Select Page

DA and MHN Relationship

Developmental Affordance Dataset Analysis

Executive Summary

This analysis investigates whether certain image categories are statistically associated with different types of developmental affordances using both raw and weighted data across 7,362 images spanning 20 independent conceptual themes. The study tests the hypothesis that visual environments embed biologically and psychologically meaningful affordances; such as human presence, natural elements, and technological artifacts that shape developmental potential and are expressed differently across conceptual image types.

Methodology

  • Dataset: 7,362 images across 20 independent image categories
  • Objects: 15,060 object instances assigned to 10 mutually exclusive DA categories
  • Weighting : Weighted Calculation Method

Developmental Affordance Groups :

  • Humans – 52.5%
  • Animals & Non-Human Organisms – 11.9%
  • Plants & Botanical Elements – 14.8%
  • Natural Landscapes & Phenomena – 17.8%
  • Built Structures & Indoor/Urban Spaces – 35.1%
  • Vehicles & Transportation – 11.7%
  • Food, Beverages & Tableware – 10.4%
  • Manual Tools & Functional Objects – 25.4%
  • Technology & Digital Media – 25.9%
  • Art, Sports & Recreational Objects – 20.8%

This distribution reflects the human-centric nature of the image dataset, with human subjects being most prevalent, followed by built environments and technological objects.

Statistical Methodology

All analyses followed rigorous statistical validation tests

Assumption Test results:
  • Normality: Shapiro-Wilk tests conducted on all domains
  • Homogeneity of variance: Levene’s tests performed
  • Independence: Satisfied through random sampling design
Non-Parametic Approach
  • Given consistent violations of normality and equal variance assumptions across all domains, non-parametric methods were employed throughout the analysis.

Assumption test results

Normality tests (Shapiro-Wilk):
  • All domains in both raw and weighted data showed significant departures from normality (p < 0.001), confirming the appropriateness of non-parametric methods.
Homogeneity of variance tests (Levene’s):
  • All domains violated the assumption of equal variances across categories (p < 0.001), further supporting the use of non-parametric approaches.

Kruskal-Wallis H-test results

Testing for differences in domain distributions across categories:

Raw Data –

  • Humans: H = 980.76, p < 0.001***
  • Animals & Non-Human Organisms: H = 280.95, p < 0.001***
  • Plants & Botanical Elements: H = 1,016.95, p < 0.001***
  • Natural Landscapes & Phenomena: H = 1,122.37, p < 0.001***
  • Built Structures & Indoor/Urban Spaces: H = 679.57, p < 0.001***
  • Vehicles & Transportation: H = 280.95, p < 0.001***
  • Food, Beverages & Tableware: H = 280.95, p < 0.001***
  • Manual Tools & Functional Objects: H = 280.95, p < 0.001***
  • Technology & Digital Media: H = 280.95, p < 0.001***
  • Art, Sports & Recreational Objects: H = 280.95, p < 0.001***

Weighted Data –

  • Weighted Kruskal-Wallis tests could not be computed for most domains due to insufficient variance across categories. This outcome is expected: the weighting system was specifically designed to equalise category influence, and the resulting lack of variation confirms that it successfully neutralised size-based bias in the data.

All domains with sufficient variance show highly significant differences across categories.

Chi-Square Test of Independence

Testing associations between categories and domain presence –

Raw Data

  • Humans: χ² = 1,193.32, p < 0.001***
  • Animals & Non-Human Organisms: χ² = 344.17, p < 0.001***
  • Plants & Botanical Elements: χ² = 986.94, p < 0.001***
  • Natural Landscapes & Phenomena: χ² = 1,122.37, p < 0.001***
  • Built Structures & Indoor/Urban Spaces: χ² = 679.57, p < 0.001***
  • Vehicles & Transportation: χ² = 344.17, p < 0.001***
  • Food, Beverages & Tableware: χ² = 344.17, p < 0.001***
  • Manual Tools & Functional Objects: χ² = 344.17, p < 0.001***
  • Technology & Digital Media: χ² = 344.17, p < 0.001***
  • Art, Sports & Recreational Objects: χ² = 344.17, p < 0.001***

Weighted Data –

Results identical to raw data, indicating binary nature of the underlying associations.

All domains demonstrate significant associations with image categories, confirming that category membership is highly predictive of object presence.

Spearman Rank Correlations –

Inter-domain relationships reveal important patterns in object co-occurrence:

Key Correlations (Raw)

  • Plants & Botanical Elements ↔ Natural Landscapes & Phenomena: ρ = 0.359, p < 0.001***
  • Natural Landscapes & Phenomena ↔ Vehicles & Transportation: ρ = 0.086, p < 0.001***
  • Humans ↔ Art, Sports & Recreational Objects: ρ = 0.072, p < 0.001***
  • Animals & Non-Human Organisms ↔ Natural Landscapes & Phenomena: ρ = 0.067, p < 0.001***
  • Animals & Non-Human Organisms ↔ Plants & Botanical Elements: ρ = 0.064, p < 0.001***

Negative Correlations –

  • Humans ↔ Natural Landscapes & Phenomena: ρ = -0.229, p < 0.001***
  • Humans ↔ Plants & Botanical Elements: ρ = -0.187, p < 0.001***
  • Humans ↔ Vehicles & Transportation: ρ = -0.139, p < 0.001***
  • Humans ↔ Manual Tools & Functional Objects: ρ = -0.114, p < 0.001***
  • Humans ↔ Technology & Digital Media: ρ = -0.109, p < 0.001***

Weighted Data Correlations –

Most correlations resulted in NaN values, indicating insufficient variance in weighted scores for meaningful correlation analysis.

The strong positive correlation between Plants and Natural Landscapes suggests these elements frequently co-occur in nature-themed images. The negative correlations between Humans and natural elements indicate a tendency for human-focused images to exclude natural content.

Descriptive Statistics by Category

Analysis of median values and interquartile ranges reveals distinct patterns across categories:

Highest Human Presence –

human_instinct, creativity, conformity, danger, happiness, play, safety (Median = 1)

Clear association with human-centered categories

Natural Elements concentration –

nature category shows highest presence of Natural Landscapes & Phenomena (Median = 1)

Plants & Botanical Elements also concentrated in nature-related categories

Technologu Distribution – 

Technology & Digital Media present across various categories but with low overall prevalence

Built Structures show moderate presence across urban-themed categories

Insights

This dataset reveals a distinctly human-centric orientation, with over half of all detected objects (52.8%) relating directly to human presence or activity. A strong inverse correlation between humans and natural elements highlights a clear thematic divide between human-focused and nature-focused imagery.

Co-occurrence patterns reinforce this divide: plants, animals, and landscapes frequently appear together, while human figures are more often accompanied by built structures and technological objects. Each image category exhibits a highly specific object signature, suggesting that object detection alone is sufficient to reliably infer thematic content.

The presence of NaN values in weighted Kruskal-Wallis analyses indicates that weighting designed to equalise category influence produced extremely low variance in some domains. While this confirms the success of the weighting strategy, it also limits the use of certain inferential tests on weighted data.

Statistical Significance

All major findings achieved statistical significance at p < 0.001 level for raw data analyses, indicating robust and reliable patterns. Weighted data analyses were limited by insufficient variance in most domains.

Heatmap

Hypothesis Evaluation:

H₁: Visual image categories are associated with systematic differences in the distribution of developmental affordances, revealing meaningful psychological or experiential patterns.

Support for Hypothesis

  • Strong raw associations across all object domains (p < 0.001)
  • Affordance-rich categories (e.g. Nature, Freedom, Connection) show consistent clustering of
    natural and social objects
  • Affordance-limited categories (e.g. Technology Addiction, Isolation) are dominated by built or digital objects
  • Weighted analysis retains most significance while reducing inflation

Challences to Hypothesis

  • Weighting reduces some effects, revealing potential sample-size dependency
  • A few conceptual themes (e.g. Instinct, Artificial) show mixed or ambiguous patterns
  • Some suppressive categories (e.g. Doomscrolling) include unexpected object variability

Conclusion

This analysis reveals strong and systematic patterns in object detection across thematic image categories. Clear statistical associations between object types and category labels suggest that object presence alone can reliably differentiate content, offering a foundation for predictive modelling and automated classification.
The dataset’s human-centric bias, alongside a distinct separation between natural and artificial content, highlights structural patterns in visual media that reflect broader cultural themes. While weighted analyses were limited by low-variance distributions an expected outcome of equalising category influence the raw data offered robust inferential power.

Through validated assumptions, non-parametric testing, and correlation analysis, this study provides a statistically sound basis for both theoretical interpretation and practical applications in visual content analysis and computer vision.

←→