Presentation on theme: "Is Free School Meal Status a Valid Proxy for Socio-Economic Status? Graham Hobbs and Anna Vignoles Centre for the Economics of Education (funded by Department."— Presentation transcript:
Is Free School Meal Status a Valid Proxy for Socio-Economic Status? Graham Hobbs and Anna Vignoles Centre for the Economics of Education (funded by Department for Education and Skills)
Family SES potentially key determinant of pupil attainment FSM proxies family SES in much UK educational research: –FSM gap –school composition effects –social segregation FSM as a control variable –Ethnicity gap –class size effects/school resource effects, –school effects CEE FSM Project
CEE Project SES measures are generally missing from administrative data sets Project assesses the biases that may arise from using FSM as a proxy in particular contexts Little in the way of formal evaluation of FSM, with notable exceptions –DfES Statistics of Education: Trends in Attainment Gaps: 2005 –Croxford 2000
CEE Project Examine FSM eligibility criteria Explore national data on benefits to determine what FSM is actually measuring Explore these relationships in ALSPAC-NPD merged data Present a framework for assessing the validity of proxy variable as a control variable in OLS regression models, recognising that the validity of FSM as a proxy for family SES is model-specifi Assess the parameter on FSM when FSM is the variable of interest.
1.Merged ALSPAC-NPD data 2.FSM eligibility rules and claimant data 3.FSM status and family SES in ALSPAC-NPD data 4.Proxy variable framework 5.Illustrative example Outline
ALSPAC eligibility: Mother resident in Avon. Expected delivery Apr 1991 to Dec ALSPAC enrolment: 14,049 live births ALSPAC data: Family income, presence of partner, mother & partner employment, mother & partner education, mother & partner social class NPD data: FSM, age, gender, ethnicity, EFL, SEN, & KS2 & KS1 attainment Merged ALSPAC-NPD data
Only 11,131 of 14,049 successfully matched to NPD –Slightly non-random sample Only 5,924 of 14,049 responded to all SES variables –Item & wave non-response –Strongly non-random sample Family SES observed in pregnancy or aged 4-4½. FSM measured aged Missing data and timing of data
FSM measure: Eligible & claiming FSM FSM eligibility 2005/6: –Eligible & claiming IS or IB-JSA –Child Tax Credit (but not Working Tax Credit) & income less than £13,480 –Guarantee element of State Pension Credit –Support under part VI of Immigration & Asylum Act 1999 FSM eligibility 2001/2: Eligible & claiming IS or IB- JSA FSM measure and eligibility rules
No-one working more than 24 hours per week, a low income & limited capital assets (benefit unit) 92% of dependents of IS & IB-JSA claimants were dependants of IS claimants 83% of IS claimants with dependents were single parents Average IS payment £106 per week for single parents 34% of lone parent claimants received IS for 5 years IS/IB-JSA eligibility and claimant data
Conclusions from eligibility rules Children in families claiming either IS or IB-JSA were eligible for FSM Majority of these children were in families with a single parent, aged 25-59, and one or two children Given IS and IB-JSA eligibility rules, all of the children should be in families without a parent in full-time employment, with low incomes and limited capital assets.
Potential reasons for non claiming Stigma effects Stigma effects varying by school Concerns about diet – differing by ethnic/ cultural background Story and Chamberlin 2001/ Croxford 2000
FSM status and family SES in ALSPAC
Ordinal associations between FSM status and family SES
Potential reasons for discrepancies Non claiming Misreporting of income etc. in ALSPAC Timing of data –For time-varying family SES variables, the true association between FSM status and family SES is likely to be stronger than reported here –Observe the family SES variables in pregnancy or between ages 4–4½ –Observe FSM status between ages 9–11 Sample selection
Structural equation: y = β 0 + β 1 x 1 + … + β K x K + γq + v Linear projection: q = θ 0 + ρ 1 x 1 + … + ρ K x K + θ 1 z + r Proxy variable regression: y = (β 0 + γθ 0 ) + (β 1 + γρ 1 )x 1 + …+ (β K + γρ K )x K + γθ 1 z + (γr + v) Proxy variable framework
q variables i.e. true SES variables Family income Parental education Social class Parental employment One parent status
Structural equation: Proxy variable framework M linear projections: m=1,…,M Proxy variable regression:
Illustrating FSM as a proxy variable The validity of FSM as a proxy for family SES is model-specific Illustrate with example from a (specific) value-added model of Key Stage 2 attainment Assessing the validity of FSM when FSM is a control variable
Findings In the context of this specific value-added model of KS2 attainment, FSM is an imperfect proxy for family SES variables Estimates of parameters on some variables are biased as a result of using FSM Large biases on SEN and, in some subjects, ethnicity (but small number of minority ethnic group children in our estimation sample prevents us from making strong inferences for other studies)
FSM as variable of interest What is the parameter on FSM in the proxy variable regression is actually measuring?
Decomposition of FSM differences in KS2 attainment
Findings Around half of conditional and unconditional FSM differences in KS2 attainment are accounted for by the eight family SES variables (with the exception of conditional FSM differences in KS2 maths) Conditional FSM differences in KS2 attainment are principally accounted for by mothers education and family income