Geometric data analysis of French cultural practices
Transcription
Geometric data analysis of French cultural practices
GEOMETRIC DATA ANALYSIS IN THE STUDY OF SOCIAL SPACES Interpreting cultural practices and cultural inequalities Symposium on statistical Learning and Data Science, Florence, 9 May 2012 Frédéric Lebaron (CURAPP, CNRS/U. Picardie J. Verne, UMR 7319) [email protected] In close collaboration with Brigitte Le Roux and Philippe Bonnet. References SLDS, Florence, 7-9 May 2012 • Bourdieu, P. (1979). La distinction. Critique sociale du jugement. Paris: Minuit. • Le Roux, B. & Rouanet, H. (2010). Multiple Correspondence Analysis (QASS Series, n°163). CA:Thousand Oaks, Sage. • Rouanet, H., Ackermann, W., Le Roux, B. (2000). The geometric analysis of questionnaires: The lesson of Bourdieu’s La Distinction. Bulletin de Méthodologie Sociologique, 65, 5-18. • Le Roux, B. (2012), Structured Data Analysis, to appear in J.Blasius, M.Greeenacre (eds), forthcoming. 2 SLDS, Florence, 7-9 May 2012 OUTLINE • 1. Sociological stakes • 2. The data • 3. Construction of the space (MCA) • 4. Interpretation of axes • 5. Structured data analysis • 6. Study of a structuring factor: social class • 7. Cluster analysis (AHC) • 8. Class Specific Analysis (CSA) • 9. Conclusions 3 SLDS, Florence, 7-9 May 2012 4 1. Sociological stakes • In line with Bourdieu’s perspective in La distinction, we investigate the relationship between cultural practices, lifestyles on one side, economic, cultural, social and symbolic capital distribution inside French society on the other side. • To continue the analysis of social space structures as Bourdieu initiated it in the 70’s (« L’anatomie du goût », 1976) and in La distinction (1979) • Not only a methodological refinement: shed a new light on sociological problems at the core of the “Distinction model”. SLDS, Florence, 7-9 May 2012 5 1. Sociological stakes • What is a social class ? Does it “exist” and how ? Against the opposition “realism” / “nominalism”, a “structuralist constructivism”. • A fundamental given is inequalities in the distribution of various species of capital. • Since species of capital are various, their distribution defines a multidimensional social space. • Classes of dispositions (habitus) are based on similar classes of living conditions and experiences. Ex : “la bonne volonté culturelle” (“cultural goodwill”) as an “ideal-type” inside the “petite-bourgeoisie”. SLDS, Florence, 7-9 May 2012 6 1. Sociological stakes • In his work, Bourdieu uses very much the socio- occupational categories of INSEE (French NIS) at a detailed level (“les CSP”). • Classically he groups them into three large “classes” (“classes populaires” / “petite-bourgeoisie” / “classes dominantes”) ; he then divides each large class into “fractions” defined by their relative amount of cultural and economic capital. • A theoretical construction which directly links classes and capital distribution (level and composition). SLDS, Florence, 7-9 May 2012 7 2. The data • INSEE, "Enquête permanente sur les conditions de vie des ménages" ("Permanent survey on households living conditions" : French Data of the SILC survey 2003). • May 2003 : supplement on cultural and sport participation. • Data on 5626 individuals (aged 15 years old and more), and on their households. • Official objective : a better knowledge of cultural and sport activities through detailed information on households and individuals. SLDS, Florence, 7-9 May 2012 8 3. Construction of the space of cultural practices • Questions on : • “legitimate” practices such as literature, classical music, reading… • practices specific of a younger lifestyle (types of music, types of radio). • More popular practices, mostly related to television and radio. • Questions revealing preferences, i.e. TV channel mostly watched, type of music mostly listened to, etc. SLDS, Florence, 7-9 May 2012 Active questions • Space constructed from : • Television : • Entertainment : • Newspapers : • Books : • Music / Radio : • Total : 10 questions 8 questions 5 questions 8 questions 2 questions 33 questions • and 5497 individuals, aged 18 or more. 9 SLDS, Florence, 7-9 May 2012 10 Specific MCA • Used here to restrict the analysis to categories of interest • Infrequent and meaningless categories are treated as passive categories • Number of questions Q=33 (104 categories) • Number of active categories K=90 • Number of individuals N=5497 SLDS, Florence, 7-9 May 2012 11 4. Diagram of eigenvalues 0.180 0.160 Eigen values 0.140 0.120 0.100 0.080 0.060 0.040 0.020 0.000 1 4 7 10 13 16 19 22 25 28 31 34 37 40 43 46 49 52 55 58 61 64 67 70 Eigenvalue num ber Interpretation of axes SLDS, Florence, 7-9 May 2012 12 Contributions of the themes to the first three axes Axe 1 Axe 2 Axe 3 TV 12.3 38.9 48.8 SP (entertainment) 24.2 12.9 5.2 Newspapers & mag. 18.1 10.7 19.6 Books 37.9 8.8 13.9 7.4 28.6 9.6 Music and Radio SLDS, Florence, 7-9 May 2012 13 Axis 1 λ1 = 0.1666 36 categories, contributing 83.5% to the axis variance Intensity of cultural practices (especially legitimate practices, reading, theater, etc.) cultural capital axis SLDS, Florence, 7-9 May 2012 14 Axis 2 λ2 = 0.0717 29 categories, contributing 80.5% to the axis variance Culture of the youth opposed to more traditional culture SLDS, Florence, 7-9 May 2012 15 Axis 3 λ3 = 0.0607 28 categories, contributing 86.3% to the axis variance "hard" (dance, rock, concerts, TV M6, etc.) at bottom versus "soft" (TV news, romance reading, etc.) at top cultural practices related to age and gender SLDS, Florence, 7-9 May 2012 16 Main dimensions of the constructed space • Axis 1: intensity of cultural practices (especially “legitimate practices”, concert, reading, theater, etc.). Cultural capital. • Axis 2: culture of the youth opposed to more traditional culture. • Axis 3: "hard" (dance, concert, etc.) versus "soft" (TV news, romance reading, etc.) cultural practices (related to age and gender). SLDS, Florence, 7-9 May 2012 17 Cloud of individuals in plane 1-2 SLDS, Florence, 7-9 May 2012 18 Cloud of individuals in plane 1-3 SLDS, Florence, 7-9 May 2012 19 5. Structured data analysis • Structuring factors: refers to the notion “factor” in the analysis of variance, i.e. a partition of the set of individuals. • General approach: double breakdown of variance. • Allows to assess the importance of the effect of a factor, and to compare two (or more) factors. SLDS, Florence, 7-9 May 2012 Level of education as a structuring factor 20 Gender as a structuring factor SLDS, Florence, 7-9 May 2012 21 Age as a structuring factor SLDS, Florence, 7-9 May 2012 22 Category mean points of age in the space of individuals SLDS, Florence, 7-9 May 2012 23 6. Social class as a structuring factor • Approach based on structuring factor and double breakdown of variance. • Here, social class can be defined in various ways : 42 PCS, 8 PCS, 9 Bourdieu’s class (CSP), 3 large classes. • Other “class-schemes” could be tested (new European class-scheme, ISCO, etc.). 24 SLDS, Florence, 7-9 May 2012 Comparison of effects Axis 1 between within Axis 2 between/ total (in%) between within Axis 3 between/ total (in%) between within between/ total (in %) PCS42 0,0508 0,1159 30,46 0,0179 0,0538 24,91 0,0076 0,0531 12,53 PCS8 0,0314 0,1353 18,82 0,0144 0,0573 20,13 0,0062 0,0544 10,27 CSP9 0,0420 0,1247 25,17 0,0080 0,0637 11,12 0,0038 0,0569 6,24 CSP3 0,0284 0,1382 17,07 0,0013 0,0704 1,83 0,0007 0,0600 1,11 SLDS, Florence, 7-9 May 2012 25 Class ellipses SLDS, Florence, 7-9 May 2012 26 Coordinates of mean points, variances of the nine classes on the first three axes and double breakdown of variance SLDS, Florence, 7-9 May 2012 Class: concentration ellipses of workers middle class eco 27 other popular middle class intellectual students/pupils SLDS, Florence, 7-9 May 2012 Class: concentration ellipses of Middle upper class retired upper class 28 managers upper class int. SLDS, Florence, 7-9 May 2012 29 Class ellipses SLDS, Florence, 7-9 May 2012 30 Coordinates of mean points, variances of the three classes on the first three axes and double breakdown of variance SLDS, Florence, 7-9 May 2012 31 Conclusions of the comparison • Very low effect with the official 8-groups “PCS”, much higher with the “level 3”, i.e. 42-groups variable. (Why not even refer to the “level 4” (497) when possible ?) • A theoretical class-scheme based on Bourdieu’s theory and an aggregation from the “level 3” (42 categories) gives rather good results in the study of cultural practices. • Three classes is a much poorer partition, except for the interpretation of Axis 1: a simple summary ? SLDS, Florence, 7-9 May 2012 32 7. Cluster analysis • Another approach is to construct the ”social classes” on the basis of the observed cultural data. It leads to the family of cluster analysis methods. • Here, the classes are constructed from questions about cultural practices. • General question: is the classification based on cultural practices (strongly or not) related to the division of society into social classes ? SLDS, Florence, 7-9 May 2012 33 7. Cluster analysis Ascending Hierarchical Clustering (AHC) using Ward’s method • Distances between individuals are the distances used in MCA (all the axes). • The aggregation index is based on the between variance. SLDS, Florence, 7-9 May 2012 4 clusters retained 34 SLDS, Florence, 7-9 May 2012 Clusters concentration ellipses in plane 1-2 35 SLDS, Florence, 7-9 May 2012 36 Coordinates of mean points, variances of the four clusters on the first three axes and double breakdown of variance SLDS, Florence, 7-9 May 2012 37 To interpret classes, one proceeds as follows: For each class c, compare the relative frequency of the category k ( ) for individuals belonging to class c to the one ( ) for all individuals. Descriptively, the deviation between category k in class c and category k in the overall set of individuals is said to be large if > 0.05 or if / >2. For the categories with large deviations, we perform the typicality test (the combinatorial test of comparison of a frequency to a reference frequency), hence a combinatorial p-value (one-sided). If p < 0.025, the frequency is statistically greater than the reference frequency with (S*) if 0.005 < p < 0.025 and (S**) if p < 0.005. The categories for which the deviation is descriptively large and statistically significant are said to be ‘over-represented’. In the same way, they are said to be ‘under-represented’ if - <-0.05 or if / <0.5, and if the result of the test is significant. The interpretation of classes based on the sectors that are overrepresented. SLDS, Florence, 7-9 May 2012 38 Cluster 1 Cluster 1 SLDS, Florence, 7-9 May 2012 39 In Cluster 1 one finds 34 categories that are over-represented. These categories concern intense cultural practices (participation and diversity): theater on TV, art programs, Arte channel, never games, sitcoms or clips; going to opera, theater, concert, dance performance, cinema; preferring classical music; reading history, art or politics books, classical literature and dayly national newspaper. The categories "high education", "above 46 years" and "high income" are overrepresented. SLDS, Florence, 7-9 May 2012 40 Sociological interpretation of cluster 1 • A class of individuals defined by legitimate cultural practices: “le goût dominant” (“the dominant taste”). Highbrow culture. • Closely related to very high levels of educational capital (“grandes écoles”). • Related to dominant classes, esp. managers. SLDS, Florence, 7-9 May 2012 41 Cluster 2 Cluster 2 SLDS, Florence, 7-9 May 2012 42 In cluster 2, there are 30 categories over represented. The categories concern the reading of various sorts of books, an active use of cultural programs on TV channels, the rejection of popular TV programs, a moderate reading of cultural magazines and national newspapers, cinema attendance, taste for rock music, comic strips... SLDS, Florence, 7-9 May 2012 43 Sociological interpretation of cluster 2 • A class of individuals defined by an attempt to be conform to a more legitimate definition of culture : “la bonne volonté culturelle” (cultural goodwill), illustrated by educational and cultural ascetic dispositions. • Closely related to intermediary and high levels of educational capital, and to the middle and upper classes. SLDS, Florence, 7-9 May 2012 44 Cluster 3 Cluster 3 SLDS, Florence, 7-9 May 2012 45 In cluster 3, one finds 21 categories that are over-represented. These categories concern cultural practices associated with youth culture: TV channel M6, watching films, clips, sitcoms, games; listening to radio for music; liking international pop music, or techno, world or rap music. SLDS, Florence, 7-9 May 2012 46 Sociological interpretation of cluster 3 • A class of individuals defined by the “common” or “popular” cultural practices of the youth, esp. around specific TV programs and music. • Closely related to the younger ages but also to people with vocational training. SLDS, Florence, 7-9 May 2012 47 Cluster 4 Cluster 4 SLDS, Florence, 7-9 May 2012 48 Categories of weak cultural practice are overrepresented: not listening to music; not going to cinema; not reading; watching TV channels FR3 (regional) or TF1 (popular); reading regional newspaper; not listening to radio except news. The categories of low educational level (primary or no diploma), age 66-76 and >76 years are overrepresented. 36% of individuals of the working class are in this cluster. SLDS, Florence, 7-9 May 2012 49 Sociological interpretation of cluster 4 • A class of individuals defined by the absence of “legitimate” cultural practices and an over-representation of individuals watching the commercial popular TV channel (TF1) or the public regional programs (France 3): “le goût de nécessité” (“necessity taste”)? • Closely related to the older ages, but also to very low educational capital and to the working class. SCUD 20/05/2012 seminar, York, 24-25 May 2011 50 8. Class-Specific Analyses - We need for more specific analyses of particular occupational groups (example: blue-collar workingclass, managers…). - The “managers” (les patrons): two PCS (« chefs d’entreprise de plus de 10 salariés », « cadres administratifs et commerciaux d’entreprise »). - Number of specific individuals: 180. 51 Thirty years after Distinction, november 2010 Motivation for the use of ClassSpecific Analysis The distance between two individuals inside a specific class (e.g. the managers) makes sense in the global space, but not in a space restricted to these particular individuals. To operationnalize this idea, the CSA consists in searching the principal dimensions of a sub-cloud without « extracting » it from the global space (keeping the distance as defined in the global space). See the sub-cloud inside the global space but projected onto its own principal dimensions. Then we apply the usual GDA methodology: interpretation of axes, structured data analysis. 52 SCUD 20/05/2012 seminar, York, 24-25 May 2011 Managers SCUD 20/05/2012 seminar, York, 24-25 May 2011 Managers: first specific axis 53 SCUD 20/05/2012 seminar, York, 24-25 May 2011 Managers: second specific axis 54 SCUD 20/05/2012 seminar, York, 24-25 May 2011 Managers: third specific axis 55 SLDS, Florence, 7-9 May 2012 56 Sociological interpretation of Axes 1 to 3 • Axis 1 is an axis of “legitimate” cultural practice. • Axis 2 is an axis of intensity of global cultural activity, opposing intense practices (including “matter-of-fact” and professional activities like reading national newspapers) to more modest practices. • Axis 3 is an axis opposing modern and traditional musical and cultural genres. SCUD 20/05/2012 seminar, York, 24-25 May 2011 Managers: age 57 SCUD 20/05/2012 seminar, York, 24-25 May 2011 Managers: age 58 SCUD 20/05/2012 seminar, York, 24-25 May 2011 Managers: education 59 SCUD 20/05/2012 seminar, York, 24-25 May 2011 Managers: gender 60 SLDS, Florence, 7-9 May 2012 61 Sociological interpretation of Axes 1 to 3 • Axis 1 first relates to educational capital. • Axis 2 first relates to gender and age . • Axis 3 first relates to age. => The structure of the specific sub-space seems relatively close (Axis 1, plane 2-3) to the structure of the global space: « structural homology ». SLDS, Florence, 7-9 May 2012 62 Conclusions • Individuals are where the information comes from: a systematic investigation of social class as a structuring factor allows to decompose and compare the effects. • Relevance of Bourdieu’s theory of capitals to refine class-schemes and investigate the cultural dimension of social classes. • A cluster analysis based on cultural practices allows to approach the consistence of particular Bourdieusian “ideal-types” (and symbolic categories). • CSA allows to complete the analysis through the study of specific sub-groups, like the managers, showing structural homologies.