Post on 01-Apr-2021
1
TWIN PEAKS: An Analysis of the Gender Gap in Pension Income in
England
Onyinye Ezeyi (University of Bath)
Sunčica Vujić (University of Antwerp and University of Bath)
Abstract:
This paper investigates the gender gap in state and private pensions in England using
data on retirees from the English Longitudinal Study of Ageing. The analysis finds that
the female distribution of state pension income is bimodal, an idiosyncrasy that arises
because a large proportion of women have pensions derived from their spouse’s
contribution record. Though there are no gender gaps in state pension coverage, the
female-male state pension income ratio is 0.75 and Blinder-Oaxaca decompositions
indicate that only a quarter of the gap in mean state pension income is explained by
observed characteristics. The analysis is also extended to quantile regressions and
decompositions reveal “sticky floors” in state pension income. The gender pension gap
is more marked for private pensions than state pensions. Decompositions suggest that
gender differentials in characteristics matter more for private pension coverage rates
than private pension income levels. Specifically, differential characteristics account for
half of the gender gap in private pension coverage but only 28% of the gender gap in
private pension income. Quantile analysis of the gap in private pension income indicate
that though there is greater volatility, the gap remains relatively close to the mean gap.
The private pension estimates are not robust to selectivity corrections.
Key words: gender pension gap, England, decomposition methods, quantile regression
JEL: C21, C24, J31, J71
Corresponding author:
Onyinye Ezeyi
ome22@bath.ac.uk
2
1. Introduction
Women have a higher risk of old age poverty than men and are more likely to claim
means-tested benefits in retirement, indicating that they do not have adequate
retirement income (Ginn, 2003; Department for Work and Pensions 2013). Women’s
higher propensity of poverty in retirement is related to their pension entitlement as
pension benefits are the main source of income for retirees in the UK. In the 2013/2014
tax year, average weekly gross income for retired households in the UK stood at £487
(in 2014 prices), of which pension benefits accounted for 67% (Department for Work
and Pensions, 2013). The pension income an individual receives in retirement is related
to their labour market outcomes and personal characteristics since individuals pay
contributions during working life to accrue pension rights and secure an income in
retirement. Despite the abundance of research into the gender wage gap, little research
has been conducted in the economic literature into the extent to which women’s lower
lifetime labour market outcomes translates into lower pension outcomes in later life.
The historical dimension of the accumulation of pension rights mean that other factors
emerge in the determination of the gender pension gap that may not be pertinent to the
gender wage gap.
An analysis of the gender gap in pensions for the UK is important for at least two
reasons. First, increasing life expectancy together with women’s greater longevity imply
that, unless women’s pension outcomes equalise with men’s, women are likely to spend
an increasing proportion of their lives at risk of poverty. Secondly, the restructuring of
the UK state pension to a one-tier system and the government’s emphasis on private
pensions in helping individuals secure adequate retirement income mean that research
on the gender pension gap can better enable the design of private pension policy in the
UK.
This paper contributes to the scant economic literature by providing a comprehensive
analysis of the gender pension gap in England. The paper first examines the factors that
determine pension coverage and pension income for state and private pensions
3
separately. The paper then uses Fairlie (1999) and Blinder (1973) Oaxaca (1973)
decomposition to respectively decompose the gap in pension coverage and income into
the explained and unexplained components. The analysis conducted in this paper is
most similar to Even and Macpherson (1994) who examine the gender gap in
occupational pensions for the US, Bardasi and Jenkins (2010) who investigate the gap in
private pensions for the UK and Hänisch and Klos (2014) who examine old age income
in Germany at the mean and for the entire pension income distribution. The results
from these studies are mixed. Even and Macpherson (1994) find that among retirees,
80% of the predicted gap in occupational pension coverage and 70% of the predicted
gap in occupational pension benefit is accounted for by gender differentials in
characteristics. However, they do not employ regression-based decomposition
techniques. Conversely, Bardasi and Jenkins (2010) find that between 33%-42% of the
gap in private pension coverage is explained, while 8%-18% of the gap in benefit level is
explained. However, unlike Bardasi and Jenkins, we are able to control for earnings and
also extend our analysis to quantile regressions. Analysing German data from 2007,
Hänisch and Klos (2014) find that only 26% of the gender pension gap is explained and
that the largest share of the explained gap occurs at the bottom 25% of the pension
income distribution. To the best of our knowledge, this is the first paper to examine the
gender gap in UK state pensions using regression-based decompositions and to apply
quantile decomposition techniques for both state and private pensions.
The results from our analysis confirm the presence of gender gaps in both state and
private pensions. While state pension coverage is virtually universal for both sexes,
there are substantial gender inequalities in the level of state pension benefit received;
the raw female-male state pension income ratio is 0.75. Blinder-Oaxaca decompositions
show that only a quarter of the gap in state pension income can be attributed to gender
differentials in mean characteristics and that women’s lower state pension income
arises because they receive state pensions derived from their spouse’s National
Insurance contribution record. Results from quantile regressions provides the first
evidence of ‘sticky floors’ in state pensions. That is, the gap is largest at the bottom 30%
of the state pension income distribution than at the top of the distribution but is
decreasing as one moves up the distribution. Unlike state pensions, the analysis shows
4
that there are gender gaps in both private pension coverage and income; only 63% of
women had private pension coverage compared to 90% of men and the raw female-
male pension income ratio conditional on coverage stands at 0.52. Decompositions of
private pension coverage suggest that gender differences in average characteristics
account for around half of the gap in probability of private pension receipt, while
Blinder-Oaxaca decompositions suggest that 72% of the gap in private pension income
is unexplained. The results for private pension income are not robust to selectivity
corrections.
The remainder of this paper is organised as follows. The next section discusses the
institutional context of the UK pension system for the sample of retirees used in this
analysis. It also provides explanations for why men and women may have differential
state pension outcomes. Section 3 provides economic rationale for the demand and
provision of pensions in the labour market and also examines the mechanisms through
which the gender gap in private pensions may arise. The methodology used to
decompose the gender gap in pensions in Section 4, while Section 5 describes the data
and sample used in this analysis. Model estimates and the results of the decompositions
are shown in Section 6. In Section 7, selectivity corrections are presented as robustness
checks. Section 8 summarises the findings and discusses the policy implications of the
results.
2. The Institutional Context
In the UK, pensions are provided by both the State and private sector. In order to
analyse the gender gap in pensions, it is necessary to understand the institutional
context of the UK pension system. In April 2016, a new state pension system was
introduced, however, because the sample of the sample of retirees analysed in this
study built up their pension entitlement under the pre-2016 system, this section
provides a brief overview of the state pension system until 2016 and also highlights
5
features of the state pension system that may have served as possible mechanisms of
gender inequalities in state pensions. This section also describes private sector
provision of pensions in the UK.
2.1 State Pensions
For individuals retiring before April 2016, the state pension was made up a flat-rate
pension known as the Basic State Pension and an earnings-related component known as
the Additional State Pension. Individuals built entitlement to the state pension during
their working lives by paying National Insurance contributions or were given National
Insurance credits. It was possible, as it is in the post April 2016 system, to claim the
state pension and continue in paid employment. Men and women of state pension age
(65 and 63 years respectively in January 2016) with sufficient years (30 years) of
National Insurance (NI) contributions or credits received the full Basic State Pension
(BSP) of £115.95 per week in the 2015-16 tax year as well as any entitlement to the
Additional State Pension. Individuals with less than 30 years of NI contributions or
credits received a pro-rata amount. The Additional State Pension received is a
composite of the various earnings-related state pension schemes that existed i.e. the
State Earnings Related Pension Scheme and the State Second Pension. The self-
employed did not build up entitlement to the Additional State Pension (ASP) and the
ASP is not payable for the years in which employees were members of a contracted-out
private pension scheme.
For the sample of retirees under consideration, there are two main reasons why
women’s state pension outcome might be worse than men’s. First, the UK state pension
system introduced in 1948 was designed to reflect prevailing societal norms. In
particular, the state pension system reinforced the male breadwinner-female
homemaker model of familial division of labour. A man’s NI contribution record bought
two state pensions; one for himself (i.e. Category A pension) and one for his wife, valued
at 60% of his entitlement to the Basic State Pension (i.e. Category B pension). In
addition, there were strong incentives for married women to claim a state pension
6
based on her husband’s NI record; until 1978, married women had the option to pay
reduced rates of NI contributions in return for forfeiting entitlement to a state pension
in their own right. This was known as the Married Woman’s Stamp and a woman that
elected to pay the lower NI contributions would receive a Category B pension on
retirement. Women that did not opt into the Married Woman’s Stamp had to pass the
Half Test which required paying sufficient NI contributions for at least half of the
number of weeks between their date of marriage and state pension age, to receive any
state pension in her own right.
Secondly, women spend a substantial period of time out of employment as due to their
caring responsibilities in the home. Thus, historically, it has been more difficult for
women to have the requisite number of years to claim entitlement to the full BSP. Since
1978, the NI system has taken into account time spent out of the labour market due to
caring responsibilities. However, individuals still needed to have paid NI contributions
for at least a quarter of their working life to receive any BSP at all and were required to
have paid 20 years of NI contributions to receive the full BSP until 2010.
2.2 Private Pensions
Private pension provision in the UK consists of occupational pension schemes and
personal pensions. Occupational pension schemes are organised by employers for their
employees, while personal pension schemes are contracts between individuals and
financial institutions. Personal pensions are the only private pension scheme available
to the self-employed. Until 2012, employees had the option to contract out of the ASP
and enrol in a private pension scheme, thereby paying a reduced rate of NI
contributions. Prior to the commencement of automatic enrolment in 2012, the
provision of an occupational pension scheme was at the discretion of the employer and
firms also had the ability to set the eligibility criteria for scheme membership.
Though there is great variety in private pension schemes, most pension schemes in
operation in the UK can be categorised as either a defined benefit pension or a defined
7
contribution pension. A defined benefit pension is an occupational pension that
typically pays beneficiaries a pension according to a predetermined formula based on
salary, tenure and an accrual rate whereas defined contribution schemes involve paying
contributions into a pension fund that is invested and the accumulated contributions
and returns at retirement is used to provide a pension for beneficiaries. All personal
pensions in the UK are defined contribution pensions while occupational pension
schemes can be of the defined benefit or define contribution type. The ASP was a
defined benefit pension.
3. The Role of Pensions in the Labour Market
Several arguments have been put forward to explain the demand and supply of
pensions by individuals and firms respectively. One strand of the economic literature
views individuals’ demand for private pensions as a form of retirement income
insurance (Bodie 1990; Gustman et al. 1994). By receiving pensions as annuity income,
recipients effectively insure against the risk that assets run out before death (i.e.
longevity risk). Private pensions in particular mitigate the impact of potential
reductions to state pensions and other pensioner benefits in future and also help ensure
adequate replacement rate. Similarly, another strand of the literature explains
individuals’ demand for pensions in terms of the desire to smooth consumption over the
lifecycle. Intertemporal models of saving and retirement posit that individuals aim to
maximise expected lifetime utility subject to their lifetime budget constraint in order to
smooth consumption over the lifecycle (Modigliani and Brumberg, 1954; Feldstein
1974). In this framework, individuals demand for pension coverage stems from the
desire to have adequate income in retirement to maintain pre-retirement standard of
living.
The theoretical literature also offers some insight into the role of occupational pensions
in the labour market. The implicit contract theory of pensions postulates that the
8
matching of workers to firms in pension contracts (of the defined benefit type) induces
strong worker-firm attachments. In the implicit contract theory, pensions are viewed as
deferred compensation that impose substantial capital losses on workers that leave the
firm before reaching normal retirement age because of pension back-loading (i.e. the
disproportionate amount of pension benefit that is accrued in later years of service in
defined benefit schemes) (Kotlikoff and Wise, 1988). By offering pensions, firms create
strong incentives for workers to stay with the firm until reaching the retirement
window. Consequently, firms offer pension schemes to attract productive workers and
regulate worker turnover due to high job turnover costs.
In this framework, occupational pensions are used as a sorting device that enable firms
to identify and attract workers with low quit propensities since the workers that take
up pension jobs are likely to be stayers (Ippolito 1985; 1987). Workers with low quit
propensities will self-select into pension jobs and enter an ‘implicit contract’ with the
firm whereby workers effectively agree to long tenure with the firm, while the firm
agree to not terminate pension schemes or lay off workers as it is concerned about their
labour market reputations (Allen et al 1993; Cornwell et al. 1991). This suggests that
while all individuals demand occupational pensions, only those with low quit
propensities will take up pension jobs. However, Gustman and Steinmeier (1993) and
Even and Macpherson (1996) have shown that turnover rates in jobs with defined
contribution pensions, which are more portable, are similar to turnover rates in defined
benefit jobs casting doubt on the effects of non-portable pensions in restricting job
mobility. Instead, Gustman and Steinmeier (1993) argue that the lower turnover rates
in pension jobs is due to the higher total compensation premium of pension jobs rather
than pension back-loading.
There are numerous channels through which a gender gap in private pensions might
arise. First, the strong worker-firm attachments created by the incentive structure of
defined benefit (DB) schemes imply that women are less likely than men to have
pension coverage since women are more likely to change employers due to their
fragmented work histories. For the majority of the sample under analysis in this study,
9
DB pensions would have been the predominant type of pension scheme offered by firms
during their working life as define contribution schemes were only incorporated into
national insurance system in the late 1980s. Consequently, the mobility restriction
embedded in DB pensions together with women’s discontinuous work histories would
have served as a barrier to women’s private pension coverage. Indeed, previous
research on the gender pension gap has shown that gender differentials in labour
market experience is the most important factor in the explained component of the gap
in private pension coverage and benefit (Even and Macpherson, 1994; Bardasi and
Jenkins, 2010; Hänisch and Klos 2014).
Secondly, men’s higher lifetime earnings than women will have implications for the
gender gap in private pensions. For instance, because the tax advantages of private
pensions rise with income, a larger proportion of men would demand private pension
coverage than women because the tax-efficiency of pension saving rises with income.
Nonetheless, given coverage, women’s lower lifetime earnings than men translates into
lower benefit levels from DB pension schemes and implies that women may face greater
financial constraints to contributing to defined contribution (DC) schemes despite the
increased portability of DC pensions relative to DB pensions.
Thirdly, the occupational segregation of men and women may have consequences for
the gender gap in private pensions since it was not uncommon for employers to have
different pension schemes for different classes of employees. For example, employers
operated senior-management schemes, staff schemes and works schemes that offered
different accrual rates (Blake 2003). Figures from the Office of National Statistics show
that in 2013 women accounted for 82% of workers in caring and leisure occupational
group, whereas only a third of individuals working in the highest paid category of
managers, director and senior officials were women. Thus, the over-representation of
women in low-paid occupations may worsen women’s pension outcome relative to men.
Lastly, eligibility for occupational scheme membership was often defined in terms of the
nature of employment and tenure. For instance, part-time workers were often excluded
10
from scheme membership until 1995, thereby a significant proportion of women were
unable to accrue occupational pension rights and instead remained contracted in the
Additional State Pension. The earnings-related component of the state pension was
primarily aimed at individuals that did not have access to an occupational pension
scheme and was generally regarded as a less generous substitute for occupational
pensions (Blake, 2003). Indeed, Barrientos (1998) finds that 60% of individuals with
SERPS membership were women and that women were less likely to contract out of
SERPS and enrol into a private pension scheme than men.
4. Methodology
4.1 Decomposition of Pension Income
The determinants of pension income is estimated using ordinary least squares for the
subsample of individuals with state (private) pension coverage separately for each
gender:
𝑌𝑖𝑚 = 𝑋𝑖𝑚�̂�𝑚 + 𝑢𝑖𝑚
𝑌𝑖𝑓 = 𝑋𝑖𝑓�̂�𝑓 + 𝑢𝑖𝑓
where, 𝑌𝑖𝑔, is weekly net state (private) pension income for individual 𝑖 sample of
gender 𝑔 = (𝑚, 𝑓). Blinder (1973) and Oaxaca (1973) decompose the difference in the
average pension income of the male and female sample as follows:
Equation 1
�̅�𝑚 − 𝑌�̅� = (∑ 𝑋𝑖𝑚
𝑁𝑚𝑖=1
𝑁𝑚−
∑ 𝑋𝑖𝑓𝑁𝑓
𝑖=1
𝑁𝑓) �̂�𝑓 −
∑ 𝑋𝑖𝑚𝑁𝑚𝑖=1
𝑁𝑚(�̂�𝑚 − �̂�𝑓)
11
where �̅�𝑔 = ∑ 𝑌𝑖𝑔
𝑁𝑔𝑖=1
𝑁𝑔. The first term on the right hand side of Equation 1 is the explained
component i.e. the portion of the gap attributed to differences in the average
characteristics of men and women. The second term is the unexplained component and
represents the portion of the gap that arises from gender differentials in the returns to
average characteristics. The unexplained component is commonly thought of as
evidence of discrimination, however, it is important to bear in mind that it captures
gender differentials in any important variables omitted from the model specification as
well as discrimination (Altonji and Blank, 1999).
4.2 Decomposition of Pension Coverage
Descriptive statistics and preliminary investigations revealed that state pension
coverage is near universal and does not significantly differ by gender. As such, the
determinants of state pension coverage are estimated for the pooled sample of men and
women. Conversely, descriptive statistics showed large discrepancies in private pension
coverage for men and women. Accordingly, the determinants of private pension
coverage is estimated separately for each sex as follows:
𝐶𝑖𝑔∗ = Z𝑖𝑔𝛾𝑔 + 𝜀𝑖𝑔 𝜀𝑖𝑔~𝑁(0, 𝜎2)
𝐶𝑖𝑔 = { 1 𝑖𝑓 𝐶𝑖𝑔
∗ > 0
0 𝑖𝑓 𝐶𝑖𝑔∗ ≤ 0
Pr(𝐶𝑖𝑔 = 1 | 𝑋𝑖𝑔 ) = Φ(Z𝑖𝑔𝛾𝑔 + 𝜀𝑖𝑔)
where 𝐶𝑖𝑔∗ is the unobserved contribution propensity for individual 𝑖 of gender 𝑔 which
is determined by observed labour market and personal characteristics plus a normally
distributed unobserved error 𝜀𝑖𝑔. 𝐶𝑖𝑔 is private pension coverage and takes the value of
1 if the individual sufficiently contributed to a private pension and 0 otherwise. Thus,
the probability that individual 𝑖 of gender 𝑔 has private pension coverage is equal to the
cumulative distribution function of the standard normal distribution of the
12
determinants of private pension coverage. Decomposition analysis is conducted using
the methodology developed by Fairlie (1999) which decomposes the difference in the
average predicted probability of private pension coverage between men and women
into the component due to gender disparities in distribution of characteristics and into
the component due to gender differentials in the process determining the probability of
private pension coverage:
Equation 2
𝐶�̅� − 𝐶�̅� = [∑Φ(𝑍𝑖𝑚𝛾𝑓)
𝑁𝑚
𝑁𝑚
𝑖=1 − ∑
Φ(𝑍𝑖𝑓�̂�𝑓)
𝑁𝑓
𝑁𝑓
𝑖=1]
+ [∑Φ(𝑍𝑖𝑚𝛾𝑚)
𝑁𝑚
𝑁𝑚
𝑖=1 − ∑
Φ(𝑍𝑖𝑚𝛾𝑓)
𝑁𝑚
𝑁𝑚
𝑖=1]
where 𝑁𝑔 is the sample size for gender 𝑔 and 𝐶�̅� is the average probability private
pension receipt for gender 𝑔. In Equation 2, differences in the distribution of
characteristics are weighted with the coefficient from the female sample and differences
in the process determining private pension coverage are weighted with the male
distribution of characteristics. The first term in the first parenthesis is the
counterfactual distribution that would arise if the female sample had the same
characteristics as the male sample i.e. women’s average predicted probability of having
a private pension if they assume mean characteristics of men.
5. Data
5.1 Sample
The empirical analyses uses a sample drawn from wave 6 of the English Longitudinal
Study of Ageing which surveys individuals aged 50 and over living in private households
in England. The English Longitudinal Study of Ageing (ELSA) began in 2002 and is
13
collected biennially, accordingly, there were six waves of ELSA between 2002 and 2014.
The estimating sample is restricted to the cross-section of retirees in wave 6 that also
participated in the wave 3 life history interview. The wave 3 life history interviews
contain retrospective data on marital, fertility and employment histories. Information
from the wave 3 life history interviews augmented with information from subsequent
waves form the basis of the independent variables used in the regression analysis.
This study focuses on pensioners in England and respondents that are not observed as
retired in wave 6 are excluded from the analyses. However, the pension system in the
UK is such that retirement cannot be explicitly defined as a discrete event. This is
because it is possible to claim the state pension and remain in employment. It is also
possible to receive a private pension and continue working provided it is not with the
same firm. Since there may be a gradual transition from employment to retirement for
some individuals, we define retirement in terms of three variables; respondents’ self-
reported employment status, a dummy variable indicating whether the respondent is of
above state pension age and the current working status variable deduced from survey
filtering. The age restriction is imposed because those that retire before state pension
age are not representative of the population of retirees as they are likely to be those
with substantial pension wealth or those with ill-health. The final sample comprises
1,577 women and 1,188 men at or above state pension age.1
There are four dependent variables in the regression equations: state pension coverage,
the log of net state pension benefit, private pension coverage and the log of net private
pension income. State pension benefit is the sum of income from the Basic State Pension
and the Additional State Pension deflated to 2014 prices, expressed in weekly terms. An
individual has state pension coverage if they receive a positive amount of state pension
benefit in retirement. Private pension coverage is defined in terms of the receipt of a
positive amount of private pension income and private pension income is defined as the
1 State pension age is 65 years for a man. For the majority of the female sample under analysis, female state pension age is 60 years. However, since 2010 the female state pension age has been increasing to equalise with the men’s. Consequently, some women in the sample have a state pension age greater than 60 years.
14
sum of pension income received from occupational pensions and annuity income.
Pension income is inextricably linked to labour market outcomes during working life.
Therefore, the explanatory variables included in the pension coverage and benefit
equations include a set of work history variables intended to capture engagement with
the labour market; years of work experience between ages 20 and 60, years spent as
self-employed between ages 20 and 60, years spent working part-time between ages 20
and 60 and starting wage of last job before retirement. The model also includes birth
cohort, educational qualifications and marital status. The full set of explanatory
variables is shown in Table 1 in the appendix.
While the English Longitudinal Study of Ageing contains many variables that are
potentially important determinants of pension coverage and income, retrospective data
on firm characteristics such as industry, union coverage status and public-private sector
distinction is unavailable. Data on firm size is only available for those that are observed
in employment in any one wave of ELSA or those that were in employment during the
Health Survey for England. Consequently, when firm size is included in the model
specification for private pensions, it is likely to be for the subsample of pensioners that
are younger than the full sample.2 Another drawback of ELSA is that retrospective
earnings data is only available for jobs where the respondent was an employee.
Including wage in the regression specification means that the analysis is restricted to
respondents whose last job of career was not in self-employment.
5.2 Descriptive Statistics
State pension coverage is near universal at approximately 99% for both the male and
female samples. Despite this, there are inequalities in state pension income.
Conditioning on coverage, average net weekly state pension income for men and
2 The distribution of birth cohort for the subsample of individuals with firm size information; born before 1934, 1934-1942 and after 1942 is 0.30, 0.35 and 0.30 for the male sample and 0.23, 0.27 and 0.50 for the female sample respectively.
15
women is £147 and £110 respectively. Accordingly, the raw gender gap in average state
pension income stands at 25%.3 That is, men’s average net weekly state pension income
is 25% higher than women’s average net weekly state pension income. Men’s median
net state pension income amounts to £146 per week compared to £119 per week for
women. Thus, the raw gap in state pension income is smaller at the median than the
mean. Descriptive statistics also show that the raw gender gap in average net private
pension income is 80% for the whole sample. Specifically, average net weekly private
pension income for the entire sample of men is almost five times greater than their
female counterpart (£68 vs. £14 respectively). Amongst those with private pension
coverage, men receive nearly twice as much private pension income as women (£123
vs. £64 respectively) with the corresponding raw gap of 48%. Such large discrepancies
in the conditional and unconditional gap highlight differences in the private pension
coverage rates of men and women. Indeed, descriptive statistics indicate that less than
two-thirds of women (63%) receive any private pension income whereas 88% of men
receive income from a private pension. Moreover, the median net private pension
income without conditioning on coverage is £122 and £25 per week for men and
women respectively, while the conditional median net private pension income is £151
and £74 per week for men and women respectively. The conditional and unconditional
raw gaps in mean private pension income do not significantly differ from the associated
median.
Figure 1 shows the distribution of log average weekly net state pension income by
gender. Several key features emerge from the kernel density plot. First, it can be seen
that for both men and women, the distribution of state pension income remains flat
until approximately £33 and becomes flat again just before reaching £403. The flat left
tail is the result of the minimum number of qualifying years needed to claim any state
pension that was in force for individuals retiring before 2010, whilst the right tail
represents the fact that there is a maximum state pension that an individual can receive
made up of the full Basic State Pension and the full Additional State Pension. Second, it
can be seen that the male state pension distribution situated to the right of the female
distribution and that there is a high degree of concentration in the upper end of the
3 Defined as [1-(average female pension income/average male pension income)]*100.
16
distribution, implying that men receive a higher amount of state pension income than
women. Finally, the distribution of state pension income is bimodal for the female
sample and presumably represents the idiosyncratic feature of the UK state pension
system that permitted female retirees to claim a state pension using their spouses’
national insurance contribution record that was worth 60% of their spouse’s
entitlement. Figure 2 depicts the distribution of state pension income for the female
sample by birth cohort and reveals that the proportion of women receiving close to the
full state pension has increased between the oldest and youngest cohorts.
[Place figure 1 here]
[Place figure 2 here]
Figure 3 shows the distribution of private pension income by sex. In addition to
receiving a lower amount of private pension income than men, the figure shows that the
variation in private pension income is greater for women than men. A larger proportion
of women receive a net private pension income that is less than £55 per week and most
women are clustered around the £90 mark whereas most men are clustered around the
£300 mark.
Table 1 reports the sample means of the variables used in the analysis of the gender
pension gap. The male sample have higher levels of education attainment than the
female sample, though the gap in educational qualifications diminishes when comparing
the distribution of education between the subsamples of men and women with private
pension coverage. The raw gender gap in wages from last job before retirement is 54%,
though this decreases to 49% when the sample is restricted to individuals with private
pension coverage. On average, women have 11 years less work experience than men,
with the gap reducing marginally to 10 years when comparing individuals with private
pension coverage. Women spend a longer period of time working part-time compared
to men irrespective of private pension coverage status. The number of years spent in
17
self-employment is small for both men and women. Nonetheless, men spend
approximately three times as long in self-employment than women who spend an
average of one year. While only 28% of women are from the three highest occupational
classification levels, this percentage is 43% for the male sample. This proportion
increases to 35% for women and 46% for men when the sample is restricted to those
with private pension coverage.
6. Results
6.1 State Pensions
State Pension Coverage
Table 2 shows the probit estimates for the pooled sample of men and women. As
descriptive statistics indicated, the estimates suggest that the probability of state
pension coverage does not significantly differ by gender.4 The probit estimates indicate
that the probability of having a state pension does not significantly differ by level of
education, occupational classification or accumulated work experience during working
life. However, birth cohort is an important determinant of state pension coverage; older
aged individuals are more likely than those from younger cohorts to receive a state
pension. The results from the pooled regression also suggest that married individuals
are less likely than unmarried individuals to receive a state pension.5 Sample means by
gender and marital status show that married individuals in the sample are younger than
individuals not currently married and so the effect of being married may be capturing
the age effect.
4 This is confirmed by the Wald test which fails to reject the null hypothesis that the female interactions are jointly equal to zero. 5 I use a dummy variable for married in the state pension coverage equation since being single, never married and being separated perfectly predict success of state pension coverage.
18
State Pension Income
Estimates of the determinants of state pension income are presented in Table 3.
Columns 1 and 3 depict the estimates for state pension income for the male and female
sample respectively, while columns 2 and 4 introduces partner’s weekly state pension
income in the model specification.6 There are strong cohort effects in state pension
income, though the effect differs by gender. While men born before 1943 receive a
higher amount of weekly net state pension income than men born on or after 1943,
women from earlier cohorts receive a lower amount of state pension income than
women born after 1942. The heterogeneity in the effect of birth cohort is likely due to
the fact that men’s labour market attachment did not change significantly in the 20th
century whereas successive cohorts of women significantly increased their labour force
participation. In addition, policies introduced to the National Insurance (NI) system to
improve women’s state pension outcomes would have had the greatest impact on the
cohort of women born after 1942.
For both sexes, education and occupational classification have no statistical association
with state pension income. The relationship between starting income of last job before
retirement and state pension income is weak for both men and women. This is
unsurprising given that the state pension is a flat-rate benefit with a small earnings-
related component. As expected, work experience has a significant and positive impact
on state pension income, though the effect is small. It is interesting to note that the
magnitude of the effect of work experience on state pension income is similar for both
men and women which is consistent with the UK NI system whereby a year’s worth of
NI contributions provides individuals with a pro rata state pension income in
retirement. For both men and women, the longer the proportion of time spent as self-
employed, the lower the amount of state pension received. This is due to the fact that
the self-employed only accrue entitlement to the BSP and not the earnings-related
component. Similar to work experience, the size of the coefficients does not differ
significantly by gender. Years spent in part-time employment have essentially no effect
on men’s state pension income but negatively impacts the amount of state pension
6 Approximately 68% of men and 41% of women have a partner.
19
income women receive in retirement, though the effect is small. This result may be due
to the fact that men in the sample only spend an average of six months in part-time
employment compared to 8 years for women and the NI system in the UK is such that
workers only accrue entitlement to the State Pension once their weekly income from
any specific job is over a threshold known as the lower earnings limit.
Current marital status is strongly associated with the amount of state pension income
women receive and is not as important for men. Married women receive a lower
amount of weekly net state pension income than unmarried women; on average, a
single woman that has never married can expect a state pension that is 35% higher than
a married woman, while a widowed woman’s weekly net state pension is 56% higher
than a married woman. Widowed women are able to inherit their deceased spouse’s
state pension, accordingly their state pension income will be higher than those of
married women on average. Controlling for partner’s state pension income (columns 2
and 4) does not change the qualitative results. However, while the coefficient on
partner’s state pension income is positive for both sexes, the coefficient on women’s
state pension income is more significant and nearly twice the effect on men’s,
suggesting that women’s state pension income is more closely linked to their partner’s
state pension income. This indicates that married women are likely to be claiming
Category B state pensions, which are derived pensions valued at 60% of their spouse’s
entitlement.
Blinder-Oaxaca Decomposition
Error! Reference source not found. presents the Blinder-Oaxaca decomposition
shown in Equation 1. The gender gap in predicted average weekly net log state pension
income is £44 (column 1). That is, men’s average weekly net state pension income is
approximately 30% higher than women’s state pension income. Differences in the
average characteristics of men and women account for only a quarter of the gap in
average weekly net state pension income, whilst gender differentials in the returns to
average characteristics account for around three-quarters of the gap. While having the
same amount of labour market experience as men increases women’s state pension
20
income, having the same distribution of current marital status as men or spending an
identical amount of time spent in self-employment is detrimental to women’s state
pension income. Although differentials in the work history variables are significant
variables in the detailed decomposition of the explained gap, it is differences in the
returns to birth cohort, marital status and occupational classification that emerge as the
most important factors in the unexplained gap. The significance of birth cohort and
marital status in the unexplained gap is consistent with the evolution of state pension
policy which initially comprised gender-specific rules that operated through women’s
marital status but were subsequently abolished.
Columns 3 and 4 in Table 4 presents the Blinder-Oaxaca decomposition when state
pension income of respondents’ partners is included in the model specification, thus the
sample is restricted to only men and women with partners.7 The gender gap in weekly
net state pension income increases to £60 with the corresponding female-male ratio of
average weekly net state pension income amounting to nearly 0.60, providing further
evidence that married women’s state pension income are Category B pensions. Again,
differences in the returns to characteristics accounts for the majority of the gender gap
in state pension income. The detailed decompositions indicate that discrepancies in
labour market engagement is the most important factor affecting the explained
component of the gender gap in state pensions.
6.2 Private Pensions
Private Pension Coverage
Probit estimates of private pension coverage with and without firm size in the model
specification care presented in Table Error! Reference source not found.5.8 The two
specifications are presented because controlling for firm size greatly reduces the
sample size and the resultant sample comprises individuals from younger cohorts. The
7 This comprises either married or cohabiting individuals. 8 Housing financial wealth variables were included in the model specification but had zero effect on private pension coverage despite being significant only for the female sample. As such, the preferred model specification excluded housing and non-housing wealth.
21
results show that there is a cohort effect in the probability of private pension coverage
for the male sample at 10% level and none at all for the female sample. Education has
statistical associations with probability of having a private pension for both sexes,
though the effect is stronger for women. For both sexes, current marital status is
associated with the probability of private pension receipt. However, similar to the
empirical findings in the gender wage gap literature, there is heterogeneity in the effect;
being married appears to be a pension premium for men and a pension penalty for
women. The estimates show that a single woman who has never married is more likely
to have private pension coverage than a married woman, whereas there is no difference
between married and single men’s probability private coverage. This result is most
likely due related to individuals’ engagement with the labour market since women that
have never married are more likely to have stronger ties to the labour market than
married women, whereas men of differential marital status have stable work histories.
The estimate of marital status also show that widowed women are more likely to have
private pension coverage than married women, which is probably due to survivor
benefits paid out from a deceased spouse’s private pension. The negative effect of
widowhood for men may be due to the fact that they are older aged men that did not
have access to occupational pension schemes during their working life.
As expected, for both men and women, more years of work experience increases the
probability of receiving a private pension income, while time spent in self-employment
significantly reduces this probability. It is possible that the negative impact of self-
employment on private pension coverage arises because the majority of private pension
income originates from occupational pensions and the self-employed typically rely on
personal savings and assets to fund their retirement consumption.9 Starting wage of last
job before retirement is only an important determinant of women’s private pension
coverage status, consistent with the idea that women with higher wages are more likely
to contribute to a private pension scheme.
9 In 2012/13, Personal pensions account for 5% of retirement income for male pensioners and 2% for female pensioners (ONS Pension Trends 2012).
22
Time spent in part-time employment and occupational classification matters more for
women’s private pension coverage than men’s; women from top-level occupations are
significantly more likely to be receiving private pension income than women from
elementary occupations. The finding aligns with the evolution of private pension policy
in the UK; at the time the sample of pensioners were still of working age, employers
were permitted to exclude part-time workers from joining their pension scheme (thus
disproportionately affecting women). In addition, UK policy enabled employers to
impose pension scheme membership as a condition of contract until 1988 and this was
particularly common for top-level occupations. Including firm size in the model
specification does not change the qualitative results; as expected, the results in columns
2 and 4 indicate that individuals that worked in smaller firms are less likely to receive a
private pension.
Decomposition of the private pension coverage
The gender gap in private pension coverage is decomposed according to Equation 2 and
the detailed decompositions of the explained component are shown in Error!
Reference source not found.6. Gender differences in the distribution of characteristics
account for 51% of the gender gap in the average predicted probability of private
pension coverage. Including firm size in the model specification marginally decreases
the explained to 50%. If women had the same distribution of characteristics as men,
their average predicted probability of private pension coverage would increase by
around 0.12. Decompositions of the private pension coverage with and without firm size
suggests that gender disparities in the distribution of years work experience and part-
time work are the two most important factors in the explained gap in private pension
coverage.
Private Pension Income
OLS estimates of private pension income for the subsample of men and women with
private pension coverage are shown in Table 7. Columns 2 and 4 includes firm size of
last job before entering retirement which imply that the resultant estimates are based
on the sample of pensioners who were observed in employment in at least one wave of
23
ELSA or HSE. Overall, it is immediately clear that labour market characteristics are more
closely linked to private pension income than state pension income. There are strong
cohort effects for the male sample; men from earlier birth cohorts receive lower private
pension income than men born after 1942. This might be due the expansion of
occupational pension schemes that occurred in the 1950s and 1960s, which meant that
men from younger cohorts would have had a longer period of time to build up pension
entitlement. For the female sample, women born before 1934 have significantly lower
private pension income than women born after 1942. Across all specifications and for
both sexes, higher levels of educational attainment are strongly associated with higher
private pension income, though there are substantial differentials in the magnitude of
the effect. The return to education is much larger for men than women; columns 1 and 3
indicate that though the average man with degree-level education receives a private
pension income that is 88% greater than a man with no qualifications, this effect is only
56% for the female counterpart. Thus, the estimates suggest that while education is
more strongly associated with men’s private pension income, education matters most
for women’s coverage status.
Marital status is an important determinant of private pension income. For both men and
women, cohabiting individuals receive a larger amount of private pension income than
those that are currently married. However, similar to finding in private pension
coverage status, there is heterogeneity in the effect of marriage. Women that have never
married receive a higher amount of private pension income than married women,
whereas men that have never married receive a lower private pension than married
men. Again, women’s marriage pension penalty may be due to their fragmented work
histories. The returns to labour market experience are more important and greater for
women than men. Specifically, on average, an extra year of work experience between
ages of 20 and 60 years of age increases female private pension income by 1.6%,
whereas there is no statistical association between men’s private pension income and
accumulated work experience or years spent working part-time. As expected, an
additional year of part-time employment reduces women’s private pension income by
2.1% but has no effect on men’s private pension income.
24
For both sexes, occupational classification is an important determinant of private
pension income though the effect is stronger for women. Women from professional
occupations receive a private pension income that is approximately twice the amount
that women from elementary occupations receive, whereas this figure is 87% for the
male counterpart. Starting wage of last job before retirement does not seem to matter
for the amount of private pension income individuals receive, which may be due to the
fact that for the sample under analysis, defined benefit pensions were the main type of
pensions available during their working life and was typically related to final salary.
Once firm size is included in the model specification, coefficient estimates increase in
absolute terms. The size of firm an individual previously worked for is an important
correlate of the amount state pension income they receive. A woman who worked in a
firm with 2 to 4 workers can expect a pension income that is half as much as a woman
that worked in a firm with more than 1000 workers, while a man that previously
worked in a firm with 2 to 4 workers can expect about 66% less. This finding is
consistent with incentive models of the wage structure in which firms wage structure
follows a steeply rising age-earnings profiles so as to discourage shirking at work
(Lazear, 1981; Akerlof and Katz, 1989). In this framework, large firms will offer
pensions because pensions can be provide a cost-efficient way of monitoring worker
effort. This is because of pension back-loading and the substantial capital losses
associated with the termination of the employment contract prior to normal retirement
age.
Blinder-Oaxaca Decomposition
Columns 1 and 2 of TableError! Reference source not found. 8 show the Blinder-
Oaxaca decomposition of private pension income of without controlling for firm size.
The estimates show that men’s average weekly net private pension income is
approximately 50% higher than women’s average weekly net private pension income.
The decomposition indicates that only 28% of the gender gap in private pension income
is attributable to differences in the mean characteristics of men and women. The most
important factor contributing to the explained gap is the number of years women spend
in part-time employment, followed by gender differentials in accumulated labour
market experience. Gender differentials in the returns to marital status emerges as the
25
most important factor in the unexplained gap. Columns 3 and 4 of Table 8 include firm
size in the regression specification for private pension income. The returns to working
in firms of varying sizes does not significantly differ by gender. However, decomposition
estimates show that gender differences in the firm size of last job before retirement is
an important source of the explained component of the gap. Controlling for the size of
firm of last job before retirement reduces the explained component of the gap to 12%.
6.3 Quantile Regression
The analysis thus far has focused on the gender pension gap at the mean of the
conditional distribution of pension income. Consequently, the estimated coefficients
presented in section 6 are assumed to be constant across the entire distribution of
pension income. Research on the gender wage gap has provided evidence of
heterogeneity in the magnitude of the gap in along the wage distribution (Albrecht et al.
2003; Arulampalam et al. 2007). Figure 4 andFigure 5 show the observed raw gap
across the whole distribution of log state pension income and log private pension
income respectively. The red line represents the mean raw gender gap.
Figure 4 demonstrates that the raw gap in state pension income is not uniform across
the entire distribution. The magnitude of the gender gap steadily decreases as one
moves up the distribution of state pension income. For instance, a woman at the 20th
percentile in the female distribution of log state pension income has a state pension that
is around 55 log points less than a man at the 20th percentile of the male distribution,
whereas a woman at the 80th percentile of the female distribution of log state pension
has a state pension that is about 18 log points less than a man at the 80th percentile of
the male distribution. That plot suggests that there are “sticky floors” in state pension
benefit; the differential is much larger for individuals at the bottom 30% of the state
pension distribution than at the top. Thus, the graph suggests that the average gender
gap in state pension is driven by the gap at the bottom of the distribution. The larger
gap at the bottom of the distribution is likely to arise because there is an effective cap on
the amount of weekly state pension income an individual can receive.
26
Figure 5 shows the raw gender gap for the entire distribution of private pension
income. Relative to state pensions, the gap is fairly constant across the pension income
distribution. The raw gap in private pension income is above the mean gap between the
12th and 53rd percentile and declines until the 80th percentile at which point it starts to
increase. The largest gap in private pension income occurs around the 25th percentile.
The gender gap in private pension income is more volatile that the gap in state pension
income and is probably due to the diverse range of private pension schemes in
operation.
The plots of the raw gap across the distribution of pension income imply that OLS
estimates may not provide an appropriate description of the gender pension gap since
OLS specifies the conditional mean function. Thus, we extend our analysis of the gender
pension gap to quantile regressions because it allows for a richer description of the
gender gap by conditioning at specific quantiles of state (pension) income distribution.
The conditional quantile function of state (private) pension income is estimated for
each sex:
𝑄𝑔𝜏(𝑌𝑔|𝑋𝑔) = 𝑋𝑔′ 𝛽𝑔(𝜏) 𝑤ℎ𝑒𝑟𝑒 0 < 𝜏 < 1
where 𝑄𝑔𝜏(𝑌𝑔|𝑋𝑔) is the conditional quantile function for gender, 𝑔 , at the 𝜏𝑡ℎ quantile
of the distribution of log state (private) pension income conditional on a set of
explanatory variables, 𝑋𝑔, and the vector of quantile-specific coefficients, 𝛽𝑔(𝜏) .
𝛽𝑔(𝜏) is estimated by minimising the weighted sum of absolute residuals (Koenker and
Basset, 1978) as follows:
min𝛽𝑔
∑ 𝜏|𝑌𝑔 − 𝑋𝑔′ 𝛽𝑔(𝜏)|
𝑛
𝑖=1: 𝑦𝑖≥𝑋𝑔′ 𝛽𝑔
+ ∑ (1 − 𝜏)|𝑌𝑔 − 𝑋𝑔′ 𝛽𝑔(𝜏)|
𝑛
𝑖=1: 𝑦𝑖<𝑋𝑔′ 𝛽𝑔
27
Table 9 andTable 10 show the quantile regression estimates of state pension income at
the 25th, 50th and 75th quantiles for the male and female sample respectively. The
estimates indicate that birth cohort and work experience are the most important
determinants of state pension income for both sexes. Education is most important for
women at the median of the state pension income distribution, while current marital
status significantly determines women’s level of state pension income irrespective of a
woman’s location in the state pension income distribution. Starting wage of last job
before retirement had no relation to men’s state pension income level but is
significantly related to the level of state pension income received by women at the
median and 75th quantile. Consistent with Barrientos’ (1998) findings, this suggest that
men contracted out of the Additional State Pension (ASP) and into a private pension
during their working life whereas women remained contracted into the ASP and only
women in the top half of the distribution state pension income receive a pension from
the ASP.
The gender gap in state (private) pension income at specific quantiles is decomposed
using the Machado and Mata (2005) technique which decomposes the gap across the
entire distribution of state (private) pension income into the explained and unexplained
components. Specifically, the Machado-Mata decomposition involves generating the
counterfactual unconditional state (private) pension income distribution that would
arise if women had the same characteristics as men but were still rewarded as women
i.e. 𝑄𝜏𝑐(𝑌𝑓) = 𝑋𝑚𝛽𝑓(𝜏). Thus, the difference in state (private) pension income can be
expressed as:
Equation 3
𝑄𝜏(𝑌𝑚) − 𝑄𝜏(𝑌𝑓) = [𝑄𝜏(𝑌𝑚) − 𝑄𝜏𝑐(𝑌𝑓)] + [𝑄𝜏
𝑐(𝑌𝑓) − 𝑄𝜏(𝑌𝑓)]
28
The first term on the right hand side is the component of the gap due to gender
differences in the distribution of coefficients, while the second term is the component of
the gap due to gender differences in the distribution of observed characteristics. Table
11 shows the estimated state pension gap and the decomposition of the observed
conditional distribution. Three main results emerge. First, the results show that the
conditional state pension income gap declines as one moves up the state pension
distribution; the estimated gap in state pension income is around 70% higher at the 10th
quantile than the gap at the conditional mean. Second, the dominance of the coefficient
effect indicates that men are better rewarded than women across all points of the
distribution. Lastly, the decompositions suggests that the share of the explained
component increases as one moves up the distribution of state pension income.
Table 12 and Table 13 present the quantile estimates of private pension income for the
male and female sample respectively. Overall, the results are qualitatively similar to the
OLS estimates of private pension income presented in the previous sub-section, though
two patterns appear that are worth emphasising. First, the effect of current marital
status is greatest for women in the bottom quartile than women at the median or top
25% of the private pension income distribution. That is, the richest female pensioners
are not as affected by their marital status as female pensioners at or below the median.
Second, unlike OLS estimates, the quantile estimates show that starting wage of last job
before retirement is significantly related to the amount of private pension income
received by men in the top 25% and women in the bottom 25% of the distribution. This
might suggest that defined benefit pensions, which are a function of wages, delivered
better private pension incomes for male pensioners but resulted in a lower private
pension income for women. This result is consistent with issues surrounding the
portability of private pensions and women’s fragmented work histories.
Table 14 shows the estimated gap in private pension income for the 10th, 25th, 50th, 75th
and 90th percentiles. The estimated gap is largest at the 25th percentile. Similar to the
Oaxaca-Blinder decompositions, the decomposition estimates of the conditional
quantile functions at specific quantiles show that the majority of the gap is due to the
gender differences in coefficients. However, identical to Hänisch and Klos (2014), the
29
results indicate that share of gender differentials in the distribution of observed
characteristics is largest at the bottom of the distribution quantile.
7. Selection Bias in Private Pensions
Decomposition estimates of the gap in private pension income tentatively showed that
most of the gap is attributed to gender differentials in the returns to characteristics.
However, descriptive statistics have shown that private pension coverage is not
widespread; nearly than two-thirds of women and 88% of men were in receipt of an
income from private pensions. The theoretical literature discussed in section 3
suggested that workers do not randomly assign themselves with firms offering
occupational pensions but instead self-select into firms whose compensation pack align
with their own preferences, implying that those with private pension income are not
representative of the entire population of retirees.
We thus estimate the determinants of private pension income using the Heckman
selection model (1979) that explicitly models the selection equation for private pension
coverage to take into account the probability of private pension receipt in the
estimation of private pension income. The model estimated is set out as follows:
𝐶𝑖𝑔∗ = Z𝑖𝑔𝛾𝑔 + 𝜀𝑖𝑔 𝜀𝑖𝑔~𝑁(0,1)
𝐶𝑖𝑔 = { 1 𝑖𝑓 𝐶𝑖𝑔
∗ > 0
0 𝑖𝑓 𝐶𝑖𝑔∗ ≤ 0
𝑌𝑖𝑔 = 𝑋𝑖𝑔𝛽𝑔+ 𝑢𝑖𝑔 𝑢𝑖𝑔~(0, 𝜎2)
Equation 4
𝑌𝑖𝑔 = 𝑋𝑖𝑔𝛽𝑔 + 𝜌𝜀𝑢𝜎𝑢𝜆𝑖𝑔(𝑍𝑖𝑔𝛾𝑔)
30
where, 𝑍𝑖𝑔 is a vector of labour market and personal characteristics determining private
pension coverage for individual 𝑖 from sample 𝑔, 𝐶𝑖𝑔 is private pension coverage, 𝑌𝑖𝑔 is
private pension income and 𝑋𝑖𝑔 is a vector of labour market and personal
characteristics determining private pension income. The model assumes that the error
terms, 𝜀𝑖𝑔 and 𝑢𝑖𝑔, are correlated with each other (𝜌𝜀𝑢) and are normally distributed
with mean 0 and variance 𝜎2. 𝜆𝑖𝑔 is the ratio of the probability density function to the
cumulative distribution function which takes account of the selection bias. Exclusion
restrictions require that 𝑋𝑖𝑔 is a subset of Z𝑖𝑔, consequently, I exclude starting wage of
last job before retirement since this variable had weak statistical associations with
private pension income.
Robustness Checks
Table 15 presents the estimates of private pension income after correcting for selection.
The positive sign on rho indicates that unobservables in the private pension coverage
equation are positively associated with private pension income. The Blinder-Oaxaca
decomposition estimates of the gap in private pension income after correcting for
selection are shown in Table 16Table 16. The results indicate that 80% of the gap in
private pension income is explained by gender differentials in average characteristics,
with discrepancies in work experience accounting for most of the explained gap.
8. Discussion
Using data from the English Longitudinal Study of Ageing, this paper finds that despite
universal state pension coverage, there are large gender gaps in state pension income.
The conditional average weekly net state pension income for men is £147,
approximately 25% higher than women’s state pension income. The gender gap in
private pension outcomes is substantially larger; less than two-thirds of retired women
in England receive an income from a private pension, yet 88% of men have private
31
pension coverage. Given coverage, women’s average weekly net private pension income
amounts to £64, around half the amount than men (£123).
For both sexes, the analysis of state pensions income indicate that, apart from years of
work experience, personal characteristics such as birth cohort and marital status are
more important in determining the amount of state pension income individuals receive
than labour market characteristics. The results show that subsequent cohorts of women
have better state pension outcomes than earlier cohorts. This is consistent with the
abolition of gender-specific policies embedded in the state pension system and the
introduction of policies acknowledging women’s caring responsibilities within the
National Insurance system. The descriptive statistics and regression results based on
samples restricted to individuals with partners suggest that women’s state pension
income is closely linked to their spouse’s state pension income. Specifically, the analysis
provides evidence that a large proportion of women have derived pensions i.e. 60% of
their husband’s entitlement known as Category B pensions. It would have been
interesting to further distinguish state pension coverage into those with only Basic State
Pension coverage and those with entitlement to both the BSP and the Additional State
Pension. However, it is not possible to identify those with Additional State Pension
coverage because the basic state pension and the earnings-related component are
reported one payment.
Blinder-Oaxaca decompositions of state pension suggests that the unexplained
component accounts for the majority of the gender gap in state pension income; around
three-quarters of the gap in state pension income is attributable to differential effects of
coefficients. However, as mentioned before, it is important to bear in mind that any
omitted variables from the state pension income equations are subsumed into the
unexplained component of the gender pension gap. Yet, even when respondents’
partner’s state pension income is controlled for, the unexplained component reduces
marginally to only 74% and the gap in state pensions widens. Gender discrepancies in
the average number of years of work experience account for the majority of the
explained portion of the gap, while the differential effect of marital status is the most
32
important factor in the unexplained gap. Estimates of the gap in state pension income at
specific quantiles provide evidence of sticky floors in state pension income. That is, the
gap in state pension income is wider at the bottom of the distribution of income than at
the top. The conditional gap estimated at various quantiles suggest that the majority of
the gap is accounted for the by the coefficient effect, though similar to Hanisch and Klos
(2014), the share of the unexplained gap decreases as one moves up the distribution of
state pension income.
The gender gap in private pension income is 48%, marginally less than the 50%
estimated by Bardasi and Jenkins (2010) for the UK. Labour market characteristics are
more closely related to the private pension outcomes of men and women than state
pension outcomes. Education, work history and occupational classification are
significant determinants of private pension coverage. As expected, firm size had strong
statistical associations with private pension coverage; those who worked in larger firms
are more likely to be receiving a private pension in retirement. Decompositions of the
gap in the average predicted probability of private pension receipt suggest that gender
differentials in the distribution of characteristics account for half of the gap in coverage
rates. While there were little cohort effects in the probability of private pension
coverage for both sexes, birth cohort is an important determinant of the amount of
private pension men and women receive. Similar to the result in private pension
coverage, marital status is significantly associated with private pension income though
there is heterogeneity in the effect; married men experience a pension premium,
whereas there is a pension penalty for being a married woman.
Blinder-Oaxaca decompositions indicate that the explained component accounts for
only 12%-28% of the gap in private pension income. Of the unexplained component,
gender differentials in the returns to marital status is the most important factor. As was
the case in the analyses conducted by Even and Macpherson (1994) and Bardasi and
Jenkins (2010), this analysis also finds that characteristics matter most in explaining
private pension coverage than private pension income. However, the Blinder-Oaxaca
decompositions are not robust to sample selection; when selection corrections are
33
accounted for 80% of the gap is attributed to gender differentials in average
characteristics. Similar to quantile decompositions of state pension income, quantile
estimates of the gap in private pension income suggest that only a small proportion of
the gap is due to gender differentials in the distributions of characteristics.
The analysis undertaken in this paper has shown that historic inequalities in the labour
market outcomes of men and women manifest in the present. From its inception in
1948, the state pension system had embedded within it deeply gender-biased rules that
created disparities in men and women’s state pension outcomes. Nonetheless, the
analysis has shown that the state pension outcomes of successive cohorts of women is
the ability to build up entitlement when taking time out of the labour market for caring
will mean that future female pensioners will have better state pension outcomes than
the current cohort of female retirees. However, the gradual retrenchment of state
pensions means individuals will have to rely on private pensions to secure a retirement
income level that is in excess of the minimum standard of living maintained by the state
pension. This shift in focus implies that while private pension coverage rates among
female workers is likely to increase over time, more needs to be done in the workplace
to address the factors restricting women’s ability to secure adequate private pension
income. In particular, employers that best accommodate for family commitments by
providing childcare vouchers, job share options are likely to deliver better pension
outcomes for their female employees since gaps are driven by differentials in labour
market experience and length of working life spent in part-time employment.
34
Bibliography
Akerlof, G. A. and Katz, L. F. (1989) Workers’ Trust Funds and the Logic of Wage Profiles.
The Quarterly Journal of Economics 104(3) pp. 525-536.
Albrecht, J.B., Anders; Vroman, Susan, 2003. Is There a Glass Ceiling in Sweden? Journal
of Labour Economics, 21(1), pp. 145-177.
Allen, S. G., Clark, R. L. and McDermed, A. A. Pensions, Bonding and Lifetime Jobs. The
Journal of Human Resources, 28(3) pp. 463-481.
Altonji, J.G and Blank, M. R., (1999). Race and Gender in the Labour Market in Handbook
of Labor Economics, 3, pp. 3143-3259.
Arulampalam, W.B., L. Alison; Bryan, L. Mark, 2007. Is There a Glass Ceiling over Europe?
Exploring the Gender Pay Gap across the Wage Distribution. Industrial and Labor
Relations Review, 60(2), pp. 163-186.
Atkinson, A. B. (1994). State Pensions for Today and Tomorrow. David Hobman Annual
Lecture, (p. 8). London.
Bardasi, E.J., Stephen, (2010). The Gender Gap in Private Pensions. Bulletin of Economic
Research 62(4) pp 343-363.
Barrientos, A. (1998). Supplementary Pension Coverage in Britain. Fiscal Studies, 19(4),
pp. 429-446.
Beveridge, W. (1942). Social Insurance and Allied Services. London: Her Majesty's
Stationery Office.
Blake, D. (2003). Pension Schemes and Pension Funds in the United Kingdom. Oxford:
Oxford University Press.
Blau, D.F.K., M. Lawrence, 1999. Analyzing the Gender Pay Gap. The Quarterly Review of
Economics and Finance 39(5) pp 625-646.
Blinder, A. S. (1973). Wage discrimination: reduced form and structural estimates.
Journal of Human Resources 8(4) pp.436-455.
35
Bodie, Z. (1990) Pensions as Retirement Insurance. Journal of Economic Literature 28(1)
pp. 28-49.
Brown, J. (1990). Social Security for Retirement. York: Joseph Rowntree Foundation.
Department for Work and Pensions (2015). The Pensioners' Income Series 2013/14
accessed 14 December 2015
[https://www.gov.uk/government/uploads/system/uploads/attachment_data/file/43
8247/pensioners-incomes-series-2013-14-report.pdf]
Even, W. E. and Macpherson, D. A. (1994). Gender differences in pensions. Journal of
Human Resources, 29, pp.555-587.
Even, W.E. & Macpherson, D.A., (1996). Employer Size and Labour Turnover: The Role of
Pensions. Industrial and Labor Relations Review, 49(4), pp. 707-728.
Fairlie, R. W. (1999) The Absence of the African-American Owned Business: An Analysis of
the Dynamics of Self-Employment. Journal of Labour Economics 17(1) pp. 80-108.
Feldstein, M. (1974) Social Security and Saving: The Extended Life Cycle Theory. Journal
of Political Economy, Vol. 82, (5) pp. 905-926.
Ginn, J. (2003). Gender, Pensions and the Lifecourse. Bristol: The Policy Press.
Gustman, A. L. and Steinmeier, T. L. (1993) Pension Portability and Labour Mobility:
Evidence from the Survey if Income and Program Participation. Journal of Public
Economics 50(2) pp. 299-323.
Gustman, A. L., Mitchell, O. S. and Steinmeier, T. L. (1994) The Role of Pensions in the
Labour Market: A Survey of the Literature Industrial and Labour Relations Review Vol.
47(3) pp.417-438.
Hänisch, C. and Klos, J. (2014) A Decomposititon Analysis of the German Gender Pension
Gap. Discussion Paper Series (4) Universität Freiburg.
Heckman, J. (1979) Sample Selection Bias As A Specification Error. Econometrica 47(1)
pp. 153-161.
Hughes, G. and McComick, B. (1984) The Influence of Pensions on Job Mobility. Journal of
Public Economics 23 pp.183-206.
36
Ippolito, R. A. (1985). The Labor Contract and True Economic Pension Liabilities.
American Economic Review 75(5) pp. 1031-1043.
Ippolito, R. A (1987) Why Federal Workers Don't Quit. Journal of Human Resources
22(2) pp. 281-299.
Ippolito, R. A. (2002) Stayers as "Workers" and "Savers": Toward Reconciling the Pension-
Quit Literature. The Journal of Human Resources 37(2) pp. 275-308.
Jann, B. (2008). The Blinder-Oaxaca decomposition for linear regression models. The Stata
Journal 8 pp.453-479.
Koenker, R. and Bassett, G. J. (1978) Regression Quantiles. Econometrica 46 pp.33-50.
Kotlikoff, L.J. & Wise, D.A., (1988). "Pension Backloading, Implicit Wage Taxes, and Work
Disincentives," in: L.H. Summers, ed., Tax Policy and the Economy. MIT Press, pp. 161-
196.
Lazear, E. P. (1981). Agency, Earnings Profiles, Producitivity, and Hours Restrictions. The
American Economic Review 71(4) pp. 606-620.
Machado, J. and Mata, J. (2005). Counterfactual Decompositions of Changes in Wage
Distributions Using Quantile Regression. Journal of Applied Econometrics 20(4) pp. 445-
465.
Ministry of Pensions and National Insurance. (1966). Financial and Other Circumstances
of Retirement Pensioners: Report on an Enquiry by the Ministry of Pensions and National
Insurance with the Cooperation of the National Assistance Board. London: Her Majesty's
Stationery Office.
Modigliani, F. and Brumberg, R. (1954) "Utility Analysis and the Cosumption Function:
An Interpretation of Cross-section Data," in K. K. Kurihara, ed., Post-Keynesian
Economics. New Brunswick.
Oaxaca. (1973). Male-Female Wage Differentials in Urban Labour Markets. International
Economic Review , 14 (3), 693-709.
Office for National Statistics. (2012). Pension Trends: 2012 Edition. London.
37
Occupational Pensions Board (1976). Equal Status for Men and Women in Occupational
Pension Schemes: A report of the Occupational Pensions Board in accordance with Section
66 of the Social Security Act 1973. London: Her Majesty's Stationery Office.
38
Appendix
Descriptive Statistics
Table 1: Sample means by sex and pension status
Men Women
Full
sample
With state
pension
coverage
With private
pension
coverage
Full
sample
With state
pension
coverage
With private
pension
coverage
With pension coverage - 0.98 0.877 - 0.985 0.634
Log weekly net state/private pension
income - 4.99 4.81 - 4.702 4.16
Birth cohort
Born before 1934 0.38 0.37 0.37 0.32 0.32 0.32
Born 1934-1942 0.38 0.40 0.38 0.30 0.31 0.31
Born after 1942 0.24 0.23 0.25 0.37 0.37 0.38
Highest educational qualification
No qualifications 0.40 0.40 0.36 0.51 0.51 0.42
O-level(s) 0.16 0.16 0.16 0.21 0.21 0.23
A-level(s) 0.25 0.25 0.27 0.19 0.19 0.23
Degree 0.20 0.19 0.21 0.9 0.9 0.13
Marital status
Married (including civil partnership) 0.73 0.73 0.75 0.51 0.51 0.43
Cohabiting 0.02 0.02 0.02 0.03 0.03 0.04
Single, never married 0.05 0.05 0.04 0.04 0.04 0.06
Widowed 0.14 0.14 0.13 0.32 0.32 0.37
Divorced 0.05 0.05 0.05 0.09 0.09 0.08
Separated 0.01 0.01 0.01 0.01 0.01 0.01
Weekly net state pension income of
partner given coverage £82 £82 - £145 £145 -
Work history
Years of work experience between
ages 20-60 32.9 32.9 33.2 21.6 21.6 23.5
Years worked as part-time ages 20-
60 0.48 0.48 0.46 8 8 7.4
Years worked in self-employment
ages 20-60 3.3 3.3 3 1.3 1.3 1
Log weekly net starting wage of last
job before retirement 5.2 5.2 5.2 4.6 4.6 4.8
Occupational classification
Managers and senior officials 0.17 0.17 0.18 0.7 0.7 0.8
Professional occupations 0.14 0.14 0.15 0.11 0.11 0.15
Associate professional and technical
occupations 0.12 0.12 0.13 0.10 0.10 0.12
Administrative and secretarial
occupations 0.7 0.7 0.8 0.29 0.29 0.32
Skilled trades occupations 0.19 0.19 0.18 - - -
39
Personal service occupations 0.3 0.3 0.2 0.11 0.11 0.10
Sales and customer service
occupations 0.3 0.3 0.3 0.11 0.11 0.8
Process, plant and machine
operatives 0.14 0.13 0.12 0.5 0.5 0.3
Elementary occupations 0.11 0.11 0.11 0.17 0.17 0.12
Number of observations 1,188 1,175 1,042 1,577 1,554 1,000
40
Figure 1: Kernel density estimates of the state pension income distribution
Figure 2: Kernel density estimates of the female state pension income distribution by birth cohort
0.5
11
.52
Den
sity
2 3 4 5 6 7Log Weekly Net State Pension Income
Women Men
Conditional Distribution of State Pension Income by Sex0
.51
1.5
Den
sity
2 3 4 5 6Log Weekly Net State Pension Income
Born before 1934 Born 1934-42
Born after 1942
Conditional Distribution of Female State Pension Income by Cohort
41
Figure 3: Kernel density estimates of the private pension income distribution
0.1
.2.3
.4
Den
sity
0 2 4 6 8Log Weekly Net Private Pension Income
Women Men
Conditional Distribution of Private Pension Income by Sex
42
Regression Estimates
Table 2: Probit estimates of state pension coverage for pooled sample
Pooled sample VARIABLES
Female -0.187 (0.203) Birth cohort Born before 1934 0.269* (0.155) Born 1934-1942 0.422** (0.154) Highest educational qualification O-level(s) 0.113 (0.194) A-levels(s) 0.109 (0.179) Degree -0.055 (0.220) Married -0.309** (0.149) Work history Years of work experience between ages 20-60 0.004 (0.008) Years worked as part-time ages 20-60 0.007 (0.010) Years worked in self-employment ages 20-60 -0.010 (0.008) Log weekly net starting wage of last job before retirement 0.012 (0.077) Occupational classification Managers and senior officials -0.227 (0.293) Professional occupations -0.162 (0.297) Associate professional and technical occupations -0.423 (0.262) Administrative and secretarial occupations -0.060 (0.250) Skilled trades occupations -0.017 (0.370) Personal service occupations 0.018 (0.331) Sales and customer service occupations 0.029 (0.330) Process, plant and machine operatives 0.297 (0.403) Constant 2.267*** (0.526) Pseudo R-squared 0.046 Log-likelihood -183.171 Observations 2,765
Notes: *** p<0.01, ** p<0.05, * p<0.1; Robust standard errors in parentheses.
43
Table 3: OLS regression estimates of state pension income
Men Women
VARIABLES (1) (2) (3) (4)
Birth cohort
Born before 1934 0.090*** 0.123*** -0.150*** -0.217***
(0.032) (0.042) (0.027) (0.046)
Born 1934-1942 0.094*** 0.132*** -0.095*** -0.112***
(0.031) (0.043) (0.023) (0.033)
Highest educational qualification
O-level(s) 0.027 0.035 -0.000 0.028
(0.027) (0.036) (0.027) (0.041)
A-level(s) 0.015 0.028 0.023 0.008
(0.029) (0.035) (0.031) (0.041)
Degree 0.029 0.063 0.070* 0.033
(0.038) (0.049) (0.038) (0.060)
Marital status
Cohabiting 0.052 -0.030 0.341*** 0.407***
(0.053) (0.056) (0.039) (0.054)
Single, never married 0.014 - 0.345*** -
(0.061) - (0.046) -
Widowed 0.065** - 0.558*** -
(0.028) - (0.026) -
Divorced 0.027 - 0.393*** -
(0.038) - (0.033) -
Separated 0.050 - 0.251*** -
(0.061) - (0.095) -
Work history
Years of work experience between ages 20-60 0.005*** 0.004* 0.007*** 0.009***
(0.002) (0.003) (0.001) (0.002)
Years worked as part-time between ages 20-60 0.000 0.003 -0.005*** -0.009***
(0.003) (0.003) (0.001) (0.002)
Years worked in self-employment ages 20-60 -0.006*** -0.006*** -0.008*** -0.008**
(0.001) (0.001) (0.002) (0.003)
Log weekly net starting wage of last job before retirement
-0.002 -0.005 0.020* 0.025
44
(0.010) (0.012) (0.012) (0.016)
Occupational classification
Managers and senior officials 0.015 0.056 -0.006 -0.082
(0.035) (0.044) (0.044) (0.066)
Professional occupations -0.009 -0.013 -0.021 -0.026
(0.039) (0.052) (0.043) (0.065)
Associate professional and technical occupations 0.004 -0.005 0.060 0.028
(0.041) (0.050) (0.043) (0.065)
Administrative and secretarial occupations -0.066 0.010 0.030 0.007
(0.041) (0.051) (0.034) (0.051)
Skilled trades occupations 0.018 0.027 - -
(0.032) (0.040) - -
Personal service occupations -0.058 -0.039 -0.044 -0.116*
(0.050) (0.068) (0.040) (0.061)
Sales and customer service occupations 0.030 0.086 0.019 -0.041
(0.114) (0.196) (0.039) (0.060)
Process, plant and machine operatives -0.046 -0.022 0.068 0.042
(0.047) (0.070) (0.048) (0.064)
Log weekly net state pension income of partner given coverage
- 0.075* 0.130**
- (0.044) (0.061)
Constant 4.755*** 4.414*** 4.309*** 3.658***
(0.092) (0.221) (0.063) (0.321)
R-squared 0.047 0.052 0.314 0.180
Observations 1,175 794 1,554 632
Notes: *** p<0.01, ** p<0.05, * p<0.1; Robust standard errors in parentheses.
45
Table 4: Blinder-Oaxaca decomposition of the gender gap in state pension income
Gap % of gap Gap % of gap VARIABLES (1) (2) (3) (4)
Predicted average log state pension income: men 4.987*** 4.976***
(0.012) (0.016)
Predicted average log state pension income: women 4.632*** 4.438***
(0.016) (0.016)
Gap 0.355*** 100 0.538*** 100
(0.019) (0.023)
Characteristics 0.087*** 25 0.140*** 26
(0.019) (0.033)
Coefficients 0.268*** 75 0.399*** 74
(0.025) (0.040)
Detailed: Characteristics
Birth cohort -0.003 -4 -0.005 -4
(0.002) (0.004)
Education 0.002 2 0.001 1
(0.003) (0.004)
Marital status -0.059*** -68 -0.002 -1
(0.006) (0.002)
Years of work experience ages 20-60 0.093*** 107 0.108*** 77
(0.012) (0.019)
Years worked as part-time ages 20-60 0.046*** 53 0.070*** 50
(0.009) (0.014)
Years worked as self-employed ages 20-60 -0.012*** -14 -0.013*** -9
(0.003) (0.003)
Log weekly net starting income of last job before
retirement
0.006 7 0.006 4
(0.004) (0.005)
Occupational classification 0.014 16 0.010 7
(0.009) (0.013)
Log weekly net state pension income of partner - - -0.036* -26
Detailed: Coefficients
46
Birth cohort 0.007*** 3 0.004 1
(0.002) (0.005)
Education -0.005 -2 0.002 1
(0.005) (0.007)
Marital status -0.103*** -38 0.061*** 15
(0.012) (0.011)
Years of work experience ages 20-60 -0.072 -27 -0.162* -41
(0.068) (0.098)
Years worked as part-time ages 20-60 -0.007 -3 0.013* 3
(0.005) (0.008)
Years worked as self-employed ages 20-60 0.001 0 0.002 1
(0.004) (0.006)
Log weekly net starting income of last job before
retirement
-0.112 -42 -0.149 -37
(0.074) (0.098)
Occupational classification 0.102*** 38 -0.009 -2
(0.019) (0.012)
Log weekly net state pension income of partner - - -0.279 -70
Constant 0.456*** 170 0.916** 230
(0.101) (0.385)
Notes: *** p<0.01, ** p<0.05, * p<0.1; Robust standard errors in parentheses; Decompositions presented use women’s coefficients to weight the differential in characteristics.
47
Table 5: Probit estimates of private pension coverage
Men Women
VARIABLES (1) (2) (3) (4)
Birth cohort
Born before 1934 -0.262* 0.026 -0.114 -0.016
(0.142) (0.278) (0.097) (0.191)
Born 1934-1942 -0.035 0.008 -0.126 0.023
(0.136) (0.217) (0.088) (0.152)
Highest educational qualification
O-level(s) 0.189 0.458** 0.343*** 0.109
(0.155) (0.223) (0.092) (0.131)
A-level(s) 0.526*** 0.622*** 0.438*** 0.541***
(0.139) (0.209) (0.112) (0.171)
Degree 0.325* 0.205 0.597*** 0.197
(0.189) (0.269) (0.164) (0.226)
Marital status
Cohabiting -0.611** -0.758* 0.583*** 0.172
(0.307) (0.443) (0.222) (0.261)
Single, never married -0.001 -0.191 0.616*** 1.561***
(0.233) (0.340) (0.228) (0.451)
Widowed -0.229 -0.768*** 0.711*** 0.753***
(0.143) (0.244) (0.091) (0.150)
Divorced -0.399** -0.537* -0.127 -0.149
(0.199) (0.314) (0.125) (0.179)
Separated -0.276 0.046 -0.046 -0.057
(0.377) (0.499) (0.368) (0.472)
Work history
Years of work experience between ages 20-60 0.049*** 0.059*** 0.042*** 0.031***
(0.010) (0.017) (0.004) (0.007)
Years worked as part-time ages 20-60 -0.026 -0.070* -0.018*** -0.017***
(0.019) (0.038) (0.005) (0.006)
Years worked in self-employment ages 20-60 -0.018*** -0.019 -0.044*** -0.029**
(0.006) (0.012) (0.008) (0.014)
Log weekly net starting wage of last job before retirement
-0.037 0.017 0.103** 0.148**
(0.047) (0.068) (0.040) (0.058)
48
Occupational classification
Managers and senior officials 0.226 0.031 0.595*** 1.006***
(0.202) (0.296) (0.161) (0.241)
Professional occupations 0.273 0.416 0.914*** 1.114***
(0.241) (0.363) (0.179) (0.269)
Associate professional and technical occupations 0.449* 0.328 0.611*** 0.971***
(0.235) (0.351) (0.156) (0.222)
Administrative and secretarial occupations 0.420 0.749 0.451*** 0.893***
(0.287) (0.514) (0.109) (0.163)
Skilled trades occupations -0.162 -0.150 - -
(0.174) (0.273) - -
Personal service occupations -0.127 -0.638 0.273** 0.595***
(0.318) (0.446) (0.137) (0.211)
Sales and customer service occupations -0.694*** -0.782** -0.053 0.198
(0.259) (0.383) (0.129) (0.187)
Process, plant and machine operatives -0.298* -0.483* -0.512*** 0.054
(0.178) (0.266) (0.197) (0.267)
Firm size
2-4 - -0.444 - -0.978***
- (0.434) - (0.269)
5-19 - -0.501* - -0.602***
- (0.283) - (0.185)
20-99 - -0.773*** - -0.838***
- (0.274) - (0.220)
100-499 - -0.120 - -0.618***
- (0.404) - (0.230)
500-999 - -0.442* - -0.513***
- (0.265) - (0.179)
Constant -0.150 -0.389 -1.464*** -1.269***
(0.433) (0.752) (0.217) (0.326)
Observations 1,188 555 1,577 805
Notes: *** p<0.01, ** p<0.05, * p<0.1; Robust standard errors in parentheses.
49
Table 6: Fairlie decomposition of private pension coverage
Gap % of gap Gap % of gap
VARIABLES (1) (2) (3)
Probability of private pension receipt: men 0.879 0.879 Probability of private pension receipt: women 0.637 0.637 Gap 0.242 0.242 Total explained 0.124 51 0.120 50 Birth cohort -0.004 -3 -0.005* -4 (0.003) (0.003) Highest educational qualification 0.015*** 12 0.015*** 13 (0.004) (0.004) Marital status -0.029*** -23 -0.029*** -24 (0.005) (0.005) Years of work experience between ages 20-60 0.137*** 111 0.136*** 113 (0.012) (0.012) Years worked as part-time ages 20-60 0.039*** 32 0.038*** 32 (0.010) (0.010) Years worked in self-employment ages 20-60 -0.023*** -19 -0.023*** -19 (0.004) (0.004) Log weekly net starting wage of last job before retirement 0.015*** 12 0.014** 12 (0.006) (0.006) Occupational classification -0.026*** -21 -0.026*** -22 (0.009) (0.009) Firm size - -0.000 0 - (0.001) Notes: *** p<0.01, ** p<0.05, * p<0.1; Robust standard errors in parentheses. Decompositions presented use women’s coefficients to weight the differential in the distribution of characteristics.
50
Table 7: OLS estimates of private pension income
(1) (2) VARIABLES Men Men firm
size Women Women
firm size
Birth cohort Born before 1934 -0.282*** -0.013 -0.186** -0.206 (0.078) (0.246) (0.093) (0.230) Born 1934-1942 -0.164** -0.066 0.006 -0.071 (0.075) (0.186) (0.082) (0.173) Highest educational qualification O-level(s) 0.274*** 0.539** 0.181** 0.224 (0.097) (0.214) (0.091) (0.161) A-level(s) 0.190** 0.552*** 0.311*** 0.813*** (0.089) (0.182) (0.092) (0.182) Degree 0.878*** 0.875*** 0.564*** 0.762*** (0.101) (0.249) (0.116) (0.250) Marital status Cohabiting 0.498*** -0.213 0.255* 0.376 (0.146) (0.530) (0.138) (0.278) Single, never married -0.414** -0.560 0.422*** 1.256*** (0.177) (0.400) (0.124) (0.223) Widowed -0.196** -0.880*** 0.602*** 1.175*** (0.097) (0.288) (0.088) (0.169) Divorced -0.183 -0.770** -0.037 -0.240 (0.170) (0.376) (0.135) (0.214) Separated -0.132 -0.311 -0.318 -0.584 (0.221) (0.484) (0.331) (0.568) Work history Years of work experience between ages 20-60
0.007 0.071*** 0.016*** 0.047***
(0.008) (0.020) (0.004) (0.008) Years worked as part-time ages 20-60 0.000 -0.027 -0.021*** -0.030*** (0.020) (0.047) (0.004) (0.008) Years worked in self-employment ages 20-60
-0.024*** -0.033** -0.036*** -0.052***
(0.005) (0.015) (0.010) (0.019) Log weekly net starting wage of last job before retirement
0.017 0.043 0.051 0.169**
(0.035) (0.072) (0.038) (0.068) Occupational classification Managers and senior officials 0.721*** 0.708*** 0.480*** 1.361*** (0.141) (0.267) (0.162) (0.278) Professional occupations 0.874*** 1.077*** 1.052*** 1.863*** (0.137) (0.269) (0.148) (0.286) Associate professional and technical occupations
0.848*** 0.996*** 0.761*** 1.543***
(0.139) (0.274) (0.152) (0.263) Administrative and secretarial occupations 0.725*** 1.102*** 0.444*** 1.341*** (0.158) (0.290) (0.120) (0.203) Skilled trades occupations 0.174 0.052 - - (0.135) (0.242) - - Personal service occupations 0.338 -0.238 0.151 0.708*** (0.206) (0.619) (0.149) (0.271) Sales and customer service occupations 0.359** -0.276 0.231 0.277 (0.177) (0.500) (0.166) (0.250) Process, plant and machine operatives 0.312** -0.206 0.128 0.156 (0.135) (0.275) (0.268) (0.394) Firm size
51
2-4 -0.655 -1.146*** (0.413) (0.303) 5-19 -0.415 -0.678*** (0.265) (0.219) 20-99 -0.911*** -1.140*** (0.309) (0.278) 100-499 -0.465 -0.924*** (0.295) (0.290) 500-999 -0.456** -0.509** (0.222) (0.214) Constant 3.985*** 1.491* 2.884*** 0.124 (0.350) (0.864) (0.220) (0.379) Observations 1,042 555 1,000 805 R-squared 0.298 0.269 0.254 0.378
Notes: *** p<0.01, ** p<0.05, * p<0.1; Robust standard errors in parentheses.
52
Table 8: Blinder-Oaxaca decomposition of the gender gap in private pension income
Gap % of gap Gap % of gap VARIABLES (1) (2) (3) (4)
Predicted average log state pension income: men 4.687*** 4.755*** (0.042) (0.085) Predicted average log state pension income: women 3.987*** 3.709*** (0.053) (0.172) Gap 0.700*** 100 1.046*** 100 (0.068) (0.192) Characteristics 0.199*** 28 0.130* 12 (0.059) (0.076) Coefficients 0.501*** 72 0.917*** 88 (0.082) (0.187) Detailed: Characteristics Birth cohort -0.012* -6 -0.014* -11 (0.007) (0.007) Highest educational qualification 0.019* 10 0.019* 15 (0.010) (0.010) Marital status 0.019 10 -0.021 -16 (0.020) (0.052) Years of work experience between ages 20-60 0.134*** 67 0.135*** 104 (0.035) (0.035) Years worked as part-time ages 20-60 0.144*** 72 0.142*** 109 (0.029) (0.029) Years worked in self-employment ages 20-60 -0.052*** -26 -0.052*** -40 (0.011) (0.011) Log weekly net starting wage of last job before retirement 0.013 7 0.012 9 (0.011) (0.011) Occupational classification -0.065** -33 -0.068** -52 (0.031) (0.030) Firm size - - -0.023*** -18 - - (0.008) Detailed: Coefficients Birth cohort -0.008*** -2 -0.009*** -1 (0.003) (0.003) Highest educational qualification -0.003 -1 -0.004 0 (0.019) (0.019) Marital status -0.219*** -44 0.366 40 (0.046) (0.231) Years of work experience between ages 20-60 -0.281 -56 -0.232 -25 (0.279) (0.282) Years worked as part-time ages 20-60 0.009 2 0.004 0 (0.019) (0.020) Years worked in self-employment ages 20-60 0.015 4 0.014 2 (0.015) (0.015) Log weekly net starting wage of last job before retirement -0.168 -34 -0.161 -18 (0.250) (0.251) Occupational classification 0.127* 25 -0.015 -2 (0.069) (0.030) Firm size - - -0.016 -2 - - (0.037) Constant 1.029*** 205 0.969** 106 (0.381) (0.391) Notes: *** p<0.01, ** p<0.05, * p<0.1; Robust standard errors in parentheses; Decompositions presented use women’s coefficients to weight
53
the differential in characteristics.
Quantile Estimates
Figure 4: Raw gender differential in state pension income by quantile conditional on coverage
Figure 5: Raw gender differential in private pension income by quantile conditional on coverage
0.2
.4.6
.81
Log
Sta
te P
en
sio
n In
com
e D
iffe
ren
tial
0 .2 .4 .6 .8 1Quantile
0.2
.4.6
.81
Log
Pri
vate
Pen
sio
n In
com
e D
iffe
ren
tial
0 .2 .4 .6 .8 1Quantile
54
Table 9: Quantile estimates of state pension income for male sample without partner’s pension income
25th Median 75th VARIABLES: (1) (2) (3)
Birth cohort Born before 1934 0.072*** 0.082*** 0.111*** (0.019) (0.017) (0.021) Born 1934-1942 0.064*** 0.057*** 0.133*** (0.015) (0.017) (0.021) Highest educational qualification O-level(s) 0.010 -0.000 -0.006 (0.016) (0.022) (0.019) A-levels(s) -0.016 -0.014 0.014 (0.018) (0.019) (0.030) Degree -0.014 -0.013 -0.030 (0.027) (0.024) (0.027) Marital status Cohabiting -0.013 -0.010 0.038 (0.019) (0.025) (0.073) Single, never married 0.001 -0.017 -0.009 (0.052) (0.059) (0.024) Widowed 0.031* 0.029 0.029 (0.016) (0.024) (0.028) Divorced -0.013 0.011 -0.009 (0.031) (0.020) (0.050) Separated 0.043 0.052 0.091 (0.069) (0.042) (0.125) Work history Years of work experience between ages 20-60 0.005*** 0.004*** 0.002*** (0.002) (0.002) (0.000) Years worked as part-time ages 20-60 0.004 0.001 -0.000 (0.003) (0.002) (0.003) Years worked in self-employment ages 20-60 -0.007*** -0.008*** -0.007*** (0.000) (0.001) (0.001) Log weekly net starting wage of last job before retirement
0.003 0.004 0.007
(0.007) (0.007) (0.000) Occupational classification Managers and senior officials -0.017 0.020 0.020 (0.025) (0.025) (0.044) Professional occupations 0.008 0.035 -0.019 (0.035) (0.028) (0.043) Associate professional and technical occupations
-0.014 0.006 -0.010
(0.024) (0.025) (0.025) Administrative and secretarial occupations -0.059 -0.021 -0.076*** (0.041) (0.034) (0.024) Skilled trades occupations -0.008 0.014 -0.004 (0.025) (0.025) (0.025) Personal service occupations -0.039 -0.023 -0.087 (0.107) (0.040) (0.000) Sales and customer service occupations -0.004 0.018 0.011 (0.048) (0.061) (0.021) Process, plant and machine operatives -0.017 0.007 -0.000 (0.024) (0.021) (0.032) Constant 4.652*** 4.786*** 4.982 (0.079) (0.066) (0.000) Pseudo R-squared 0.056 0.049 0.044
55
Observations 1175 1175 1175 Notes: *** p<0.01, ** p<0.05, * p<0.1; Robust standard errors in parentheses.
56
Table 10: Quantile estimates of state pension income for female sample without partner’s state pension income
25th Median 75th VARIABLES: (1) (2) (3)
Birth cohort Born before 1934 -0.132*** -0.177*** -0.121*** (0.020) (0.022) (0.027) Born 1934-1942 -0.074*** -0.145*** -0.092*** (0.015) (0.023) (0.022) Highest educational qualification
O-level(s) -0.016 -0.007 -0.008 (0.014) (0.023) (0.028) A-levels(s) 0.031* 0.090*** 0.041* (0.019) (0.029) (0.023) Degree 0.040* 0.101*** 0.031 (0.023) (0.038) (0.037) Marital status Cohabiting 0.458*** 0.359*** 0.200*** (0.045) (0.027) (0.027) Single, never married 0.470*** 0.363*** 0.271*** (0.066) (0.028) (0.039) Widowed 0.645*** 0.614*** 0.520*** (0.019) (0.023) (0.025) Divorced 0.497*** 0.411*** 0.300*** (0.012) (0.032) (0.024) Separated 0.076 0.284** 0.352*** (0.089) (0.122) (0.037) Work history Years of work experience between ages 20-60
0.005*** 0.008*** 0.007***
(0.001) (0.001) (0.001) Years worked as part-time ages 20-60
-0.003*** -0.006*** -0.005***
(0.001) (0.001) (0.001) Years worked in self-employment ages 20-60
-0.005*** -0.009*** -0.007***
(0.001) (0.001) (0.003) Log weekly net starting wage of last job before retirement
0.002 0.020** 0.024**
(0.005) (0.009) (0.010) Occupational Classification Managers and senior officials -0.033 -0.025 -0.019 (0.032) (0.047) (0.040) Professional occupations -0.017 -0.015 -0.043 (0.031) (0.039) (0.043) Associate professional and technical occupations
-0.002 -0.034 -0.008
(0.018) (0.037) (0.048) Administrative and secretarial occupations
0.038** 0.009 0.011
(0.017) (0.028) (0.036) Skilled trades occupations - - - - - - Personal service occupations -0.023 -0.043 -0.087* (0.021) (0.034) (0.046) Sales and customer service occupations
0.028 -0.015 -0.042
(0.020) (0.026) (0.048)
57
Process, plant and machine operatives
0.031 0.049 -0.007
(0.026) (0.032) (0.045) Constant 4.214*** 4.332*** 4.565*** (0.030) (0.053) (0.061) Pseudo R-squared 0.294 0.257 0.202 Observations 1,554 1,554 1,554
Notes: *** p<0.01, ** p<0.05, * p<0.1; Robust standard errors in parentheses.
58
Table 11: Decomposition of the gender gap in state pension income using quantile regression
Total gap Effects of: Characteristics Coefficients
10th 0.606 [100%] -0.036 [-6%] 0.642 [106%] (0.032) (0.059) (0.062) 25th 0.645 [100%] -0.016 [-3%] 0.660 [103%] (0.013) (0.043) (0.042) Median 0.610 [100%] -0.015 [-3%] 0.626 [103%] (0.021) (0.044) (0.047) 75th 0.460 [100%] 0.024 [5%] 0.437 [95%] (0.024) (0.053) (0.053) 90th 0.406 [100%] 0.119 [29%] 0.286 [71%] (0.027) (0.068) (0.067)
Note: Bootstrap standard errors with 100 replications in parentheses. All pension gaps are significant at the 1% level.
59
Table 12: Quantile estimates private pension income male sample without firm size
(1) (2) (3) VARIABLES: 25th Median 75th
Birth cohort Born before 1934 -0.390*** -0.177** -0.193*** (0.089) (0.069) (0.062) Born 1934-1942 -0.176** -0.100* -0.105** (0.079) (0.057) (0.053) Highest educational qualification
O-level(s) 0.218* 0.232** 0.204** (0.122) (0.094) (0.082) A-levels(s) 0.238** 0.239*** 0.208*** (0.120) (0.076) (0.048) Degree 1.034*** 0.791*** 0.628*** (0.126) (0.079) (0.085) Marital status Cohabiting 0.452** 0.322** 0.219 (0.188) (0.140) (0.258) Single, never married -0.458*** -0.358*** -0.331* (0.164) (0.125) (0.180) Widowed -0.279** -0.226** -0.160** (0.132) (0.109) (0.077) Divorced -0.287** -0.092 -0.284*** (0.139) (0.136) (0.052) Separated -0.005 -0.204** -0.489*** (0.212) (0.083) (0.069) Work history Years of work experience between ages 20-60
0.018*** 0.014** 0.005
(0.007) (0.007) (0.007) Years worked as part-time ages 20-60
-0.020** 0.002 0.028***
(0.010) (0.032) (0.010) Years worked in self-employment ages 20-60
-0.030*** -0.025*** -0.020***
(0.003) (0.006) (0.003) Log weekly net starting wage of last job before retirement
0.009 0.041 0.062***
(0.036) (0.026) (0.022) Occupational classification Managers and senior officials 0.905*** 0.684*** 0.724*** (0.221) (0.143) (0.085) Professional occupations 0.949*** 0.611*** 0.544*** (0.208) (0.140) (0.097) Associate professional and technical occupations
0.966*** 0.645*** 0.628***
(0.209) (0.146) (0.100) Administrative and secretarial occupations
0.710*** 0.545*** 0.523***
(0.245) (0.136) (0.138) Skilled trades occupations 0.105 0.058 0.073 (0.213) (0.144) (0.078) Personal service occupations 0.699 0.257 -0.008 (0.599) (0.175) (0.227) Sales and customer service occupations
0.475** 0.136 0.012
(0.207) (0.189) (0.120)
60
Process, plant and machine operatives
0.370* 0.132 0.139*
(0.216) (0.148) (0.076) Constant 3.184*** 3.888*** 4.580*** (0.370) (0.299) (0.279) Pseudo R-squared 0.195 0.204 0.217 Observations 1,042 1,042 1,042
Notes: *** p<0.01, ** p<0.05, * p<0.1; Robust standard errors in parentheses.
61
Table 13: Quantile estimates private pension income female sample without firm size
(1) (2) (3) VARIABLES: 25th Median 75th
Birth cohort Born before 1934 -0.243* -0.251*** -0.154 (0.133) (0.093) (0.000) Born 1934-1942 -0.096 -0.053 0.115** (0.092) (0.079) (0.049) Highest educational qualification
O-level(s) 0.136 0.158 0.231*** (0.112) (0.109) (0.080) A-levels(s) 0.420*** 0.260** 0.290*** (0.117) (0.101) (0.054) Degree 0.837*** 0.418*** 0.456*** (0.148) (0.106) (0.085) Marital status Cohabiting 0.441*** 0.139* 0.082 (0.096) (0.082) (0.123) Single, never married 0.625*** 0.265** 0.247*** (0.184) (0.129) (0.063) Widowed 0.732*** 0.536*** 0.554*** (0.117) (0.088) (0.070) Divorced 0.135 0.000 -0.000 (0.192) (0.155) (0.063) Separated -0.597 -0.703 -0.337 (0.673) (0.637) (0.000) Work history Years of work experience between ages 20-60
0.016*** 0.019*** 0.017***
(0.006) (0.003) (0.000) Years worked as part-time ages 20-60
-0.022*** -0.028*** -0.025***
(0.004) (0.004) (0.004) Years worked in self-employment ages 20-60
-0.048** -0.050*** -0.036**
(0.024) (0.010) (0.015) Log weekly net starting wage of last job before retirement
0.105** 0.038 0.063
(0.045) (0.036) (0.000) Occupational classification Managers and senior officials 0.444** 0.540** 0.642*** (0.190) (0.229) (0.132) Professional occupations 1.209*** 1.177*** 0.828*** (0.126) (0.148) (0.063) Associate professional and technical occupations
0.809*** 0.900*** 0.720***
(0.152) (0.157) (0.087) Administrative and secretarial occupations
0.568*** 0.448*** 0.483***
(0.130) (0.139) (0.073) Skilled trades occupations - - - - - - Personal service occupations 0.100 0.160 0.131 (0.145) (0.187) (0.122) Sales and customer service occupations
0.119 0.485** 0.349***
(0.203) (0.224) (0.083)
62
Process, plant and machine operatives
-0.009 0.318 0.169
(0.230) (0.315) (0.242) Constant 1.960*** 3.133*** 3.533 (0.297) (0.183) (0.000) Pseudo R-squared 0.166 0.179 0.157 Observations 1,000 1,000 1,000
Notes: *** p<0.01, ** p<0.05, * p<0.1; Robust standard errors in parentheses.
63
Table 14: Decomposition of the gender gap in private pension income (without firm size) using quantile regression
Total gap Effects of: Characteristics Coefficients
10th 0.731 [100%] 0.018 [3%] 0.713 [97%] (0.135) (0.337) (0.383) 25th 0.831 [100%] 0.130 [16%] 0.701 [84%] (0.062) (0.255) (0.269) Median 0.701 [100%] 0.022 [3%] 0.679 [97%] (0.062) (0.140) (0.151) 75th 0.583 [100%] -0.055 [-9%] 0.638 [109%] (0.054) (0.104) (0.089) 90th 0.545 [100%] -0.094 [-117%] 0.639 [117%] (0.054) (0.115) (0.081)
Note: Bootstrap standard errors with 100 replications in parentheses. All pension gaps are significant at the 1% level.
64
Robustness Checks
Table 15: Estimates of Private Pension Income with Selectivity Corrections
Men Women
Log PPI Log PPI
VARIABLES
Birth cohort
Born before 1934 -0.301*** -0.167*
(0.085) (0.095)
Born 1934-1942 -0.160** 0.014
(0.080) (0.083)
Highest educational qualification
O-level(s) 0.298*** 0.170*
(0.096) (0.094)
A-level(s) 0.110*** 0.167***
(0.042) (0.046)
Degree 0.304*** 0.189***
(0.035) (0.038)
Marital status
Cohabiting 0.494** 0.285**
(0.238) (0.137)
Single, never married -0.415*** 0.420***
(0.156) (0.122)
Widowed -0.181* 0.590***
(0.096) (0.090)
Divorced -0.183 -0.076
(0.150) (0.146)
Separated -0.135 -0.289
(0.303) (0.335)
Work history
Years of work experience between ages 20-60 0.006 0.018***
(0.008) (0.004)
Years worked as part-time ages 20-60 0.001 -0.021***
65
(0.015) (0.004)
Years worked in self-employment ages 20-60 -0.026*** -0.036***
(0.004) (0.010)
Socio-economic classification
Managers and senior officials 0.702*** 0.542***
(0.127) (0.164)
Professional occupations 0.868*** 1.100***
(0.140) (0.149)
Associate professional and technical occupations 0.848*** 0.811***
(0.134) (0.154)
Administrative and secretarial occupations 0.722*** 0.505***
(0.149) (0.123)
Skilled trades occupations 0.163 -
(0.120) -
Personal service occupations 0.341 0.120
(0.221) (0.158)
Sales and customer service occupations 0.350 0.218
(0.219) (0.172)
Process, plant and machine operatives 0.317** 0.169
(0.128) (0.267)
Constant 4.073*** 3.004***
(0.317) (0.178)
ρ 0.029 0.031
(0.178) (0.056)
σ 0.988 1.044
(0.021) (0.030)
Average mills value 0.223 0.605
Log likelihood -1862.765 -2327.335
Observations 1188 1577
Notes: *** p<0.01, ** p<0.05, * p<0.1; Robust standard errors in parentheses.
66
Table 16: Blinder-Oaxaca + Heckman Private Pension
Gap % of gap VARIABLES (1) (2)
Predicted average log private pension income: men 4.739*** (0.038) Predicted average log private pension income: women
4.113***
(0.048) Gap 0.625*** 100 (0.062) Characteristics 0.497*** 80 (0.092) Coefficients 0.128 20 (0.103) Detailed: Characteristics Birth cohort -0.022** -4 (0.011) Highest educational qualification 0.006 1 (0.013) Marital status -0.154*** -31 (0.028) Years of work experience between ages 20-60 0.605*** 122 (0.051) Years worked as part-time ages 20-60 0.272*** 55 (0.040) Years worked in self-employment ages 20-60 -0.093*** -19 (0.017) Socio-economic classification -0.117** -24 (0.047) Size - - Detailed: Coefficients Birth cohort 0.001 1 (0.009) Highest educational qualification 0.009 7 (0.022) Marital status 0.094 73 (0.081) Years of work experience between ages 20-60 -0.818*** -639 (0.284) Years worked part-time ages 20-60 -0.117*** -91 (0.039) Years worked in self-employment ages 20-60 0.048*** 38 (0.019) Socio-economic classification 0.074 59 (0.048) Constant 0.835*** 652 (0.288)
Notes: *** p<0.01, ** p<0.05, * p<0.1; Robust standard errors in parentheses. Decompositions presented use women’s coefficients to weight the differential in characteristics.
67
Table of the Existing Literature on the Gender Pension Gap
Study Country of
Analysis Methodological Approach Main Results
Even and Macpherson (1994)
United States Probit OLS Estimates from the above models are used to predict occpational pension coverage rates and benefit income levels when men and women assume the characteristics of one another.
Gender differences in observed characteristics explain 80% of the gap in occupational pension income. Around 70% of the gap in occupational pension benefit level is explained by gender differentials in characteristics.
Bardasi and Jenkins (2010)
United Kingdom
Heckman selection model. Blinder-Oaxaca decomposition of private pension income. Gomulka-Stern decomposition of private pension income receipt probabilities.
Differences in returns account for at least 80% of gap in private pension income. Differences in characteristics account for between 33% to 42% of the gap in private pension income receipt probabilities.
Hänisch and Klos (2014)
Germany OLS Blinder-Oaxaca Quantile regression Firpo decomposition
Blinder-Oaxaca decompositions show that the explained component accounts for 26% of the gap in mean pension income, with employment and education contributing most to the explained gap. The magnitude of the gender pension gap declines for increasing quantiles of the pension income distribution. Across the entire distribution, the unexplained gap is larger than the explained gap in pension income. The proportion of the gap attributed to the explained component is largest at the bottom of the distribution.