Amynah Gangji, Université Libre de Bruxelles, Department of Applied Economics (DULBEA), Brussels, Belgium
Robert Plasman, Université Libre de Bruxelles, Department of Applied Economics (DULBEA), Brussels, Belgium
The authors thank Khalid Sekkat, Ilan Tojerow and participants of the DULBEA-ETE internal seminar for helpful comments and discussions.
Purpose – The purpose of this paper is to investigate the causes of unemployment persistence among the Belgian labour force. The underlying issue is to determine the eventual existence of a true causal relationship between successive unemployment spells.
Design/methodology/approach – The model used is a dynamic random effects probit model controlling for unobserved heterogeneity and the initial condition problem. It was applied to the Panel Study on Belgian Households (1994-2002).
Findings – The results suggest that while observed and unobserved heterogeneity explain between 57 per cent and 82 per cent of unemployment persistence, the remainder is induced by the presence of state dependence. All else being equal, an individual unemployed this year will be between 11.4 and 33 percentage points more likely to be unemployed next year as compared with an employed person.
Practical implications – The presence of a stigmatisation effect of unemployment means that the costs of unemployment are much higher than the simple loss of income and human capital associated with the current job loss. The study demonstrates the importance of concentrating efforts on the prevention of unemployment.
Originality/value – The paper's contribution is to test again the hypothesis of the presence of state dependence in unemployment using a different technique, allowing, among other things, to control for exogenous variables. The paper demonstrates its existence and measures its contribution in the explanation of unemployment persistence in Belgium, besides that of observed and unobserved characteristics.
Unemployment; Mathematical modelling; Belgium; Labour; Social welfare.
International Journal of Manpower
Emerald Group Publishing Limited
Since the beginning of the 1970s, following the oil crisis, most European countries have faced high and persistent unemployment rates. As a small, open economy, Belgium was not exempt. Many economists have tried to better understand the causes of this phenomenon in order to define adequate policies to curb it. Some have advanced the idea that there could be a persistence phenomenon by which the unemployment rate today would be related to its past achievements. Nowadays, unemployment persistence in Europe is a well-known fact. If, at the beginning, the tools used to study this phenomenon referred primarily to macroeconomic fundamentals, some economists thereafter were rather interested in individual behaviours. They wanted to determine how past unemployment experiences influence future individual labour market prospects. If unemployment persistence is actually present, it can come either from differences in characteristics, observable or not, between individuals influencing their propensity to experience unemployment spells, or from a true causal relationship between past and present unemployment spells. The second possibility is called state dependence of unemployment.
The purpose of this paper is to analyse the determinants of unemployment persistence in terms of its occurrence at the individual level within the Belgian labour force. Among other things we will measure how a previous unemployment experience increases the probability of an individual being unemployed today. The model used is a dynamic random effects probit model controlling for unobserved heterogeneity and the initial condition problem. It is applied to the Panel Study on Belgian Households (PSBH), covering the years 1994 to 2002.
A better understanding the causes of persistence in the occurrence of unemployment could be very useful for the implementation of appropriate policies against unemployment. Indeed, their success is largely conditional on to what extent unemployment incidence in itself has a damaging effect on future labour prospects and to what extent unemployment incidence depends on unfavourable individual characteristics. If there is existence of state dependence in unemployment persistence, unemployment costs are much higher than the current loss of wage. There is therefore a need to include these future effects in any costs assessments. The presence of a stigmatisation effect also highlights the need for active labour market policies in order to decrease the long-term unemployment rate. Finally, in that case, the prevention of an initial unemployment experience becomes an important policy objective, indicating the need to focus on education and training.
The remainder of the paper is organised as follows. The next section presents the theoretical framework explaining the existence of state dependence as well as empirical results. The sample design and the data are described in the third section. The fourth section develops the dynamic panel data model allowing for the introduction of state dependence. The main empirical results of the determinants of unemployment incidence are presented in the fifth section. The sixth section displays the marginal effects of state dependence on unemployment persistence. The final section concludes.
Theoretical framework and some empirical results
Heckman and Borjas (1980) were amongst the first to model the impact of a former unemployment experience on future individual behaviours in the labour market. Following this work a vast literature on the subject has been developed, particularly regarding the analysis and the measurement of duration and occurrence dependences. Generally, most economists agree on the fact that unemployment impairs an individual's future employment prospects: observations suggest that an individual who has recently experienced an unemployment spell is more likely to be also observed unemployed in the near future than someone who has never been unemployed. This observed correlation between successive unemployment spells at the individual level is explained in the economic literature either because of the presence of a structural state dependence or because of underlying characteristics making individuals more vulnerable to unemployment.
True state dependence can be defined as “a genuine behavioural effect in the sense that an otherwise identical individual who did not experience unemployment would behave differently in the future than an individual who experienced unemployment” (Heckman and Borjas, 1980).
Several explanations can be found in the economic literature that clarify the causal relationship existing between successive unemployment spells (see, for example, Arulampalam, 2002; Arulampalam et al., 1999; Corcoran and Hill, 1985; Narendranathan and Elias, 1993; Heckman and Borjas, 1980).
They are mostly based on the human capital and segmented market theories. A first justification for the existence of a stigmatisation effect of unemployment is on the firms' side. Indeed, during recruiting procedures, firms are not able to observe the worker's future productivity. Therefore, they will use various job criteria to sort applicants. In particular, they can judge them through their past history in the labour market. In that case, unemployment incidence can be used as an indicator of lower productivity or less reliability if firms place sufficient confidence in this information to make their recruiting decision. Consequently, the unemployed will be systematically worried when applying (Lockwood, 1991; Heckman and Borjas, 1980; Hämäläinen, 2003; Taylor, 2002). Another firm practice that may create a causal relation between consecutive unemployment spells is the “last in first out” procedure (Narendranathan and Elias, 1993).
A second set of justifications lies on the supply side. Some authors argue that unemployment may modify the characteristics or the behaviour of individuals, which in turn will influence their labour market status (Heckman and Borjas, 1980). For instance, according to the human capital theory, unemployment prevents the accumulation of human capital within a firm but also implies a deterioration of more general skills and knowledge with, as a consequence, a degradation of future wages and prospects in the labour market. Because of this more precarious situation, the individual could rapidly return to unemployment (Heckman and Borjas, 1980). Another example can be found in the segmented market theory. Indeed, it postulates that unemployment may induce individuals to develop a greater attraction for leisure or less assiduity to work. Therefore, it could involve the exclusion of the primary market, with stable and better remunerated jobs. Consequently, unemployed workers are especially confronted with the secondary market, with shorter periods of employment followed by increasingly longer unemployment spells (Lollivier, 2000). Finally, after long unemployment spells, the unemployed might be discouraged and finally accept weaker wages than initially hoped for or a poorer quality job, implying a loss of qualification and experience and thus increasing the probability to experience unemployment again.
Even though theoretical models suggest a causal link from past to future unemployment, it is not straightforward to establish this in empirical works since some statistical artefacts induce spurious state dependence.
Indeed, the observation of unemployment persistence could be explained by individual characteristics affecting the arrival of job offers or job-retention rate, and consequently influencing the propensity to be unemployed. Those characteristics, observable or not, could lead to spurious correlation if they are not taken into account correctly. As far as the observed characteristics are concerned, they are generally inserted in the model as control variables. However, it may be that these differences are unobservable for the analyst (punctuality or responsibility, for example). If the unmeasured differences among individuals remain uncontrolled and are correlated over time, a previous unemployment spell may appear to determine further unemployment solely because it acts as a substitute for temporally persistent characteristics. Those unobserved individual characteristics are called unobserved heterogeneity and could partly explain the observation of unemployment persistence. The only way to separate state dependence and heterogeneity is to use panel data techniques (Corcoran and Hill, 1985; Hämäläinen, 2003).
State dependence could also be biased by two other sources of spurious correlation. First, a spurious correlation could emerge because of the overlapping of the same unemployment spell over several periods. Therefore, during the analysis, one could conclude that the individual has known several unemployment spells, whereas only one unique experience is concerned. Second, spurious correlation could appear if the past history of the individual on the labour market before their entry in the sample is not taken into account correctly (Flaig et al., 1993).
Some economists have therefore tried to measure the impact of state dependence on unemployment persistence by controlling for these different statistics artefacts.
American studies have found little evidence of state dependence in unemployment even if unemployment persistence is found in the raw data (Corcoran and Hill, 1985; Heckman and Borjas, 1980). According to Corcoran and Hill (1985), the observation of unemployment persistence is only explained by unobserved heterogeneity and data collection.
In contrast, state dependence seems to be present in Europe. As far as the UK is concerned, Narendranathan and Elias (1993) and Gregg (2001) focussed on the youth labour force, using the National Child Development Study, a birth cohort panel survey. They both demonstrated the presence of state dependence in unemployment. Narendranathan and Elias (1993) estimated that an unemployed person may be 2.3 times more likely to be unemployed the following year than an individual who is currently employed. Gregg (2001) demonstrated that individuals experiencing unemployment represent a minority of the labour force, indicating that it is always the same individuals who experience unemployment. However, even if both papers demonstrate the presence of state dependence, its magnitude is weak. Arulampalam et al. (2000) and Arulampalam (2002), using the British Household Survey, found a stronger state dependence for the entire active male labour force. However, when differentiating the results by age categories, they found that younger people face weaker state dependence than older people. They explained this result by the fact that “job-shopping” among the youth labour force is more accepted by employers, since younger people are at the beginning of their careers.
A stigmatisation effect in unemployment occurrence was also found in Germany by Flaig et al. (1993) and Muhleisen and Zimmerman (1994), both using the German Socio Economic Panel (GSOEP) (1984-1989). Flaig et al. (1993) found that the probability to become unemployed is ten times higher for individuals who were unemployed in the preceding period than for those who were in employment. State dependence has also been demonstrated in Finland and Austria (Hämäläinen, 2003; Winter-Ebmer and Zweimuller, l992). Most of those studies have demonstrated the importance of introducing unobserved individual heterogeneity in order not to overestimate the role played by the stigmatisation effect in the observed persistence in unemployment.
As far as Belgium is concerned, Cahuzac (1998) found that past spells of unemployment are not informative about future labour market status as far as white collar workers are concerned. He concluded that, as for American cases, the observed persistence in unemployment is partly due to data collection procedures or unmeasured personal characteristics.
We have decided to test the hypothesis again using a different technique allowing us, among other things, to control for exogenous variables. Indeed, the methodology used by Cahuzac (1998), which is a fixed-effects conditional logit model, only controls for time-invariant characteristics through the inclusion of the heterogeneity term.
Data and sample
The data used are drawn from the Panel Study on Belgian Households (PSBH). This consists of a harmonised questionnaire submitted each year to a representative sample of individuals and households in Belgium, and covers the years 1992-2002. The original sample in 1992 consisted of approximately 4,400 households and 8,700 adults (aged 16 years or older). It ended with about 3,000 households and 5,300 individuals in 2002. Moreover, the database covers a wide range of topics such as family structures, economic activity, housing, health, education, income, geographical mobility, living conditions, values and opinions, etc. In particular, it includes many targeted questions related to individual labour market trajectories.
The sample used for the study corresponds to the last nine waves of the database (1994-2002). It includes men and women aged between 18 and 57 years in 1994 who were active in the labour market for the same year. Individuals remain in the sample until the end of the period studied (2002) except if they become inactive, if they have missing values, or are no longer interviewed. Moreover, new entries (after 1994) will not be accepted since the econometric specification of the model requires a common date of entry for all individuals. The sample is therefore unbalanced and the sample decreases from 3,815 individuals in 1994 to 1,508 in 2002.
The variable of interest is the experience of unemployment. The individual is considered as having experienced unemployment if he declares himself to be unemployed for at least one month during the year considered. About 15 per cent of the sample was unemployed for at least one month in 1994. This proportion decreased during the period studied to 7 per cent in 2002. This fall may be explained by the attrition rate of the sample as well as by the trend in the labour market.
Table I shows that unemployment persistence in Belgium is quite high. On average about 77 per cent of individuals who experienced unemployment in the previous period are also unemployed in the current period. This high unemployment persistence could be explained partly by the presence of state dependence. However, it should be kept in mind that spurious correlation can emerge if the same unemployment spell overlaps several periods (years) or if the underlying characteristics (observable or not) influencing the propensity to be unemployed are not correctly taken into account. The objective of the study will therefore be to disentangle spurious correlations from true state dependence in order to better understand unemployment persistence in Belgium.
The econometric methodology used in order to test the eventual existence of a true state dependence in unemployment occurrence is a dynamic random-effects probit model. It incorporates a correction for unobserved heterogeneity as well as for the initial-condition problem.
The reduced form model of unemployment incidence will be specified for individual i at time t as: Equation 1 where 1(.) is an indicator function equal to 1 if the enclosed statement is true and 0 otherwise; y it is binary and takes the value 1 if the individual is unemployed at the time of interview and 0 otherwise; x it represents exogenous and observable individual and environmental characteristics affecting unemployment probabilities, which vary through time; z i symbolizes all the observable time-invariant variables; y it−1 denotes the previous labour market status, the inclusion of which allows for testing for the eventual presence of state dependence in the occurrence of unemployment; and νit is the error term, which is independent of observable characteristics such that νit is independently and identically distributed.
However, as we have seen from the previous section, several corrections have to be made to the model in order to avoid biased estimation of the true state dependence arising from spurious correlation.
First, the unobservable individual-specific characteristics are taken into account by decomposing the error term into two components: Equation 2 where μ it is the random error term which varies among individuals and through time, independent of observable characteristics such that μ it ∼IIN(0,σ μ 2); and ɛi represents the individual-specific unobserved heterogeneity, varying across individuals but assumed to be time-invariant. This specification allows observationally identical individuals to face different probabilities of experiencing unemployment given unobservable characteristics such as motivation, responsibility, punctuality, etc. Assuming that this individual-specific term is treated as randomly distributed such that ɛ i ∼IIN(0,σ ɛ 2) and is supposed to be independent of x it , z i and μ it leads to a random effects probit model (discussed in Heckman, 1981a, b; Guilkey and Murphy, 1992). Among other things, since the unobserved heterogeneity persists over time, it implies that the composite error term is correlated across cross-section units in time. It is supposed that the correlation between successive error terms for the same individual is a constant: Equation 3 However, assuming independence between unobserved heterogeneity and observable characteristics should be further discussed. If this hypothesis is violated, maximum likelihood estimates will be biased. Mundlak (1978) suggested parameterising ɛi in order to allow a correlation between ɛi and observable characteristics, assuming a linear regression function of ɛi in the means of all time varying covariates.
Second, because we are estimating dynamic models, we need to take into account the individual pre-sample history. Otherwise, an endogeneity problem may arise since the start of the observation period does not correspond to the start of the stochastic process generating the propensity to be unemployed. This could lead to a correlation between individual effects and the initial observation and therefore give biased and inconsistent estimates (Arulampalam et al., 2000). Different techniques exist in order to solve this so-called initial condition problem (see, for example, Hsiao, 2003; Heckman, 1981c; Orme, 1997; Wooldridge, 2005). Wooldridge (2005) proposed an attractive methodology that models the distribution of the unobserved effect on the initial observation value.
Combining the two corrections mentioned above results in modelling the distribution of the individual effects as: Equation 4 where αi∼IIN(0,σ α 2) and is independent of the explanatory variables in equation (1).
Consequently the following equation entirely specifies the unemployment behaviour: Equation 5 According to Heckman (1981a, b), one can estimate the parameters by marginalisation of the likelihood function with respect to the α's if assuming that the conditional distribution of y it on αi , x it , z i , and yit−1 is independent normal. Marginalising the likelihood with respect to α gives: Equation 6 where α˜=α/σ α and σν is set equal to 1. Φ and φ are, respectively, the distribution and the density functions of a standard normal.
A last issue needs to be corrected, which is the spurious correlation between successive unemployment spells that may appear if the individual observed as unemployed in successive waves is in the same unbroken spell of unemployment. According to Arulampalam et al. (2000), as long as the average duration of unemployment is lower than the period between two successive interviews, one can expect that most individuals observed in unemployment over two consecutive periods experience two distinct spells rather than an unbroken spell. In that case the spurious correlation should not be too important. However, in our case, the mean duration of unemployment spells is quite high. Therefore the unemployment equation will be estimated according to three different specifications. Specification 1 will consist of the basic unemployment equation in which the one-year lagged dependent variable is added as explanatory variable. Specification 2 is based on the first unemployment equation in which unemployment spells lasting longer than 12 months have been excluded. Finally, specification 3 consists of the unemployment equation in which the two-years lagged dependent variable is added as explanatory variable. This gives a measure of state dependence over two years.
The results of the estimations of equation (5) are presented in Table II, for the three different specifications of the model.
The observable characteristics introduced in the econometric specification reflect individual job search intensity, the arrival of job offers as well as marginal productivity and job retention rate which, in turn, influence job-to-unemployment and unemployment-to-job transitions. They can be grouped in three different categories.
The first category concerns individual and household characteristics. We can observe a concave and negative relationship between age and unemployment likelihood. Older people have accumulated more experience on the labour market and therefore have higher human capital, implying higher marginal productivity. Also, their job retention rate should be superior since they have higher seniority. We can also observe that men are less likely to experience unemployment than women. Household characteristics (marital status and the presence of children) do not have a strong influence on unemployment probability. When the number of children is significant, it has a positive sign, meaning that the presence of numerous children in the household could be interpreted by the employer as a sign of less attachment to the firm or greater absenteeism (Narendranathan and Elias, 1993). Having Belgian nationality decreases the probability of experiencing unemployment. Also, being disabled reduces job search intensity and can be considered as a signal of weaker productivity involving a weaker job retention rate or fewer job offers (Taylor, 2002).
A second category of variables measures the individual human capital through educational levels. The higher the level of education, the higher is the probability to be in employment. Various explanations may be advanced to explain this phenomenon. Educated people may have more efficient job search methods and more motivation. They can also appear more attractive from the point of view of potential employers. Lastly, the higher recruiting and dismissal costs of graduates involve a higher job retention rate (Taylor, 2002).
The third variables category is related to labour demand, which can be approached by employment growth rate and unemployment rate. Both variables are stated in function of age, gender, corresponding year and the region where individuals are living. While the unemployment rate is insignificant, the employment growth rate has a positive influence on the probability of experiencing an unemployment spell. Lastly, year dummies were also added to the model in order to take into account the global economic trend over the period 1995-2002. When significant, they have a negative sign. These results show that the likelihood for an individual to find a job is strongly related with tensions on the labour market and the business cycle.
An important result is the significance of the coefficient attached to the initial condition for the three specifications. We can therefore reject the null hypothesis that the initial condition is not independent of the heterogeneity term. It is thus necessary to endogenise it.
However, the main question remains. Is there evidence of a scary effect in unemployment occurrence amongst the Belgian labour market? Results show that the probability to be unemployed today is positively and significantly correlated to the presence of unemployment in the previous period, and this stands whatever the specification considered and after controlling for the initial condition problem and unobserved heterogeneity. The importance of the individual heterogeneity term can be demonstrated by the value taken by the share of the total variance attributed to the heterogeneity term (ρ) which is relatively important (respectively 28 per cent, 25 per cent and 51 per cent for the three specifications). Moreover, the test associated with this value, whose null hypothesis is that this coefficient is not significantly different from 0, is rejected for all specifications. This means that estimating the unemployment equation by panel brings more information than a simple probit.
All else being equal, an individual experiencing unemployment in the previous period has a higher propensity to be unemployed today than somebody who was at work. The coefficients attached to this variable lies between 0.9 and 1.9 according to the specification considered. This is consistent with other studies on the subject, which obtain relatively similar coefficients. The explanation of this observed causal relationship is the presence of a stigmatisation effect. Unemployment incidence reduces individual human capital and may be considered as an indicator of lower productivity or less reliability by recruiters. This experience may also influence individual behaviour such as motivation or ambition. Lastly, after some time, the unemployed might be discouraged and finally accept poorer quality jobs or weaker wages involving a loss of qualification and experience, and therefore increasing the probability of living an unemployment spell in the future.
Lastly, it should be noticed that the coefficient associated with state dependence is higher in the first specification than in the other two. This result can be explained by the fact that in specification 1, the effect of state dependence could be overestimated given the presence of a significant number of individuals for which unemployment spells are longer than the period considered. However, even when unemployment spells longer than one year are rejected (specification 2), the coefficient attached to the lagged dependent variable remains high and significant.
The coefficients estimated with the random effects probit model do not lend directly to a marginal effects interpretation. However, it is interesting to measure the impact of the stigmatisation effect on the probability to be unemployed. One way to proceed is to compare the predicted probabilities that are conditional on the different labour market status on the previous period (employed in t−1 or unemployed in t−1) (Arulampalam, 2002). However, since we are working in a panel data framework, it is necessary to take into account the fact that individuals observed as identical may have different unemployment propensities considering the presence of unobserved heterogeneity. Chamberlain (1984) suggested first computing the marginal effects for each individual and thereafter taking the mean on the whole sample, giving us a mean effect for a randomly chosen individual (Arulampalam and Booth, 2000; Hämäläinen, 2003).
The mean effect of changing the covariate yit−1 from y it−1 a to y it−1 b on the probability of a randomly chosen individual to experience unemployment is given by: Equation 7 Also, the probability distribution of y it conditional on y it−1, x it and z i and marginal on ɛ has the following form: Equation 8 Consequently, the formula allowing computing the marginal effects of state dependence on the unemployment probability is given by: Equation 9 where the denominator represents the root square of the total variance and the parameters are replaced by their estimators.
Practically, we first compute the predicted probability for each individual assuming that they have all known an spell of unemployment in the preceding period. Secondly, we compute the predicted probabilities for each individual assuming that they were all in employment in the previous period. Then the difference in the predicted probabilities is computed for each individual. When taking the mean of these differences we obtain a state dependence measure, all else being equal, as well as the part that it explains in the observed persistence. Table III reports the marginal effects of the lagged dependent variable computed for each of the three specifications of the unemployment equation.
The results show that observed and unobserved heterogeneity explain between 57 per cent and 82 per cent of unemployment persistence calculated from the raw data. The results vary according to the specification considered. The remainder is explained by the presence of state dependence between successive unemployment spells.
According to the first specification, in Belgium, an individual who has experienced unemployment in the previous period will be 33 percentage points more likely to be in this situation again one year later than a person who was in employment. However, in this case, state dependence is overestimated since the sample incorporates lots of individuals remaining in unemployment for more than one year. However, even when rejecting those individuals, state dependence remains positive, amounting to 11.4 percentage points (specification 2). Finally, the measure of state dependence over two years amounts to 17.7 percentage points.
This study analyses the determinants of unemployment persistence among the Belgian labour force through the end of the 1990s using the Panel Study on Belgian Households (PSBH). The main purpose was to measure the eventual presence of state dependence in terms of occurrence in unemployment persistence. To test the hypothesis we used a random effects probit model controlling for unobserved heterogeneity, the initial condition problem as well as the overlapping of the same unemployment spell over several periods.
The results strongly suggest the presence of state dependence in unemployment persistence even after controlling for unobserved heterogeneity and the overlapping of the same unemployment spell over several periods. State dependence explains between 18 per cent and 43 per cent of the observed persistence, the remainder being induced by observed and unobserved heterogeneity. Moreover, all else being equal, a randomly chosen individual experiencing unemployment in t−1 will be between 11.4 per cent and 33 per cent more likely to be unemployed again in t than an individual who was in employment in t−1. Other important unemployment determinants are sex, region, age and nationality. Educational level also has a significant role since the chances to hold a job increase as the level of education rises. It also seems useful to control for labour demand characteristics, employment growth rate and time dummies variables all being significant. Finally, the results also show the importance of using panel data in order to incorporate unobserved individual heterogeneity, as well as to endogenise the initial condition.
The presence of state dependence involves that unemployment has long lasting and fatal consequences on future labour market prospects and that the unemployed will not behave on the labour market in the future in the same way that someone who has never faced unemployment. Different interpretations may be advanced to explain this phenomenon. On the one hand, unemployment generates the depreciation of human capital, specific or not to the firm, and can consequently be interpreted as a signal of lower productivity from the employer's point of view. On the other hand, individuals experiencing unemployment will not have the same behaviours as before. According to economic theory, unemployment could induce the individual to develop a greater attraction for leisure as well as less assiduity to work and less motivation. The individual may also, if he is strongly discouraged, accept weaker wages than initially hoped for or a poorer quality job, implying a loss of qualification and experience and thus increasing the probability of being unemployed again.
The presence of a stigmatisation effect of unemployment involves that the costs of unemployment are much higher than the simple loss of income and human capital associated with the current job loss. The study demonstrates the importance of concentrating efforts on the prevention of unemployment. Among other things, it is important to accompany young people correctly in their transition between school and employment in order to avoid unemployment being their first experience of the labour market. It is necessary to increase their employability by providing them with as much experience and attachment on the labour market as possible. Attention should also be given to older unemployed people, i.e. it is important to correctly accompany workers suffering from involuntary job termination and help them to find a new job as quickly as possible, so that unemployment is avoided. There is consequently huge breathing space for public intervention in the fight against unemployment.
Table IUnemployment persistence in the sample
Table IIResults of the estimation of the unemployment equation
Table IIIMarginal probabilities of state dependence
Arulampalam, W. (1999), "A note on estimated effects in random effect probit models", Oxford Bulletin of Economics and Statistics, Vol. 61 No.4, pp.597-602.
Arulampalam, W. (2002), "State dependence in unemployment incidence: evidence for British men revisited", Institute for the Study of Labor, Bonn, IZA Discussion Paper No. 630, .
Arulampalam, W., Booth, A.L. (2000), "Union status of young men in Britain: a decade of change", Journal of Applied Econometrics, Vol. 15 No.3, pp.289-310.
Arulampalam, W., Booth, A.L., Taylor, M.P. (2000), "Unemployment persistence", Oxford Economic Papers, Vol. 52 No.1, pp.24-50.
Cahuzac, E. (1998), "Past unemployment influences on the incidence of Belgian employees' unemployment", Cahiers économiques de Bruxelles, No.159, pp.231-54.
Chamberlain, G. (1984), "Panel data", in Griliches, S., Intriligator, M. (Eds),Handbook of Econometrics, North Holland, Amsterdam, .
Corcoran, M., Hill, M.S. (1985), "Reoccurrence of unemployment among adult men", Journal of Human Resources, Vol. 20 No.2, pp.165-83.
Flaig, G., Licht, G., Steiner, V. (1993), "Testing for state dependence effects in a dynamic model of male unemployment behaviour", in Bunzel, H., Jensen, P., Westergaard-Nielsen, N. (Eds),Panel Data and Labour Market Dynamics, North Holland, Amsterdam, .
Gregg, P. (2001), "The impact of youth unemployment on adult unemployment in the NCDS", The Economic Journal, Vol. 111 No.475, pp.F626-53.
Guilkey, D., Murphy, J. (1992), "Estimation and testing in the random effects probit model", Journal of Econometrics, Vol. 59 pp.301-17.
Hämäläinen, K. (2003), "Education and unemployment: state dependence in unemployment among young people in the 1990s", Government Institute for Economic Research, Helsinki, VATT Discussion Paper No. 312, .
Heckman, J.J. (1981a), "Statistical models for discrete panel data", in Manski, C.F., McFadden, D. (Eds),Structural Analysis of Discrete Data with Econometric Applications, MIT Press, Cambridge, MA, .
Heckman, J.J. (1981b), "Heterogeneity and state dependence", in Rosen, S. (Eds),Studies in Labor Markets, University of Chicago Press, Chicago, IL, .
Heckman, J.J. (1981c), "The incidental parameters problem and the problem of initial conditions in estimating a discrete time-discrete data stochastic process", in Manski, C.F., McFadden, D. (Eds),Structural Analysis of Discrete Data with Econometric Applications, MIT Press, Cambridge, MA, .
Heckman, J.J., Borjas, G. (1980), "Does unemployment cause future unemployment? Definitions, questions and answers from a continuous time model for heterogeneity and state dependence", Economica, Vol. 47 pp.247-83.
Hsiao, C. (2003), Analysis of Panel Data, 2nd ed., Cambridge University Press, Cambridge, .
Lockwood, B. (1991), "Information externalities in the labour market and the duration of unemployment", Review of Economic Studies, Vol. 58 No.4, pp.733-53.
Lollivier, S. (2000), "Récurrence du chômage dans l'insertion des jeunes: des trajectoires hétérogènes", Economie et Statistique, Vol. 334 No.4, pp.49-63.
Muhleisen, M., Zimmerman, K.F. (1994), "A panel analysis of job changes and unemployment", European Economic Review, Vol. 38 No.3/4, pp.793-801.
Mundlak, Y. (1978), "On the pooling of time series and cross section data", Econometrica, Vol. 46 No.1, pp.69-85.
Narendranathan, W., Elias, P. (1993), "Influences of past history on the incidence of youth unemployment: empirical findings for the UK", Oxford Bulletin of Economics and Statistics, Vol. 55 No.2, pp.161-85.
Orme, C.D. (1997), "The initial conditions problem and two step estimation in discrete panel data models", Department of Economics, University of Manchester, Manchester, mimeo, .
Taylor, M. (2002), "Labour market transitions in the context of social exclusion: a study of the EU", Directorate-General for Employment and Social Affairs, European Commission, Brussels, .
Wooldridge, J. (2005), "Simple solutions to the initial conditions problem for dynamic, nonlinear panel data models with unobserved heterogeneity", Journal of Applied Econometrics, Vol. 20 No.1, pp.39-54.
Arulampalam, W., Gregg, P., Gregory, M. (2001), "Unemployment scarring", The Economic Journal, Vol. 111 No.475, pp.F577-84.
Butler, J., Moffitt, R. (1982), "A computationally efficient quadrature procedure for the one-factor multinomial probit model", Econometrica, Vol. 50 No.3, pp.761-4.
Winter-Ebmer, R. (1991), "Some micro evidence on unemployment persistence", Oxford Bulletin of Economics and Statistics, Vol. 53 No.1, pp.27-44.
Winter-Ebmer, R., Zweimuller, J. (1992), "Do they come back again? Job search, labour market segmentation and state dependence as explanations of repeat unemployment", Empirical Economics, Vol. 17 No.2, pp.273-92.
Appendix 1. Long term unemployment in Belgium
Appendix 2. Results of the literature review
Appendix 3. Descriptive statistics
About the authors
Amynah Gangji holds a Master's degree in Political Economics from the Free University of Brussels. She is currently a research fellow at the Department for Applied Economics, ULB (DULBEA) and a PhD student at the DULBEA Doctoral School. Her fields of interest are labour economics, unemployment, poverty and the housing market. Amynah Gangji is the corresponding author and can be contacted at: firstname.lastname@example.org
Robert Plasman (PhD) is Full Professor of Political Economics and Industrial Relations at the Free University of Brussels (ULB) and Co-director of the Department of Applied Economics, ULB (DULBEA). He is the author of books and articles on gender, labour and employment.