P-Values, Statistical Significance, and Correlation Interpretation

Fundamentals of P-Values and Statistical Significance

  • P-Value Definition: A pp-value serves as a direct quantitative measure of error in hypothesis testing and experimental analysis.
  • Acceptable Error Threshold: The maximum acceptable level of error to establish statistical significance is 0.050.05, which corresponds to a 5%5\% error rate.
    • A pp-value of 0.050.05 indicates that there is a statistically significant difference between two comparison groups.
    • Any error rate greater than 5%5\% (p>0.05p > 0.05) exceeds the maximum tolerable threshold for declaring a meaningful effect or difference.
  • Directionality Rule for P-Values: Unlike standard mathematical metrics where larger values are preferred, pp-values operate on an inverse rule where smaller values indicate better statistical outcome:
    • Smaller pp-values represent lower probability of error and stronger evidence of a statistically significant difference.
    • Larger pp-values represent higher error rates and a lack of statistical significance.
  • Values Lacking Statistical Significance (p>0.05p > 0.05):
    • 0.060.06
    • 0.070.07
    • 0.080.08
    • 0.090.09
    • 0.100.10
    • 0.200.20
    • 0.300.30
    • 0.400.40
    • 0.500.50
    • 0.600.60
    • 0.800.80
    • 0.900.90
  • Values Demonstrating Statistical Significance (p0.05p \le 0.05):
    • 0.050.05
    • 0.040.04
    • 0.030.03
    • 0.020.02
    • 0.010.01
    • 0.0050.005
    • 0.0010.001

Interpreting Correlations and Significance Cutoffs

  • Standard Decision Convention: By universal academic and statistical convention, an experimental outcome or correlation is statistically significant if and only if p0.05p \le 0.05.
  • Step-by-Step Correlation Analysis Workflow:
    1. Interpret the relationship direction (positive or negative) and strength (weak, moderate, or strong) using the correlation coefficient (rr).
    2. Evaluate the accompanying probability value (pp-value) against the 0.050.05 threshold to determine if the relationship is statistically significant.
  • Case Study 1: Bipolar No More:
    • Observed Relationship: There is a strong, negative correlation between administration of the treatment Bipolar No More and a decrease in symptoms of bipolar disorder.
    • Significance Assessment: Statistically significant.
    • Justification: The probability value falls directly at the threshold where p0.05p \le 0.05.
  • Case Study 2: Lithium:
    • Observed Relationship: Data yields a correlation coefficient of r=0.60r = -0.60, indicating a moderate, negative relationship between taking lithium and a decrease in bipolar disorder symptoms.
    • Significance Assessment: Not statistically significant.
    • Justification: The probability value is greater than 0.050.05 (p>0.05p > 0.05).
  • Testing and Coursework Guidelines:
    • Although p>0.05p > 0.05 represents an error rate greater than 5%5\%, formal evaluations (quizzes, assignments, and activities) require explicitly writing that the pp-value is "greater than 0.050.05" (or p>0.05p > 0.05) rather than stating a percentage.

Application and Presentation of Probability Values

  • Analytical Expectations:
    • Computation or hand calculation of probability values is not required in standard introductory interpretation workflows; probability values are provided directly for evaluation.
    • The primary task is accurate conceptual interpretation of the provided pp-value.
  • Representation in Scientific Literature:
    • In academic research publications and empirical papers, pp-values are commonly presented in one of two formats:
    1. As an exact equality statement (e.g., p=0.04p = 0.04).
    2. As an inequality relative to alpha thresholds (e.g., p<0.05p < 0.05 or p0.05p \le 0.05).

Questions & Discussion

  • Inquiry Regarding Scope (Todd):
    • Question: Is the material covered up to this point inclusive of all $t$-test procedures?
    • Response: Coverage of $t$-tests is deferred to the beginning of the next class session because detailed introduction and analysis of $t$-tests cannot be adequately completed in a brief 3-minute window remaining at the end of class (at 2:16 PM).