Multilevel modelling of social networks and occupational structure Dave Griffiths¹, Paul S. Lambert¹ & Mark Tranmer¹ ² ¹ School of Applied Social Science, University of Stirling ² CCSR& Mitchell Centre for Social Network Analysis, University of Manchester Work for this paper is supported by the ESRC as part of the project ‘Social Networks and Occupational Structure’, see http://www.camsis.stir.ac.uk/sonocs/
23
Embed
Multilevel modelling of social networks and occupational structure Dave Griffiths¹, Paul S. Lambert¹ & Mark Tranmer¹ ² ¹ School of Applied Social Science,
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Multilevel modelling of social networks and occupational
structureDave Griffiths¹, Paul S. Lambert¹ & Mark Tranmer¹ ²
¹ School of Applied Social Science, University of Stirling² CCSR& Mitchell Centre for Social Network Analysis, University of
Manchester
Work for this paper is supported by the ESRC as part of the project
‘Social Networks and Occupational Structure’, see http://www.camsis.stir.ac.uk/sonocs/
Registered nurses 3.9% 4.5% 56.8Nursing, psychiatric and home healing assistants
3.9% 1.9% 42.6
Secretaries 3.9% 5.3% 55.5Customer service representatives 3.6% 1.7% 51.8Receptionists 3.2% 1.6% 53.2Cashiers 3.2% 1.8% 41.3Labourers 2.9% 0.4% 32.0Janitors and building cleaners 2.5% 1.7% 32.5Maids and housekeeping cleaners 2.2% 0.3% 27.4Retail salespersons 2.2% 1.9% 51.9Tellers 2.2% 0.6% 46.3
Most common occupations for the wives of lawyers and labourers in the USA
Source: Current Population Survey 2010.
Contexts and identities
• Family– Are identities formed by family background?
• Social background?
• Wider social network– Are identities shaped by those we associate with?
• Social capital?
• Occupation– Are identities formed through our choice of
occupation?• Social status?
British Household Panel Survey• Ran from 1991 to 2008• Selected 5,500 initial households (plus later
booster households for regions/minorities)• All initial sample members interviewed each year
– Any they cohabit with also interviewed• Around 30,000 different people interviewed• Personal identifiers (PID) are for life; household
identifiers (HID) alter each year• This enables us to link together individuals into
networks
Geller Household: Initial household
Geller households:(up to 1995(ish))
Grouped by cohabitation networks
Grouped by family ties
Grouped by occupation
BHPS respondents 26,090
People cases 90,784 Largest Mean
People-job cases 347,542 22 3.8
Occupations (SOC) 374 5,176 235
Networks identified (NID) 9,846 36 2.7
Families identified (FID) 12,096 19 2.2
Data extracted from the British Household Panel Survey, 1991-2008 waves
20 40 60 80 100CAMSIS of SOC
0 10 20 30 40subjective wellbeing (ghq) 1: likert
1 2 3 4 5how often: walk/swim/play sport
1 2 3 4 5financial situation
1 2 3 4 5strong trade unions protect employees
1 2 3 4 5family suffers if woman works full time
CAMSIS score of occupational advantageSelf-rated healthParticipation in exerciseFeeling financial secureAttitudes towards trade unionismAttitudes towards motherhood and employment
• 7 models to measure outcome:Controls: age, gender and CAMSIS scoresLevels:
1. None2. Family3. Network4. Occupation5. Family and network6. Family, network and occupation7. Family, network, occupation and occupation-by-gender
Outcome 1: CAMSIS score (scale from 1 to 99, modelled as linear scale)
n of FID groups 12096 12096n of NID groups 9846 9846Analysis for BHPS respondents (panel 1) and for all strong and weak ties identified (panel 2) (scale from 1 to 99, modelled as linear scale)
Outcome 2: GHQ score (scale from 0 to 36, 36=healthiest, modelled as linear scale) (1) (2) (3) (4) (5) (6) (7)
ID variance ICC 100% 89.3% 91.2% 99.8% 89.3% 89.1% 89.3%FID variance ICC 10.7% 9.4% 9.3% 9.4%NID variance ICC 8.8% 1.3% 1.3% 1.3%SOC variance ICC 0.2% 0.2% 0.1%Fem | soc variance 0.03%Notes: For model (7), the ICC estimates refer to variance proportions for males at the intercept (due to the ‘random coefficients’ formulation of that model).
Outcome 3: Scale ranking for self-rated sports participation level (scale from 1 to 5, 1=very active, modelled as linear scale)
ID variance ICC 100% 71.6% 74.2% 99.1% 71.4% 71.2% 70.9%
FID variance ICC 28.4% 19.8% 19.3% 19.9%
NID variance ICC 25.8% 8.7% 8.9% 8.3%
SOC variance ICC 0.9% 0.6% 1.0%
Fem | soc variance 0.3%
Notes: For model (7), the ICC estimates refer to variance proportions for males at the intercept (due to the ‘random coefficients’ formulation of that model).
Outcome 4: Scale ranking for self-rated level of ‘financial security’ (scale from 1 to 5, 5=lowest security, modelled as linear scale)
(1) (2) (3) (4) (5) (6) (7)Intercept 2.26* 2.27* 2.25* 2.24* 2.26* 2.25* 2.25*(Age – 40)/100 -0.74* -0.67* -0.67* -0.73* -0.66* -0.66* -0.67*(CAMSIS -50)/10 -0.13* -0.11* -0.12* -0.13* -0.11* -0.11* -0.11*(Age*CAMSIS)/1000 -0.08* -0.10* -0.10* -0.08* -0.10* -0.09* -0.10*Deviance 70424 70606 71407 70406 70351 70324AIC 71511 70436 70618 71419 70420 70367 70344ID variance ICC 100% 75.1% 80.3% 98.9% 75.0% 74.8% 74.5%FID variance ICC 24.9% 20.2% 19.7% 19.8%NID variance ICC 19.7% 4.8% 4.7% 4.6%SOC variance ICC 1.2% 0.8% 1.0%Fem | soc variance 0.1%Notes: For model (7), the ICC estimates refer to variance proportions for males at the intercept (due to the ‘random coefficients’ formulation of that model).
Outcome 5: Scale ranking for attitudes towards ‘families suffer if the mother works full time’ (scale from 1 to 5, 1=strongly agree, modelled as linear scale)
ID variance ICC 100% 83.5% 85.2% 99.3% 83.4% 83.2%
FID variance ICC 16.5% 11.8% 11.6% 11.6%
NID variance ICC 14.8% 4.8% 4.7% 4.4%
SOC variance ICC 0.7% 0.5% 0.3%
Fem | soc variance 0.2%
Notes: For model (7), the ICC estimates refer to variance proportions for males at the intercept (due to the ‘random coefficients’ formulation of that model).
Outcome 6: Scale ranking for attitudes towards ‘strong trade unions protect employees rights’ (scale from 1 to 5, 1=m, modelled as linear scale)
ID variance ICC 100% 80.3% 81.2% 96.0% 80.0% 78.3% 77.3%
FID variance ICC 19.7% 8.5% 7.3% 7.0%
NID variance ICC 18.8% 11.5% 10.9% 10.9%
SOC variance ICC 4.0% 3.5% 4.3%
Fem | soc variance 0.5%
Notes: For model (7), the ICC estimates refer to variance proportions for males at the intercept (due to the ‘random coefficients’ formulation of that model).
CAMSIS Health Sports Financial security
Working mothers
Trade unions
ID variance ICC 71.3% 89.1% 71.2% 74.5% 83.2% 77.3%
FID variance ICC 7.9% 9.3% 19.3% 19.8% 11.6% 7.0%
NID variance ICC 20.8% 1.3% 8.9% 4.6% 4.7% 10.9%
SOC variance ICC 0.2% 0.6% 1.0% 0.5% 4.3%
Fem | soc variance 0.1% 0.5%
• Weak ties associated with occupational position and attitudes towards trade unionism
• Strong ties associated with health, fitness and financial perspective outcomes and attitudes to family roles
• Occupational role not overly important in measuring outcomes– But, is this captured by controlling for CAMSIS
position?
Next steps
• Distinguishing between family positions?• Multiple-membership models
– families? – households? – occupations?
• Controlling for types of initial households – are those consisting of one family different?