Instructions
User Manual:
Open the PDF directly: View PDF 
.
Page Count: 10
How to work with the MCVL - An application to
unemployment
Cristina Lafuente
University of Edinburgh
January 22, 2019
This document provides technical guidance on how to turn the raw text files from the
Muestra Cont´ınua de Vidas Laborales (MCVL thereafter) into a panel dataset in Stata.
The code allows the researcher to then do two things:
•LFS-like panel: Turn the data into a quarterly (or monthly) panel similar to the
Labour Force Survey (LFS) panelling format.
•Spell panel: Keep the data as an “spell panel” (one observation corresponds to
one spell) and link it to tax registry data. This link allows to add wage, benefits,
and self-reported profits into the spell panel.
There are four main blocks: formatting, binding, implementing unemployment ex-
pansions and panel formatting. All steps up to ‘panel formatting’ are required for both
LFS-like panel and Spell panel.
Please see the README file in this GitHub repository for more details on the order of
execution of files. The unemployment extensions and the panneling process are explained
in detail in Lafuente (2019).
Obtaining the data
The data is available upon request and subject to approval at the Spanish Social Security
website (Seguridad Social).1Please notice that the application instructions and docu-
ments are in Spanish.
1http://www.seg-social.es/wps/portal/wss/internet/EstadisticasPresupuestosEstudios/Estadisticas/EST211/1459
1
In the Seguridad Social website you will find 4 files. ‘COMO SOLICITARLOS’ gives
you the full application instructions. ‘FICHA USUARIO’ should be completed, detail-
ing the name and affiliation of the researcher and giving specific details about your re-
search project and how you plan to sue the data. Then, you should sign one declaration
of confidentiality for each year of the MCVL you are requesting. This form is either
“CONDICIONES MCVL SDF” for the version without fiscal data and “CONDICIONES
MCVL CDF” for the fiscal data version. Once you have read the conditions, signed the
confidentiality agreement(s) and filled the ‘FICHA USUARIO’, you should remit these
documents to the address in the ‘COMO SOLICITARLOS’ file.
Once approved, you will be sent a copy of the data. Then you can proceed to download
this repository and place the files in the right folders as instructed in the README file.
1. Formatting
There are two ways of formatting the MCVL: year-by-year panel and retrospective panel:
•The year-by-year panel uses the information in all waves of the MCVL separately.
This allows to take into account workers who are in some waves but not in others
(see Garc´ıa P´erez (2008)) and keeps the representatively of the population in every
year.
•The retrospective panel uses information from the latest available year only.
Although some representative of the sample is sacrificed, it is easier to study unem-
ployment duration as there are no “cuts” or overlapping spells as in the year-by-year
version.
As described in Lafuente (2019), there may be applications where one or the other
is preferable. In what follows I describe the year-by-year panel approach, as it is more
complex and appropriate for unemployment measurement. The same steps are needed
for the retrospective panel, but there is no need to limit spell duration to be within the
year they are reported, which makes things simpler.
1.1 Formatting Affiliation files (format all.do)
Open the ASCII files for each year, name the variables according to the MCVL guide. Be
careful as the position of variables as it changes through years. Then format the dates of
start (alta) and and end (baja) of each spell, which together with the personal identifiers
define each spell.
Next proceed to clean the overlapping spells (spells beginning before the end of the
previous spell) as in Garc´ıa P´erez (2008) cases a (total overlap) and b (partial overlap).
2
For total overlaps, keep the longest continuous spell and drop smaller spells that happen
at the same time. For partial overlaps, make the continuing spells start when the previous
ends.
It is time to create the most important variable: labour market status during the
spell. I chose to make this a string variable - which will be convenient when creating
flows - but a labelled numerical value would work as well. For this we need to combine
the information of 4 other variables:
•Tipo de relacion con la seguridad social, codes 700-800 correspond to unemployment
benefit claimants. Mark as unemployed (“U”).
•Tipo de contrato, codes 400-900 correspond to temporary contracts. Mark as tem-
poray, “T”.
•Tipo de contrato, codes 99-400 correspond to permanent contracts2. Mark as per-
manent, “P”. There is an exception though: code 540 corresponds to partial retire-
ment, so I mark this as out-of-the-labour-force (OLF). Also those whose variable
regimen de cotizacion is 140 are in early retirement - so I mark them as OLF as
well.
•Regimen de cotizacion, codes 500-600 correspond to self-employment. I mark them
as ”A” for Spanish autonomos3.
You may also want to create other auxiliary variables, such as an indicator for part-time
contracts. For this you can refer to the accompanying .do file or refer to the official guide.
The variable for this is Tipo de contrato.
Important: the variables Empleador (forma Judirica) - Letra NIF de la Entidad Pa-
gadora and Identificador (NIF/CIF) anonimado de la entidad pagadora uniquely identify
firms both in the affiliation files and in the fiscal file. If you want to use wage information,
make sure to create a variable that joins both into one string variable. I call this variable
firmID and move it right after the worker identifier.
1.2 Formatting pension files (format all pension.do)
Name the variables according to the official guide. If using for labour market flows, most
of this variables are irrelevant, but keep the dates (and format them accordingly) and the
2Note that some of these contracts may be fijos discontinuos, that is, permanent workers that only
work for part of the year. They are different than temporary workers because they don’t have a contract
expiration rate and are protected by severance payments. If the reserchers wants to treat them differently,
their contracts correspond to codes 300-400.
3If you want to be really precise, you should mark those whose Regimen de cotizacion is equal to
700-800 and 824-840 as self employed. These are the cases of farmers and sea captains.
3
personal identification number.
1.3 Formatting personal information files (format all personal.do)
Name the variables according to the official guide.
Special care should be taken with the variable fecha de defuncion that marks the death
date of some workers who passed. The birth date should also be considered carefully as
there are some likely mistakes - most famously a worker who was supposedly born in
1906 and was still working in 2005 - likely a coding error for 1960.
Ideally the 2011-2012 personal file should be the most up-to-date information as there
was a census in 2011. It is recommend to impose the values of the education variables
from 2012 onwards over earlier years, whenever possible.
There are some exceptional cases of repeated personal identifiers. I chose to keep the
youngest of the two, but whatever criteria you use, keep only one so it can merge easily
with the affiliation file.
2. Binding
2.1 Binding the files together: pensions (format all pension.do)
Year by year, open the formatted affiliation file and append the pension file to it. sort
by date. If the final observation of a person is an entry from the pension file (easily
identifiable because all affiliation file variable will be blank) then fill in their labour
market status variable as OLF. Then fill in all the information missing in this last entry
from the previous spell (place of residence, for example). Delete all other pension entries
if you are only interested in labour market flows.4
2.2 Binding the files together: personal information (format all personal.do)
Merge the formatted personal file and the affiliation file (with pension information) to-
gether, using the personal identifier as joining variable. It should match virtually all
cases. For the rare exceptions, I choose to keep the affiliation registered without personal
information but drop the personal information entries without a matching affiliation en-
try.
Now it is a good time to drop all spells that happen before the current wave year - so
2006 only has spells active in the period 1st January 2006 onwards. This would ease the
binding process below. Skip this step for the earliest year in the sample if you want keep
4For example, Some disability or widowhood pensions can be received while in the labour force. If
these are of no interest for the study, they can be committed.
4
retrospective information before the start of the sample5. You can keep retrospective
information for all years, but that would make the merging years together process more
cumbersome. Whenever in doubt, keep all spells.
Important: create a variable called year and set it equal to the file year. This will
help you identify the information that each wave brings to the unified panel. Save the
enriched affiliation files.
2.3 Binding all years together (Patchwork.do / Patchwork prelim.do)
Start with the earliest year in the dataset.6Drop all spells starting further than the
31st of December - modify the end date of the spell so it is 31st December. This right
censoring should ensure the years are well matched together, so the 2006 file brings only
spells active in 2006, the 2007 in 2007 etc. Append the next affiliation file (with pension
and personal information). Trim the spells over the end of the year and add the next file.
Continue until you are left with the last wave. Do not censored this last year.
If you choose to keep retrospective information in each year, as you append each
year erase duplicated spells. Here having created a variable for each year will come in
handy, as it would help to identify identical spells but in different years (waves): they
must share the same start date and the same firmID value. For the rare cases where
firmID is not available (for example in unemployment spells) use the variable Codigo
de Cuenta de Cotizacion Principal (CCCP) (right after the fiscal identifier variable) to
identify duplicates spells.
At this point you should have one unique affiliation file with all waves joined together
and no duplicated spells except from those that last beyond a calendar year - for example,
a job that starts in May 2006 and ends in June 2009 should have 4 entries: one each for
2006, 2007, 2008 and 2009.
Depending on what you are interested in, it may be a good idea to create a effective
end date variable that matches the latest end date for each spell - in the example above,
set the effective end date as June 2009 in all entries. This way if you want to get statistics
on tenure, you can either consider tenure up to the current year or total tenure in the
sample. In the previous example, the first variable would be 8 months (May-December
2006) and the second 3 years (May 2006- June 2009). As a general rule it is better to
create new variable than modify old ones, and this is particularly true with dates.
5In practice this trimming will affect all observations that are active in later years, so if you are using
many waves together keep all retrospective information and drop the repeated cases later, during the
Biding all years together phase.
6It is recommended the 2005 wave as the absolute earliest to keep the sample representative of the
population in that year. However you can use the 2005 file but name an earlier year as your starting
point. However the further away from 2005, the worse the representativity of the sample.
5
3. Expansions
3.1 Contract modification adjustment (cma.do / cma panel.do)
In many cases contracts change across the years - this is the case of temporary workers
promoted to permanent contracts. The way these cases are recorded in the MCVL is
not easy to deal with. Ideally, for the purpose of job market flows we would like to have
separate entries for each kind of contract.
Look at the varaible Fecha de modificacion del tipo de contrato inicial o del coeficiente
de tiempo parcial inicial, towards the end of the affiliation file variables. If this variable
is filled with a date, then there was a change in contract. The next variable, Tipo de
contrato inicial, contains the original contract code of the job. Use the guide in step 1 to
interpret its value.
Create an indicator variable that is equal to 1 if (1) the current wave year equals the
year of the contract modification date AND (2) the type of contract is not the same as the
original type of contract (for example, if there is a change from temporary to permanent
(or vice-versa) or to part-time).7
Duplicate the spell in which the indicator variable is equal to 1. Change the type of
contract of the first copy to be the original type of contract. Change the end date of
this first copy to coincide with the modification date, and change the start date of the
second copy to the modification date. Depending on how you want to treat tenure, you
may want to extend this last change to the start date of all the other entries in posterior
years to the contract modification. Now you have two spells for each job: one before the
contract change and one after.
Repeat these steps with the variable Fecha de modificacion del tipo de contrato segundo
o del coeficiente de tiempo parcial segundo and Tipo de contrato segundo. This is the
second contract modification variable.
Be careful when recording the length of each spell before and after the contract change.
In some applications you may be interested in the whole period (for example for tenure)
but if you want to count temporary and permanent job experience separately you may
want to treat the two contracts differently.
3.2 Unemployment Expansions
Before proceeding, sort all spells by worker id, labour market state and date (in this
order). Number the spells in separate variables for each state - so for example, if a worker
was unemployed in two separate periods, create a variable called number of unemployment
7In case you are interested on recalls or temporary contract renewals, you may wish to also create a
new entry even if the contract type doesn’t change.
6
spell (NoU) and set it equal to 1 for the first one and 2 for the second. Or if a worker
had 9 temporary jobs, create a variable NoT and number them chronologically.
Sort again the sample by id and date of entry and exit. Fill in all the blanks in NoU
equal to the previous NoU value and set 0 for all spells before the first unemployment
value. Using this variable (NoU), create another variable counting the days the worker is
employed at each year in between unemployment spells. This will give us the total number
of days contributed to the social security, which we will use to calculate unemployment
benefit entitlements.8Remember to reset this counter to zero each time there is a new
unemployment spell.9
Create a variable equal to the end of each spell (call it original ending) that will be
of use later to calculate the expansion period.
The LTU expansion (coru ltu.do)
First, join consecutive unemployment spells within the year: if both unemployment spells
came from the same wave, and one starts immediately after the other, I consider them
one single spell10. There are many cases of workers that received more than one subsidy
at the same time (because of illness or family reasons) but they are part of the same
unemployment spell.
Second, if there is a gap between the end of an unemployment spell and the beginning
of the next job, extend the end date of the unemployment spell as to join the two. Make
sure that the next spell is employment or self-employment, and not retirement. The
reason being that we cannot be sure that these workers are looking for a job - if they
transition to retirement probably they were out of the labour force to start with. I
choose to extend the spells of workers whose last entry is unemployment to the end of
the sample.11 This is crucial to account for all the workers whose benefits have expired
and are still unemployed at the end of the sample. If your final year is beyond 2009, you
should definitely do this as the number of unemployed workers without benefits reaches
50% in 2012.
Third, if the previous expansion meant that the unemployment spell extended over
the year of its original wave, duplicate the unemployment spell and set the wave year
equal to the next year. If as a result the spell extends over two years, create two copies.
8Before 2013, self-employed time does not count as contributions towards unemployment insurance,
so do not count self-employment spells.
9Some workers can choose to “save” part of their unconsumed unemployment benefits for next un-
employed period, in which case the time contributed by the next job won’t count towards the total. By
resetting after each unemployment spell by default we make sure we only count the minimum possible
time a worker could have contributed to the social security.
10Some authors want to make distinctions between unemployment benefits and unemployment sub-
sidies - the latter referring to reduced amounts that some long term unemployed workers receive after
running out of unemployment insurance. If so you may want to skip this step.
11 See section 5.1 in Lafuente (2019) for a more detailed discussion on the effect of this extension and
some alternatives.
7
For example, if a spell started and ended in 2009 but after the expansion it ends in 2010,
create a duplicate of the original spell and modify its year so it belongs to 2010. This
way there would be two copies: one in 2009 and one in 2010 - as it would be the case if
unemployment benefits wouldn’t have expired.
The STU expansion (coru stu.do)
In addition to the previous expansion, create a new unemployment spell if there is a
gap between two jobs that lasts more than 15 days12 and at least one of the following
conditions are met:
1. the first job was self-employment
2. the first job ended in a quit (if the variable Causa de Baja en Afiliacion is 51)
3. by the end of the first job, the worker hasn’t accumulated 12 months of continuous
employment
In all of the previous conditions, the worker is not legally entitled to unemployment ben-
efits, and thus we can interpret the period between jobs as unregistered unemployment.13
You can further restrict these conditions by imposing that the firm identifiers of the two
firms are different, so the worker is not being recalled to the same firm. I follow this
approach in the paper. Set the end and start dates to fill in the gap between jobs.
Expand also unfinished spells that start towards the end of the sample. For my period
of study, 2005-2013, I choose to expand all unfinished spells that satisfy the requirements
above and start within the last 3 years of the sample. As before, accounting for spells that
are not yet done is important to accurately reflect unemployment. See section ?? for a
more detailed discussion on the effect of this extension and some alternative assumptions.
As before, if this expansion takes the unemployment spell over the year of the wave,
duplicate the affected observations. There should be one copy for each year the spell
takes place in.
Finally, stata creates a new variable to identify copies and originals every time you
duplicate observations. If your software of choice does not do that, make sure you have
an indicator variable for these unemployment spells so you can identify them later.
12This threshold is arbitrary. Results do not change much when the limit is put at 10 days or 1 month.
Garcia-Perez (2008) also sets 15 days as a reasonable threshold.
13 See section 4 of Lafuente (2019) for more information on this assumption.
8
4. Panel Formatting (LFS-like)
(panel/quarterly panel U0.do)
4.1 Select the window
The LFS runs interviews during the reference quarter, and so it gets its answers from
replies in an unknown reference day within the quarter. This is inevitably going to lead
to discrepancies in the results, as if the reference day in the MCVL does not coincide
with the LFS the answers can be different. The extent of the discrepancy is discussed in
section 5 of Lafuente (2019) at length. Overall the window of choice does not seem to
make much of a difference. The approach here is to select a window period within the
quarter (or the month if interested monthly transitions). I chose the 1st to the 15th of the
first month of each quarter. That is, the 1st-15th of January, April, July and October.
Allowing for 2 week windows is important, especially in the case of January as many jobs
start after the Christmas break - which in Spain can last up until the 6th of January.
4.2 Create quarterly state variables
Once the window has been chosen, I focus on the spells whose entry date is after the
beginning of the window period, but before the end. It is important to clear completely
overlapping spells (with the same start and end dates) before proceeding. If there is more
than one spell within a window, I chose to keep the one that continues onwards - that
is, the last one. Another approach is to take the one with longest duration, but this can
prove difficult. For example, take a long employment spell that ends the 22th of January,
followed by 6 months of unemployment. The first spell would be selected if we apply
the longest duration rule, but the second spell would be selected instead if we keep the
continuing spell. This is particularly important if in the next window period the worker
is employed again, as not counting unemployment can understate the flows in and out of
unemployment. See Lafuente (2019) section 5 for further discussion.
Once we have one spell per window, all that is left is to fill in a variable for a spell-
period. In many cases you will have to create copies of the same spell, when it appears in
two different windows. For example, an employment spell that features in the first and
second quarter would need to be duplicated. The original will be assigned to quarter 1
and the copy to quarter 2.
If you have applied any of the unemployment expansions, you may also want to create
a different state for unregistered unemployment spells. For example, if an unemployment
spell that originally lasted for a quarter now lasts for two - because of the LTU expansion
- then you can label the first observation ”U” and the second ”0”. For this you must
use the expansion date variable we created before modifying the start and end dates of
9
spells. All unemployment spells generated from the gaps expansion can also be labelled
differently.
4.3 Create stocks and flows
The last step is to only keep spells that feature in a quarter and discard all other spells.
We will be left with a panel dataset that mimics the structure of the LFS, with one
observation per quarter. In this case however we will more detailed information - precise
tenure and experience data for example. This can be used to calculate unemployment
stocks (with or without benefits) and temporary share of employment, for example.
To create flows, just link two consecutive quarters for the same worker. String vari-
ables are best for this, as you can simply add two strings to form a new variable. Here
you may want to relabel “TUT” or “UTU” flows, conditioning on the duration of un-
employment (or the temporary contract). This can also be achieved by following “the
longest spell” rule when choosing one observation per quarter. Note that this brings the
MCVL flows closer to the LFS, but they are not classification errors - which is the reason
these flows are modified when working with surveys (see Elsby et al. (2015) for more
details).
References
Garc´ıa P´erez, J. I. (2008). La muestra continua de vidas laborales: una gu´ıa de uso para
el an´alisis de transiciones. Revista de Econom´ıa Aplicada 16 (1).
Lafuente, C. (2019). Unemployment in administrative data using survey data as a bench-
mark.
10
