Content uploaded by Rense Nieuwenhuis
Author content
All content in this area was uploaded by Rense Nieuwenhuis on Jun 17, 2016
Content may be subject to copyright.
COMPARATIVE RESEARCH WITH NET AND GROSS INCOME DATA:
AN EVALUATION OF TWO NETTING DOWN PROCEDURES FOR THE
LIS DATABASE
by Rense Nieuwenhuis
Swedish Institute for Social Research (SOFI), Stockholm University
Teresa Munzi
LIS
and
Janet C.Gornick
LIS and The Graduate Center, City University of New York
Researchers seeking to perform country-comparative and trend analyses using income data have to
account for the fact that income surveys differ in whether income is measured gross or net of taxes and
contributions. We discuss, develop, and evaluate two `netting down procedures' for data in the LIS
Database. Evaluations of these netting down procedures indicate that comparisons across gross and
net datasets can be greatly improved when netting down procedures are applied. In several cases, how-
ever, substantial amounts of bias remain.
JEL Codes: D3, P5, C8
Keywords: comparative research, data harmonization, income, LIS, netting down
1. Introduction
A common challenge in country-comparative and trend analyses of income
using microdata, is that income surveys differ in whether income is measured
gross or net of taxes and contributions. The issue of comparability between gross
and net datasets is common in comparative datasets on income, including the
data contained in the LIS Database.
LIS acquires existing income surveys, and harmonizes them into a pre-
defined template for comparative analysis. All LIS datasets provide fully compara-
ble measures of disposable household income.
1
However, comparability problems
arise with other income variables, because LIS provides income data that are net
of (income) taxes in some countries or years, while providing gross income data in
others. For users of LIS who seek to perform country-comparative analyses and/
*Correspondence to: Rense Nieuwenhuis, Swedish Institute for Social Research (SOFI),
Stockholm University, SE-106 91 Stockholm, Sweden (rense.nieuwenhuis@sofi.su.se).
1
Disposable household income is the income concept usually adopted for poverty and distribution
analysis. It is also used for calculation of the LIS Key Figures: http://www.lisdatacenter.org/data-
access/key-figures/disposable-household-income/.
V
C2016 International Association for Research in Income and Wealth
1
Review of Income and Wealth
Series 00, Number 00, Month 2016
DOI: 10.1111/roiw.12233
bs_bs_banner
or analyses of trends within countries, this results in the challenge that their
selected income variables refer to different (net vs. gross) concepts across datasets
and therefore in most applications should not be compared directly. Of the 262
LIS datasets available at the time of writing, 74 (28 percent) were classified net,
175 (67 percent) as gross, and 13 (5 percent) as `mixed'.
2
Datasets on the U.S. or
the U.K. have always been gross, while Austria has been net. Other countries, such
as Ireland and Luxembourg were covered by both net and gross datasets, at differ-
ent points in time. Mixed datasets are a special case in which income variables are,
for instance, as in France, net of mandatory contributions but gross of income tax.
Such `mixed' datasets are beyond the scope of this paper.
Researchers working with the LIS data have applied at least four different
strategies for comparing gross and net datasets. The first is to include both types
of datasets in the same (comparative) analysis, acknowledging incomparabilities
that could lead to biased results. The second strategy is to restrict all analyses to
either gross or net datasets. This results in accurate findings, but clearly limits the
scope of the research. Third, LIS users sometimes present separate analyses using
gross and net datasets. The limitation of this strategy is that differences in the
results based on gross and net datasets could originate from the different earnings
concepts, or from real differences across countries, or both. The fourth strategy is
to modify the gross income data to approximate net income data. This process is
referred to as netting down, and entails subtracting observed or estimated taxes
from the gross income amounts. Such netting down procedures, however, have
not been evaluated empirically for their capacity to produce measurements of
income that are comparable across datasets.
This technical note presents and evaluates two netting down procedures. We
present background information on the comparison of income in gross and net
LIS datasets.
3
We then present the two procedures for netting down, and evaluate
their performance.
We introduce the rationale of netting down with reference to comparing
income. In the empirical part, however, we focus on netting down the more nar-
row concept of earnings from dependent employment; as we will explain, the chal-
lenges of comparing gross and net data are more apparent with earnings.
Program syntax is available in an online appendix.
2. Comparing Gross And Net Income When Using The Lis Data
Comparing results based on gross versus net income data can be of substan-
tive interest; carrying out research using a mix of the two types of data can also
present challenges with respect to comparability.
The difference between gross and net income is of substantive interest, and
can be assessed directly, when a single dataset contains information on both gross
and net income, when information on taxes is available, and when additional
2
For a continuously updated overview, see: http://www.lisdatacenter.org/our-data/lis-database/
datasets-information/.
3
Although the authors are associated with LIS, the presentation of this method does not represent
official LIS work nor an official LIS recommendation; LIS data users should feel free to utilize the
method presented here or any other.
Review of Income and Wealth, Series 00, Number 00, Month 2016
V
C2016 International Association for Research in Income and Wealth
2
information on social transfers is available. This allows LIS users to compare pre-
tax-pre-transfer income to post-tax-post-transfer income, and thereby to answer
a set of research questions about how taxes, and also social transfers, affect
income distributions. This approach has been applied to study the effects of social
welfare policies on poverty rates (see, e.g., Kenworthy, 1999). Other studies have
evaluated how taxes and transfers affect poverty rates among specific subgroups,
such as among children (see, e.g., Gornick and J€
antti, 2012), working-age popula-
tions (Gornick and Milanovic, 2015), single parents (Maldonado and Nieuwen-
huis, 2015a,b), and migrant households (Sainsbury and Morissens, 2012). For
such “redistribution studies,” the actual differences between gross and net income
are of substantive interest, and both are compared within a single dataset.
The difference between gross and net income becomes a challenge in compa-
rability, when comparing datasets of which some are gross and others are net.
This is the case in country-comparative analyses and/or in trend analyses. It has
been shown that country-comparative studies based on different earnings con-
cepts across countries can be “seriously misleading” (Atkinson and Brandolini,
2001, p. 777).
The issue of comparability between gross and net datasets in LIS also applies
to the above-mentioned redistribution studies. This is clarified using an example of
a typical redistribution study on the comparison between pre-tax-pre-transfer
income (referred to as “market household income”) to post-tax-post-transfer
income (referred to as “disposable household income”) (Gornick and J€
antti,
2012). Market income is reported gross of income taxes and contributions in some
LIS datasets, and net of income taxes and contributions in other LIS datasets.
Without correction, this would have understated the poverty/inequality reduction
in the net datasets, as the comparison between market income and disposable
household income in these datasets only captures the effects of transfers, whereas
in gross datasets this comparison would capture the combined effect of taxes and
transfers. Thus, here too, comparisons of redistributive efforts drawing on a mix of
gross and net datasets can be improved by netting down the gross datasets.
2.1. Netting Down, or Grossing Up?
An alternative to netting down gross income data would be to “gross up” net
income data. With LIS, however, grossing up is not possible, drawing on the
microdata alone, as most net datasets do not contain information on taxes. To
then estimate the gross income would require country-specific details on the tax
system, which is beyond the scope of this technical note. Detailed simulations to
this end are available for European countries through the Euromod project
(Sutherland and Figari, 2013).
3. Netting Down Person-Level Earnings
So far, we have discussed netting down gross income. In this section, we intro-
duce practical complexities involved with netting down one specific type of
income: that is, earnings. Earnings, of course, are typically studied at the level of
individuals rather than households. Hence, we shift our focus to developing and
Review of Income and Wealth, Series 00, Number 00, Month 2016
V
C2016 International Association for Research in Income and Wealth
3
evaluating two procedures designed specifically for netting down person-level
earnings.
As noted earlier, when researchers studying redistribution across households
are faced with a mix of gross and net datasets, they sometimes “net down” the
gross income data to enable meaningful comparisons across the two types of data-
sets. In short, by shifting to only net income they restrict their comparisons to the
effects of transfers only (not of transfers combined with taxes).
Likewise, researchers concerned with earnings are often faced with a mix of
gross and net datasets. They could simply mix the two (as some researchers have
done), but that risks arriving at results that are incomparable across datasets. For
example, if one is studying gender gaps in earnings and comparing results in a
country with gross data (e.g., the U.S.) with results in a country with net data
(e.g., Hungary), the results will be problematic. In the U.S., with gross data, the
researcher is capturing gaps in pay levels as set by employers; in countries with
net data, such as Hungary, these same gender pay gaps have been reduced (in
most cases) due to progressive taxation. The researcher thus is working with
“apples and oranges,” and cannot know how much of the difference between the
two countries is “real” versus an artefact of the data. We argue that researchers
ought never mix gross and net earnings data whether they are studying earnings
disparities between groups, or other outcomes related to the distribution of earn-
ings. The techniques laid out in this paper offer researchers one option for avoid-
ing that kind of “mixing”. In short, “netting down” restricts comparisons across
all datasets to after-tax earnings; thus, researchers are comparing “apples to
apples”.
In addition, in some cases, researchers want to assess net earnings. For exam-
ple, some argue that net earnings are more appropriate than gross earnings in
studies of intra-household bargaining. If partners are negotiating (for example,
who does what domestic work) on the grounds of how much money each brings
in, it is arguably net earnings that matter. Our netting down procedure offers
researchers a way to “net down” gross earnings, including in those cases when
researchers prefer to work with net earnings for substantive reasons, and not just
to maximize cross-country comparability.
3.1. Earnings as a Specific Income Source
There are three basic sources of income: labor, capital, and transfers. Netting
down the income from a specific source, such as earnings, is challenging both con-
ceptually and practically. The conceptual problem with net earnings lies in the
fact that countries can apply different tax rates to income from, for instance, labor
and capital. As these tax rates are progressive based on total income, and only the
total amount of paid income taxes is recorded in the data, it must be assumed
that the average tax rate applied equally to all separate sources of income.
The practical problem with net income from separate sources is that as a
result of the above, only information on total taxes is available. Therefore, to cal-
culate the net income from a separate source, given the information available in
the microdata, the assumption that income from each source was taxed at the
same rate is required. This assumption is likely violated as most countries have
Review of Income and Wealth, Series 00, Number 00, Month 2016
V
C2016 International Association for Research in Income and Wealth
4
different tax rates for income from labor and capital. However, as few households
pay taxes on capital income, it remains to be seen how much this violated assump-
tion leads to a biased approximation of net earnings.
4
3.2. Earnings as a Person-Level Concept
In order to calculate net earnings at the person level, person-level income
taxes and social contributions must be subtracted from person-level gross earn-
ings. In countries with joint taxation, however, this is conceptually challenging
because the amount of taxes to be paid is determined at the level of the house-
hold. This also means that individual “personal” earnings depend on the earnings
of other household members. Joint taxation often pertains to the head of the
household and her/his spouse, with separate taxation of the income from addi-
tional earners such as older children or relatives living in the household.
5
If no
person-level tax variables are available, netting down person-level earnings
requires the assumption that the taxes paid at the household level were paid by
each household member proportionally to the share of the total household
income received by that member. This assumption is likely violated in joint taxa-
tion regimes, but it remains an empirical question to what extent this leads to a
biased approximation of net earnings.
3.3. Two Netting Down Procedures
We developed two programs that perform netting down procedures, available
for STATA, SPSS, SAS, and R. One procedure uses information on taxes at the
person-level. If these are not available, the other procedure can be used based on
household-level tax information. The LIS website has a table providing informa-
tion on whether datasets are gross or net. Datasets classified as mixed should be
treated with more caution, as the earnings reported in these datasets can be gross
of income taxes but net of contributions, or vice versa. This is reported in detail
in the LIS data documentation per country. All LIS datasets also contain a vari-
able named grossnet, providing information on how earnings (and other
income variables) were reported.
The person-level netting down procedure can only be used when person-level
variables on taxes (LIS variable pmxiti) and (self-paid) social security contribu-
tions (pmxitss) are available. It first calculates the proportion of earnings in the
total taxable income: gross earnings (pmile), self employment (pmils), unem-
ployment compensation benefits (pmitsisun), short-term sickness and work
injury benefits (pmitsissi), family leave benefits (pmitsisma), and pensions
(ppension). Then it calculates net earnings by subtracting from the gross earn-
ings the value of taxes paid, proportional to the amount of total income obtained
from earnings (propearnings). This assumes that the total amount of taxes
4
In the country-samples used in our analyses only 3 percent of individuals lived in a household in
which capital income represented more than 10 percent of gross household income, with 1 percent in
Estonia 2004 and 7 percent in Belgium 1997.
5
In the country-samples used in our analyses approximately 16 percent of individuals, other than
the household-head or spouse, contributed more than 10 percent of total household earnings. This per-
centage ranged from 12 percent in the U.K. 2004 to 25 percent in Ireland 2004.
Review of Income and Wealth, Series 00, Number 00, Month 2016
V
C2016 International Association for Research in Income and Wealth
5
was distributed proportionally over all sources of income. As taxable income is
made up of different components across countries, this procedure is based merely
on an approximation of taxable income. The calculation is shown in equation (1).
propearnings5pmile
pmile1pmils1pmitsisun1pmitsissi1pmitsisma1ppension
net5pmile2ðpmxiti1pmxitssÞpropearningsðÞ
(1)
The household-level netting down procedure can be used when tax information is
available only at the household level. It calculates the percentage of the total mon-
etary household income (hmi) that remains after taxes (hmi - hmxit) and multi-
plies gross person-level earnings (pmile) by this percentage. This assumes that
this percentage is equal across all members of the household, and applies equally
to all sources of income. The calculation is shown in equation (2).
net5pmile hmi2hmxit
hmi
(2)
It should be noted that these netting down procedures are deliberately simple, in the
sense that no country-specific rules were applied. The benefit is that these procedures
can be applied to all gross LIS datasets. At the same time, if users wish to modify
these procedures to account for specific countries' tax systems, they can do so.
4. Method And Data
4.1. Method
A select number of LIS datasets has both gross and net earnings variables, as
well as information on taxes and social security contributions on both the person-
level and the household-level. This provides a unique opportunity for evaluating
netting down procedures.
To evaluate a netting down procedure, we applied it to a gross earnings vari-
able, and compared the resulting “netted-down” variable to the original net earn-
ings variable in the LIS dataset. We calculated bias for each percentile in the
earnings distribution:
Biasð%Þ5Xnd 2Xn
Xn
3100%(3)
in which X
nd
represents the earnings in the “netted-down” earnings variable, and
X
n
represents the net earnings reported in the LIS dataset. The resulting bias is
expressed as a percentage of the reported net earnings. So, a bias of 0 percent
means that the results based on the “netted-down” earnings variable are identical
to those reported in the original net earnings variable. If the bias % is larger than
0, this means that the netted-down earnings are higher than those based on the
reported net earnings; a percentage below 0 indicates that the netted-down results
are lower.
Review of Income and Wealth, Series 00, Number 00, Month 2016
V
C2016 International Association for Research in Income and Wealth
6
In addition to calculating bias for the earnings levels of different percentiles,
we also calculated bias (again, based on equation (3)) for commonly used measures
of inequality: the ratio of the 75th to the 25th percentile of earnings, the Theil
index, the Coefficient of Variation, the Gini, the low earnings rate (defined as the
percentage of earners with earnings below 2/3 of median earnings), and the gender
gap in earnings (defined as: (male earnings – female earnings)/male earnings).
4.2. Data
The netting down procedures described here can be applied to LIS datasets
harmonized using the “new” (post-2011) template that are classified as gross. The
evaluation of these netting down procedures, however, required the availability of
both gross and net earnings variables in the data, which could only be the case
when using the “old” template (pre-2011). The required earnings variables, as well
as person- and household-level variables on taxes and social contribution were
available in seven datasets: Austria 2004, Belgium 1992, Belgium 1997, Estonia
2004, Ireland 2004, U.K. 1999, and U.K. 2004. We restricted our analyses to those
observations with valid information on both the gross and net earnings variables.
This ensured that our measurement of bias was not affected by the possibility
that gross and net earnings variables were based on different observations.
5. Results
Figure 1 shows bias incidence curves, representing the amount of bias associ-
ated with a netting down procedure for each percentile in the earnings distribution.
The solid lines represent the scenario in which no netting down was applied, i.e., a
direct comparison between gross and net earnings. Of course, the difference
between gross and net does not necessarily indicate bias, as they represent different
earnings concepts. However, the lines represent a reference point to evaluate the
performance of netting down procedures compared to no netting down at all. In all
Austria '04 Belgium '92 Belgium '97 Estonia '04 Ireland '04 UK '04 UK '99
0
40
80
120
0 25 50 75 100 0 25 50 75 100 0 25 50 75 100 0 25 50 75 100 0 25 50 75 100 0 25 50 75 100 0255075100
Percentile
% Bias
procedure No Netting Down Person Level Netting Down Household Level Netting Down
Figure 1. Bias Incidence Curves for Two Netting Down Procedures Compared to no Netting
Down
Review of Income and Wealth, Series 00, Number 00, Month 2016
V
C2016 International Association for Research in Income and Wealth
7
countries, the results show that the differences between gross and net earnings
increase at higher percentiles, which of course results from progressive taxation.
The dotted lines represent the bias resulting from the netting down procedure
that used person-level tax information. The results suggest that this procedure
typically underestimates earnings levels at lower percentiles (bias <0) and overes-
timates at higher percentiles (bias >0). Bias levels are substantially smaller than
not correcting at all (the solid line), but reaches levels above 25 percent at higher
percentiles in the Belgium 1997 and both UK datasets. In the other datasets, levels
of bias are close to 0 at all percentiles.
The dashed lines represent the bias resulting from the netting down proce-
dures that used household-level tax information. While the patterns are similar as
described above, the household-level netting down procedure typically performs
less well than the person-level procedure.
These results demonstrate that applying netting down procedures is prefera-
ble over not correcting for the difference between gross and net earnings in com-
parative research, and that preferably the person-level procedure is applied.
Nevertheless, in some datasets substantial amounts of bias remained, particularly
at the higher percentiles. As the net earnings tend to be under-estimated at lower
percentiles and over-estimated at higher percentiles, the bias incidence curves fur-
ther suggest that various estimates of inequality based on netted down earnings
variables would be biased upwards. The extent of this bias, however, is difficult to
assess from these curves. Therefore, Tables 1 and 2 present estimates of bias for
six commonly applied measures of inequality based on the person-level and
household-level netting down procedures, respectively.
The bias of the person-level netting down procedure (in Table 1) typically
was below 10 percent, with the clear exception of the Theil index and the Coeffi-
cient of Variation in the U.K. In that country, the bias incidence curve of the
person-level netting down procedure continued sloping upwards at higher income
percentiles. The bias of the household-level netting down procedure (in Table 2)
was typically higher, with many estimates upwards of 10 percent. Again, the Theil
index and Coefficient of Variation in the U.K. show exceptionally high levels of
bias, up to 51 percent. In some cases the netting down procedures were associated
with a negative bias, indicating that using the netted down earnings variable
resulted in a under-estimate of inequality. An example is the share of low earners
TABLE 1
Quantifying Bias (%) InPerson-Level Netting Down Procedure
Dataset
75p/
25p
Theil
Index
Coefficient
of
Variation GINI
Low
Earnings
Gender
Gap
Austria '04 20.8 20.1 0.2 20.1 20.2 0.0
Belgium '92 0.0 20.8 20.3 20.5 21.3 20.5
Belgium '97 0.4 8.4 4.8 4.3 1.4 2.0
Estonia '04 21.2 21.2 20.8 20.5 22.3 21.3
Ireland '04 20.8 20.5 20.5 20.2 20.4 20.2
U.K. '99 4.2 13.9 17.3 5.0 3.5 2.9
U.K. '04 4.8 30.7 48.7 7.9 4.9 5.0
Review of Income and Wealth, Series 00, Number 00, Month 2016
V
C2016 International Association for Research in Income and Wealth
8
in Estonia 2004: the person-level netting down procedure is associated with a bias
of 22 percent. Overall, comparing the results presented in Tables 1 and 2 suggests
that the person-level netting down procedure outperformed the household-level
netting-down procedure.
6. Conclusion
This technical note presented guidelines for comparing gross and net income
datasets, which were tailored to use with the LIS data but which apply to a wider
range of income datasets. Two netting down procedures were introduced that
approximate net earnings from information regarding gross earnings, in conjunc-
tion with data on taxes and social contributions paid by the household. Using
these netting down procedures reduced bias in comparisons of earnings between
gross and net LIS datasets. Generally, this suggests that applying a netting down
procedure is preferable over not netting down. Using the person-level procedure
was desirable over using the procedure based on household-level tax variables.
Data availability will often dictate which of the two netting down procedures
users can apply. It should be noted, however, that it is to be expected that the
household-level netting down procedure performs better in a country with joint-
taxation, relative to countries with separate taxation. Furthermore, in both net-
ting down procedures it is assumed that all sources of income are taxed at the
same rate. From this, the expectation follows that the procedures will perform bet-
ter in countries with a single, rather than a dual tax system in which separate tax
rates exist for capital income and other income.
The netting down procedures performed more poorly in the datasets for Bel-
gium and the U.K., compared to other datasets. The LIS dataset on Belgium in
1997 was based on the Socio-Economic Panel. In this original dataset, the infor-
mation on holiday- and end-of-year bonuses was available net of taxes, and
unavailable gross of taxes. Hence, in calculating the yearly gross earnings when pre-
paring the LIS dataset, the monthly earnings were multiplied by 13,85 (approxi-
mating the average bonuses). For the net yearly earnings the information on
bonuses was available in the original data. Hence, whereas in the harmonized LIS
dataset the net yearly earnings account for person-level variation in bonuses inde-
pendent of other earnings, in the gross yearly earnings the bonuses that affect such
TABLE 2
Quantifying Bias (%) InHousehold-Level Netting Down Procedure
Dataset
75p/
25p
Theil
Index
Coefficient
of
Variation GINI
Low
Earnings
Gender
Gap
Austria '04 5.7 6.2 2.8 3.2 6.0 6.4
Belgium '92 5.9 8.8 3.0 5.5 29.0 8.1
Belgium '97 6.4 12.2 3.6 7.7 20.5 13.2
Estonia '04 1.4 1.9 0.5 1.2 22.5 2.3
Ireland '04 6.6 5.9 1.4 3.4 3.3 2.8
U.K. '99 11.0 21.7 20.8 8.8 9.2 8.2
U.K. '04 10.0 38.0 51.3 11.5 9.6 10.0
Review of Income and Wealth, Series 00, Number 00, Month 2016
V
C2016 International Association for Research in Income and Wealth
9
person-level variation were not accounted for. Second, the LIS datasets from the
U.K. (both in 1999 and 2004) were based on the Family Resources Survey. During
the recoding of these datasets to the LIS templates, the gross earnings were speci-
fied to include income from odd jobs, while net earnings could not be specified to
include this source of income. Hence, the difference between gross and net yearly
earnings is an overestimation of the “real difference”. Therefore, the netted down
results may actually be a better representation of persons' true net earnings than
the net earnings reported in the data. It should be noted, that within the scope of
this paper it was not possible to empirically test this statement.
When comparing a large number of gross and net datasets, users may want
to statistically control for the different netting down procedures used. In
regression-based analyses, for instance, this could be done by adding dummy vari-
ables indicating the observations derived from datasets netted down with the
person-level procedure, and another dummy for the observations from datasets
that were netted down using the household-level procedure (with observations
from net datasets as the reference category).
To conclude, country-comparative and trend analyses of earnings based on
both gross and net datasets should be done with caution. The netting down proce-
dures presented here typically improve comparability in studies based on the LIS
data. However, depending on the outcome measure of interest, and especially
when no person-level tax variables are available, netting down procedures based
solely on household income can sometimes result in substantially biased approxi-
mations of net earnings.
References
Atkinson, A. and A. Brandolini, “Promise and Pitfalls in the Use of Secondary Data-Sets: Income
Inequality in OECD Countries as a Case Study,” Journal of Economic Literature, 39, 771–799, 2001.
Gornick, J. C. and M. J€
antti, “Child Poverty in Cross-National Perspective: Lessons from the Luxem-
bourg Income Study,” Children and Youth Services Review, 34, 558–568, 2012.
Gornick, J. C. and B. Milanovic, “Income Inequality in the United States in Cross-National Perspec-
tive: Redistribution Revisited,” LIS Center Research Brief, 1, 2015.
Kenworthy, L., “Do Social-Welfare Policies Reduce Poverty? A Cross-National Assessment,” Social
Forces, 77, 1119–1139, 1999.
Maldonado, L. C. and R. Nieuwenhuis, “Family Policies and Single Parent Poverty in 18 OECD
Countries, 1978–2008,” Community, Work & Family, 18: 395–415, 2015a.
———. “Single-Parent Family Poverty in 24 OECD Countries: A Focus on Market and Redistribution
Strategies,” LIS Center Research Brief, 2, 2015b.
Sainsbury, D. and A. Morissens, “Immigrants' Social Rights across Welfare States,” in Sainsbury, D.,
Welfare States and Immigrant Rights. The Politics of Inclusion and Exclusion, Chapter 6, Oxford
University Press, Oxford, 2012.
Sutherland, H. and F. Figari, “EUROMOD: the European Union Tax-Benefit Microsimulation Mod-
el,” International Journal of Microsimulation, 6, 4–26, 2013.
Supporting Information
Additional Supporting Information may be found in the online version of this article at the
publishers web-site:
Appendix A1: Comparative Research with Net and Gross Income Data: An Evaluation of
Two Netting Down Procedures for the LIS Database.
Review of Income and Wealth, Series 00, Number 00, Month 2016
V
C2016 International Association for Research in Income and Wealth
10
