Select Page

DA and MHN Analysis

DA, MHN and Category Relationship Analysis

The previous section established that DA–MHN associations are statistically robust at the whole-dataset level (60 tests), but many significant results were likely inflated by the very large sample size, and strong effect sizes (|r| > 0.7) were comparatively rare, especially in weighted analyses.

To address these limitations and uncover more meaningful, context-sensitive relationships, the present analysis examines DA–MHN alignment within each of the 20 search categories, greatly expanding the granularity of the assessment.

Methodology

Design – 60 DA–MHN pairs per category × 20 categories = 1,200 tests

Binary data analysis (Raw): Each cell recoded as 1 (category present, i.e., any non-empty cell) or 0 (category absent).

Weighted analysis: Cell value = pre-weighted object count per image (See weighted calculation methods)

Filtered Analysis: Pearson and cosine computed only on rows where at least one of the two vectors is non-zero.

Reporting thresholds:

      • Strong: |r| > 0.7 and p < 0.05
      • Medium: 0.3 < |r| ≤ 0.7 and p < 0.05
      • Significant: p < 0.05

Results

DA–MHN pairs per category meeting each criterion (out of 60)

[wpdatatable id=3]

Table 1: Number of strong (|r| > 0.7, p < 0.05) and significant (p < 0.05) DA–MHN pairs per category for binary and weighted analyses.

Discussion

Binary co-occurrence is widespread. Many categories show high counts of significant and strong binary associations.

Strong weighted effects are rare. Only Nature shows a single strong weighted pair.

Medium weighted effects carry most of the signal. Notable medium and significant weighted counts appear in several categories, for example:

      • Medium weighted: Instinct 40, Happiness 37, Artificial 35, Danger 35, Conformity 34
      • Significant weighted: Danger 54, Happiness 53, Instinct 52, Standardisation 51, Artificial 51

Sparse weighted coverage in some categories. Connection, Creativity, and Education show no medium or significant weighted pairs in this run.

Conclusion

Presence/absence captures broad co-occurrence, but object-richness alignment is mostly moderate rather than strong.

Focus interpretation on effect sizes, not p-values alone. Medium weighted relationships appear to reflect the more realistic structure of affordance–need alignment within categories.
Where weighted counts are zero, consider data coverage and merging choices before drawing substantive conclusions.

←→