Investigating the impact of housework on wages: longitudinal evidence for Britain

1
Investigating the impact of housework on wages
longitudinal evidence for Britain

• Mark Bryan
• Institute for Social and Economic Research,
  University of Essex
• Almudena Sevilla Sanz
• University of Oxford
• ESRC Research Methods Festival
• Oxford, 2nd July 2008

2
Introduction

• Gender pay gap in the UK is (still) about 18
  (full time workers).
• Traditional explanation (based on differences in
  human capital and job characteristics) can only
  explain part of the gap.
• Theory (e.g. Becker, 1985) suggests housework may
  lower wages (holding constant human capital and
  hours of work) because it restricts energy and
  flexibility available for labour market activity.
• So womens greater housework burden may
  contribute to gender pay gap.

3
Our aims and contribution

• Investigate effect of housework on wages in
  Britain (evidence already exists for US and
  Denmark).
• Investigate channels by which housework effects
  operate. Amount, type or timing of housework?
• We know (and show) that marriage involves
  specialisation into different types of housework
  tasks done at different times. So investigate
  housework effects for men and women, and for
  marrieds and singles.
• See Bryan and Sevilla Sanz (2008) for full
  details http//www.iser.essex.ac.uk/pubs/workpaps
  /pdf/2008-03.pdf
• Focus on methodology in this presentation.

4
Data

• British Household Panel Survey (BHPS), waves 2-14
  (1992-2004)
• Full time employees (16-59 years for women and
  16-64 years for men) with full work history (and
  other vars) 2574 men (observed over 7.0 waves on
  average) and 2191 women (5.5 waves).
• Separate equations by gender and marital status
  to allow for specialisation in housework. Married
  married or cohabiting
• Housework question About how many hours do you
  spend on housework in an average week, such as
  time spent cooking, cleaning and doing the
  laundry?

5
Wages and housework
Variable Women Men Single women Single men Married women Married men
Log wage 2.040 (0.532) 2.207 (0.555) 1.939 (0.559) 1.941 (0.558) 2.094 (0.510) 2.309 (0.519)
Housework (hours) 10.638 (7.856) 4.659 (4.576) 7.125 (6.685) 4.382 (4.633) 12.507 (7.793) 4.764 (4.550)
Observations 12123 18030 4209 4979 7914 13051
Individuals 2191 2574 1115 1179 1585 1903
6
Descriptive wage equation

• wit xit'ß ?hit vit (1)
• wit is log hourly wage
• xit is human capital, experience, year, region
  (and job characs)
• hit is weekly hours of housework
• vit is random error term

7
Descriptive wage equation

• wit xit'ß ?hit vit (1)
• Estimate (1) by OLS using pooled data (all waves
  combined). So we treat differences in wages (etc)
  between people the same as changes in wages (etc)
  over time for same person. Advantage of panel
  data here is just to give us more observations.
• In (1), ? tells us difference in wages associated
  with a one-hour difference in housework, within
  groups of similar workers same quals,
  experience, industry etc (depending on variables
  in x).

8
Association of housework with wages (OLS)
Single women Single men Married women Married men
Basic spec -0.0017 (-0.82) 0.0036 (1.79) -0.0108 (-8.81) -0.0091 (-5.25)
Basic spec job characs -0.0007 (-0.48) 0.0015 (0.85) -0.0074 (-7.66) -0.0066 (-4.97)

• Basic specification includes experience
  (squared), education, region and year.
• Job characteristics are one-digit occupation and
  industry, public sector employment, establishment
  size, trade union coverage, and temporary and
  fixed-term employment.
• t-statistics in parentheses

9
Percentage of gender wage gap explained by
characteristics (OLS)
Coefficients from Womens Equation Coefficients from Mens Equation Coefficients from Pooled Equation
Excluding housework 19.2 -15.0 30.5
Including housework 35.9 12.0 55.7
10
Descriptive wage equation

• wit xit'ß ?hit vit (1)
• Can ? tell us what would happen if housework
  changed (but nothing else did)? Probably not
  (though depends on how well x controls for
  differences between workers).
• Specifically, ? will not give us the causal
  effect of housework on wages if there are
  unobserved differences between workers which
  affect wages and are also related to housework.
  Example career orientation.

11
Wage equation with unobserved effect

• wit xit'ß ?hit µi eit (2)
• To progress further, need to think about and
  investigate unobservables in more depth. Panel
  data enable us to do this.
• Panel data structure enables us to split error
  term into two components
• µi is unobserved individual effect
  (heterogeneity), e.g. unobserved skills,
  motivation, labour mkt resources. Assume constant
  over time!
• eit is random error. Assume (for now) unrelated
  to explanatory variables.

12
Wage equation with unobserved effect

• wit xit'ß ?hit µi eit (2)
• Estimate (2) as fixed-effect (FE) model using
  within estimation (xtreg,fe in Stata)
• Only uses variation (changes) within
  individuals over time.
• In FE model, ? tells us change in wages
  associated with a one-hour change in housework
  (and no other changes) for a given individual
  (holding constant unobserved effect).
• Intermediate solution is random effects (RE),
  but makes similar assumption to descriptive
  model µi unrelated to x and h

13
Effect of housework on wages (FE)
Single women Single men Married women Married men
Basic spec 0.0011 (1.11) -0.0019 (-1.42) -0.0016 (-3.12) -0.0014 (-2.03)
Basic spec job characs 0.0012 (1.28) -0.0019 (-1.41) -0.0014 (-2.73) -0.0013 (-2.04)

• Basic specification includes experience
  (squared), education, region and year.
• Job characteristics are one-digit occupation and
  industry, public sector employment, establishment
  size, trade union coverage, and temporary and
  fixed-term employment.
• t-statistics in parentheses

14
Between and within variation in housework
Single women Single men Married women Married men
Within variance 12.9 7.8 20.6 8.9
Between variance 31.8 13.7 40.2 11.8
Overall variance 44.7 21.5 60.8 20.7
15
Simultaneity and measurement error

• wit xit'ß ?hit µi eit
• FE controls for permanent heterogeneity (µi), but
  not for correlation of eit with (measured)
  housework.
• Wage increase may reduce housework time, e.g. use
  bonus or pay rise to hire a cleaner. Then eit is
  negatively correlated with hit.
• If ? is negative, classical measurement error in
  housework (only) implies that measured housework
  and eit are positively correlated.
• Overall effect ambiguous...

16
Alternative estimates and specifications

• FE IV but which instruments? Try spousal labour
  market behaviour and total number of employed
  household members.
• Use panel to find instruments?
• Estimate FD equation instrument housework by 2nd
  lag of housework in levels.
• More generally use further lags panel GMM.
• Use lagged housework as alternative measure. No
  ambiguity in timing of wage vs housework changes.
  Less endogeneity bias(?). Estimates are total
  effect of lagged housework (inc effect of current
  housework correlated with past housework).

17
Effect of lagged housework on wages (FE)
Single women Single men Married women Married men
Basic spec -0.0003 (-0.30) -0.0011 (-0.79) -0.0015 (-3.19) -0.0002 (-0.24)
Basic spec job characs -0.0001 (0.09) -0.0008 (-0.63) -0.0014 (-2.94) 0.0002 (0.25)

• Basic specification includes experience
  (squared), education, region and year.
• Job characteristics are one-digit occupation and
  industry, public sector employment, establishment
  size, trade union coverage, and temporary and
  fixed-term employment.
• t-statistics in parentheses

18
Robustness checks

• Include children variables to check estimated
  housework effect is not due to (omitted)
  childcare.
• Include PT workers. Stronger effects (inc married
  men), but are FT and PT workers comparable?
• Including age instead of experience (larger
  samples). V similar estimates.
• Combined married-single equations with
  interactions (but fixed effect constrained to be
  same across marital status).

19
Conclusions

• Comparing people with similar human capital (and
  in similar jobs), we find a strong association of
  housework with wages among married people. This
  can explain about quarter of gender pay gap.
• But cannot interpret causally, since data suggest
  that results may be affected by omitted factors
  like career orientation/path. Descriptive results
  are perhaps indicative of potential long-term
  changes, if housework reduction accompanied by
  changes in career orientation and prospects.
• Controlling for unobserved factors, effects are
  much smaller and concentrated among married women
  (probably due to specialisation into certain
  housework tasks - see paper for details). If they
  reduced their housework to mens levels, womens
  wages would increase by about 1 (but still gt5
  of pay gap).

20
Extra slides
21
Effects of children

• We do not have data on childcare. But we know
  children are associated with lower wages for
  women (family gap, e.g. Waldfogel, 1998). What
  about men?
• Childcare is correlated with housework in our
  data children are associated with 3 hours more
  housework for women and 0.5 hour for men.
• Does omitted childcare bias housework
  coefficients?
• Test whether housework coefficients are picking
  up childcare effect.

22
Effects of children (FE)
Single women Single men Married women Married men
Housework 0.0009 -0.0020 -0.0012 -0.0014
(0.93) (-1.50) (-2.28) (-2.14)
No of own children in household 0.0373 0.0401 -0.0401 0.0197
No of own children in household (2.12) (0.69) (-6.45) (4.72)
23
IV estimation

• IV is potentially a solution to simultaneity and
  measurement error.
• Have already controlled for µi using FE. So need
  instruments strongly correlated with (changes in
  ) hit and uncorrelated with eit ? FE IV
  estimation.
• Focus on married individuals (as seemingly no
  effect for singles), allowing use of spousal
  characteristics.
• Instruments spousal labour market participation,
  hours of work, occupation and wage, and the total
  number of employed household members.
• Also tried spousal attitudes towards domestic
  roles (limited number of waves, and low variation
  over time).

24
Effect of housework on wages (FE IV)
Basic spec Basic spec Basic spec job characs Basic spec job characs
Women Men Women Men
Housework coeff 0.0098 -0.0253 0.0106 -0.0230
(1.74) (-3.72) (1.90) (-3.42)
First-stage partial R2 0.011 0.018 0.011 0.012
First-stage F-statistic p-value 4.45 0.00 9.13 0.00 4.39 0.00 8.92 0.00
Sargan statistic ?2(12) p-value 38.5 0.00 16.6 0.17 32.3 0.00 16.8 0.16
25
IV results

• Test indicate that changes in womens labour
  market behaviour are valid instruments for
  (changes in) spouses housework.
• But changes in mens labour market behaviour do
  not appear to be valid instruments. Due to
  differences in labour market dynamics of men vs
  women?
• Focus on mens results one hour increase in
  housework leads to 2.5 reduction in wages
  (compared to 0.14 in FE equation).

26
IV results contd.

• FE IV estimate is larger in magnitude than FE
  estimate, suggesting measurement error dominates
  simultaneity bias.
• Test endogeneity (either measurement error or
  simultaneity) by comparing FE and IV FE results
  using a Hausman test.
• Test of housework coefficients only ?2(1)11.74
  ? reject exogenous housework (IV preferred).
• Test of all coefficients ?2(30)11.72 ? do not
  reject exogeneity (FE preferred).
• We take this as suggestive evidence that
  housework is measured with error for men and that
  FE could be seen as a lower bound (in magnitude)
  on the true effect.

27
IV results

• Instruments (spousal labour market behaviour and
  total number of employed household members) only
  test as valid in (married) mens equation.
• FE IV estimate is larger in magnitude than FE
  estimate (-2.5 compared to -0.14), suggesting
  measurement error dominates simultaneity bias.
• But exogeneity test (comparing FE and IV FE) does
  not reject ?2(30)11.72 ? FE preferred.
• FE remains our preferred specification. But
  suggestive evidence that housework is measured
  with error and that FE could be seen as a lower
  bound (in magnitude) on the true effect.

28
Including part-time workers

• Previous results excluded PT workers to maintain
  homogenous sample (and data show that PT workers
  earn less and do more housework)
• Re-estimate including PT workers and adding
  (endogenous?) PT dummy.
• PT 6-30 hours (exclude v short hours to
  alleviate measurement error problems)

29
Including part-time workers (FE)
Single women Single men Married women Married men
Housework -0.0006 -0.0020 -0.0012 -0.0020
(-0.73) (-1.49) (-3.23) (-2.95)
Part-time 0.0202 0.1974 0.0282 -0.0370
Part-time (1.27) (7.10) (3.54) (-2.02)
30
Dimensions of housework amount, type and timing

• Do only large amounts of housework have an
  effect?
• Could explain lack of effect for singles and
  married men?
• Try quadratic and spline functions (nodes at 5
  and 10 hours).
• Find no evidence of non-linear effects.

31
Non-linear effects (FE)
Single women Single men Married women Married men
Housework 0.0019 -0.0002 -0.0033 -0.0014
(1.23) (-0.08) (-2.66) (-1.19)
Housework squared / 100 -0.0020 -0.0085 0.0045 0.0000
Housework squared / 100 (-0.68) (-0.75) (1.48) (0.01)
32
Type and timing of housework

• No information on type and timing in BHPS, but
  use complementary descriptive data from UK Time
  Use Survey 2000 .
• UK TUS is based on time diaries.
• What doing in 10 min slots throughout day
  primary, secondary activity with whom where.
• But limitations in earnings data (net, some
  banded, last pay period only) . Cross section
  only.

33
Specialisation in housework activities
Food Up-keep Laun-dry Gard-en Rep-airs Shop-ping Mana-ge-ment Child care
Men Men Men Men Men Men Men Men Men
Married 20.9 9.6 1.4 10.1 9.2 7.9 1.4 15.1
Single 22.2 5.5 2.2 6.3 2.0 8.8 0.8 1.2
 Women  Women  Women  Women  Women  Women  Women  Women  Women
Married 43.7 22.5 14.3 9.9 1.5 16.0 1.0 15.0
Single 30.4 14.6 6.5 9.3 2.4 13.0 1.1 4.4
Source UKTUS. Full time workers minutes of housework on weekday. Source UKTUS. Full time workers minutes of housework on weekday. Source UKTUS. Full time workers minutes of housework on weekday. Source UKTUS. Full time workers minutes of housework on weekday. Source UKTUS. Full time workers minutes of housework on weekday. Source UKTUS. Full time workers minutes of housework on weekday. Source UKTUS. Full time workers minutes of housework on weekday. Source UKTUS. Full time workers minutes of housework on weekday. Source UKTUS. Full time workers minutes of housework on weekday.
34
Specialisation in housework activities
Food Up-keep Laun-dry Gard-en Rep-airs Shop-ping Mana-ge-ment Child care
Men Men Men Men Men Men Men Men Men
Married 20.9 9.6 1.4 10.1 9.2 7.9 1.4 15.1
Single 22.2 5.5 2.2 6.3 2.0 8.8 0.8 1.2
 Women  Women  Women  Women  Women  Women  Women  Women  Women
Married 43.7 22.5 14.3 9.9 1.5 16.0 1.0 15.0
Single 30.4 14.6 6.5 9.3 2.4 13.0 1.1 4.4
Source UKTUS. Full time workers minutes of housework on weekday. Source UKTUS. Full time workers minutes of housework on weekday. Source UKTUS. Full time workers minutes of housework on weekday. Source UKTUS. Full time workers minutes of housework on weekday. Source UKTUS. Full time workers minutes of housework on weekday. Source UKTUS. Full time workers minutes of housework on weekday. Source UKTUS. Full time workers minutes of housework on weekday. Source UKTUS. Full time workers minutes of housework on weekday. Source UKTUS. Full time workers minutes of housework on weekday.
35
Specialisation in housework activities

• Married women specialize in more routine and time
  intensive activities, e.g. food, laundry,
  shopping.
• Married men specialize in gardening, household
  management and repairs.
• Consistent with other studies, e.g. Hersh and
  Stratton (2002) US and Fernandez and
  Sevilla-Sanz (2006) Spain.
• Lack of threshold effects suggests effect on
  wages is not because married womens housework
  activities are time-intensive
• More likely, these activities need to be done
  routinely, usually during work-days, and cannot
  be postponed until the weekend.
• Do not know whether these types of housework are
  more tiring but can check timing relative to
  market work.

36
Timing of paid work and housework
37
Timing of paid work and housework

• Married men and women do more housework between
  4pm and 8pm than singles.
• But both married and single men spend about the
  same time on work (44 between 4pm-8pm)
• Whereas married women do less work than singles
  over these times (32 of time vs 40 of time for
  single women).
• Married womens housework appears to be done at
  times that interfere with market work.