Poverty dynamics, family background and attainment: BHPS evidence

Stephen P. Jenkins

(Institute for Social and Economic Research, University of Essex)

ESRC Research Methods Festival, Oxford, 19 July 2006

2

British Household Panel Survey (BHPS)• Representative sample of GB private household population in

1991• First wave of interviews in Autumn 1991 c. 5,500 households

with c. 10,000 adult respondents, plus children• Original sample members followed and re-interviewed annually

thereafter• New sample members:

– children turning 16– others joining panel households (e.g. by marriage)

• Retrospective work, job and partnership histories covering pre-panel period

• Youth self-completion questionnaire (11–15 year olds)– the ‘British Youth Panel’, 1994–

• Extension samples – Scotland and Wales, 1999 (1,500 households each)– Northern Ireland, 2001 (2000 households)

3

The BHPS provides both short- and long-term perspectives on life course dynamics

• Short-term (using data from the annual interviews): – Example 1: describing and modelling how incomes change

from one year to the next, especially movements in and out of low income

• Long-term/intergenerational (combine annual interview data from the main panel survey with retrospective history data for parents covering pre-panel years)– Example 2: how various dimensions of family background

during a respondent’s childhood are related to outcomes later in the respondent’s life

Policy relevance: the outputs discussed in the two examples have been widely used and cited by research users in government and elsewhere

Short-term perspectives from the BHPS

Example 1: poverty dynamics

5

Equ

ival

ent h

ouse

hold

net

inco

me

Wav

e 2

Equivalent household net income Wave 10 100 200 300 400 500 600 700 800 900 1000

0

100

200

300

400

500

600

700

800

900

1000

Cross-sectional stability hides longitudinal flux

• Lots of short-distance mobility; little long-distance mobility

• Substantial turnover in the low income population • … or is it just measurement error or transitory variation?

Low incomethreshold

6

Income mobility (2-year averaged incomes)

w1-w2income

w3-w4 income relative to w1 mean

relative tow1 mean

< 0.5 0.5–0.75

0.75–1.0

1.0–1.25

1.25–1.5

>1.5

All (col%)

< 0.5 50 41 7 1 1 1 100 (15)0.5–0.75 14 55 23 6 1 2 100 (25)0.75–1.0 5 22 46 19 5 3 100 (21)1.0–1.25 3 6 19 45 17 10 100 (15)1.25–1.5 2 4 6 28 32 29 100 (10)> 1.5 1 1 4 7 16 71 100 (14)

All 13 26 20 16 9 16 100 (100)

Transition matrix (row percentages)

• 50% of those with income below half 1991 average moved out• 25% of those with income above half 1991 average moved in • 51% of sample remained in same income group; 92% remained in same or adjoining income group• Lots of short-distance mobility; little long-distance mobility, a.k.a. the “rubber band” model of income dynamics

7

Major approaches to examining the ‘causes’ of poverty dynamics

(all have been applied to BHPS: see reference list at end)

• Trigger event classifications of poverty entries and exits– According to whether poverty transitions associated with different

types of income event or demographic event:1. Hierarchical classifications2. Non-hierarchical classifications

E.g. Jenkins & Rigg (2001)

• Multivariate regression models1. Hazard regression models 2. Markovian transition probability models (variant of 1)3. Variance component models4. ‘Structural’ models

E.g. Jenkins & Rigg (2001), Devicienti (2001), Cappellari & Jenkins (2004), Aassve et al. (2005)

8

Trigger events and poverty exits/entries

• Both income events and demographic events are important

• Demographic events more important for entries

• Earnings changes very important – not only the head’s, but also those of others in household

• Importance varies by household type (not shown)

Main event associated with poverty transition

Poverty Exits (%)

Poverty Entries (%)

HH head’s labour earnings rose/fell 33 26 Spouse’s / other labour earnings rose/fell 29 18 Non-labour income rose/fell 20 17 Demographic event 19 39 All 100 100 Source: Jenkins and Rigg (DWP Research Report 157, 2001), applying methods of Bane and Ellwood, JHR, 1986. BHPS waves 1–9. Poverty line = 60% contemporary median

9

The importance of repeat poverty spellsNumber of years poor out of the next eight for a

cohort entering povertyPredicted

Number of years

Singlespell(%)

Repeatedspell(%)

Actual(%)

1 53.7 19.0 25.02 30.6 17.7 17.43 15.6 15.1 12.04 7.5 13.5 13.55 3.9 11.3 6.66 2.7 8.7 10.57 2.7 6.9 9.88 7.8 7.8 5.2

Total 100.0 100.0 100.0Mean (# years poor) 2.4 3.7 3.6

Predictions based on lifetable estimators of exitrate (col. 2) and exit and re-entry rates (col. 3).Jenkins and Rigg (2001), BHPS waves 1–9.Poverty line = 60% contemporary median.

Only by takingaccount of repeatpoverty spells,does one get goodestimates of the total time spentpoor over a period

10

Mean number of years poor out of eight for a cohort of poverty entrants, by household type

(mixture hazard regression model estimates)Household type of individual(characteristics measured at start of spell)

Mean

Two adult household1. Both working, no children 1.32. Both working, 1 child aged under 6 1.63. Both working, 2 children, 1 aged under 6 1.94. Head working but no others, no children 1.65. Head working but no others, 2 children, 1

aged under 62.8

6. No one working, no children 2.37. No one working, 2 children, 1 aged under 6 3.8Lone parent household8. Not working, 1 child aged under six 3.69. Not working, 2 children, 1 child aged under

six4.2

10. Not working, 2 children, 1 child aged undersix, no A-levels

4.8

11. Working, 2 children, 1 aged under six 3.3Elderly household12. Head aged 60–64, head working 3.013. Head aged 60–64, no one working 3.514. Head aged 80+ years, no one working 4.6Predictions based on mixture hazard regression modelsof poverty exit and entry rates. Jenkins and Rigg (2001),BHPS waves 1–9. Poverty line = 60% contemporarymedian.

Substantial heterogeneityin experience of povertyover time

More time spent poor is associated with:• having more kids• not working• being a lone parent• being elderly

11

Policy implications• Importance of labour market as a route out of poverty for

those of working age

• Policy based around labour market insufficient for all vulnerable groups (e.g. elderly, sick, disabled, carers)

• Important to see individuals in terms of their household context (affects number of income-bringers, and needs)

• Spell repetition (and decline in re-entry rates with time non-poor) remind us of the importance of measures preventing entries into poverty (not just helping exits) – ‘real’ jobs and job retention, not just promoting moves into employment

• Exit rates fall with time poor identify potential long stayers early and target them

Intergenerational perspectives from the BHPS

Example 2: the effects of family background during childhood on later-life outcomes

13

Effect of childhood poverty on young adults’ outcomes

Stage when ‘poverty’ was measured Outcome

Sex

All ages 0–5 6–10 11–15

Leave parental home M +++ +++ F + Education M – – – – – – – F Economic inactivity M +++ +++ +++ F +++ +++ Early childbearing F +++ +++ Smoking M + F + Psychological distress M F +(++)/–(– –): (large) positive/negative effect of poverty on the probability of the outcome. (Large/moderate refer both to the magnitude and to statistical significance of the effect.) Digest of results from various models. ‘Poverty’: both parents have 1+ months of unemployment in same childhood year. Source: Ermisch, Francesconi, and Pevalin (2001).

• No effects for some outcomes (blanks),but • some large effects

Variation of effects by:• outcome• sex of young adult• childhood stage when poverty experienced

14

Effect of living in a lone parent family during childhood on young adults’ outcomes

Individual sample Sibling sample

(1) (2) (3) Outcome M.E. |t| M.E. |t| Coeff |t| Education (A-level +) –0.14 2.9 –0.08 1.7 –0.146 1.9 Economic inactivity 0.06 3.9 0.04 2.6 0.018 0.9 Birth before 21 (women)

0.02 2.3 0.01 1.5 0.024 1.7

Distress (GHQ > 2) 0.06 2.5 0.06 2.5 0.036 1.4 Smoking 10+ a day 0.07 2.8 0.06 2.3 0.068 2.7 Regression (1) also includes age, sex, birth year, mother’s education. (2) as (1) plus % time mother worked, mother’s age at kid’s birth, family income when kid aged 16, home ownership, … (3) Within-mother FE regression. Source: Ermisch and Francesconi (2000).

• Experience of life in lone parent family usually associated withdisadvantageous outcome

• Most of unfavourable effect linked to when kid aged 0–5• For most outcomes, adverse effect persists when control for

economic conditions of family (model 2)

15

Effect of living in a lone parent family during childhood on young adults’ outcomes:

Britain versus Germany

Outcome Britain West German sample

Educational qualifications to university entry level

Adverse Adverse?

Non-employment Adverse Adverse?

Smoker Adverse Adverse“Adverse”: adverse effect is statistically significant according to both level and sibling

difference estimates.“Adverse?”: adverse effect according to level estimate but no statistically significant

effect according to sibling difference estimate.Sources: Ermisch, Francesconi & Pevalin (JRSSA, 2004), Francesconi, Jenkins and

Siedler (Anglo-German Foundation, 2005)

16

Policy implications

• Both poverty and family structure during childhood appear to have effects on selected later-life outcomes in Britain (though not necessarily in Germany)

• Why poverty has these effects not yet clear e.g. is it what money buys, or is money a proxy for other things?

• You need to be careful when drawing conclusions – Example: Ermisch and Francesconi (JRF, 2000) found that kids with

mothers who worked full-time during childhood were later disadvantaged in some respects. Different readers drew different conclusions from same ‘fact’ !

• ‘mothers should stay at home’?

• ‘what about more input from fathers’?

• ‘increase availability of quality child care facilities’?

17

Data sources

BHPS data are available from the UK Data Archive :http://www.data-archive.ac.uk/findingData/bhpsTitles.asp

• Main release data for waves 114 (survey years 19912004)

• Derived ‘net income’ variables to accompany the BHPS (Jenkins et al. data)

18

Selected referencesPoverty dynamics

Aassve, A., Burgess, S., Dickson, M. and Propper, C. (2005). ‘Modelling poverty by not modelling poverty: an application of a simultaneous hazards approach to the UK’, ISER Working Paper 2005-26, University of Essex, Colchester. http://www.iser.essex.ac.uk/pubs/workpaps/pdf/2005-26.pdf

Cappellari, L. and Jenkins, S.P. (2004). ‘Modelling low income transitions’, Journal of Applied Econometrics, 19, 593–610.

Devicienti, F. (2001). ‘Estimating poverty persistence in Britain’, LABORatorio Riccardo Revelli (Centre for Employment Studies), Working Paper 1, Torino. http://www.labor-torino.it/workingpapers/wp1.htm

Jenkins, S.P. (2000). ‘Modelling household income dynamics’, Journal of Population Economics, 13, 529–567.Jenkins, S.P. and Rigg, J.A. (2001). The Dynamics of Poverty in Britain, DWP Research Report No. 157.

Corporate Document Services, Leeds. http://www.dwp.gov.uk/asd/asd5/rrep157.asp

Intergenerational transmission

Ermisch, J.F. and Francesconi, M. (2001a). ‘Family structure and children’s achievements’, Journal of Population Economics 14, 249–70.

Ermisch, J.F. and Francesconi, M. (2001b). ‘Family matters: impacts of family background on educational attainments’. Economica 68, 1372–56.

Ermisch, J.F., Francesconi, M., and Pevalin, D.J. (2001). Outcomes for Children of Poverty, Department for Work and Pensions Research Report No. 158. Corporate Document Services, Leeds. http://www.dwp.gov.uk/asd/asd5/rrep158.asp

Ermisch, J., Francesconi, M., and Pevalin, D.J. (2004). ‘Parental partnership and joblessness in childhood and their influence on young people’s outcomes’, Journal of the Royal Statistical Society A, 167(1): 69–101.

Francesconi, M., Jenkins, S.P., and Siedler, T. (2005) The Impact of Family Structure during Childhood on Later-Life Attainment, Anglo-German Foundation, London. http://www.agf.org.uk/pubs/pdfs/1414web.pdf