The Effect of Outplacement on Unemployment Duration in Spain* by F. Alfonso Arellano** DOCUMENTO DE TRABAJO 2007-16

June, 2007

SERIE Capital humano y empleo CÁTEDRA Fedea - Santander

I would like to thank to César Alonso and Juan José Dolado (University Carlos III of Madrid) for the supervision of the paper, and Manuel Arellano (CEMFI), Juan F. Jimeno (Central Bank of Spain), Florentino Felgueroso (University of Oviedo) and participants of the XXX Simposio de Análisis Económico of the Spanish Economic Association in Murcia and the 2006 EEA-ESEM Congress in Vienna for very useful comments. Also thanks to Josep Pau Hortal (Creade), Eugenio Sanz (Creade) and Antonio Hernando (INEM) for data bases. Financial support from the Spanish Ministry of Education for Predoctoral Fellowship AP2000-0853 and from the Spanish DGI for Grant BEC2003- 03943 is gratefully acknowledged. The usual disclaimer applies.

University of Alicante and FEDEA. Address for correspondence: F. Alfonso Arellano. Departamento de Fundamentos del Análisis Económico. Facultad de Ciencias Económicas y Empresariales. Universidad de Alicante, Código 99, 03080 Alicante. Tel.: +0034 965 903263. E-mail: arellano@merlin.fae.ua.es.

Los Documentos de Trabajo se distribuyen gratuitamente a las Universidades e Instituciones de Investigación que lo solicitan. No obstante están disponibles en texto completo a través de Internet: http://www.fedea.es. These Working Paper are distributed free of charge to University Department and other Research Centres. They are also available through Internet: http://www.fedea.es.

Jorge Juan, 46 28001 Madrid -España Tel.: +34 914 359 020 Fax: +34 915 779 575 infpub@fedea.es

Abstract

The paper analyses the effects of individual and group outplacement services for a group of unemployed workers in Spain on their unemployment spells. Two data bases are used, one from the Spanish Department of Employment (INEM) and other from one of the most important outplacement firms (Creade), between 1998 and 2003. Using (non-parametric) matching methods and unemployment duration as outcome variable, the results suggest outplacement produces a “reservation wage” effect for men, increasing unemployment spell by three and two months for individual and group outplacement, respectively. Women who receive the services increase very slightly unemployment duration, showing also (non-significant) spell reductions for individual outplacement.

Resumen

El documento analiza los efectos de los servicios de recolocación individual y colectiva para un grupo de trabajadores parados en España en sus respectivos periodos de desempleo. Se usan dos bases de datos, una perteneciente al Instituto Nacional de Empleo (INEM) y otra proveniente de una de las empresas de servicios de recolocación más importante (Creade), entre 1998 y 2003. Utilizando métodos de emparejamiento (no paramétricos) y la duración en el desempleo como variable de estudio, los resultados sugieren que la recolocación produce un efecto “salario reserva” para los hombres, aumentando el periodo de desempleo en tres y dos meses para la recolocación individual y colectiva, respectivamente. Las mujeres que reciben los servicios aumentan la duración del desempleo ligeramente, presentando también reducciones del periodo (no significativas) para la recolocación individual.

Keywords: outplacement, unemployment duration, (non-parametric) matching methods

JEL Codes: J68, C14

1. Introduction

Unemployed workers have gone occasionally to public organisms (Spanish Department of Employment –INEM– or Regional Employment Services) to receive basic and general career guidance services. However, these services are not provided for all workers or they do not receive the support that each particular case requires, since these organisms show limited budgets or time limit.

The term “outplacement” was not associated to employment creation in Spain until a few years ago. Mergers and reorganizations for the first years of the XXI century have forced firms to make a part or all of their staffs redundant. Outplacement services consist of helping the candidates (unemployed workers) to discover their abilities and use effective tools for job search, according to professional's characteristics. The consultant who is devoted to the service is not a recruiter neither a head-hunter, and complete confidentiality is guaranteed.

The beginning of outplacement dates back as far as the deep social and political changes of United States in the 1960s . The outplacement sector in USA was constituted by around fifty companies that turned over about 50 million dollars in 1980, while the number of companies ascended at 230 and the turnover overcame 650 million dollars in 1991 (Cowden, 1992). In 2000, a thousand outplacement firms had a turnover of around two billion dollars (Mendels, 2001).

Development of outplacement has spread over USA thanks to labour market legislation and cultural factors (higher propensity to geographical and job changes and dynamism of managerial fabric). Considering EU Member States, apart from UK (with a similar system to USA), France is one of the countries that shows a higher outplacement growth, since legislation has regulated this activity. Outplacement must be incorporated in the social plan which companies carry out and use in employment regulations. In spite of the existence of outplacement firms for years in Spain, their importance has increased since the stock market downturn in 2002.

Both the firm and the professional can enjoy several advantages of outplacement services, besides generating positive effects for policy-makers. The negative influence of firing on the relationship between the rest of employees and the firm is mitigated, motivation and productivity in the organization are maintained and the firm’s image improves. The worker who leaves the firm will get a psychological and professional help to be ready for a new labour stage. Finally, Public Sector obtains a cost reduction from working population's loss. The reduction becomes significant for older professionals because the misuse of early retirement diminishes.

¹ British officers were helped to get a job after returning from the colonies in 1908, but there was not a methodology for outplacement services (Hortal, 1999).

Despite the extensive literature and description of experiences, research on outplacement has been related to psychological and sociological aspects (Wolfer and Wong, 1988). Theoretical models emphasize the effect of outplacement on job change and worker's human capital growth. Each phase of outplacement process is analysed to clarify its influence on unemployed worker's state of mind (Lattack and Dozier, 1986).

Applied studies justify the utility of outplacement, although only variance-covariance analysis has been used as statistical tools (Gowan and Nassar-Mc Millan, 2001). Westaby (2004) evidences that outplacement facilitates high level professionals the return to employment situation. The effect is also positive for white-collar workers compared to other similar social services (Davy, Anderson and DiMarco, 1995). At European level, there are interesting examples as the group Usinor Sacilor in France (Jacquier, 1996) and British Coal Enterprise in United Kingdom (Furness and Lewis, 1996). These firms had to face up to hard reorganizations, sometimes concentrated on regions. The creation of subsidiary outplacement firms mitigated the effect of factory closure. The most interesting references in Spain are also focused on the study of cases, as instruments for the promotion of outplacement companies’ activities³.

Given the importance of outplacement, the aim of the paper consists on the analysis of this active labour market measure developed by one of the most important outplacement firms in Spain, Creade. The remainder of the paper is organised as follows: The next section outlines important details about outplacement for the purpose of the paper. In Section 3, identification and estimation strategy are introduced. Sections 4 and 5 provide descriptive statistics and complete information of the data bases. Section 6 presents the main results and Section 7 concludes. Appendix A contains selected descriptive statistics of data base. Finally, estimates are found in Appendixes B and C.

² Aquilanti and Leroux (1999), Meyer and Shadle (1994), and De Ramos and Hernández (1999) include further references.

See examples and further information in http://www.e-creade.com, http://www.uniconsult.es, http://www.adecco.es and Jiménez (2000).

2. Description of outplacement

Hortal (1999) defines outplacement as “the set of services provided by a consultant to the firm (client) and the professional (candidate) when they negotiate the rupture of the contractual relationship (…) in order to facilitate consultancy and support to assure the continuation of the professional's career.”

Outplacement services can be requested by private and public companies. The firm that dismisses the professional usually pays the cost of the programme. This expenditure is not related to the dismissal compensation of the worker. The outplacement firms move between the client and the candidate, knowing the restrictions of each party. They cannot guarantee the candidate a new job. If the programme concludes and the candidate does not have found a job, the contact between the worker and the outplacement firm does not get lost, since the prestige of the firm will be clearly committed. The failure can take place, but only a small percentage of candidates gives up the methodology before achieving the goal.

Although outplacement services are custom-designed, there exist general stages describing the process: study on the situation, professional project, action plan, search strategy and integration process. The consultant initially discovers and identifies the capacities, abilities, ambitions, motivations and knowledge of the candidate. In the second step, the results are assessed to give coherence to the candidate' professional profile and develop worker’s potential according to the labour supply-demand. An action plan and search strategies are carried out according to the project, and the consultant helps the candidate to face up to work interviews. The consultant guides the worker and provides job offers according to the professional profile in the fourth phase. Finally, integration process involves beginning of the new professional activity and supervision is made, finishing approximately after one year, or when the candidate is satisfied and integrated.

According to De Ramos and Hernández (2000), the programme can be applied to a professional (individual outplacement), the candidate’s spouse, or a set of workers. In the latter option, the groups of workers usually belong to homogeneous organization levels and areas, and the service is defined as group outplacement.

The basic causes of individual outplacement are related to individual motivations, as inadequacy to the position or irreconcilable differences. The reasons for using group outplacement are not only associated to economic difficulties of the firms, but mergers, takeovers and strategic plans of the firms.

These facts generate duplicities which force to undertake staff reorganizations. Hence, group outplacement implies a bigger activity and complexity than individual outplacement. The firm that finances the programme has the initiative to choose the type of service. This fact is more evident in the case of group outplacement. The candidate's influence is bigger in the individual case, since the service constitutes a soft and conventional solution between both parties.

Individual outplacement is designed to managers. They hold further financial resources and time does not represent an important restriction to examine the available options. Nevertheless, the supply of similar jobs to the last one is smaller and age plays an outstanding role to intensify the job search. Group outplacement is associated to sets of workers in the medium-low level of firms, who make redundant for reorganization reasons. Time is a key factor, not only for financial reasons but the labour supply is wider than the previous case.

According to the descriptive statistics carried out about outplacement services in Spain (Sáenz, 2000; Group MOA, 2000; and Creade, 2003), individual outplacement is usually used by men with university education, between 35 and 45 years. The firms which hire the services are focused on Service Sector and belong to big multinational firms. These results are similar to group outplacement, although the education level decreases and workers get a new job earlier. The unemployment spell is below six months on average, and the new job is achieved through the use of contacts, with a similar or higher wage and it belongs to an analogous professional level.

The service cost depends on type, complexity and duration, albeit the mean rate is around fifteen percent of the worker's gross wage for programmes with unlimited duration (De Ramos and Hernández, 2000).

Outplacement firms in Spain have been constituted given the commitments of the Constitution, even though the legislation is not explicit enough. They are defined as Service-Sector companies, since legislation forbids the existence of profit-making outplacement agencies. In the so-called Workers’ Statute of 1995, there is a generic mention to use these services for employment regulations. However, there are not general agreements about outplacement between firms and unions.

The consultants specialized in outplament arose from multinational companies in the middle of the 1980s, but their methodology was not very extended among the Spanish companies until the last years. Given this context, the most important outplacement consultant companies in Spain - Creade, Lee Hecht Harrison, MOA Groupe BIS, Right Management Consultants and Uniconsult- helped almost 7,000 professional from 800 firms to get a job in

2003. The firms also render other services, as career management, coaching and transition consultancy.

3. Identification strategy and estimation methods

The theoretical approach is based on the terminology of Heckman, Lalonde and Smith (1999) about Roy (1951) and Rubin (1974) model. Workers belong to one of two mutually excluding states at the same time, “1” denotes the treatment state and “0” denotes the non-treatment state. Let Y be the outcome variable (i.e., unemployment spell in days), so is the value of the outcome variable for the individual i at time t if the worker has received the treatment, and in the case of non-treatment. In the paper, the treatment group is formed by workers who received outplacement services of the firm Creade, the rest of workers are included in the control group. The impact for individual i at period t of the measure is . However, this difference is unknown because these two terms cannot be observed for any individual at the same time:

\[Y _ {i t} = D _ {i} \cdot Y _ {i t} ^ {1} + (1 - D _ {i}) \cdot Y _ {i t} ^ {0}\]

where is a dummy variable equals to one if the individual i receives the treatment and zero otherwise. The difficulty is known by the Fundamental Evaluation Problem. The solution is related to the available data bases. In this case, the alternative consists of using non-parametric matching methods. Lechner (2000) comments the advantages of the non-parametric matching methods versus other parametric and non-parametric methods. With respect to the first group, matching methods are robust to functional forms of the conditional means, so the individual causal effect is free of restrictions, as well as the unobserved heterogeneity of the population. Comparing to other nonparametric methods, matching methods are easy to use and intuitive.

The aim is to determine the average treatment effect on the outcome variable Y. The value of the outcome variable for individual i is assumed to be independent of the rest of individuals and the assignment mechanism to the treatment. If there are interactions among individuals or idiosyncratic effects, the changes of the outcome variable could be motivated by the treatment, the influence of other individual or other external effects. This condition devised by Rubin (1980), is known as stable unit treatment value assumption.

One of the most important parameters of interest is the average treatment effect on the treated (Heckman, Lalonde and Smith, 1999, and Blundell and Costa-Dias, 2002). This effect determines the average treatment value of the treatment group in the hypothetical case that the control group had also received the treatment:

\[E \left(Y _ {1} \mid D = 1\right) - E \left(Y _ {0} \mid D = 1\right)\]

Notwithstanding, the second term is unobserved. An alternative to overcome the problem is the Conditional Independence Assumption (CIA). CIA settles down that the assignment to the treatment or control group is independent of the potential values of the outcome variable, conditioning for the observed characteristics (X):

\[Y _ {1}, Y _ {0} \perp D | X\tag{1}\]

CIA is satisfied whether the outcome variable and the variables having influence over the selection process are used in the matching process. Lechner (2000) shows empirically the importance of a data base with enough information in order to satisfy the property. Frölich (2004) proves that including the variables which affect both to treatment choice and the outcome variable is also necessary to achieve consistent estimates. Considering CIA then,

\[E \left(Y _ {0} \mid D = 1, X\right) = E \left(Y _ {0} \mid D = 0, X\right)\]

Selection bias prevents the equality, since there would be some unobserved reason to justify the membership to the treatment group. The alternative to this bias is the use of matching techniques among elements of each group. Workers may be assumed to be similar enough to minimize the unobserved characteristics, although the condition is restrictive. A more suitable option is to accept that available characteristics are enough to capture unobserved information, at least partially.

Given the previous conditions, the treatment effect is defined as the difference of two observed terms:

\[E \left(Y _ {1} \mid D = 1, X\right) - E \left(Y _ {0} \mid D = 0, X\right)\]

An initial choice using this methodology is Exact Matching. Each treated unit is associated to a control unit with the same characteristics. This process presents a dimensionality problem. A solution is the use of matching through the propensity score:

\[P (X) = \operatorname * {P r} (D = 1 | X) = E (D | X)\tag{2}\]

Propensity score demands the Common Support Condition (CSC). CSC is satisfied when the generating data processes are random. If the treated workers have systematic differences with respect to the control workers, the propensity score could be equal to zero or one. This fact implies the existence of a relationship between the treatment group and the outcome variable, breaking condition (1).

According to Rosenbaum and Rubin (1983), there are two conditions to determine the average treatment effect using the propensity score. First,

\[Y _ {1}, Y _ {0} \perp D | P (X)\tag{3}\]

This assumption is satisfied whether condition (1) is accepted. The second requirement is known as the Balancing Property (BP). This condition postulates that the propensity score is a balancing score⁴,

\[X \perp D | P (X)\tag{4}\]

Considering conditions (3) and (4), the average treatment effect is equal to the difference:

\[E \left[ Y _ {1} \mid D = 1, P (X) \right] - E \left[ Y _ {0} \mid D = 0, P (X) \right]\]

The set of control units matched to the treated element i is defined as The four matching methods incorporated to the paper consider a different version of , using expression (2) as instrument. A common advantage of the methods consists on the lack of a parametric functional structure on the distribution of the outcome variable or about the conditional mean⁵.

The Stratification method splits the propensity score into Q blocks to the treatment and control groups. In each block , the treated and control workers show the same propensity score on average. Hence, C(i) is equivalent to . In each block, there are treated units and control units. The difference of the mean outcome variable is computed among treated and control units for each block:

\[\tau_ {q} ^ {S} = \frac {\sum_ {i \in I (q)} Y _ {i} ^ {1}}{N _ {q} ^ {1}} - \frac {\sum_ {j \in I (q)} Y _ {j} ^ {0}}{N _ {q} ^ {0}}\]

b(X)

X ⊥ D b ( X )

⁴The balancing score is defined by Rosenbaum and Rubin (1983) as a function b( X ) such that the distribution of X conditional on this function is the same for treatment and control group: .

⁵ Becker and Ichino (2002) and Heckman, Lalonde and Smith (1999) provide further information on matching methods.

The global effect is the weighted average of the differences:

\[\tau^ {S} = \sum_ {q = 1} ^ {Q} \tau_ {q} ^ {S} \frac {\sum_ {i \in I (q)} D _ {i}}{\sum_ {i} D _ {i}}\]

On condition that there is independence of the values of the outcome variable among units, the variance of the estimate is equal to:

\[\operatorname{Var} \left(\tau^ {S}\right) = \frac {1}{N ^ {1}} \left[ \operatorname{Var} \left(Y _ {i} ^ {1}\right) + \sum_ {q = 1} ^ {Q} \frac {\left(N _ {q} ^ {1}\right) ^ {2}}{N ^ {1} N _ {q} ^ {0}} \operatorname{Var} \left(Y _ {j} ^ {0}\right) \right]\]

However, this method eliminates the treated units as any control unit is not available in a block. With the Nearest Neighbour method, each treated unit is matched to the control unit whose propensity score is the nearest:

\[C (i) = \left\{j \left| \min _ {j} \left\| P _ {i} - P _ {j} \right\| \right. \right\}\]

This method calculates the mean value of the differences:

\[\tau^ {M} = \frac {1}{N ^ {1}} \sum_ {i \in T} Y _ {i} ^ {1} - \frac {1}{N ^ {1}} \sum_ {i \in T} \sum_ {j \in C (i)} \omega_ {i j} Y _ {j} ^ {0}\]

where if and zero otherwise, and is the number of control workers matched with treated worker . Assuming the weights are fixed and the values of the outcome variable are independent among observations, the variance is:

\[\operatorname{Var} \left(\tau^ {M}\right) = \frac {1}{\left(N ^ {1}\right) ^ {2}} \left[ \sum_ {i \in T} \operatorname{Var} \left(Y _ {i} ^ {1}\right) + \sum_ {j \in C} \left(\omega_ {j}\right) ^ {2} \operatorname{Var} \left(Y _ {j} ^ {0}\right) \right] = \frac {1}{N ^ {1}} \operatorname{Var} \left(Y _ {i} ^ {1}\right) + \frac {1}{\left(N ^ {1}\right) ^ {2}} \sum_ {j \in C} \left(\omega_ {j}\right) ^ {2} \operatorname{Var} \left(Y _ {j} ^ {0}\right)\]

This procedure is usually carried out with replacement, so a control unit can be matched to several treated units. The choice of replacement depends on the researcher. Although the advantages of replacement are obvious, this option also increases the variance of the treatment effect and only several control units can be used to be compared with the treated observations.

Matching always takes place, despite of the distance between the observations of the treatment and control groups, affecting matching quality. The Radius method and the Kernel method qualify the trade-off between matching quality and quantity in the estimates. The Radius method establishes a predetermined limit r which accepts or discards the matching:

\[C (i) = \left\{j \mid \left\| P _ {i} - P _ {j} \right\| < r \right\}\]

The treatment effect is estimated by the same formula as the Nearest Neighbour method.

The Kernel method considers the matching of each treated unit with a weighted mean of the control units. This weight is inversely proportional to the distance of the propensity score:

\[\tau^ {K} = \frac {1}{N ^ {1}} \sum_ {i \in T} \left\{Y _ {i} ^ {1} - \frac {\sum_ {j \in C (\cdot)} Y _ {j} ^ {0} G \left(\frac {P _ {j} - P _ {i}}{h _ {n}}\right)}{\sum_ {k \in C (\cdot)} G \left(\frac {P _ {k} - P _ {i}}{h _ {n}}\right)} \right\}\]

where is the kernel Gaussian function and is a bandwidth parameter. The standard errors are derived using bootstrapping.

Each method presents advantages and inconveniences for matching quantity and quality. Hence, none of the methods prevails over the others a priori.

4. Data

I use administrative data from two different sources: INEM and the outplacement firm Creade. Creade was created in 1988. The firm is founding member of the Spanish Association of Consulting Outplacement Firms (AECO) since 1993 and member of the Association of Career Management Consulting Firms International (AOCFI) since 1997. In 1992, Creade organized the first important group outplacement programme in Spain and coordinated the first outplacement programme for INEM in 2002. Creade’s data set presents personal characteristics of the candidates in the outplacement programmes. INEM’s data set includes workers whose contract has registered in the INEM offices in the region of Madrid. The latter data set shows daily information about contracts and monthly information about workers’ characteristics. The two data sets were merged to obtain an integrated data base. The time limits of the data bases correspond to the period between October 1998 and September 2003.

Given the large number of observations in INEM’s data set, an exhaustive homogeneity process is carried out regarding the treatment group. The workers who belong to special groups have been eliminated, because of personal, professional or social characteristics. Disabled workers, non-unemployed workers and workers who were carrying out training courses (or similar labour market policies) are removed. The age limits were established between 21 and 61 years.

Appendix A includes selected descriptive statistics for several samples, distinguishing between the region of Madrid and pooled data. Table A1 shows differences on average between treatment and control groups for the intermediate sample, constituted by 262,983 workers with complete information (494 individuals belong to the treatment group)⁷. The treatment group is usually made up of male workers with high education level and knowledge of other languages. Consequently, the proportion of workers specialized in academic qualification with high degree level (especially Medicine-Health, Economics and Management) is bigger. The cause of dismissal is focused on firm’s behaviour.

The proportion of workers whose residence belongs to Madrid is (obviously) higher in the control group. The sample reduction to the region of Madrid shows the same characteristics as the full sample, but the potential unobserved effects from geographical factors are eliminated.

With respect to types of outplacement services (Table A5), the workers who receive individual outplacement are usually men, middle-aged, with university degree (related to Economics and Management) and know other idioms. These characteristics usually define high level professionals. The workers who have participated in group outplacement programmes are (on average) middle-aged, although the proportion is smaller than the previous case. Education level and knowledge of other idioms are lower and their job dismissals are owed exclusively to the firm. This description usually corresponds to workers in the medium-low level of firms’ structure.

Given the information about the sample and identification strategy, two factors become critical in the evaluation process, the existence of a suitable outcome variable and the creation of a control group to be compared with a well-defined treatment group.

Unemployment spell is the outcome variable used in the paper. In order to define the days of unemployment period, the starting-point of unemployment situation for the treatment group will coincide with the date of the treatment beginning. The professionals who begin outplacement programme are supposed to be unemployed, because the company dispenses with their services. Unemployment spell finishes as the worker gets a new job. The date of contract beginning and the number of days of active job search will be used to determine the unemployment spell for controlled workers.

⁶ Heckman and Smith (1999) comment the importance of the workers’ labour market history on the estimates.

⁷ Further description about the creation of the intermediate sample is presented in Section 5.

Unemployment duration represents a fair measure to evaluate the effect of outplacement as workers and jobs present similar properties. With respect to the quality of the new job, wages are not available for the full sample. Nevertheless, there is information about other important features, like contract type and membership of an economic activity. Apart from personal characteristics of the workers, general economic information is introduced, such as macroeconomic context and dates of labour market legislation changes.

The initial date of the unemployment spell constitutes an important factor to limit unbiasedness of the estimates, because workers with similar characteristics at the same time are matched. The beginning of the unemployment situation coincides with the initial point of the service for treated individuals. Outplacement programmes are provided to the professional as the rupture with the firm is clear, so the probability of getting a new job is not affected by any effect in advance from the worker (Ashenfelter, 1978) . The date depends on the previous job for the control group. There are not difficulties whether the previous contract is temporary, because the ending of the contract coincides with the initial unemployment date. When the information is unknown, worker’s active job search period is used to determine unemployment spell.

Given the range of the period, macroeconomic variables are included in the estimation process because outplacement services can produce different effects according to the economic cycle, as Heckman, Ichimura and Todd point out (1997). Quarterly regional unemployment rate is used as economic situation index. Since transitions from unemployment to permanent employment are analyzed, labour market reforms and law modifications are also introduced to control their effects, as Kugler, Jimeno and Hernanz (2002) and Arellano (2005) evince.

Regarding the treatment group, the programmes provided by Creade are not supplemented jointly, so the effect of each programme can be analysed separately. The treatment is suspended for a period occasionally due to either holidays or other personal motivations. They are eliminated to avoid potential biased results.

⁸ For individual outplacement, this argument is weaker, but the final decision of using outplacement services should be taken by the firm which pays the programme. Hence, there are clear limitations of the Ashenfelter´s dip.

The creation of a suitable control group prevents several potential difficulties, like the existence of serious inconsistencies between groups, selfselection problems and other substitutive outplacement services in the control group. The solution to the first problem does not come from the worker's identification code, so personal characteristics are used to avoid this incompatibility. The control workers who share the same characteristics as any treated worker are eliminated (discarding around half a dozen observations).

Outplacement is a voluntary measure the company provides to the participant. Hence, the worker’s implication and full collaboration are required. This fact would suppose an important estimation bias (Westaby, 2004). The difficulty decreases on the assumption that workers did not know (or take into account) the existence of outplacement services as they incorporated to the firm, and they are also induced (for their own good) to accept outplacement programme after the dismissal.

Any variable does not show the existence of other services, so the treatment effect is certain whenever no controlled worker has received outplacement services. Some characteristics guarantee this fact, as the information about the worker’s labour market history. The workers who incorporate to the data base with a permanent contract or those who got a temporary contract previously do not receive outplacement services. These requirements are introduced into the control group, so the possible bias of the estimates is not especially outstanding with the available information.

The previous comments on transition from unemployment to permanent employment allow the creation of a subset of workers from the intermediate sample. The new sub-sample, defined as permanent employment sample, includes all treated workers with complete information (494 observations) and 44,096 controlled workers who get a new permanent job given the restrictions of their labour market history.

Comparing descriptive statistics of Table A1 and the part of Tables A2, A3 and A4 called “Permanent Employment” (representing the latter sample), a slight convergence of control group’s characteristics to treatment group’s ones is achieved, specially for women.

5. Who receives outplacement services? Bayesian approach

Derived from the information of the previous section, there are reasonable doubts about the random choice of observations of the treatment and control group. As commented previously, a potential criterion to homogenize the control group consists of job transitions. However, there is not any reason to exclude the rest of observable characteristics from the selection process. Taking into account all available information, other learning process about the treatment and control group takes place, calculating a new set of probabilities and modifying initial probabilities (Ledermann, 1984). The study of individual characteristics conditional on belonging to the treatment group fosters the discovery of distinctive features about workers who receive outplacement services.

Let be a combination of observable characteristics for a worker, and let D be the event of interest (the binary variable indicating outplacement status). D happens under any hypothesis , where countable, disjoint class of events with positive probability. Using Bayes’ theorem, the probability of having a combination of characteristics given that the worker belongs to the treatment group is:

\[P \left(A _ {k} \mid D = 1\right) = \frac {P (D = 1 \mid A _ {k}) \cdot P \left(A _ {k}\right)}{P (D = 1)}\]

The inference of elements of Bayes’ rule comes from the intermediate sample of workers described in the previous section. The denominator represents the proportion of treated workers in the sample. The probability of the combinations (or prior probability) is derived from the values of the variables which are used in the matching process. The conditional probability is estimated by a discrete choice model.

Initially, every observation represents a contract for a particular worker. But the event D refers to workers specifically, so the application of Bayes’ rule requires the use of workers as realizations in the partition of combinations, presenting a one-to-one relationship between observations and individuals.

The choice of a contract for any worker depends on job contract quality. A permanent contract is preferred to a temporary contract (excluding renewals or conversions of temporary contracts); the first temporary contract is selected in lack of permanent contracts.

Taking the intermediate sample as reference and posterior probability as instrument, a second subset of observations with the same size as the permanent employment sample (44,590 workers, including 494 treated workers) is created. The new sub-sample (known as posterior probability sample) includes controlled workers who show the highest posterior probability values.

Tables A2, A3 and A4 present the characteristics of the two final samples. The descriptive statistics of the treatment group coincide because both of them share the same group. The differences between the treatment and control groups in the posterior probability sample are smaller than those in the permanent employment sample. The use of the sample limited to the region of Madrid reasserts the previous comment.

The homogeneity level of the control group with respect to the treatment group in the posterior probability sample is not identical for all characteristics. Age and residence show similar figures in both samples. Women are underrepresented in the posterior probability sample, according to the treatment group. Average education level and knowledge of idioms approaches figures of the treatment group but the differences are still significant, especially for University Education and knowledge of English. The convergence is also incomplete for the position in last job and academic qualification, as the weights of managers, supervisors and Economics and Management show. The dissimilarities about the reason of the last dismissal become smaller. These properties considering only men and women in the sample are maintained, although similarities are slightly higher for women.

6. Results

Since treatment group is observed between the beginning of the outplacement programme and the date of the new job, the analysis of the results focuses on the short run.

The use of different samples to create a reliable homogeneity degree allows the comparison and analysis of multiple estimates. Given the comments of the previous section, two set of estimates are elaborated depending on the two final samples, permanent employment sample and posterior probability sample. As Gowan and Nassar-McMillan point out (2001), women usually accept different types of services than men, so distinction by gender is considered in the estimates. Because of the existence of two great outplacement programmes, individual and group outplacement services are also analyzed separately.

Appendix B presents the estimates of the mean increase of unemployment spell (in days) of the treatment group compared to the control group in the posterior probability sample. Appendix C shows the estimates of the treatment effect for the permanent employment sample.

Tables of Appendixes B and C present the following structure: estimates with their corresponding standard deviation and significance level for each matching method are shown in columns. The figures of the upper part correspond to the number of treated and controlled workers included in the estimation process, respectively. The last column incorporates information about BP. Given the comments on sample of the previous sections, the rows are divided into two groups depending on workers’ residence: the full sample and workers living only in the region of Madrid.

As commented in Section 3, BP is required to accept consistency of the estimates. Only the first moment condition of the property is analyzed in the paper, so this weaker version is necessary but not sufficient. Several options have been considered because of the difficulties of achieving BP. Option A incorporates all possible variables, although BP is not usually satisfied. The alternatives are focused on elimination of variables⁹, decline in the significance level¹⁰ and sample reduction (using gender, outplacement services and geographical information). The estimates of the models which do not satisfy BP constitute only a reference regarding the rest of options. Options B satisfy BP at different significance levels with the highest number of groups of variables. Option C incorporates all variables but the sample is reduced to the workers of the region of Madrid. This model does not usually satisfy BP either. Options D use the same sample reduction as Option C and they are created using the same methodology as Options B.

The more homogeneous the sample shows, the easier the fulfilment of BP turns out. Moreover, there is an inverse relationship between the number of observations involved in the estimation process and the result of BP. The basic fulfilment process of BP works in both the permanent employment sample and the posterior probability sample similarly, and the posterior probability sample demands a greater deal of flexibility to satisfy the property as the sample size does not incorporate any constraint.

Qualitative results remain unchanged, but important quantitative differences among options using the two final samples arise, although the restricted sample of Madrid shows smaller estimates than those from the full sample. By construction, the Nearest Neighbour and Stratification matching methods are more sensitive to changes of the number of observations in the control and treatment groups, which are more likely as both men and women are included in the sample. This fact justifies the differences of estimates and significance levels among options and samples in the tables. Only the Radius method shows robust estimates among options.

⁹ Given the important number of variables used in estimates, the elimination process is produced with groups of dummy variables which come from the same original variable.

¹⁰ Other two weaker significance levels are used apart from the standard value equal to 0.01, 0.005 and 0.001.

The discrepancy becomes evident between the Radius method and the rest of methods as the estimates are positive, high and significant (in the full sample), such that the former produces the smallest estimates. With respect to the latter ones, the Nearest Neighbour method presents the highest values frequently and the Kernel method generates smaller values especially for the sample restricted to the region of Madrid.

Tables of Appendix B suggest that outplacement services enlarge unemployment spell. For the pooled data, the period is near to three months except for the Radius method, whose estimates are reduced to one month regardless of worker’s residence (Table B1). The increase of unemployment duration is also reduced by one month for the sample of Madrid. The distinction between individual and group outplacement (Tables B4 and B7) does not show relevant discrepancies in the full sample, although group outplacement estimates are more similar among options. The figures indicate an increase of the unemployment spell around eighty days, except for the Radius method (the period increases only around one month). In the case of Madrid, group outplacement increases unemployment spell between one and two months (below a month for the Radius method) while the effect of individual outplacement is analogous to the pooled data set and assessed at a period between two and three months.

The distinction by gender shows particular features. Using the full sample, outplacement programmes enlarge significantly unemployment spell for men. Although the results are not robust to changes in options regardless of the estimation method, qualitative results are not affected (except for the Nearest Neighbour method). Apart from the Radius method, the estimates in the full sample suggest that general outplacement (Table B2) has risen unemployment spell for men by three months. The increase is around four and two months for individual and group outplacement, respectively (Tables B5 and B8). The figures halve for the Radius method.

The estimates for men using the sample of Madrid do not show high variability among options for general treatment and individual outplacement, and share qualitative properties for group outplacement as occurs in the pooled data set. The estimates confirm the negative effect of outplacement on unemployment duration although the quantities are slightly smaller than those derived from the full sample. In the case of group outplacement, both the Kernel and Radius methods present an increase of unemployment spell by a month.

With respect to women, the figures of BP suggest an outstanding homogeneity degree especially for individual outplacement, regardless of the number of observations. The options of general treatment (Table B3) show a great variability of estimates. The results of the full sample are not usually significant, except for the Radius method, indicating the small relevance of outplacement on women’s unemployment duration. Only the Nearest Neighbour and Stratification methods present significant positive values in the restricted sample. The Kernel and Radius methods, whose estimates are robust to changes of options, suggest a negligible effect. The negative estimates of individual outplacement (Table B6) are not significant, so only group outplacement (Table B9) for the full sample seems to affect women, increasing unemployment spell by two months (one month for the Radius method) in the pooled data set. The results for the sample of Madrid confirm the irrelevant role of group and individual outplacement on women’s unemployment duration.

The estimates using the permanent employment sample (Appendix C) are lower than those presented in Appendix B, but the characteristics of the figures by matching methods and options remain constant. The estimates derived from the full sample show higher dissimilarities between the permanent employment sample and the posterior probability sample than those from the region of Madrid. The average difference among estimates of the two samples in the pooled data set is around three weeks for general treatment (Table C1) and (at most) two weeks for individual outplacement (Table C4). The difference for the restricted data set is not usually above ten days. This property is outstanding in the case of group outplacement (Table C7), because the average difference amounts to two months and two weeks for the pooled and restricted data set, respectively. The same behaviour is observed for women (Table C9), while men (Table C8) show the opposite version of this relationship. The Radius method constitutes an exception as both samples are compared, because its estimates in the posterior probability sample at least double the values obtained in the permanent employment sample.

The tables focused on women (Tables C3, C6 and C9) present better results than male partners. There are negative estimates in many cases (around half a month) although the figures are not statistically significant. As the sample is restricted to the region of Madrid, the results improve and confirm the characteristics of outplacement services, since group outplacement show better figures than individual outplacement.

7. Concluding remarks

Outplacement is defined as the process of placing employees in other positions once they have been separated from a job. The aim of the paper is to analyze the effect of outplacement services provided to workers by the outplacement firm Creade on unemployment spell.

Outplacement services belong to the measures Private Sector has developed to improve access of unemployed workers to the active labour market. A particular case is the group of older workers, because outplacement programmes constitute an effective and constructive tool to prevent early retirements. Outplacement has begun to represent an outstanding solution in the Spanish labour market for the last years.

The study of unemployment spell constitutes a fundamental factor in the analysis of outplacement services for policy-makers, given the importance of unemployment benefits on State Budget. Using unemployment duration as outcome variable, the estimates suggest that outplacement services do not generate positive effects except for women in a limited way. A potential explanation is the “reservation wage” effect. Unemployed workers face up to a trade-off between unemployment spell and job quality. The workers who receive outplacement services are more demanding in the job selection process. Help and advice of the consultant qualifies for a more suitable job. Hence, the unemployment period tends to be higher, sacrificing potential immediate future wage in order to get a job more appropriated to unemployed worker’s characteristics.

Therefore, wage is a measure of employment quality and constitutes an important variable in the outplacement process. An approximation to the monetary effect of outplacement services is evaluated using workers’ mean wage. The information about labour costs from the Spanish Department of Statistics (INE) indicates that the mean wage of a worker in the third quarter of 2005 was 1,489.74 Euros in Spain, and 1,783.11 Euros in the region of Madrid. Considering the best results (the reduction of women’s unemployment spell is around two weeks), the mean profit per worker will be around 744.87 Euros in Spain and 891.56 Euros in the region of Madrid. In the case of men, an average increase of the unemployment period by eighty days means losses quantified between 3,907.51 and 4,677 Euros, respectively.

In spite of the previous comments, the paper deals with a great part of the effects outplacement programmes generate on unemployment process. Heterogeneity of outplacement services in terms of measures and time dedicated to the candidate could be taken into account, as Westaby indicates (2004). Apart from the distinction between individual and group outplacement, future research should be focus on duration of outplacement services and differentiation among several age groups, since young workers are more confident of their future and psychological help is less important for them than older workers, with higher family and financial responsibilities.

References

1. Aquilanti, T. M. and J. Leroux (1999), “An integrated model of outplacement counselling”, Journal of Employment Counselling, 36, 177-192.
2. Arellano, A. (2005), “Evaluating the effects of labour market reforms “at the margin” on unemployment and employment stability: the Spanish case”, Working Paper 05-12, February 2005, Universidad Carlos III de Madrid.
3. Ashenfelter, O. (1978), “Estimating the effect of training programs on earnings”, Review of Economic Studies, 60, 47-57.
4. Becker, S. O. and A. Ichino (2002), “Estimation of average treatment effects based on propensity scores”, The Stata Journal, 2, 358-377.
5. Blundell, R. and M. Costa-Dias (2002), “Alternative approaches to evaluation in empirical microeconomics”, Portuguese Economic Journal, 1, 91-115.
6. Cowden, P. (1992), “Outplacement services assessment”, HRMagazine, 37, 69- 70.
7. Creade (2003), available in http://www.e-creade.com/esp/Conocenos/igest.asp.
8. Davy J. A., J. S. Anderson and N. DiMarco (1995), “Outplacement comparisons of formal outplacement services and informal support”, Human Resource Development Quarterly, 6, 275-288.
9. De Ramos, M. F. and C. Hernández (1999), Outplacement: Principios de éxito y reorientación laboral, Griker.
10. De Ramos, M. F. and C. Hernández (2000), “Tipología y metodología de los programas de outplacement”, Capital Humano, 133, 40-44.
11. Frölich, M. (2004), “Programme evaluation with multiple treatments”, Journal of Economic Surveys, 18, 181-224.
12. Furness, W. M. and W. T. Lewis (1996), “Un ejemplo de regeneración económica regional: la experiencia de British Coal Enterprise”, Economía Industrial, 309, 145-150.
13. Gowan, M. A. and S. C. Nassar-McMillan (2001), “Examination of individual differences in participation in outplacement program activities after a job loss”, Journal of Employment Counselling, 38, 185-196.
14. Grupo MOA (2000), “El outplacement: La perspectiva de los candidatos y las empresas contratantes”, Capital Humano, 133, 46-50, May 2000.
15. Heckman, J. J., H. Ichimura and P. Todd (1997), “Matching as an econometric evaluation estimator: evidence from evaluating a job training program”, Review of Economic Studies, 64, 605-654.
16. Heckman, J. J., R. J. Lalonde and J. A. Smith (1999), “The Economics and Econometrics of Active Labour Market Programs”, in Ashenfelter A. and D. Card (eds.) Handbook of Labour Economics, vol. 3, Elsevier, Amsterdam.
17. Heckman, J. J. and J. A. Smith (1999), “The pre-program earnings dip and the determinants of participation in a social program: implications for simple program evaluation strategies”, Economic Journal, 109, 313-348.
18. Hortal, J. P. (1999), “Pero… ¿qué es y en qué consiste el Outplacement?”, Revista Infor25, February, 1999, Arcocreade.
19. Jacquier, P. (1996), “Procesos de Reconversión y desarrollo de las PYMES: La experiencia de Sodie (Grupo Usinor Sacilor)”, Economía Industrial, 309, 113-116.
20. Jiménez, A. (2000), “Un caso práctico de outplacement: La integración laboral de ex deportistas de élite”, Capital Humano, 133, 52-54, May 2000.
21. Kugler, A., J. F. Jimeno and V. Hernanz (2002), “Employment consequences of restrictive permanent contracts: evidence from Spanish labour market reforms”, IZA Discussion Paper, 657.
22. Latack, J. C. and J. B. Dozier (1986), “After the ax falls: Job loss as a career transition”, Academy of Management Review, 11, 375-392.
23. Lechner, M. (2000), “Some practical issues in the evaluation of heterogeneous labour market programmes by matching methods”, Discussion Paper Nº 2000-14, Department of Economics, University of St. Gallen.
24. Ledermann, W. (1984), Handbook of applicable mathematics, vol. 6, Part B, John Wiley & Sons, New York.
25. Meyer, J. L. and C. C. Shadle (1994), The changing outplacement process, Quorum Books, Westport.
26. Mendels, P. (2001), “The best of times for outplacement firms”, Business Week, 8, May 2001.
27. Rosenbaum, P. R. and D. R. Rubin (1983), “The central role of the propensity score in observational studies for causal effects”, Biometrika, 70, 41-55.
28. Roy, A. D. (1951), “Some thoughts on the distribution of earnings”, Oxford Economic Papers, 3, 135-146.
29. Rubin, D. B. (1974), “Estimating causal effects of treatments in randomized and nonrandomized studies”, Journal of Educational Psychology, 66, 688-701.
30. Rubin, D. B. (1980), “Comment on ‘Randomization analysis of experimental data: the Fisher randomization test’ by D. Basu”, Journal of American Statistical Association, 75, 591-593.
31. Sáenz, M. T. (2000), “Outplacement: Una renovación para el futuro que consolida el presente”, Capital Humano, 133, 28-38, May 2000.
32. Westaby, J. D. (2004), “The impact of outplacement programs on reemployment criteria: a longitudinal study of displaced managers and executives”, Journal of Employment Counselling, 41, 19-28.
33. Wolfer, K. and R. G. Wong (1988), The outplacement solution: getting the right job after mergers, takeovers, layoffs, and other corporate chaos, John Wiley & Sons, New York.

Appendix A Descriptive Statistics

Table A 1 Descriptive statistics for the intermediate sample

Notes : The table reports averages and percenta ppropriate. No Education includes any kind ofges for the indicated group . Standard deviations are in parenthesis where a education which does not satisfy Primary education Technical College (TC) education is divided into three levels (B asic Medium and Superior TC) and University Education incorporates Lower degree and Higher degree The most important options for Idioms Residence Position in Last Job Reasons of the Last Dismissal and Academic Qualification are included in the table Position in Last Job follows the National Classification of Occupation (CNO-94) and Academic Qualification follows the National Classification of Economic Activities (CNAE)

Table A2: Descriptive statistics for homogeneous samples

Table A3: Descriptive statistics for men

Notes: The table reports averages and percentages for the indicated group. Standard deviations are in parenthesis where appropriate. No Education includes any kind of education which does not satisfy Primary Education. Technical College (TC) education is divided into three levels (Basic, Medium and Superior TC), and University Education incorporates Lower degree and Higher degree. The most important options for Idioms, Residence, Position in Last Job, Reasons of the Last Dismissal and Academic Qualification are included in the table. Position in Last Job follows the National Classification of Occupations (CNO-94) and Academic Qualification follows the National Classification of Economic Activities (CNAE).

Table A4: Descriptiv statistics for womene

Table A5: Descriptive statistics for individual and group outplacement

Notes: The table reports averages and percentages for the indicated group. Standard deviations are in parenthesis where appropriate. No Education includes any kind of education which does not satisfy Primary Education. Technical College (TC) education is divided into three levels (Basic, Medium and Superior TC), and University Education incorporates Lower degree and Higher degree. The most important options for Idioms, Residence, Position in Last Job, Reasons of the Last Dismissal and Academic Qualification are included in the table. Position in Last Job follows the National Classification of Occupations (CNO-94) and Academic Qualification follows the National Classification of Economic Activities (CNAE).

Appendix B: Estimates using the posterior probability sample

Table B1: General treatment - Full sample

NOTES:

Option A: All variables (age, gender, education level, professional level, academic qualification, knowledge of other idioms, reasons of the last dismissal, week, quarter and year of beginning of the unemployment spell, quarterly regional unemployment rate, worker's province and date of the labour market reforms) are included with the full sample.

Option B1: Option A excluding knowledge of other idioms and date of the labour market reforms.

Option B2: Option A eliminating knowledge of other idioms and reasons of the last dismissal.

Option B3: Option A without professional level and reasons of the last dismissal.

Option C: All variables are included with the sample restricted to the region of Madrid.

Option D1: Option C except week of beginning of the unemployment spell.

ption D2: OO ption C ruling out knowledge of other idioms.

ption D3: Option C omitting professional level.O

* significant at 10%, ** significant at 5%, *** significant at 1%.

Table B2: General treatment - Only men

NOTES:

Option A: All variables (age, education level, professional level, academic qualification, knowledge of other idioms, reasons of the last dismissal, week, quarter and year of beginning of the unemployment spell, quarterly regional unemployment rate, worker's province and date of the labour market reforms) are included with the full sample.

Option B1: Option A excluding reasons of the last dismissal.

Option B2: Option A eliminating age.

Option B3: Option A without quarter of beginning of the unemployment spell and quarterly regional unemployment rate.

Option C: All variables are included with the sample restricted to the region of Madrid.

Option D1: Option C except week of beginning of the unemployment spell.

Option D2: Option C ruling out academic qualification.

Option D3: Option C omitting education level and academic qualification.

nt at 1%.* significant at 10%, ** significant at 5%, *** significa

Table B3: General treatment - Only women

NOTES:

vel, professional level, academic qualification, knowledge of other idioms, reasons of the lastOption A: All variables (age, education le week, quarter and year of beginning of the unemployment spell, quarterly regional unemployment rate, worker's province and datedismissal, of the labour market reforms) are included with the full sample.

Option B1: Option A excluding professional level and reasons of the last dismissal.

Option B2: Option A eliminating professional level and knowledge of other idioms.

regional unemployment rate.Option B3: Option A without week and year of beginning of the unemployment spell, and quarterly

Option C: All variables are included with the sample restricted to the region of Madrid.

Option D1: Option C except date of the labour market reforms.

Option D2: Option C ruling out quarterly regional unemployment rate.

Option D3: Option C omitting week of beginning of the unemployment spell.

* significant at 10%, ** significant at 5%, *** significant at 1%.

Table B4: Individual treatment – Full sample

NOTES: Option A: All variables (age, gender, education level, professional level, academic qualification, knowledge of other idioms, reasons of the last dismissal, week, quarter and year of beginning of the unemployment spell, quarterly regional unemployment rate, worker's province and date of the labour market reforms) are included with the full sample. Option B1: Option A excluding academic qualification and date of the labour market reforms. professional level and quarter of beginning of the unemployment spell.Option B2: Option A eliminating age, Option B3: Option A without gender, professional level and academic qualification. Option C: All variables are included with the sample restricted to the region of Madrid. Option D1: Option C except year of beginning of the unemployment spell. ell.Option D2: Option C ruling out quarter of beginning of the unemployment sp Option D3: Option C omitting academic qualification. * significant at 10%, ** significant at 5%, *** significant at 1%.

Table B5: Individual treatment – Only men

NOTES:

of other idioms, reasons of the lastOption A: All variables (age, education level, professional level, academic qualification, knowledge ent spell, quarterly regional unemployment rate, worker's province and datedismissal, week, quarter and year of beginning of the unemploym of the labour market reforms) are included with the full sample.

ployment rate.Option B1: Option A excluding professional level and quarterly regional unem

ployment spell and quarterly regional unemployment rate.Option B2: Option A eliminating quarter of beginning of the unem

ption B3: Option A without professional level and worker's province.O

ption C: All variables are included with the sample restricted to the region of Madrid.O

ption D1: Option C except year of beginning of the unemployment spell.O

Option D2: Option C ruling out academic qualification.

ption D3: Option C omitting ageO .

significant at 10%, ** significant at 5%, *** significant at 1%.*

Table B6: Individual treatment – Only women

NOTES:

Option A: All variables (age, education level, professional level, academic qualification, knowledge of other idioms, reasons of the last dismissal , week, quarter and year of beginning of the unemployment spell, quarterly regional unemployment rate, worker's province and date of the labour market reforms) are included with the full sample.

ant at 10%, ** significant at 5%, *** significant at 1%.* signific

iables are included with the sample restricted to the region of Madrid.Option C: All possible var

Table B7: Group treatment – Full sample

NOTES:

Option A: All variables (age, gender, education level, professional level, academic qualification, knowledge of other idioms, week, quarter and year of beginning of the unemployment spell, quarterly regional unemployment rate, worker's province and date of the labour marke reforms) are included with the full sample.

Option B1: Option A excluding quarter of beginning of the unemployment spell.

ption B2: Option A eliminating year of beginning of the unemployment spell.O

: Option A without knowledge of other idioms.Option B3

Option C: All variables are included with the sample restricted to the region of Madrid.

Option D1: Option C except knowledge of other idioms and quarter of beginning of the unemployment spell.

lification and quarterly regional unemployment rate.Option D2: Option C ruling out professional level, academic qua

ployment spell and date of the labour marketOption D3: Option C omitting knowledge of other idioms, quarter of beginning of the unem reforms.

* significant at 10%, ** significant at 5%, *** significant at 1%.

ent – Only menTable B8: Group treatm

NOTES:

ption A: All variables (age, education level, professional level, academic qualification, knowledge of other idioms, week, quarter and yearO beginning of the unemployment spell, quarterly regional unemployment rate, worker's province and date of the labour market reforms) areof cluded with the full sample.in

ption B1: Option A excluding worker's province and quarterly regional unemployment rate.O

ption B2: Option A eliminating professional level and worker's province.O

ption B3: Option A without professional level and quarterly regional unemployment rate.O

ption C: All variables are included with the sample restricted to the region of Madrid.O

ption D1: Option C except professional level and date of the labour market reforms.O

ption D2: Option C ruling out professional level and academic qualification.O

ption D3: Option C omitting academic qualification and quarter of beginning of the unemployment spell.O

significant at 10%, ** significant at 5%, *** significant at 1%.*

Table B9: Group treatment – Only women

NOTES:

Option A: All variables (age, education level, professional level, academic qualification, knowledge of other idioms, week, quarter and year unemployment rate, worker's province and date of the labour market reforms) areof beginning of the unemployment spell, quarterly regional included with the full sample.

Option B1: Option A excluding quarter of beginning of the unemployment spell.

Option B2: Option A eliminating knowledge of other idioms and quarter of beginning of the unemployment spell.

Option B3: Option A without professional level and quarter of beginning of the unemployment spell.

Option C: All variables are included with the sample restricted to the region of Madrid.

Option D1: Option C except date of the labour market reforms.

Option D2: Option C ruling out academic qualification.

ption D3: Option C omitting quarterly regional uneO mployment rate.

References

1. * significant at 10%, ** significant at 5%, *** significant at 1%.

Appendix C: Estimates using the permanent contract sample

Table C1: General treatment - Full sample

NOTES:

ption A: All variables (age, gender, education level, professional level, academic qualification, knowledge of other idioms, reasons oO f the week, quarter and year of beginning of the unemployment spell, quarterly regional unemployment rate, worker's province andlast dismissal, date of the labour market reforms) are included with the full sample.

Option B1: Option A excluding gender and year of beginning of the unemployment spell.

Option B2: Option A eliminating professional level, week of beginning of the unemployment spell and date of the labour market reforms.

pell and date of the labour market reforms.Option B3: Option A without gender, quarter of beginning of the unemployment s

Option C: All variables are included with the sample restricted to the region of Madrid.

Option D1: Option C except year of beginning of the unemployment spell.

Option D2: Option C ruling out knowledge of other idioms.

Option D3: Option C omitting professional level.

* significant at 10%, ** significant at 5%, *** significant at 1%.

l treatment - Only menTable C2: Genera

NOTES: Option A: All variables (age, education level, professional level, academic qualification, knowledge of other idioms, reasons of the last dismissal, week, quarter and year of beginning of the unemployment spell, quarterly regional unemployment rate, worker's province and date of the labour market reforms) are included with the full sample. Option B1: Option A excluding date of the labour market reforms. Option B2: Option A eliminating age. Option B3: Option A without professional level and knowledge of other idioms. Option C: All variables are included with the sample restricted to the region of Madrid. Option D1: Option C except quarter of beginning of the unemployment spell. Option D2: Option C ruling out reasons of the last dismissal. Option D3: Option C omitting professional level. * significant at 10%, ** significant at 5%, *** significant at 1%.

Table C3: General treatment - Only women

NOTES:

Option A: All variables (age, education level, professional level, academic qualification, knowledge of other idioms, reasons of the last dismissal, week, quarter and year of beginning of the unemployment spell, quarterly regional unemployment rate, worker's province and date of the labour market reforms) are included with the full sample. of the labour market reforms) are included with the full sample

Option B1: Option A excluding quarter of beginning of the unemployment spell.

Option B2: Option A eliminating professional level.

Option B3: Option A without week of beginning of the unemployment spell.

Option C: All variables are included with the sample restricted to the region of Madrid.

Option D1: Option C except reasons of the last dismissal.

Option D2: Option C ruling out knowledge of other idioms.

Option D3: Option C omitting academic qualification.

* significant at 10%, ** significant at 5%, *** significant at 1%.

Table C4: Individual treatment – Full sample

NOTES:

Option A: All variables (age, gender, education level, professional level, academic qualification, knowledge of other idioms, reasons of the last dismissal, week, quarter and year of beginning of the unemployment spell, quarterly regional unemployment rate, worker's province and .date of the labour market reforms) are included with the full sample date of the labour market reforms) are included with the full sample

of beginning of the unemployment spell.Option B1: Option A excluding quarter

Option B2: Option A eliminating academic qualification.

Option B3: Option A without education level.

Madrid.Option C: All variables are included with the sample restricted to the region of

oyment spell.Option D1: Option C except week of beginning of the unempl

dismissal.Option D2: Option C ruling out reasons of the last

Option D3: Option C omitting academic qualification.

* significant at 10%, ** significant at 5%, *** significant at 1%.

Table C5: Individual treatment – Only men

NOTES:

Option A: All variables (age, education level, professional level, academic qualification, knowledge of other idioms, reasons of the last arterly regional unemployment rate, worker's province and datedismissal, week, quarter and year of beginning of the unemployment spell, qu of the labour market reforms) are included with the full sample.

Option B1: Option A excluding reasons of the last dismissal.

Option B2: Option A eliminating academic qualification.

d worker's province.Option B3: Option A without reasons of the last dismissal an

the region of Madrid.Option C: All variables are included with the sample restricted to

Option D1: Option C except date of the labour market reforms.

ption D2: Option C ruling out year of beginO ning of the unemployment spell.

Option D3: Option C omitting reasons of the last dismissal.

significant at 10%, ** significant at 5%,* *** significant at 1%.

Table C6: Individual treatment – Only women

NOTES:

Option A: All variables (age, education level, professional level, academic qualification, knowledge of other idioms, reasons of the last dismissal, week, quarter and year of beginning of the unemployment spell, quarterly regional unemployment rate, worker's province and date of the labour market reforms) are included with the full sample.

Option C: All possible variables are included with the sample restricted to the region of Madrid.

* significant at 10%, ** significant at 5%, *** significant at 1%.

Table C7: Group treatment – Full sample

NOTES:

Option A: All variables (age, gender, education level, professional level, academic qualification, knowledge of other idioms, week, quarte ear of beginning of the unemployment spell, quarterly regional unemployment rate, worker's province and date of the labour markand y et ed with the full sample.reforms) are includ

Option B1: Option A excluding quarterly regional unemployment rate.

Option B2: Option A eliminating quarter of beginning of the unemployment spell.

Option B3: Option A without knowledge of other idioms.

the region of Madrid.Option C: All variables are included with the sample restricted to

ualification.Option D1: Option C except professional level and academic q

Option D2: Option C ruling out education level and professional level.

Option D3: Option C omitting gender and professional level.

* significant at 10%, ** significant at 5%, *** significant at 1%.

oup treatment – Only menTable C8: Gr

OTES:N

ption A: All variables (age, education level, professional level, academic qualification, knowledge of other idioms, week, quarter and yearO beginning of the unemployment spell, quarterly regional unemployment rate, worker's province and date of the labour market reforms) areof cluded with the full samplein

ption B1: Option A excluding week of beginning of the unemployment spell.O

ption B2: Option A eliminating knowledge of other idioms and year of beginning of the unemployment spell.O

ption B3: Option A without professional level and quarterly regional unemployment rate.O

ption C: All variables are included with the sample restricted to the region of Madrid.O

ption D1: Option C except week of beginning of the unemployment spell.O

ption D2: Option C ruling out academic qualification.O

ption D3: Option C omitting knowledge of other idioms and week of beginning of the unemployment spell.O

significant at 10%, ** significant at 5%, *** significant at 1%.*

Table C9: Group treatment – Only women

NOTES: Option A: All variables (age, education level, professional level, academic qualification, knowledge of other idioms, week, quarter and year of beginning of the unemployment spell, quarterly regional unemploym ker's province and date of the labour market reforms) areent rate, wor included with the full sample. Option B1: Option A excluding quarterly regional unemployment rate. Option B2: Option A eliminating professional level. Option B3: Option A without date of the labour market reforms. ion of Madrid.Option C: All variables are included with the sample restricted to the reg Option D1: Option C except professional level. loyment spell.Option D2: Option C ruling out quarter of beginning of the unemp ption D3: Option C omitting education level.O * significant at 10%, ** significant at 5%, *** significant at 1%.