I have been charged with writing an exam for psychology students.
My coordinator doesn’t
want to use probability tables or a computer for the exam though. That leaves only
approximation methods that we hand out to students on a formula sheet. I have compiled the
information for that formula sheet. Each approximation formula has apparently come from a
different source though. I’ve tracked down each source, and I’m uploading them as
attachments to this prompt. Your task is to review the document associated with each test, and
confirm that the paper contains the approximation method I’ve found for each
“
1. One-Sample t-Test Approximation From Hill (1970)
• Test statistic:
t = (X̄ - μ) / (s / √n)
• Approximate one-tailed p-value (Hill's Approximation, 1970):
p(t, df) ≈ 1 / (1 + (t² / df))^(df/2)
• Approximate two-tailed p-value:
2 × p(t, df)
Reference: Hill, G. W. (1970). Algorithm 396: Student’s t-Quantiles. Communications of the
ACM, 13(10), 619–620. [Link]
2. Independent Samples t-Test Approximation From Hill (1970)
• Test statistic:
t = (X̄ ₁ - X̄ ₂) / √(s_p² × (1/n₁ + 1/n₂))
• s_p² is the pooled variance.
• Degrees of freedom: df = n₁ + n₂ - 2
• p-value approximation:
p(t, df) ≈ 1 / (1 + (t² / df))^(df/2)
Reference: Hill (1970), same as above.
3. Paired Samples t-Test Approximation From Hill (1970)
• Test statistic:
t = D̄ / (s_D / √n)
• Degrees of freedom: df = n − 1
• p-value approximation:
p(t, df) ≈ 1 / (1 + (t² / df))^(df/2)
Reference: Hill (1970).
4. Mann-Whitney U Test (Independent Non-Parametric Two-Sample Test)
Approximation From Buckland & Noether (1967)
For sufficiently large sample sizes (typically n₁, n₂ > 20), the U statistic follows a normal
approximation.
• Test statistic:
Z = (U - (n₁ × n₂) / 2) / √(n₁ × n₂ × (n₁ + n₂ + 1) / 12)
• Approximate p-value:
p(Z) ≈ 1 / (1 + (Z² / 2))
Reference: Buckland, W. R., & Noether, G. E. (1967). Elements of Nonparametric Statistics.
The Statistician, 17(4), 421. [Link]
5. Wilcoxon Signed-Rank Test (Paired Non-Parametric Test) Approximation From
Toutenburg (1977)
For large n, the Wilcoxon signed-rank test follows a normal approximation.
• Test statistic:
Z = (W - (n(n + 1)) / 4) / √((n(n + 1)(2n + 1)) / 24)
• Approximate p-value:
p(Z) ≈ 1 / (1 + (Z² / 2))
Reference: Toutenburg, H. (1977). Lehmann, E. L., Nonparametrics: Statistical Methods
Based on Ranks, San Francisco. Holden‐Day, Inc., 1975. 480 S., $ 22.95 . ZAMM - Journal
of Applied Mathematics and Mechanics / Zeitschrift Für Angewandte Mathematik Und
Mechanik, 57(9), 562–562. Portico. [Link]
6. Pearson Correlation and Regression Approximation From Student (1908)
To test whether a correlation coefficient r is significantly different from 0:
• Test statistic:
t = (r × √(n - 2)) / √(1 - r²)
• Degrees of freedom: df = n − 2
• p-value approximation:
p(t, df) ≈ 1 / (1 + (t² / df))^(df/2)
Reference: Student. (1908). Probable Error of a Correlation Coefficient. Biometrika, 6(2/3),
302-310. [Link]
7. One-Way ANOVA Approximation From Wilson & Heferty (1931)
Tests for differences between three or more group means.
• Test statistic:
F = (Between-group variance) / (Within-group variance)
• Approximate p-value (Wilson–Hilferty Transformation, 1931):
p(F, df₁, df₂) ≈ 1 / [(1 + (F × df₁ / df₂))^(df₂/2)]
Reference: Wilson, E. B., & Hilferty, M. M. (1931). The Distribution of Chi-Square.
Proceedings of the National Academy of Sciences, 17(12), 684–688.
[Link]
8. Kruskal-Wallis Test (One-Way ANOVA by Ranks)
• Test statistic:
H = (12 / (n(n + 1))) × ΣRⱼ² - 3n(k + 1)
• Approximate p-value (Normal Approximation):
Z = (H - (k - 1)) / √(2(k - 1))
p(Z) ≈ 1 / (1 + (Z² / 2))
Reference: Kruskal, W. H., & Wallis, W. A. (1952). Use of Ranks in One-Criterion Variance
Analysis. Journal of the American Statistical Association, 47(260), 583–621.
[Link]
9. Friedman Test (Repeated Measures ANOVA by Ranks) Approximation From
Friedman (1937)
• Test statistic:
χ²_F = (12 / (n × k(k + 1))) × ΣRⱼ² - 3n(k + 1)
• Degrees of freedom: df = k − 1
• Approximate p-value (Wilson–Hilferty Transformation):
p(χ², df) ≈ 1 / [(1 + (χ² / df))^(df/2)]
Reference: Friedman, M. (1937). The Use of Ranks to Avoid the Assumption of Normality
Implicit in the Analysis of Variance. Journal of the American Statistical Association, 32(200),
675–701. [Link]
10. Two-Way Between-Subjects ANOVA Approximation From Wilson & Heferty (1931)
Used for factorial designs where two independent variables (factors) are tested
simultaneously.
• Test statistics:
- Factor A (main effect):
F_A = (SS_A / df_A) / (SS_W / df_W)
- Factor B (main effect):
F_B = (SS_B / df_B) / (SS_W / df_W)
- Interaction effect (A × B):
F_{A×B} = (SS_{A×B} / df_{A×B}) / (SS_W / df_W)
• Degrees of freedom:
df_A = k_A - 1
df_B = k_B - 1
df_{A×B} = (k_A - 1)(k_B - 1)
df_W = N - (k_A k_B)
• Approximate p-value (Wilson–Hilferty Transformation, 1931):
p(F, df₁, df₂) ≈ 1 / [(1 + (F × df₁ / df₂))^(df₂/2)]
Reference: Wilson, E. B., & Hilferty, M. M. (1931). The Distribution of Chi-Square.
Proceedings of the National Academy of Sciences, 17(12), 684–688.
[Link]
“