Time Use, Consumption Expenditures and

Time Use, Consumption Expenditures and
Employment Status: Evidence from the
LISS Panel
Fabrizio Colella∗ , Arthur van Soest†
Preliminary Version for the 7th MESS Workshop
Den Haag - August, 2013
Abstract
In this paper we analyze individual and household consumption expenditure patterns and time use patterns in the Netherlands using three waves
of panel data collected in the LISS Panel, containing detailed retrospective
information on time use in the past seven days and consumption expenditures in the past month. We discuss the extent to which variation in
individuals’ work status leads to variation in time use profile and household and individual consumption patterns, controlling for observed and
unobserved household and individual specific characteristics. Applying a
Dynamic Random Effects Tobit model to the longitudinal data, we find
changes in consumption expenditure allocation due to the improvement of
income and occupation status of the subjects in the sample. More specifically, we point out that the non-employed tend to use both time and wealth
in other productive activities that are not included in the computation of
GDP. We revisit the retirement consumption puzzle proving both a decline
in expenditures for food consumed away from home and an increase in time
spent on home production for retired individuals and other non-workers.
Furthermore we show that household consumption is strongly related to
family characteristics but not much to employment status and earnings of
its household head.
Keywords:
∗
†
Time use - Consumption expenditure - Retirement consumption
puzzle - Home production - Household resources allocation
Bocconi University
Tilburg University
1
1
Introduction
The most of the economic theory and policy requires a good understanding of household
or individual consumer behavior. Moreover, the necessity to give empirical evidence supporting the theoretical framework on consumer behavior, together with the necessity to
compare the different models applications to different countries according with their intrinsic characteristics, makes the analysis of the economic agents’ habits an interesting
area for research (see Blundell, 1988).
The attractiveness of this research framework increases the availability of data on
consumer behavior, giving to economists different avenues for research. Among others,
surveys about consumer expenditures and time use became very spread and popular.
Using these categories of surveys, facts about how people spend their time and wealth
in relation to the gender, whether they live with a partner or not and whether they are
employed, unemployed or out of the labor force, are often analyzed for different purposes.
The related topics concern: the use of resources and the time allocation between paid work
and home production of the household components in relation to their personal income or
employment status; the comparison of the work rhythms between the agents in different
countries and their evolution over the last decades; changes in the individual expenditure
upon retirement (the literature refers to such a finding as ”the retirement consumption
puzzle”).
In this paper we consider different subjects related to the individual and household
behavior and, through empirical applications, we carry out evidences from a novel and
unique dataset: the time use and consumption module of the Longitudinal Internet Studies
for the Social sciences (LISS) panel, which is a regular survey representative for the Dutch
population, gathered by the CentERdata of Tilburg University.
The paper is structured as follows: section II contains a review of the different studies
about similar topics and using similar surveys. Section III contains a description of the
LISS panel, a brief evaluation of the time use and consumption module and some characteristics of the subsample used. Section IV illustrates the empirical model used for the
investigation. In section V we report the main findings of our regressions dividing them in
terms of both the time use and the consumption expenditure analysis. Finally, in section
VI conclusions are drawn.
2
Literature Review
The way in which individuals spend their time away from work, has been a theoretical
issue faced by many economic studies over the past 50 years. Contextually, micro data
measuring, in a consistent way, the daily individual time allocation became very useful for
empirical researches (see Aguiar, Hurst, and Karabarbounis (2012) for a review of recent
trends in a time use context). Since Becker (1965), with his general theory, provided an
interesting framework to analyze the allocation of time between different activities, many
economists started to modeling the individual and household behavior and to produce
2
empirical evidences about it.
For example, Aguiar and Hurst (2007) and Aguiar, Hurst, and Karabarbounis (2012)
examine the trends in allocation of time over the past five decades in the U.S. documenting
a large increment in leisure time, hand in hand with a decline in market work hours and
in home production hours - the largest increment in leisure time address the less educated
portion of population, and the downfall of the no-market work (home production) is greater
for females. In a research on seven European Industrialized Countries over the same
period, Gimenez-Nadal and Sevilla (2012) find that the decrement in men’s market work
is also driven by increments in men’s no-market work and childcare, while the decreases
in women’s home production is accompanied by an increase in market work. Similar
results are found by Ramey (2008) studying time allocation of Americans during the
entire twentieth century.
Focusing particularly on employment status in a household dimension in Italy, Pasqua
and Mancini (2012) show a substantial reduction in care time, both quality and basic,
that parents spend with their children, resulting from the choice of the mothers to enter
in the job market; this is partially compensated by a higher share of time devoted from
the fathers to their children. According with their results, the mother, even if employed
in market work, still remains the spouse more in charge for the childcare. The American
Time Use Survey (ATUS) also reveals that even self-employed women spent less amount of
time working and more on childcare activities, even after controlling for a variety of other
demographic characteristics - this result is reported in a research developed by GurleyCalvez et al. (2009). On the other hand, Cherchye et al. (2010), in a study using the LISS
Panel, do not find relevant differences in time spent in childcare among Dutch mothers and
fathers. Moreover, exploiting data from national-time use surveys in France, Sweden, Italy
and United States, a study by Anxo et al. (2011), highlights large gender discrepancies
in allocation of time in each country at all life stage, motivated by different institutional
contexts, family policies and employment regimes - consistent with the previous literature,
Sweden results the country with the lowest gender gap.
An important strand of research using different dataset on time use addresses changes
in time devoted to different productive activities on different life stages, in particular in
the life period just before and right after the retirement. In the book ”Time for Life:
The Surprising Ways Americans Use Their Time”, Godbey and Robinson (1999) stressed,
among others, also this interesting topic; according to them, the way in which Americans
use their time at an advanced stage of their life varies as a function of their devotion to
the paid job rather than their age. Similar conclusions are drawn by Krantz-Kent and
Stewart (2007), in an investigation using the ATUS: they find that the time allocation to
different activities changes dramatically with age and that what matters most for this is the
lower employment rates at older ages, rather than age itself. With reference to the same
issue, in a recent work on French population, Stancanelli and van Soest (2011) stressed
the causal effect of retirement on time use of both partners in a couple. Exploiting data
from a time use diary, they conclude that no-market work time increases with retirement,
in particular semi-leisure activities increase most for the husbands and house chores and
weekdays cooking for the wives.
3
The changes in individuals’ behavior upon retirement (i.e.: retirement consumption
puzzle) have been handled by lifecycle models, as household transition into retirement.
Moreover, different patterns have been tested exploiting also data collected from surveys
on household and individuals consumption expenditures. Banks, Blundell, and Tanner
(1998) have been the first to investigate on this topic with a study based on a panel
data from the U.K. Family Expenditure Survey. They find that the retirement spending
declined more rapidly than could be explained by a simple life-cycle model, motivating this
drop as the result of ”unanticipated shocks occurring around the time of retirement” such
as an over-estimate of pensions. Recently, Hurst (2008), in his paper ”The Retirement of
a Consumption Puzzle”, summarizes different facts about consumption behavior during
retirement, as emerged from the recent literature. The author stressed the issue about
the inconsistence between the standard lifecycle model of consumption and the household
behavior during retirement. Combining results from several researches, he shows that for
most households the hasty decline in consumption at the exit of the labor force is due to
the expenditure on food; the home production model could explain this - the retired has
plenty of time and use it for ”food production”, which is less expensive -, thus there is
no puzzle. A similar conclusion is drawn from an investigation carried out by Hurd and
Rohwedder (2008): examining data from the Health and Retirement Study (HRS), they
show that there is retirement consumption puzzle only in few subpopulations, and not
in a population level; in addition, the explanation to these declines can be provided by
conventional economic theory.
Another issue related to the expenditures patterns is about the link between wage
inequality and consumption inequality in a household context. Applying a lifecycle model
that incorporates dynamics of consumption, hours, and earnings to the Panel Study of
Income and Dynamics (PSID) for the period 1999-2009, Blundell et al. (2012), find strong
evidence of consumption smoothing of male’s and female’s permanent shocks to wages.
Remaining in the household context, a relevant focus in empirical studies during the last
decades is related to the intrahousehold allocation of time and money (Lundberg and
Pollak, 1996, Apps and Rees, 1997). Recently, Browning and Gørtz (2012), tested their
own model, which assume efficiency in the household decisions about expenditures, to
control whether the spouse with a relative higher wage (or income in general) has also a
stronger power. The authors present results from the Danish Household Survey (DES)
and the Danish Time Use Survey (DTUS), showing that Danish wives who spend more
time in leisure compared to their husbands, also use to expend a relative higher share of
their wealth in consumption goods. Furthermore, they also find that relative wages have
an impact on family expenditure decisions. Browning et al. (2008) model the differences
in standard living and consumption expenditure among people living alone and couples,
taking into account economies of scale and bargaining power within the household. They
estimate their model on the Canadian Survey of Family Expenditures (Famex) from 1974
to 1992, obtaining the following results: living in a couple generates saving of one third
of the total expenditure through joint consumption, wives on average control almost two
thirds of the household consumption expenditure.
4
In this section, we addressed different topics - retirement consumption, home production, childcare, gender differences, household allocation - and we revised how recent
literature referred to these issues applying different models on different datasets, in particular on data exploited from surveys on time use or on consumption expenditures. The
contribution to the literature of this paper is to revise similar issues using a new and unique
dataset containing data on time use, consumption expenditures and employment status
of single males, single females and household in The Netherlands, carrying out several
evidences.
3
Data
Our data come from a new Dutch panel, the Longitudinal Internet Studies for the Social Sciences (LISS) panel, administered by the CentERdata. A sample of households,
representative of the Dutch population, is drawn from individuals registered by Statistics
Netherlands who have been asked to join the panel by Internet interviewing. The LISS
consists in a panel of 5000 households (8000 individuals). It is based on a true probability
sample of individuals and it collects yearly information in order to capture changes in
standard living conditions of its members.1
In 2009, the new module on time use and consumption has been added to the LISS
panel. The availability of rich information on time use and a detailed list of different
kinds of expenditure for the same individuals at different stages of their life, added to the
detailed background variables collected each month, make the dataset new and unique.
Data have been collected through surveys submitted to all household members registered
older than sixteen years.
To fill the time use module questions about ”normal” hours spent on a set of defined
time use categories during the past seven days have been asked to the interviewees. This
is consistent with the indications reported by Browning and Gørtz (2012) such questions
are informative and avoid infrequency problems associated to diary-based surveys. The
result is a detailed set of retrospective information on time use during the past seven days
distinguished on thirteen exhaustive categories measured in hours/minutes.
The consumption module is based on survey questions about an exhaustive list of
highly disaggregated expenditure items in order to derive an accurate outline of the individual and household expenditures during the previous month. The questionnaire follows
the recommendations suggested by Browning et al. (2003), underlining the useful indications on individual and household consumption, even taking into account the higher level
of ”noise” implicit in this type of data than in diary measures. Information is grouped
into three macro areas, corresponding to three sets of questions: expenditures on twelve
categories of goods and services that can be argued to be publicly consumed by the household, private expenditures on own personal goods (nine categories), and expenditures on
large durables - only for this category data are yearly instead of monthly.
1
See de Vos (2009) for an evaluation of the LISS panel and Scherpenzeel (2011) and Scherpenzeel and
Das (2010) for further information on the design of the LISS panel.
5
In addition to the time use and consumption module, for this study we exploit also
yearly information from the background variables and from the work and schooling form.
We refer to three waves of data carried out in 2009, 2010, 2012. We drop individuals with
no observations in both modules and individuals that are not always in the sample across
the three waves. This restriction is because we estimate a dynamic model of order one,
which requires, at least, three time observations per individual. After the application of
these sample selection criteria, we are left with a balanced dataset containing 8,220 observations equally split among the three years. During the process, some other individuals,
with missing values in the variables used in the econometric analysis, have been dropped
and the dataset has been re-balanced. The result is a panel dataset with 2,681 individuals
for a total of 8,043 observations on the three years.
The additional information gathered from the other LISS modules refers to: the households composition, the age and the gender of the individuals in the subsample, their education level, their employment status and their income. At the end of this subsection,
we report some summary statistics describing some general individual characteristics related to their work status to give a first overview of the subsample composition (the other
variables will be analyzed in the following sections).
As is clear from the tables 1 and 2, the presence of males is more or less equal to
that of females. Concerning ages, almost 50% are older than 55, and the 35% has an age
between 35 and 55 years; younger people are a very small part of the sample, but this is
not an issue considering that we focus our investigation particularly on a household level.
As expected, less than 10% of respondents has a university degree and only the 30% has
the higher education diploma. In table 2 emerges that the majority performs a paid job
(55%), followed by the retired (30%); It is also possible to observe that one third of the
females declared to have as main productive activity ”taking care of the house”; this value
is only the 0.5% between males. Considering only the workers: the 75% of the sample is
employed in a permanent work and the rest is almost equally split between independent
workers and performer of temporary activities. Here, there are no important differences
related to the gender.
Table 1: Age
Variable
From 15 to 24
From 25 to 34
From 35 to 44
From 45 to 54
From 55 to 64
More than 65
Total
Mean
0.071
0.072
0.136
0.195
0.262
0.263
Male
Std. Dev.
0.257
0.259
0.343
0.396
0.44
0.44
6
N
271
276
520
744
998
1003
3812
Mean
0.079
0.083
0.169
0.229
0.244
0.197
Female
Std. Dev.
0.269
0.276
0.375
0.42
0.429
0.398
N
331
349
712
965
1027
829
4231
Table 2: Work Status
Variable
Worker
Unemployed
Student
Homemaker
Retired
Disable
Total
Between ”Workers”:
Independent
Permanent
Temporary
Total
4
Mean
0.679
0.031
0.023
0.005
0.229
0.033
Male
Std. Dev.
0.467
0.173
0.151
0.07
0.42
0.178
0.098
0.785
0.117
0.298
0.411
0.321
N
2588
117
89
19
874
125
3812
Mean
0.629
0.046
0.022
0.122
0.144
0.037
Female
Std. Dev.
0.483
0.209
0.148
0.327
0.351
0.189
254
2032
302
2588
0.085
0.758
0.157
0.279
0.428
0.364
N
2649
192
94
514
608
156
4213
226
2008
415
2649
Empirical Method
We are primarily interested in understanding how different profiles, mostly related to
the employment status, can drive the individual and household behavior. We divide
our analysis in three parts: time use, personal expenditure and household expenditure.
Individual characteristics of the respondents are regressed on different observed dependent
variables each time.2
As reported from Heckman and Macurdy (1980) and McDonald and Moffitt (1980),
to analyze survey data as the one we use for this study - in which the observations often
have values clustered at zero - the standard techniques used in the estimation of linear
data model are inappropriate. The Tobit model is preferred over the other alternative
techniques to deal with the censoring nature of our data. Moreover, the serial correlation
in the error terms, due to the panel nature of the dataset, and the introduction of the lagged
variables, complicate the estimation procedure of the Dynamic Tobit Model. Consequently,
we apply to the standard Tobit model a random effects specification, which allows the
evaluation of the unobserved heterogeneity and serial correlation of the error terms. We
follow the Wooldridge approach to deal with the initial condition problem related to the
dynamicity of the model.
In a dynamic random effects framework, the model introduced by Tobin (1958) can be
represented as follows:
2
See the Appendix for detailed lists of the time use categories, individual expenditure categories and
household expenditure categories.
7
y∗it = βxit + δy∗i,t−1 + ξit , (1)
(
yit =
y∗it if
0
if
ξit = αi + it ;
y∗it > 0
, (2)
y∗it = 0
t = 1, . . . , T,
i = 1, . . . , N (3)
∗ is a latent dependent variable, x is a vector of exogenous variables, y
where yit
it
it
represents an observed dependent variable, and yi,t−1 is the lagged dependent variables.
The exogenous variables include different individual characteristics of the respondents and
the observed dependent variables, that incorporate time use and consumption expenditures
categories, vary in each of the three specifications. The component αi is an unobserved
individual specific random disturbance which is constant over time; it is an idiosyncratic
error which varies across time and individuals:
it ≈ N (0,σ2 ) independent of xi ,
Estimating the model through the maximum likelihood estimation, the theoretical and
practical problem regarding the initial condition comes up. We face this issue following
the method proposed by Wooldridge (2005) which is based on unobserved individual-effects
as conditional on the initial values. This method uses the following auxiliary distribution
of the unobserved individual effects, which is conditioned on the initial values, yi1 :
ei
αi = λyi1 + α
2
e i ≈ N (0,σ ) independent of xi , i and yi1 (4)
α
α
e
e i is the new unobserved individual effect. After this assumption we are able
Where α
to write the LogLikelyhood and to estimate consistently the parameters through the maximization as in standard static random-effect tobit model. Nevertheless, we are not interested on the vector of parameters β, but on the effects of the changes in the exogenous
variables to the dependent variables. Consequently we compute the marginal effects as
the conditional value of y given the latent variable y ∗ > 0.
8
5
5.1
Results
Time Use
In this section, we apply the theoretical dynamic random effects Tobit model on our data
to analyze the ”use of time” by the panel components during the three waves. Due to
missing, or no-reasonable, values in the time-use variables, we have been forced to drop
some observations, keeping a balanced panel dataset with 2432 individuals. For this part,
we use four time use categories, measuring (in hours per week) the amount of time spent
by each individual doing a single activity; table 3 shows some summary statistics about
these variables. In average, time spent in leisure is the most relevant, followed by marketworking time and then by time spent doing household chores. Helping others is the less
relevant category with only 6 hours per week spent on it.
Table 3: Time Use - summary statistics
Variable
Mean Std. Dev. Min. Max.
Work
20.526
21.783
0
90
Household Chores 17.064
14.046
0
88
Helping Others
6.077
10.12
0
72
Leisure
36.802
21.073
0
125.75
N
7296
7296
7296
7296
Time measured in hours per week
In order to capture the differences in the time use profiles, we regress some dummy
variables, regarding personal individual characteristics, on these four dependent variables.
The results of the time use investigation are reported in table 4.
Focusing on employment status, it is not surprising to observe that the less amount
of time spent performing a job, from unemployed and retired, is compensated by time
dedicated to the home production rather than the other categories. This result support
the thesis that not-employed tend to use their time performing activities not included in
the computation of the GDP. Moreover, the table shows a dramatic fall in the working
time, hand in hand with an increase of leisure and household chores, for over-65 individuals.
This is consistent with the result of Krantz-Kent and Stewart (2007): in fact, they find that
the allocation of time to different activities changes with age, and that the employment
rate at older age matters for this alteration too.
Consistent with previous theories, we observe a strict correlation between job responsibility and working time. In effect, there are no relevant differences among permanent
and temporary workers, while the self-employed tend to work in average four hours more
than the employees do, and to have less leisure time.
Looking at the household composition part, the table 4 shows that individuals that use
to live with parents and members of a couple can dedicate to their work a larger amount of
time; this is probably due to the benefits of the household’s economies of scale Browning
et al. (2008). It is also possible to point out a positive correlation between income level
and working time, together with a decrease in home production, probably due to a greater
use of external house work or food away from home.
9
Table 4: Time Use
Dynamic Random Effect Tobit Model on Time Use - Marginal Effects
Expected Value: E(y|y ∗ > 0)
Working
HH Chores
Help Others
dy/dx
dy/dx
dy/dx
Work Status
Worker
—
—
—
Unemployed (d)
-10.507∗∗∗
4.544∗∗∗
1.345∗
Retired (d)
-12.279∗∗∗
3.012∗∗∗
2.412∗∗∗
Disable (d)
-12.677∗∗∗
0.278
2.601∗∗∗
Student (d)
-4.211∗∗∗
-0.244
0.034
Homemaker (d)
-10.291∗∗∗
6.069∗∗∗
1.722∗∗
Between the ”Workers”:
Permanent contract
—
—
—
Temporary contract (d)
0.064
-0.154
0.557
Independent worker (d)
3.815∗∗∗
-1.522∗∗
0.659
No disability
—
—
—
Part-disable (d)
-7.732∗∗∗
2.902∗∗
0.285
Household Status
Living w/o parents
—
—
—
Living w/ parents (d)
4.198∗∗∗
-2.054∗∗
2.897∗∗∗
Between the ”Living w/o parents”:
Single
—
—
—
Couple (d)
1.064∗
-0.251
0.733∗∗
No children
—
—
—
One child (d)
1.590∗∗
0.505
-0.138
Twochildren (d)
0.940
1.097∗∗
-0.135
Three or more (d)
0.508
1.963∗∗
-0.215
Gender
Male
—
—
—
Female (d)
-1.382∗∗∗
3.493∗∗∗
0.426
Age
15 - 24 years old (d)
3.872∗∗∗
-3.620∗∗∗
-0.702
25 - 34 years old (d)
1.369
-0.006
-0.047
35 - 44 years old
—
—
—
45 - 54 years old (d)
-0.798
1.691∗∗∗
1.838∗∗∗
55 - 64 years old (d)
-5.719∗∗∗
3.352∗∗∗
4.112∗∗∗
∗∗∗
∗∗∗
more than 65 years (d)
-14.514
4.667
3.029∗∗∗
Education
Primary or No Educ. (d)
2.586∗∗∗
-0.066
-0.465
Intermediate Educ.
—
—
—
Higher Educ. (d)
-0.810∗
-0.387
-0.645∗∗
University Educ. (d)
1.027
-0.280
-0.784
Income level
Less than 500 E/m
—
—
—
From 501 to 1000 (d)
4.962∗∗∗
-1.791∗∗∗
-0.191
From 1001 to 1500 (d)
8.217∗∗∗
-2.757∗∗∗
0.383
From 1501 to 2000 (d)
9.698∗∗∗
-3.499∗∗∗
-0.240
From 2001 to 2500 (d)
12.912∗∗∗
-3.917∗∗∗
0.177
From 2501 to 3000 (d)
13.098∗∗∗
-2.690∗∗∗
0.389
∗∗∗
More than 3000 (d)
12.929
-4.389∗∗∗
-0.288
Unknown income (d)
8.710∗∗∗
-2.058∗∗
0.145
lag work-time
0.212∗∗∗
work-time 0
0.145∗∗∗
lag hhchores-time
0.094∗∗∗
hhchores-time 0
0.248∗∗∗
lag help-time
0.103∗∗∗
help-time 0
0.237∗∗∗
lag leisure-time
leisure-time 0
N
4864
4864
4864
∗
p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01 - (time measured in hours per week)
Four different Dynamic RE Tobit Regressions on Time Use - Marginal Effects are reported
Year dummies are included in each regression - Dynamic model using the Wooldridge approach
10
Leisure
dy/dx
—
3.324∗∗
2.037∗
3.294∗∗
1.385
0.047
—
2.074∗
-2.700∗∗
—
1.207
—
-2.764
—
-1.337∗
—
-3.431∗∗∗
-4.731∗∗∗
-5.645∗∗∗
—
-3.006∗∗∗
1.139
-2.166
—
2.070∗∗
3.659∗∗∗
6.539∗∗∗
-2.631∗∗
—
1.567∗∗
-1.233
—
-0.666
-0.533
-0.567
-2.513∗
-1.262
-1.348
-3.133∗∗
0.122∗∗∗
0.194∗∗∗
4864
5.2
Consumption Expenditure
In order to investigate in detail the consumption patterns we perform two different analysis, respectively related to: individual consumption and household consumption. We
are interested in capturing the variation in the allocation of wealth among different kinds
of expenditure, so that, we decide to transform our dependent variables, from euros per
month, to shares of total consumption expenditure.
For the individual consumption part, we exclude individuals without information on
consumption of assignable goods (five categories) and outliers - in terms of no-reasonable
share of expenditure (100%) only on one item. The result is a subsample composed by
4185 observations. The five classes of expenditure for this sample are summarized in table
5. The item in which people tend, in average, to use the highest part of their consumption
budget is the food at home (almost half of the whole consumption expenditure); it is
also relevant the amount spent on personal care - it includes clothes and shoes as well.
Expenditure on food out of home and on leisure is somewhat less considerable, amounting
7% of the total each. Even lower is the budget share consumed on medical products.
Table 5: Individual Expenditure - summary statistics
Variable
Mean Std. Dev. Min. Max.
Food at home (%)
44.579
17.081
0
95.238
Food out (%)
7.129
8.301
0
61.224
Personal (%)
21.355
11.794
0
79.439
Medical (%)
4.661
6.238
0
57.613
Leisure (%)
7.309
6.733
0
54.545
Other personal exp. (%) 14.968
12.052
0
89.13
N
4185
4185
4185
4185
4185
4185
Shares calculated on total individual consumption expenditure - data in percentages
Regarding the aggregate household consumption expenditure, it is possible to identify
two different branches of explanatory variables: the ones related to the household, and the
others related to personal status of the respondents. We believe that the individual characteristics are crucial for this investigation; for this reason, we consider only the observations
in which the interviewed is the ”household head”.3 Consequently, the household in which
the respondent is not the household head have been dropped from the sample and 2736
observation left. Also for the household consumption we use five classes of expenditure,
the table 6 shows some summary statistics about these categories. Most of the family
expenses is connected to the house:4 on average, households spend about 40% of their
budget on it. As expected, expenditure for food at home (with a share equal to the 20%
of the total household’s consumption) is among the highest, immediately followed by the
one for insurance fees. Relatively less relevant, compared with other items, are transport
costs as well as holiday expenditures.
3
4
When it is not explicitly declared, household head is considered the one with the higher income.
This category includes rent, mortgages and other general expenditures for the house
11
Table 6: Aggregate Household Expenditure - summary statistics
Variable
Mean Std. Dev. Min. Max.
N
Housing (%)
40.971
14.562
0
85.333 2736
Transport (%)
7.136
5.254
0
37.736 2736
Insurance (%)
13.973
7.41
0
53.922 2736
Holiday (%)
7.118
7.67
0
57.112 2736
Eating at home (%)
19.015
8.672
0
62.5
2736
Other household exp. (%) 11.787
10.006
0
65.076 2736
Shares calculated on total household consumption expenditure - data in percentages
Tables 7 and 8 show the results of the two estimations, respectively: the first one
reports the marginal effects of the regression of some dummy variables regarding individual
characteristics on five personal expenditure categories and the second one is about the
household consumption analysis.
Observing the table 7, the first result regards the unemployed, which have a higher
share of expenditure in food at home compensated by a smaller amount spent on personal
care and to eat outside the home; this is an index of a lower ”life status” compared with
the workers and it is consistent with the results of the time use analysis. The retired
follow, even with a lower magnitude, a similar pattern signaling, also in line with the time
use review, changes in consumption allocation upon retirement; in particular, a sort of
substitution effect between food at home and food outside the home. Further, we do not
find noteworthy differences connected to different type of job contract.
The outcome resulting from the household composition part is in line with the expectations: having children implies less expenditure on food at home; besides, living in
couple means expend less on food and the possibility to use the wealth for personal care,
consistently with the findings about the households’ economies of scale emphasized by
Nelson (1988).
While there are no gender differences in expenses for leisure and medical products, it
is found out that women utilize, in average, their consumption budget mostly for items
related to personal care and less for buying food (both eating at home and out of home),
compared with what men use to do.
From table 8, it results that the household consumption allocation is not heavily relied
individual characteristics of its household head, while there is something more to say about
the dependence from the household composition. Regarding insurances expenditure, for
example, data reveals a sort of breaking point between one-member households and the
ones with two or more members; the latter tend to spend, in average, more than the former
do. Furthermore, we notice a direct positive correlation between household dimension and
expenditure on food at home.
There are two interesting evidences referred to the age of the household head. The
younger (16 to 24 years old) have a smaller share of consumption spent on housing: this
is probably due to house-sharing for young workers or university accommodation for students. The second result is about the over 65: their family rate of consumption allocated
on housing is also lower than the average, therefore, they increment the expenditure on
12
Table 7: Individual Consumption Expenditure
Dynamic Random Effect Tobit Model on Individual Cons. Exp. - Marginal Effects
Expected Value: E(y|y ∗ > 0)
Food at home Food out Personal Medical Leisure
dy/dx
dy/dx
dy/dx
dy/dx
dy/dx
Work Status
Worker
—
—
—
—
—
Unemployed (d)
3.769∗∗
-2.243∗∗∗
-2.866∗∗∗
1.293∗∗
-1.497∗∗∗
Retired (d)
1.872∗
-1.800∗∗∗
-0.853
0.128
-0.635
Disable (d)
-2.682
-1.582∗∗
-1.625
1.869∗∗
-1.292∗∗
∗∗∗
∗∗∗
∗
Homemaker (d)
3.533
-1.596
-1.636
-0.654
-2.410∗∗∗
Between the ”Workers”:
Permanent contract
—
—
—
—
—
Temporary contract (d) 0.100
-0.411
-0.630
0.562
0.245
Independent worker (d) 0.224
-0.120
-1.673∗
0.668
-0.266
No disability
—
—
—
—
—
Part-disable (d)
2.355
-0.963
-1.396
1.066
-0.856
Household Comp.
Single
—
—
—
—
—
Couple (d)
-1.408∗
-0.584∗
2.723∗∗∗
1.275∗∗∗
0.175
No children
—
—
—
—
—
One child (d)
-2.039∗
0.149
0.790
-0.630
0.851∗
Two children (d)
-1.981∗∗
0.135
0.746
-0.051
0.433
Three or more (d)
-3.173∗∗
0.751
-0.489
0.203
0.589
Gender
Male
—
—
—
—
—
Female (d)
-2.792∗∗∗
-1.384∗∗∗
4.825∗∗∗
0.380
-0.460∗
Age
15 - 24 years old (d)
-6.215∗∗
2.893∗
3.228
-1.149
1.131
25 - 34 years old (d)
-0.781
0.232
-0.503
0.051
0.938
35 - 44 years old
—
—
—
—
—
45 - 54 years old (d)
1.335
-0.925∗∗
-0.449
0.960∗∗
0.132
55 - 64 years old (d)
1.779∗
-0.596
-1.560∗∗
0.656
0.244
more than 65 (d)
1.787
-0.267
-1.111
1.631∗∗∗
-0.511
Total Expenditure
ln(total exp)
65.014∗∗∗
-8.189∗∗
-22.466∗∗∗ 6.396∗∗
-7.264∗∗
∗∗∗
∗∗∗
∗∗∗
∗
ln(total exp) sq
-6.014
0.872
1.916
-0.461
0.753∗∗∗
∗∗∗
lag sh food home
0.113
0.248∗∗∗
sh food home 0
lag sh food out
0.010
sh food out 0
0.328∗∗∗
lag sh personal
-0.003
sh personal 0
0.358∗∗∗
lag sh medical
-0.050
sh medical 0
0.305∗∗∗
lag sh leisure
-0.002
sh leisure 0
0.244∗∗∗
N
2790
2790
2790
2790
2790
∗
p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01 - (shares of the total expenditure - measured in euro per month %)
Five different Dynamic RE Tobit Regressions on Individual Consumption - Marginal Effects are reported
Year Dummies are included in each regression - Dynamic model using the Wooldridge approach
13
Table 8: Household Consumption Expenditure
Dynamic Random Effect Tobit Model on Household Cons. Exp. - Marginal Effects
Expected Value: E(y|y ∗ > 0)
Housing Transport
Insurance Holiday Eating home
dy/dx
dy/dx
dy/dx
dy/dx
dy/dx
Household Characteristics
Household Comp.
One members
—
—
—
—
—
Two members (d)
-0.434
0.507
2.454∗∗∗
0.940∗∗
2.799∗∗∗
Three members (d)
-0.540
-0.519
2.740∗∗∗
0.835
2.552∗∗∗
Four members (d)
-0.525
0.330
2.777∗∗∗
0.336
3.844∗∗∗
∗∗
Five or more (d)
1.126
-0.227
2.206
-0.600
4.145∗∗∗
Place of residence
Urban place
—
—
—
—
—
Slightly or No Urban(d)
-0.455
0.543∗∗
0.465
0.040
-0.097
Total Expenditure
ln(total exp)
38.136∗∗∗ -3.137
20.888∗∗
-17.775∗∗ 25.580∗∗
∗∗∗
∗∗
ln(total exp) sq
-2.918
0.234
-1.642
1.476∗∗
-1.877∗∗∗
Household Head Characteristics
Work Status
Worker
—
—
—
—
—
Unemployed (d)
0.855
-0.847
0.715
-1.520∗∗
0.133
Retired (d)
-0.038
-0.659∗
0.650
-0.053
-0.322
Disable (d)
-1.580
-1.405∗∗∗
2.502∗∗∗
-1.891∗∗∗ -0.578
Homemaker (d)
-0.186
-0.321
-1.951∗
-0.813
2.233∗
Between the ”Workers”:
Permanent contract
—
—
—
—
—
Temporary contract (d) -0.593
-0.362
0.390
0.756
-0.031
Independent worker (d) -0.537
-0.924
-0.073
-0.344
0.911
No disability
—
—
—
—
—
Partdisable (d)
0.398
-0.263
2.999∗∗
-3.462∗∗∗ -1.313
Gender
Male
—
—
—
—
—
Female (d)
1.060
-0.571∗
0.537
-0.162
-0.456
Age
15 - 24 years old (d)
-9.801∗∗∗
2.603∗
-0.963
2.349
2.894
25 - 34 years old (d)
1.524
-0.008
-0.935
0.220
-0.633
35 - 44 years old
—
—
—
—
—
45 - 54 years old (d)
-2.001∗∗
0.410
0.337
0.459
1.571∗∗
55 - 64 years old(d)
-1.497
0.569
0.717
1.185∗∗
0.962
∗∗∗
more than 65 (d)
-3.172
0.625
0.889
1.086
3.001∗∗∗
∗∗
lag sh fam housing
0.130
sh fam housing 0
0.573∗∗∗
lag sh fam transport
-0.012
sh fam transport 0
0.549∗∗∗
lag sh fam insurance
0.044
sh fam insurance 0
0.317∗∗∗
lag sh fam holiday
-0.015
sh fam holiday 0
0.397∗∗∗
lag sh fam eat home
0.137∗∗
sh fam eat home 0
0.319∗∗∗
N
1824
1824
1824
1824
1824
∗
p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01 - (shares of the total expenditure - measured in euro per month %)
Five different Dynamic RE Tobit Regressions on Household Consumption - Marginal Effects are reported
Year dummies are included in each regression - Dynamic model using the Wooldridge approach
14
food at home. This last remarkable result, together with the evidences of the previous
section, supports the theory that imputes a higher level of home production to older and
retired population, Hurst (2008).
6
Conclusion
In the sections above, we have presented a general analysis concerning time use and consumption expenditure allocation at individual and household level. We applied a simple
Dynamic Random Effects Tobit model to data exploited from the LISS panel, which is representative of the Dutch population. The paper contributes to the literature, giving some
evidences about three different popular research topics, i.e. the retirement consumption
puzzle, the home production and the resources allocation within the household.
Our analysis reports a change in both time use and consumption patterns upon retirement. In particular, a decrease in time spent performing a job, compensated by time
dedicated to household chores, and a higher share of expenditure in food at home offset
by a lower consumption of food outside the house. Therefore, our data support the theory
according to which there is a systematic alteration in consumption allocation at retirement
inconsistent with the life cycle/permanent income hypothesis. However, the evidence that
retired people use a part of their free time in home production can explain the shift in
allocation of expenditures, partially clearing up the retirement consumption puzzle.
For what concerns the productive activity not included in the GDP, as expected, our
findings show a greater level of home production for non-employed together with a larger
expenditure share on goods requiring more handling (e.g.: food at home instead of food
outside home).
Considering the household context, we find that people living alone are not able to
devote the same amount of time to their work as the others do and that they tend to expend
a higher share of their consumption budget on food. The results of the two analysis are
consistent with each other and in line with the theory about the existence of households’
economies of scale.
References
Mark Aguiar and Erik Hurst. Measuring trends in leisure: The allocation of time over five
decades. The Quarterly Journal of Economics, 122(3):969–1006, 2007.
Mark Aguiar, Erik Hurst, and Loukas Karabarbounis. Recent developments in the economics of time use. Annual Review of Economics, 4(1):373–397, 07 2012.
15
Dominique Anxo, Letizia Mencarini, Ariane Pailhe, Anne Solaz, Maria Letizia Tanturri,
and Lennart Flood. Gender differences in time use over the life course in france, italy,
sweden, and the us. Feminist Economics, 17(3):159–195, 2011.
Patricia F. Apps and Ray Rees. Collective labor supply and household production. Journal
of Political Economy, 105:178–190, 1997.
James Banks, Richard Blundell, and Sarah Tanner. Is there a retirement-savings puzzle?
American Economic Review, 88(4):769–88, September 1998.
Gary S. Becker. A theory of the allocation of time. The Economic Journal, 75:493–517,
1965.
Richard Blundell. Consumer behaviour: Theory and empirical evidence–a survey. Economic Journal, 98(389):16–65, March 1988.
Richard Blundell, Luigi Pistaferri, and Itay Saporta-Eksten. Consumption inequality
and family labor supply. NBER Working Papers 18445, National Bureau of Economic
Research, Inc, October 2012.
Martin Browning and Mette Gørtz. Spending time and money within the household.
Scandinavian Journal of Economics, 114(3):681–704, 09 2012.
Martin Browning, Thomas F. Crossley, and Guglielmo Weber. Asking consumption questions in general purpose surveys. Economic Journal, 113(491):F540–F567, November
2003.
Martin Browning, Pierre-Andre Chiappori, and Arthur Lewbel. Estimating consumption
economies of scale, adult equivalence scales, and household bargaining power. Economics
Series Working Papers 289, University of Oxford, Department of Economics, October
2008.
Laurens Cherchye, Bram De Rock, and Frederic Vermeulen. Married with children: a
collective labor supply model with detailed time use and intrahousehold expenditure
information. Center for Economic Studies - Discussion papers ces10.24, Katholieke
Universiteit Leuven, Centrum voor Economische Studiën, September 2010.
M. Knoef & K. de Vos. Representativeness in online panels: how far can we reach. mimeo,
Tilburg University, 2009.
Jose Ignacio Gimenez-Nadal and Almudena Sevilla. Trends in time allocation: A crosscountry analysis. European Economic Review, 56(6):1338 – 1359, 2012. ISSN 0014-2921.
Geoffrey Godbey and John Robinson. Time for Life: The Surprising Ways Americans
Use Their Time. The Pennsylvania State University Press, second edition, 1999.
Tami Gurley-Calvez, Amelia Biehl, and Katherine Harper. Time-use patterns and women
entrepreneurs. American Economic Review, 99(2):139–44, 2009.
James J Heckman and Thomas E Macurdy. A life cycle model of female labour supply.
Review of Economic Studies, 47(1):47–74, January 1980.
16
Michael D. Hurd and Susann Rohwedder. The retirement consumption puzzle: Actual
spending change in panel data. NBER Working Papers 13929, National Bureau of
Economic Research, Inc, April 2008.
Erik Hurst. The retirement of a consumption puzzle. NBER Working Papers 13789,
National Bureau of Economic Research, Inc, February 2008.
Rachel Krantz-Kent and Jay Stewart. How do older americans spend their time? Monthly
Labor Review, 130:8–26, May 2007.
Shelly Lundberg and Robert A. Pollak. Bargaining and distribution in marriage. Journal
of Economic Perspectives, 10(4):139–158, Fall 1996.
John F McDonald and Robert A Moffitt. The uses of tobit analysis. The Review of
Economics and Statistics, 62(2):318–21, May 1980.
Julie A Nelson. Household economies of scale in consumption: Theory and evidence.
Econometrica, 56(6):1301–14, November 1988.
Silvia Pasqua and Anna Laura Mancini. Asymmetries and interdependencies in time use
between italian parents. Applied Economics, 44(32):4153–4171, November 2012.
Valerie A. Ramey. Time spent in home production in the 20th century: New estimates
from old data. NBER Working Papers 13985, National Bureau of Economic Research,
Inc, May 2008.
A. Scherpenzeel. Data collection in a probability based internet panel: How the liss panel
was built and how it can be used. Bulletin of Sociological Methodology, 109:5661., 2011.
A. Scherpenzeel and M. Das. True Longitudinal and Probability-Based Internet Panels:
Evidence from the Netherlands. in M. Das, M.P. Ester, and L. Kaczmirek, eds., Social
and Behavioral Research and the Internet: Advances in Applied Methods and Research
Strategies, 2010.
Elena G. F. Stancanelli and Arthur van Soest. Retirement and home production: A
regression discontinuity approach. IZA Discussion Papers 6229, Institute for the Study
of Labor (IZA), December 2011.
James Tobin. Estimation of relationships for limited dependent variables. Econometrica,
26:24–36, 1958.
Jeffrey M. Wooldridge. Simple solutions to the initial conditions problem in dynamic,
nonlinear panel data models with unobserved heterogeneity. Journal of Applied Econometrics, 20(1):39–54, 2005.
17
Appendix
Dataset Description
In this paper use is made of data of the LISS (Longitudinal Internet Studies for the Social
sciences) panel administered by CentERdata (Tilburg University, The Netherlands).See
the website http://www.lissdata.nl for details.
For this investigation, we made use of the following modules of the LISS data:
1. Background Variables
2. Work and Schooling
3. Time Use and Consumption.
The data from these modules were merged by means of household and individual
specifiÂc. Data were carried out in 2009, 2010, 2012. Individuals with observations only
about one study, individuals that are not always in the sample across the three waves
and other outliers have been dropped. The result is a balanced dataset contains 2681
individuals, so 8043 observations equally splitted among the 3 different years.
Independent Dummy Variables
The following variables are the ones used to ”describe” the individual profile in the three
analysis: Time Use, Individual Consumption Expenditure and Household Consumption
Expenditure.
Work Status
There are 6 different categories regarding the employment status. In addition, between
people performing a stable work, other 3 different tipologies of contract have been analyzed.
For people who gave several answers to the question about their work situation, the one
that, according to them, better describes their situation has been used.
The first variable costructed is ”worker” and it contains all the individuals who specify
their labour contract (plus the ones that are performing a paid work and did not fill the
other form).
The not-workers are divided in other 5 different categories:
1. Unemployed (this category contains people that: are not working but performed a
paid job in the past, or are retaining benefits and are performing an unpaid work
or are looking for a new job following the loss of the previous one or after lengthy
interruption or are the first-time job seeker),
2. Retired (pensioner, people that live off private means or have taken an early retirement or perform a voluntary job - this aggregation has been guided mostly looking
at the age of people in the subcategory),
3. Disable (People that because of the disability do not perform any stable work),
4. Student.
5. Homemaker (the ones who affirm that the task that fits better with their status is
”taking care of the Household”).
18
Between the ”Workers”, five others subcategories:
(a) Independent (Self-Employed / Freelancer / Independent professional / Director or
major shareholder of a company),
(b) Permanent contract (employee in permanent employement),
(c) Temporary worker (employee in temporary employement).
And:
(a) No disability,
(b) Partly Diable (People with a partial disability but with a stable work ).
Household Status
- Only for the Individual Sections
1. Living with parents,
2. Living without parents.
Between the ”Living without parents”:
(a)
i) Single (with or without children),
ii) Couple (with or without children, married or not, with a partner or with
an housemate);
(b)
i) No children,
ii) One child,
iii) Two children,
iv) Three or more children.
Number of Household Components - Only for the Household Section
1. One member
2. Two members
3. Three members
4. Four members
5. Five or more members
Gender
1. Male,
2. Female.
19
Age
1. 15 - 24 years old,
2. 25 - 34 years old,
3. 35 - 44 years old,
4. 45 - 54 years old,
5. 55 - 64 years old,
6. 65 years old or older.
Education
1. Primary school or Not Education,
2. Intermediate education (secondary and vocational),
3. Higher education (secondary and vocational),
4. University education.
Urban characteristics of the place of residence
1. Urban (extremely urban, very urban, moderate urban),
2. Slightly or not urban (slightly urban, not urban).
Income (euro/month)
- For the Time Use Section
1. Less than 500,
2. From 501 to 1000,
3. From 1001 to 1500,
4. From 1501 to 2000,
5. From 2001 to 2500,
6. From 2501 to 3000,
7. From 3001 to 3500,
8. More than 3500,
9. Unknown income.
Income (euro/month)
- For the Consumption Expenditure Sections
1. ln(total exp),
2. ln(total exp) squared.
20
Time Use Dependent Variables
Four dependent variables about the amount of time (measured in hours per week) have
been constructed using 7 of the 14 different classes of the Time Use and Consumption
section. In particular, time spent traveling to and from work has been added to time
spent performing a paid work, and time spent helping parents, time spent helping family
members and time spent helping non-family members form the category helping others.
The four category are the following:
1. Time spent performing a paid work,
2. Time spent doing household chores,
3. Time spent helping others,
4. Time spent doing leisure activities.
Individual Consumption Expenditure Dependent Variables
Five dependent variables regarding the individual expenditure (measured in euro per
month) have been constructed using 6 different classes of the Time Use and Consumption
section. In particular the variables clothes and personal care have been added togheter
in the variable Personal. In order to capture the variation in the allocation of the total
individual expenditure, the dependent variable are the shares of the total expenditure.
The four category are the following:
1. Food at home,
2. Food out (restaurants, bars),
3. Personal (clothes, shoes, hairdresser, etc...),
4. Medical.
5. Leisure.
Household Consumption Expenditure Dependent Variables
Five dependent variables about the family expenditure (measured in euro per month)
have been constructed using 7 different classes of the Time Use and Consumption section.
In particular mortgages, rent and general expenditure for the house have been summed
up togheter in the variable Housing. Also here, the dependent variable are the shares of
the total expenditure.The four category are the following:
1. Housing,
2. Trasport (sum of expenditure of each household component),
3. Family Insurance,
4. Holiday (tickets, hotel, restaurant bills, etc...).
5. Eating at home.
21