Select Page

Hypothesis Analysis

H4 – Education Under-Supplies Higher Needs

H1:The mean L5–L6 share in education is lower than in play, nature, and creativity.

H0: he mean L5–L6 share in the education category equals that in play, nature, and creativity.

Question: Does the education category have a lower proportion of high-level Maslow needs (Self-Actualisation + Self-Transcendence) than play, nature, and creativity?

Data

Data subset: 1,564 images from the four target categories (education, play, nature, creativity) after filtering to retain only these categories.

Variables:

      • Predictor(s): Category (4 levels: education, play, nature, creativity).
      • Outcome(s): L5–L6 share per image (proportion of high-level needs).

Controls: None (comparisons are direct between category groups).

Normalisation & Weighting: See Weighting and Normalisation Methods reference page.

N: 379–404 images per category; total N = 1,564.

Methods

Test type:

  • Descriptive statistics for each category (means, medians, SDs).
  • Kruskal–Wallis H-test for overall group differences.
  • Dunn’s post-hoc tests (Bonferroni-adjusted p-values, one-sided for “education lower” hypothesis).
  • Permutation tests (5,000 shuffles) for mean differences.
  • Effect size estimation via Hodges–Lehmann median difference and Cliff’s δ, with bootstrap 95% CIs.

Rationale:

  • Kruskal–Wallis detects differences in medians without assuming normality.
  • Dunn’s tests identify which pairs differ and allow one-sided directional testing.
  • Permutation tests provide non-parametric resampling confirmation.
  • HL and Cliff’s δ quantify effect size and direction.

Implementation:

  • Analyses run on both raw and weighted data.
  • Weighting affects descriptive means only; ranks preserved for non-parametric tests.
  • Sensitivity checks compared raw vs weighted results, tested consistency across multiple statistical methods, and checked sample balance.

Results – Weighted

Descriptive statistics (Weighted means):

Category n_images Mean Median SD
Creativity 379 0.4088 0.3636 0.3540
Education 404 0.1205 0.0000 0.1981
Nature 387 0.8600 0.9167 0.1879
Play 394 0.2854 0.2000 0.2942

Weighted and raw descriptive statistics are identical for this subset. This occurs because the per-object weights in maslow_weighted_per_object.csv are normalised within each image.

The relative proportions across Maslow levels within an image are therefore unchanged from the binary counts in the raw dataset, producing the same L56 share values.

Results — Raw

Category n_images Mean Median SD
Creativity 379 0.4088 0.3636 0.3540
Education 404 0.1205 0.0000 0.1981
Nature 387 0.8600 0.9167 0.1879
Play 394 0.2854 0.2000 0.2942

Descriptive statistics:

Global difference (Kruskal–Wallis)

H = 709.40, p ≈ 1.92×10⁻¹⁵³ — strong evidence that categories differ.

Pairwise comparisons (Dunn & permutation, one-sided)

Comparison Mean diff p-value (Dunn) p-value (Perm) HL diff (95% CI) Cliff’s δ (95% CI)
Edu vs Play −0.165 <0.001 <0.001 −0.125 (−0.155, −0.086) 0.356 (0.284, 0.427)
Edu vs Nature −0.739 <0.001 <0.001 −0.833 (−0.856, −0.809) 0.945 (0.919, 0.968)
Edu vs Creativity −0.288 <0.001 <0.001 −0.250 (−0.333, −0.200) 0.481 (0.409, 0.548)

Combined Interpretation

Education’s high-level share is consistently and substantially lower than in play, nature, and creativity.

This pattern holds in both raw and weighted data, is highly significant in all statistical tests, and shows large effect sizes unlikely to be due to chance. The ordering of categories (Nature > Creativity > Play > Education) is identical under both weighting schemes. The identical weighted and raw statistics indicate that, for this subset, the weighting scheme did not alter proportional distributions across Maslow levels within images.

Conclusion

Outcome: Reject H₀ and support H.

 The education category has a significantly lower proportion of high-level needs than play, nature, and creativity. This result is robust across raw and weighted analyses, multiple non-parametric and resampling tests, and consistent effect size measures.

←→