new Doctor Who or disliking the final episode of Game of Thrones), and physical height.
  • The foundational assumption underlying psychological and social measurement is that on these dimensions, people are not alike.

• Definition of Constant: A constant is any characteristic or attribute that does not vary within a given group or analytical context.

  • In a university course composed entirely of students enrolled at the University of Central Florida, the designation "UCF student" functions as a constant because every single participant shares that identical designation.
  • In historical demographic sets, such as the complete group of United States senators serving prior to $1920$, "male" represents a constant because no variation in sex existed within that specific group.

• The Core Research Question of Variability: When observation confirms that people vary, the primary scientific inquiry becomes determining the underlying reason: Why is that?

  • Investigation targets why individuals differ across observable capacities—such as being taller, smarter, faster, jollier, or expressing elevated levels of "internet troll rage."
  • The standard scientific method for examining why individuals differ in one capacity involves measuring another capacity that also varies, hypothesizing that variation in one variable can predict or explain variation in another.

• Predictive Investigation Example (Statistics Examination Grades):

  • In an academic course, instructional elements such as the specific lessons, the instructor, the instructional videos, and the course textbooks remain constants across all students.
  • In contrast, numerous individual factors vary and may predict differences in exam outcomes:
  • Total amount of sleep obtained prior to testing.
  • Degree of personal stress experienced.
  • Baseline level of interest in statistics.
  • Motivation or desire to attend graduate school.
  • Prior mathematical or statistical experience.
  • Total duration of study time allocated.
  • Engagement with entertainment media (e.g., whether engaging programming aired on television the evening preceding the exam).
  • All of these factors represent independent variables that may co-vary with, or serve to predict, the primary variable of interest (examination score).

Classification of Variables and Types of Statistical Analyses

• Discrete Variables:

  • Characterized by separate, indivisible categorical units ($1, 2, 3, 4 \text{ ...}$, containing no intermediate gradations such as $1.000001$ or $1.000002$).
  • Values are restricted to whole numbers and fixed discrete states.
  • Examples include survey items scored along a Likert scale (comprising the specific choices Strongly Agree, Agree, Neutral, Disagree, Strongly Disagree) and political party affiliations.
  • Dichotomous Variables: A specialized subclass of discrete variables characterized by having exactly $2$ available categories (e.g., mutually exclusive options such as Yes or No).

• Continuous Variables:

  • Characterized by an infinite number of potential numerical values that fall between any two observed points.
  • Measurement precision is limited only by the measuring instrument rather than the construct itself.
  • Examples include physical duration (such as the exact time required to run around a city block) and verbal production metrics (such as the total number of words in an essay that exceed six letters in length).

• Broad Classifications of Statistical Analyses:

  • Descriptive Statistics: Analytical methods developed to summarize, organize, and simplify raw quantitative data into interpretable indices.
  • Inferential Statistics: Advanced mathematical techniques that evaluate empirical data gathered from observed samples to form valid probabilistic generalizations and conclusions about the broader populations from which the samples were drawn.

Populations, Samples, and Inferential Decision-Making

• Definitions of Population and Sample:

  • Population: The complete, all-inclusive collection of individuals to whom a researcher wishes to generalize empirical findings.
  • In a hypothetical investigation exploring whether instructional humor improves lecture comprehension, the target population consists of all college students throughout the entire world.
  • Sample: A measured subset of the population actively selected and measured within an empirical investigation.
  • In the instructional humor study, the sample may consist exclusively of students enrolled across two specific statistics classes at the University of Central Florida (one taught by the principal researcher and one taught by a departmental colleague).

• Sampling Error and Methodological Mitigations:

  • Non-representative or biased ("bad") samples threaten inferential validity.
  • To minimize bias, researchers prioritize gathering large, randomly selected samples, though logistical constraints frequently restrict sample acquisition.
  • Because random sampling cannot entirely eliminate sampling variability, inferential procedures calculate the exact mathematical probability that an observed outcome occurred purely as a consequence of random sampling error or chance.

• Statistical Error Parameters:

  • Alpha Level ($\alpha$): The pre-established acceptable probability threshold of committing an inferential error by drawing an incorrect conclusion from sample data. In standard social science research, the alpha level is traditionally set a priori at $\alpha = .05$.
  • $p$-value: The empirically observed alpha level computed directly from sample data, representing the exact probability that an outcome of equal or greater magnitude would emerge assuming the null hypothesis were true.
  • Decision Rule: For a finding to achieve statistical significance, the observed $p$-value must be strictly lower than the established alpha level ($p < \ne .05$). A lower observed $p$-value indicates greater confidence that the observed pattern is not the byproduct of random chance.

Nature and Properties of Bivariate Correlations

• Core Definition: Correlation is a non-experimental statistical analysis employed to determine whether an empirical association or systematic mathematical relationship exists between two existing variables.

  • Bivariate correlation is typically applied to two quantitative, continuous variables, though specialized statistical adaptations exist for discrete and nominal data.

• Primary Characteristics of the Correlation Coefficient:

  • Calculating a correlation coefficient establishes two independent mathematical attributes of a bivariate relationship:
  1. Direction: Whether variables move concurrently in the same direction or invert in opposing directions.
  2. Magnitude: The absolute degree of strength, predictability, or convergence characterizing the relationship.

• Methodological Properties:

  • Lack of Experimental Manipulation: Correlational designs involve zero researcher manipulation or control of independent variables; variables are merely observed, measured, and recorded in their natural state (e.g., recording natural heights and weights without experimental intervention).
  • Quantitative Variable Requirement: Both measured variables must be quantitative to compute a standard Pearson correlation coefficient. When working with categorical or nominal attributes, alternative non-parametric or specialized statistical techniques must be utilized.

Direction and Magnitude of Correlation Coefficients

• Directional Categories:

  • Positive Relationship: Variables change systematically in the identical direction.
  • Mathematical condition: As $X$ increases, $Y$ increases ($X \ne \rightarrow \ne \text{up}$, $Y \ne \rightarrow \ne \text{up}$); correspondingly, as $X$ decreases, $Y$ decreases ($X \ne \rightarrow \ne \text{down}$, $Y \ne \rightarrow \ne \text{down}$).
  • Real-world examples:
    • As human height increases, body weight typically increases.
    • As an individual consumes an increasing quantity of doughnuts, dress size increases; conversely, as doughnut consumption declines, dress size correspondingly decreases.
  • Negative (Inverse) Relationship: Variables change systematically in opposite directions.
  • Mathematical condition: As $X$ increases, $Y$ decreases ($X \ne \rightarrow \ne \text{up}$, $Y \ne \rightarrow \ne \text{down}$); conversely, as $X$ decreases, $Y$ increases ($X \ne \rightarrow \ne \text{down}$, $Y \ne \rightarrow \ne \text{up}$).
  • Real-world examples:
    • As television viewing time increases, academic exam grades decrease.
    • As physical walking steps increase, dress size decreases; conversely, as physical step count decreases, dress size increases.

• Magnitude and Predictability:

  • Magnitude (also termed degree or strength) reflects the precision with which one can predict values of $Y$ from known values of $X$.
  • If a negative correlation between television viewing and examination performance is strong, an observer can predict with high accuracy that for every single additional hour spent watching television (such as binge-watching Lucifer on Netflix), the exam score drops by a reliable, mathematically consistent amount.
  • If the relationship is weak, predictability deteriorates, and values show high dispersion.
  • Independence of Strength and Direction: The magnitude of an association is completely independent of its directional sign. A negative correlation of $-.90$ reflects an exceptionally strong relationship, identical in strength and predictability to a positive correlation of $+.90$.

• Theoretical Extremes versus Empirical Realities:

  • Perfect Correlation ($+1.00$ or $-1.00$): Represents the absolute maximum relationship strength. Every change in one variable produces an exact, invariant proportional change in the other (e.g., the direct metric conversion between inches and centimeters).
  • Zero Correlation ($.00$): Represents the absolute absence of any systematic association (e.g., the relationship between human height and Social Security numbers).
  • Empirical Norms: In biological and social sciences, measured relationships virtually never attain perfect values of $+1.00$, $-1.00$, or exactly $.00$. Individuals sharing an identical height (such as $5\text{' }2\text{"}$) display a natural range of corresponding body weights. As height shifts upward or downward, the corresponding weight range shifts, yielding an intermediate correlation coefficient.

• Magnitude Interpretation Standards:

• Applied Interpretative Exercises:

Calculating the Pearson Product-Moment Correlation Coefficient

• Bivariate Dataset: An empirical evaluation of rehearsal duration in hours ($X$) and the cumulative number of acting awards won ($Y$) across a sample of $n = 8$ actors.

• Summation and Computational Matrix:

• Summary Computational Totals:

  • Sample size: $n = 8$
  • Sum of $X$: $\sum X = 460$
  • Sum of $Y$: $\sum Y = 30$
  • Sum of squared $X$ values: $\sum X^2 = 29000$
  • Sum of squared $Y$ values: $\sum Y^2 = 224$
  • Sum of cross-products: $\sum XY = 2230$

• Correlation Output and Substantive Meaning:

  • Application of the computational formula yields a Pearson bivariate correlation coefficient of $r = .95$.
  • Interpretation: An $r$ value of $.95$ demonstrates an exceptionally strong positive linear association.
  • Directional dynamic: As hours of rehearsal increase, the number of awards won increases substantially; conversely, as rehearsal hours decrease, the number of awards won decreases predictably.

Causality Limitations: Third-Variable and Directionality Problems

• The Impermissible Causal Inference: Correlation does NOT equal causation.

  • Even when a correlation coefficient reaches near-perfect levels (such as $r = .95$), stating that rehearsal causes winning awards, leads to awards, or generates award success is scientifically invalid.
  • Causal terminology is strictly inaccurate within correlational research designs.

• The Third-Variable Problem:

  • An unobserved or unmeasured extraneous variable may influence both measured variables simultaneously, producing an apparent statistical association.
  • Classic example: The robust observed positive correlation between city-wide ice cream sales and violent crime rates is not causal; rather, a third variable—ambient outdoor temperature (hot weather)—independently increases both ice cream consumption and interpersonal encounters/temper-related crimes.
  • Socioeconomic example: An observed correlation between physical health and social status might suggest that health elevates status; however, an unmeasured third variable—financial wealth/money—may independently provide superior healthcare access while directly elevating social status.

• The Directionality Problem:

  • Correlational data fails to establish temporal precedence or causal vectoring, leaving unresolved which variable influences the other.
  • Directional dilemma: Does robust physical health improve one's social status, or does having high social status provide the psychological confidence, financial stability, and elite social connections required to pursue and maintain optimal physical health?

Partitioning Variance: Coefficients of Determination and Alienation

• Coefficient of Determination ($R^2$):

  • Definition: The proportion of total variance in the dependent/criterion variable ($Y$) that is directly accounted for, explained by, or predictable from knowledge of the predictor variable ($X$).
  • Calculation: Square the Pearson correlation coefficient ($R^2 = r^2$).
  • Example 1 (Interpersonal Hugs and Self-Reported Happiness):
  • Observed correlation: $r = .72$
  • Calculation: $R^2 = (.72)^2 = .5184$
  • Mathematical interpretation: Exactly $51.84\%$ of the variance in self-reported happiness is accounted for by the number of hugs given. Stated conversely, in this sample, over half ($> 50\%$) of individual differences in happiness can be determined from hugging behavior.
  • Example 2 (Rehearsal Hours and Acting Awards):
  • Observed correlation: $r = .95$
  • Calculation: $R^2 = (.95)^2 = .9025 \approx .90$
  • Mathematical interpretation: Exactly $.90$ ($90\%$) of the total variance in the number of awards won ($Y$) is explained by hours of rehearsal ($X$).

• Coefficient of Alienation ($1 - R^2$):

  • Definition: The proportion of total variance in the criterion variable ($Y$) that remains unexplained, unaccounted for, and independent of the predictor variable ($X$).
  • Calculation: Subtract the coefficient of determination from unity ($1 - R^2$).
  • Example 1 (Hugs and Happiness):
  • Determination: $R^2 \approx .52$
  • Alienation: $1 - R^2 = 1 - .52 = .48$
  • Mathematical interpretation: Exactly $48\%$ of the variance in happiness cannot be explained by knowing the number of hugs given, reflecting unmeasured individual, situational, and physiological differences.
  • Example 2 (Rehearsal Hours and Awards):
  • Determination: $R^2 = .90$
  • Alienation: $1 - R^2 = 1 - .90 = .10$
  • Mathematical interpretation: Exactly $10\%$ of the variance in acting awards won remains completely unexplained by rehearsal duration (attributable to factors such as innate talent, script quality, network politics, or random variance).

Visualizing Data with Scatterplots

• Construction Methodology:

  • A scatterplot visually represents bivariate data on a Cartesian plane.
  • For every individual participant or observation in a study, the researcher identifies the intersection of their quantitative score on the horizontal abscissa ($X$) and their quantitative score on the vertical ordinate ($Y$), marking that coordinate intersection as an individual plotted data point.

• Structural Patterns and Linear Convergence:

  • The geometric tightness with which plotted data points converge toward a straight diagonal line directly reflects the magnitude of the correlation.
  • Strong Positive Correlation (e.g., $r = .95$): Plotted coordinates cluster tightly along a straight diagonal linear trajectory ascending from the lower-left quadrant to the upper-right quadrant.
  • Strong Negative Correlation (e.g., $r = -.99$): Plotted coordinates cluster almost perfectly along a straight diagonal linear trajectory descending from the upper-left quadrant to the lower-right quadrant.
  • Weak Correlation (e.g., $r = .20$): Plotted coordinates demonstrate pronounced dispersion across the coordinate space, showing negligible alignment along a diagonal linear path and reflecting minimal predictive precision.