Musical Genres and Socio-Demographic Labels as Fuzzy Categories

Genres and Categories

• Notion of genre fundamental to the Sociology of Taste
• Key organizing principle
• Perennially contested Category
• Basic Problem:
  • Our intuitions imply fuzziness and distributional thinking
  • But our methods mostly rely on crisp classifications and central tendencies
  • Calls to either reject or radically revise the notion

Genres and Categories

• Notion of genre fundamental to the Sociology of Taste
• Key organizing principle
• Perennially contested Category
• Recent work on Measuring Culture has begun to make headway on the issue
  • Relational approaches applying network methods
    • Focus on fuzziness and overlapping categories
  • Geometric Data Analysis
    • Focus on distributions of judgments in a relational space

The sociology of Taste

• Focused on the objective dual linkage between two systems of categories:
• Genre categories developed in fields of cultural production (scenes, industries)
• Social categories endowed with ritual potency and social validity constitutive of status orders
• Basic intuition:
  • Genres are defined by the categories of people who choose them
  • Categories of persons are defined by the genres they choose
• Yet examining folk construals of these dual linkages not a common focus

Data

• Data on cultural tastes among a (weighted) representative sample of Americans collected in Summer of 2012
  • \(N = 2,250\)
• Like GSS 1993 survey included items assessing respondents’ likes and dislikes for 20 categories of musical style**
  • Inclusion of a perceptual module
  • Which of these characteristics describe the typical fans of [genre category] music? Choose all that apply.***

Approach

• We can treat the association data provided by respondents as cognitive two-mode data
  • In analogy with Krackhardt-style data for one-mode networks
  • Persons provide perceived associations between musical genres and social labels
• Use relational techniques to extract the backbone of the perceived associations isolating the most important ones:
  • Dual Projection**
  • Binarizing via Backbone extraction***

Approach

• We can treat the association data provided by respondents as cognitive two-mode data
  • In analogy with Krackhardt-style data for one-mode networks
  • Persons provide perceived associations between musical genres and social labels
• Model the backbone of the person-specific projected networks jointly using:
  • Stacked Correspondence Analysis**
  • Geometric Data Analysis of the various clouds of individuals***
  • Regression, clustering and other methods

Mondo Breiger Approach**

• Each individual provides a “personal” two mode network of genres by social labels with corresponding affiliation matrix \(\mathbf{A}(p)\)
  • \(a^{(p)}_{gl} = 1\) if individual \(p\) associates genre \(g\) with label \(l\)
• The personal genre projection is given by:
  • \(\mathbf{G}(p) = \mathbf{A}(p)\mathbf{A}(p)^T\)
  • \(g^{(p)}_{ij}\) records the \(p^{th}\) individual’s perceived similarity between genres \(i\) and \(j\) based on their shared labels
  • \(g^{(p)}_{ii}\) records the number of times individual \(p\) associates genre \(i\) with a label

Mondo Breiger Approach**

• Each individual provides a “personal” two mode network of genres by social labels with corresponding affiliation matrix \(\mathbf{A}(p)\)
  • \(a^{(p)}_{gl} = 1\) if individual \(p\) associates genre \(g\) with label \(l\)
• The personal label projection is given by:
  • \(\mathbf{L}(p) = \mathbf{A}(p)^T\mathbf{A}(p)\)
  • \(l^{(p)}_{kl}\) records the \(p^{th}\) individual’s perceived similarity between labels \(k\) and \(l\) based on their shared genres
  • \(l^{(p)}_{kk}\) records the number of times individual \(p\) associates genre \(i\) with a label

Mondo Breiger Approach**

• Each individual provides a “personal” two mode network of genres by social labels with corresponding affiliation matrix \(\mathbf{A}(p)\)
  • \(a^{(p)}_{gl} = 1\) if individual \(p\) associates genre \(g\) with label \(l\)
• We take each dual projection and transform them:
  • \(\mathbf{G}(p) \rightarrow \mathbf{G}^{*}(p)\)
  • \(\mathbf{L}(p) \rightarrow \mathbf{L}^{*}(p)\)
• Where \(\mathbf{G}(p)^{*}\) and \(\mathbf{L}(p)^{*}\) are the (binarized) backbone of the network projections using the Stochastic Degree Sequence Model (SDSM)**

Individual Matrix Examples

Example Row Projection

Example Column Projection

Example Row Projection

Example Column Projection

Example Row Backbone

Example Row Backbone

Example Column Backbone

Example Column Backbone

Stacked CA Analyses

• Three sets of scores:
  • Row scores are person-specific judgments of the relative similarity of genres and labels
  • Column scores represent the aggregate judgment of the relative similarity of genres (with respect to labels) and labels (with respect to genres)
  • Supplementary scores for genres and labels represent the centroid of the cloud of person-specific judgments
• Multiple clouds of individuals:
  • Projected into genre space (one cloud per genre)
  • Projected into social label space (one cloud per label)
    • Individual specific clouds (across all genres and labels)
    • Sub-clouds in each space based on individual socio-demographic characteristics

Mondo Breiger Workflow

Correspondence Plot of Genre Column Scores

First and Second Dimensions

Correspondence Plot of Genre Column Scores

First and Third Dimensions

Correspondence Plot of Label Column Scores

First and Second Dimensions

Correspondence Plot of Label Column Scores

First and Third Dimensions

Correspondence Plot of Genre Clouds

First and Second Dimensions

Correspondence Plot of Genre Clouds

First and Second Dimensions

Correspondence Plot of Genre Clouds

First and Second Dimensions

Correspondence Plot of Genre Clouds

First and Second Dimensions

Correspondence Plot of Genre Clouds

First and Second Dimensions

Correspondence Plot of Genre Clouds

First and Second Dimensions

Correspondence Plot of Genre Clouds

First and Second Dimensions

Correspondence Plot of Label Clouds

First and Second Dimensions

Correspondence Plot of Label Clouds

First and Second Dimensions

Correspondence Plot of Label Clouds

First and Second Dimensions

Correspondence Plot of Label Clouds

First and Second Dimensions

Correspondence Plot of Label Clouds

First and Second Dimensions

Individual Genre Cloud Typicality

Group-Specific Genre Clouds

Group-Specific Genre Clouds

Group-Specific Genre Clouds

Group-Specific Genre Clouds

Group-Specific Genre Clouds

Group-Specific Genre Clouds

Group-Specific Genre Clouds

Group-Specific Genre Clouds

Group-Specific Genre Clouds

Group-Specific Genre Clouds

Group-Specific Genre Clouds

Group-Specific Genre Clouds

Group-Specific Genre Clouds

Group-Specific Genre Clouds

Group-Specific Genre Clouds

Discussion and Conclusion

• Exploiting the sociological duality of genres and social labels to extract associational schemas for both

• Combining relational and geometric methods to examine central tendencies (centroids) and heterogeneity (variance) simultaneously

• Move across multiple levels

  • Individual (typical/atypical)
  • Category
  • Group (Demographics)
  • System

• Examining variation in meaning consensus across levels

Final Slide

• Thanks!