13.2 Descriptive and Inferential Statistics for Rehabilitation Counselors

13.2 Descriptive and Inferential Statistics for Rehabilitation Counselors

Core Focus: Certified Rehabilitation Counselors must interpret statistical analyses in peer-reviewed literature, clinical assessment reports, and program evaluations. Mastering measurement scales, descriptive dispersion, normal curve dynamics, hypothesis testing errors, parametric vs. non-parametric decision trees, correlation, and regression allows counselors to make rigorous, evidence-based vocational decisions.

---

1. Scales of Measurement and Data Classification

Statistical analysis begins by identifying the mathematical properties of the data being evaluated. S.S. Stevens established four distinct scales of measurement (NOIR), each determining which mathematical operations and statistical procedures are permissible:

                      STEVENS' HIERARCHY OF MEASUREMENT SCALES

  ┌────────────────────────────────────────────────────────────────────────┐
  │ RATIO SCALE (Highest Rigor: Absolute True Zero Point)                  │
  │ • Hourly Wages ($), Weekly Earnings, Hours Worked, Reaction Time       │
  │ • Operations: All mathematical operations; ratios are meaningful       │
  │ • Statistics: Mean, SD, Geometric Mean, Coefficient of Variation       │
  └───────────────────────────────────┬────────────────────────────────────┘
                                      │
                                      ▼
  ┌────────────────────────────────────────────────────────────────────────┐
  │ INTERVAL SCALE (Equal Units; Arbitrary Zero Point)                     │
  │ • Standardized IQ Scores (WAIS-IV), Temperature (°F / °C), GRE Scores  │
  │ • Operations: Addition & Subtraction; no meaningful ratios             │
  │ • Statistics: Mean, Standard Deviation, Pearson r, t-test, ANOVA       │
  └───────────────────────────────────┬────────────────────────────────────┘
                                      │
                                      ▼
  ┌────────────────────────────────────────────────────────────────────────┐
  │ ORDINAL SCALE (Ranked Order; Unequal / Unknown Intervals)              │
  │ • Likert Satisfaction Scales (1-5), Disability Severity Ranks (Mild/Mod│
  │ • Operations: Greater than / Less than; no addition or subtraction     │
  │ • Statistics: Median, Interquartile Range, Spearman rho, Mann-Whitney  │
  └───────────────────────────────────┬────────────────────────────────────┘
                                      │
                                      ▼
  ┌────────────────────────────────────────────────────────────────────────┐
  │ NOMINAL SCALE (Lowest Rigor: Qualitative Mutually Exclusive Categories)│
  │ • Disability Type (Orthopedic, Psychiatric, Sensory), Gender, Race     │
  │ • Operations: Counting frequencies / percentages; no ranking           │
  │ • Statistics: Mode, Frequency Counts, Chi-Square (χ²) Tests            │
  └────────────────────────────────────────────────────────────────────────┘

---

2. Descriptive Statistics, Distribution Shapes, and Dispersion

Descriptive statistics summarize, organize, and describe the characteristics of a sample dataset.

Measures of Central Tendency

• Mean (X̄ or μ): The arithmetic average of all scores. Calculated as the sum of all scores divided by sample size (N). Highly sensitive to extreme outlier scores.
• Median (Mdn): The exact middle score (50th percentile) dividing an ordered distribution into two equal halves. Robust against extreme outliers; preferred measure for skewed economic and vocational data.
• Mode: The most frequently occurring score in the dataset. Useful for bimodal distributions and the only central tendency measure for nominal data.

Skewness and Distribution Symmetry

                  DISTRIBUTION SKEWNESS & CENTRAL TENDENCY

    POSITIVE (RIGHT) SKEW             NORMAL (SYMMETRIC)            NEGATIVE (LEFT) SKEW
   (e.g., Client Wages)             (e.g., Standard IQ)            (e.g., Easy Mastery Exam)

          ▲                                ▲                                 ▲
         ╱ ╲                              ╱ ╲                               ╱ ╲
        ╱   ╲                            ╱   ╲                             ╱   ╲
       ╱  |  ╲                          ╱  |  ╲                           ╱  |  ╲
      ╱   |   ╲                        ╱   |   ╲                         ╱   |   ╲
     ╱    |    ╲                      ╱    |    ╲                       ╱    |    ╲
    ╱     |     ╲────                ╱     |     ╲                 ────╱     |     ╲
   Mode  Mdn   Mean                 Mode=Mdn=Mean                    Mean   Mdn  Mode
   (Mode < Mdn < Mean)                                              (Mean < Mdn < Mode)

• Normal (Symmetrical) Distribution: Bell-shaped and perfectly symmetrical. Mean = Median = Mode.
• Positively Skewed (Right-Skewed): The tail extends toward the high positive values (right). Outliers pull the mean upward: Mode < Median < Mean. (Typical of VR client earnings, where a few high earners inflate the mean while most earn entry-level wages; the median represents typical earning capacity).
• Negatively Skewed (Left-Skewed): The tail extends toward the low negative values (left). Low-scoring outliers drag the mean downward: Mean < Median < Mode. (Typical of an easy mastery test where most participants score high).

Measures of Variability and Dispersion

• Range: Simple distance between maximum and minimum raw scores (Xmax - Xmin). Extremely sensitive to outliers.
• Variance (s² / σ²): The average squared deviation of scores from the mean.
• Standard Deviation (s / SD / σ): The positive square root of the variance. Expresses dispersion in the original raw units of measurement.

Properties of the Standard Normal Curve (Empirical Rule)

In a normal distribution with mean μ and standard deviation σ:

• 68.26% (~68%) of all scores fall within ±1 SD of the mean (μ ± 1σ).
• 95.44% (~95%) of all scores fall within ±2 SD of the mean (μ ± 2σ).
• 99.74% (~99.7%) of all scores fall within ±3 SD of the mean (μ ± 3σ).

---

3. Inferential Statistics, Hypothesis Testing, and Decision Errors

Inferential statistics allow researchers to draw conclusions about population parameters based on sample data, testing whether observed differences reflect true effects or random sampling error.

                      HYPOTHESIS TESTING DECISION MATRIX

                                          ACTUAL POPULATION REALITY
                                  ┌────────────────────────┬────────────────────────┐
                                  │   Null Hypothesis H0   │   Null Hypothesis H0   │
                                  │        is TRUE         │        is FALSE        │
  ┌───────────────────────────────┼────────────────────────┼────────────────────────┤
  │ DECISION: Reject H0           │     TYPE I ERROR       │    CORRECT DECISION    │
  │ (Declare Significant Effect)  │       (Alpha α)        │   Power = (1 - Beta)   │
  │                               │    "False Positive"    │    "True Positive"     │
  ├───────────────────────────────┼────────────────────────┼────────────────────────┤
  │ DECISION: Fail to Reject H0   │    CORRECT DECISION    │     TYPE II ERROR      │
  │ (Declare No Significant Diff) │    Confidence Level    │        (Beta β)        │
  │                               │       (1 - Alpha)      │    "False Negative"    │
  └───────────────────────────────┴────────────────────────┴────────────────────────┘

Core Inferential Concepts

• Null Hypothesis (H0): States there is no true difference, relationship, or effect in the population (μ1 = μ2). Any observed sample difference is due to chance.
• Alternative / Research Hypothesis (H1): States there is a real, non-random difference or relationship in the population (μ1 ≠ μ2).
• Type I Error (α / Alpha): False Positive. The researcher rejects a true null hypothesis, claiming an intervention is effective when it actually has no effect. The alpha level (typically set at α = .05 or .01) defines the maximum acceptable risk of making a Type I error.
• Type II Error (β / Beta): False Negative. The researcher fails to reject a false null hypothesis, concluding an intervention has no effect when a real difference actually exists.
• Statistical Power (1 - β): The probability of correctly rejecting a false null hypothesis (detecting a real effect). A prospective power target of .80 (80%) is a common planning convention, not a universal adequacy rule. Power is increased by:
  1. Increasing sample size (N).
  2. Increasing the effect size magnitude.
  3. Increasing the alpha level (e.g., from .01 to .05).
  4. Utilizing sensitive, reliable measurement tools to reduce error variance.
• p-Value: The exact probability that the observed sample difference occurred purely by random chance if the null hypothesis were true. If p <= α (p < .05), the finding is declared statistically significant.

---

4. Parametric vs. Non-Parametric Statistical Tests

Statistical test selection is governed by measurement scale, normality of distribution, and sample independence:

Parametric Inferential Tests

Requirements: Continuous dependent variable (interval/ratio), normal distribution, and homogeneity of variance (equal group variances via Levene's test).

• Independent Samples t-Test: Compares the means of two independent groups (e.g., comparing mean weekly wages between clients receiving Individual Placement and Support [IPS] vs. traditional VR).
• Paired / Dependent Samples t-Test (Repeated Measures): Compares two related means from the same group measured across two time points (pretest vs. posttest) or matched pairs.
• One-Way Analysis of Variance (ANOVA): Compares the means of three or more independent groups on one continuous dependent variable. Calculates the F-ratio (Between-Group Variance divided by Within-Group Variance). If significant, post-hoc tests (Tukey HSD, Scheffé, Bonferroni) identify which specific group pairs differ.
• Factorial ANOVA (2x2, 2x3, etc.): Evaluates the simultaneous effects of two or more independent variables (factors) on a continuous outcome, yielding Main Effects for each factor and an Interaction Effect (whether the effect of one factor depends on the level of another).
• Analysis of Covariance (ANCOVA): Compares group means on a dependent variable while statistically removing the confounding influence of an extraneous continuous variable (covariate), such as baseline pretest scores or client age.

Non-Parametric Inferential Tests

Requirements: Distribution-free tests used when data are nominal or ordinal, or when parametric normality assumptions are severely violated.

• Chi-Square (χ²) Test of Independence: Evaluates whether two categorical (nominal) variables are related by comparing observed cell frequencies against expected frequencies (e.g., examining employment status [Employed vs. Unemployed] across disability categories [Sensory, Physical, Psychiatric]).
• Mann-Whitney U Test: The non-parametric counterpart to the independent t-test; compares ranked ordinal data between two independent groups.
• Wilcoxon Signed-Rank Test: The non-parametric counterpart to the paired samples t-test; evaluates pre-post ranked changes within a single group.
• Kruskal-Wallis Test: The non-parametric counterpart to the one-way ANOVA; compares ranked data across three or more independent groups.

---

5. Correlation, Regression, and Effect Size

Correlation Analysis

• Pearson Product-Moment Correlation (r): Quantifies the direction and strength of the linear relationship between two continuous (interval/ratio) variables. Ranges from -1.00 (perfect negative relationship) through 0.00 (no relationship) to +1.00 (perfect positive relationship).
• Spearman Rank Correlation (rho / ρ): Non-parametric correlation measuring the monotonic relationship between two ranked (ordinal) variables.
• Coefficient of Determination (r²): The proportion of variance in the dependent variable explained by or shared with the independent variable. Calculated by squaring Pearson's r (e.g., if r = .60, then r² = .36, indicating that 36% of the variance is shared, while 64% is unexplained).
• Correlation Does NOT Imply Causation: High correlation indicates association but cannot prove direct causal direction due to potential confounding third variables.

Regression Analysis

• Simple Linear Regression: Models the predictive relationship between one continuous predictor (X) and one continuous outcome (Y) using the linear equation: Y = a + bX, where a is the Y-intercept and b is the regression slope coefficient.
• Multiple Linear Regression: Evaluates how multiple independent predictor variables (X1, X2, ..., Xk) jointly and uniquely predict a single continuous criterion (Y): Y = a + b1X1 + b2X2 + ... + bkXk. Enables CRCs to determine which vocational services uniquely contribute to post-rehabilitation earnings while controlling for client demographic variables.

Effect Size: Clinical Magnitude vs. Statistical Significance

While p-values indicate whether an effect is statistically non-zero (heavily inflated by large sample sizes), effect size measures the practical magnitude of an intervention:

• Cohen's d (Standardized Mean Difference): d = (Mean1 - Mean2) / SD_pooled
  • d = 0.20: Small effect (subtle clinical change)
  • d = 0.50: Medium effect (noticeable, moderate clinical impact)
  • d = 0.80+: Large effect (substantial, highly impactful clinical change)