The Stepping Stone Effect of Temporary Help Agency Work for Non-Western Immigrants in Norway by

Kristine von Simson Institute for social research October 2009 Preliminary, please do not quote

Abstract This paper investigates the stepping stone hypothesis of temporary help agency (THA) work for later employment prospects for non-western immigrants in Norway. Is an unemployed worker more likely to find and to keep a regular job if he first accepts a job in a THA? Using the timing-of-events duration model approach, which assumes no anticipation of treatment, I find that the individual transition rate from unemployment to regular employment is significantly and substantially raised as a result of THA work. This is not due to selection into THA work, as this has been taken account for through the inclusion of correlated random effects. The estimated results show that women, older individuals, individuals with small children and lower education and who have stayed in Norway only a few years are less likely to exit unemployment to both regular and THA work.. Married individuals are more likely to find a job in the regular labor market, while those living in Oslo are more prone to enter a THA from unemployment. Immigrants from Africa have a larger probability of experiencing a transition to THA work, but are less likely to have a direct transition to regular employment. Immigrants from South America are more likely to exit unemployment for both destination states.

Key words: Temporary help agency work, duration models, treatment effect

1.

Introduction

The weak labor market performance of immigrants in many developed countries is now well documented. For example, numbers from Statistics Norway from August 2008 show that the unemployment rate among immigrants in Norway was 4.7 percent, compared to 1.4 percent among the natives. Worst off are immigrants from non-western countries, with unemployment rates ranging from 5 to 10 percent. This group of immigrants is also represented with lower employment rates and is frequently found among welfare recipients.

The Norwegian labor market may be a particular challenge to enter for non-western immigrants. Minimum wages are among the highest in the world, and wage dispersion is relatively small. Many non-western immigrants have no or undocumented education when they arrive in Norway, and work-specific skills acquired in the origin country may not be transferable to the Norwegian labor market. The weak labor market performance may thus be a result of the general shortage of jobs for low-skilled workers. In addition many immigrants lack social networks, connections and references that hamper their possibility to signal the actual productivity they hold. If high productivity demands and low signaling capacities are parts of the problem, different kinds of productivity and information enhancing measures may be a significant part of the solution.

This paper sheds additional light on this issue by analyzing the effect of temporary help agency (THA) employment on later employment prospects for non-western immigrants in Norway. Is an unemployed worker more likely to find and to keep a regular job if he first accepts a job in a THA? THA employment represents a special type of temporary work which has experienced a tremendous growth across OECD-countries the last decade, including Norway1. Empirical evidence for Europe suggests that THA work often serves as an intermediary between unemployment and regular work (Zijl and van Leeuwen (2005)), as a substantial share of people move from THA work towards regular employment. Although this does not indicate a stepping stone effect of THA work per se, THA work can be looked upon as an alternative job search strategy. Instead of staying in unemployment until a regular job is found, the unemployed individual can take up work in a THA as an intermediate position while waiting for a regular job. If the intermediate THA spell speeds up the job-

1

Arrowsmith (2006) gives a survey of the extent of THA work in Europe. For the case of Norway, see Nergaard and Svalund (2008).

search process and increases the probability of obtaining regular employment, the stepping stone hypothesis of THA work is supported.

Some groups seem to be overrepresented in the THA-industry, among them women and young people. Torp et al (1994) find that 1 out of 4 THA workers enter THA employment directly from unemployment, giving some support to the hypothesis that THA work mobilizes marginal groups in the labor market. A study using Swedish register data (Andersson and Wadensjö) show that immigrants from outside of Europe are overrepresented in the THA sector. The study also finds evidence for the stepping stone hypothesis of THA work for immigrants, as immigrants more often leave THA employment for another type of employment compared to natives.

Economic theory offers several explanations why THA work may be a gateway to the primary labor market for unemployed. According to human capital theory (Becker (1964)) investments in training and education have an influence on both labor market participation and productivity differences. Individuals invest in training today to produce income in the future. By working in a THA the worker gets the chance to gain experience and to build social networks, thus increasing human capital. Closely linked is job search theory (Mortensen (1986)), which considers job search as a result of rational behavior of individuals with imperfect information about jobs and wages. In this setting, THA work provides an opportunity to gather information necessary to improve career prospects.

The theory of statistical discrimination (Arrow (1972)) offers yet another explanation for the potential stepping stone effect of THA work, which may be especially relevant in the context of immigrants. Statistical discrimination refers to the phenomenon that employers, in lack of other signals, attribute to the job seeker the average abilities of the group to which the individual belongs. Since immigrants often lack important documentation of education and work experience, THA work may serve as a positive signal of the worker’s motivation and abilities for prospective employers. At the same time the employer’s risk of hiring the wrong person is reduced through an on-the-job screening for permanent positions. This may be particularly important in the Norwegian labor market, which is strictly regulated with strong employment protection and high firing costs.

However, there are factors pointing in the other direction. The dual labor market theory (Doeringer and Piore (1971)) divides the labor markets into different segments: While jobs in the primary labor market are characterized by stability, high wages and chances of advancement, jobs in the secondary labor market tend to have low wages, little career advancements and high labor turnover. There have been expressed concerns that jobs in the THA sector constitute a secondary labor market. Short spells of THA work are often interrupted by extended periods of unemployment, and many THA workers seem to be trapped within the industry. THA workers often receive lower wages and less training compared to regular workers, and career opportunities are worse. In addition, THA employment may stigmatize workers, signaling low productivity or lack of other alternatives.

With the use of register data covering the entire immigrant population in Norway from 19952003, I analyze whether THA employment serves as a stepping stone for unemployed nonwestern immigrants in Norway. The data contain detailed information about individual labor market histories and make it possible to analyze transitions from unemployment to regular employment and from unemployment to THA work. I use the timing-of-events duration model by Abbring and van den Berg (2003), which utilizes information about the timing of the treatment to distinguish treatment effects from pure selection effects. The results indicate that unemployed non-western immigrants may benefit from THA-employment compared to staying unemployed until a job-offer is received.

The rest of the paper is organized as follows: Section 2 gives an overview of the THA industry in Norway followed by a literature review in section 3. Section 4 presents the econometric model, and section 5 describes the data. Section 6 presents the results and section 7 concludes.

2.

The THA Industry in Norway

THA work has been the most rapidly growing form of temporary contracts throughout Europe over the last decades, including Norway. Even though the industry employs only a small share of the working population (around 1.4% in Norway in 2006) 2, the sector has experienced a large increase both in terms of the number of employees, number of THA firms and overall economic value. Figure 1 shows the development in Norway from 1995 to 2006, 2

Nergaard and Svalund (2007)

based on register data from Statistics Norway. The number of employees in THA-firms during a year increased from 13.000 in 1993 to 33.000 in 2006.

Figure 1. The number of THA-employees and -firms in Norway from 1993-2006. 1200

40000 35000

1000

30000 800

25000

600

20000 15000

400

10000 200

5000

0

0 1993 1994 1995

1996 1997 1998

1999 2000

Number of THA-firms

2001 2002 2003

2004 2005 2006

Number of THA-employees

Source: Nergaard and Svalund (2007), p.19

The number of firms providing THA-employment has also experienced a steady growth, following a deregulation in 2000. Before this deregulation, THA-employment was generally banned, with an exception for providers of labor for office work. In 2000 there was a liberalization of the Work Environment Act regarding employers’ use of temporary contracts, permitting THA work in situations where the law permits fixed-term contracts. The law was further revised in 2005, making it easier for firms to use temporary contracts in general and widening the opportunities for THA employment. The growth in the number of THA firms supplying unregistered (often immigrant) labor has led to a retightening of the law. Recently, a regulation introducing a registration requirement for THA firms has been adopted. Firms who want to provide THA employment must satisfy certain financial and organizational requirements, and must submit regular details of their activities to a public register.

3.

Literature Review

Research on the effect of THA work is scarce. The main reason for this is lack of appropriate data to identify firms in the THA industry. But there exist a number of studies in the literature focusing on the stepping stone effect of temporary work in general. The main finding of these

is that temporary contracts are stepping stones into permanent employment. Such evidence is available for several European countries, among others the United Kingdom (Booth, Francesconi and Frank (2002)), the Netherlands (Zijl, van den Berg and Heyma and (2004)) and Belgium (Göbel and Verhofstadt (2008)). Longva (2002) finds support of the stepping stone hypothesis for Norway, as two out of three temporary workers are in permanent employment two years later.

Existing analyses on the THA industry reach opposite conclusions about the effect of THAwork for later employment prospects. An Italian study by Ichino et al (2006) is in favor of the stepping stone hypothesis of THA-work, while Amuedo-Dorantes et al (2008) find that Spanish THA workers have a lower probability of being hired on a permanent basis. Kvasnicka (2008), using German register data, finds neither positive nor negative effects of THA-work. However, unemployed workers who enter a THA have higher employment chances and lower unemployment chances in general compared to similar non-THA workers. An American study by Autor and Houseman (2005) find no evidence for the stepping stone function of THA-work. On the contrary, they claim that THA-work may hurt workers’ career prospects in the long run.

There can be given several explanations of why the above mentioned studies differ in their empirical findings. Firstly, they use different methods in the process of identifying the causal effect. A common characteristic of all the European studies is that they are based on nonexperimental data and methodologies, and most of them rely on the assumption that selection into THA-work depends on observable characteristics only. The Autor and Houseman (2005) analysis has a quasi-experimental design, using random assignment of workers into THAwork as an exogenous source of variation. While the non-experimental studies may suffer from self-selection problems, questions can be posed about the external validity of the Autor and Houseman results as they focus on a specific group of welfare recipients.

Secondly, the studies are not measuring the same effects. Ichino, Mealli and Nannicini (2006) are measuring the effect for the average worker, while Autor and Houseman are looking at the marginal effects of THA-work. They also focus on different groups: Welfare recipients (Autor and Houseman (2005); Heinrich, Mueser and Troske (2005)), unemployed workers (Kvasnicka (2008)) or temporary workers in general (Amuedo-Dorantes et al. (2008)). Lastly, there may be institutional reasons for the different effects of THA-work. European labor

markets are characterized by stronger employment protection compared to the US, and screening through temporary work might play a more important role in Europe than in the US labor market. In the light of these opposing results, it is difficult to draw any general conclusions about the effect of THA-work for later employment prospects.

4.

Econometric framework

An important issue in all non-experimental treatment evaluation analyses is the potential selection bias. If individuals for whom it is easy to experience the event of interest tend to self-select into the treatment, a pure selection effect may be wrongly interpreted as a positive treatment effect. In my setting this can be expressed as follows. If unemployed individuals who enter temporary help agency work have higher hazard rates to regular employment than those who enter regular employment directly from unemployment, this can be for two reasons. (1) The treatment effect of temporary help agency work is positive, or (2) these individuals have some characteristics that influence the hazard rate to regular employment, and would have relatively fast transitions to regular employment even in the absence of treatment. It is unlikely that that the reasons for this variability in the hazard will be fully captured by observable covariates, and unobserved heterogeneity must be included in the model.

I use the timing-of-events framework formalized by Abbring and van den Berg (2003). This framework exploits the fact that variation in the duration until treatment relative to the duration until the outcome conveys useful information on the causal treatment effect in the presence of selection effects. For instance, if the individual experiences a transition to regular employment relatively fast after having entered a temporary help agency, for all constellations of explanatory variables, then this is evidence of a positive causal treatment effect. This model framework is also referred to as endogenous shocks (van den Berg (2001)). Two durations are being considered: Duration until treatment, t1, and duration until event of interest, t2. In my case, t1 is time until exit to a THA, and t2 is time until exit to regular employment. I only observe t1 if t1 is completed before t2. This ensures that the cause precedes the effect. If t1 is realized, this may have an impact on the hazard rate until the event of interest. For instance, if an individual enters a THA, this affects the transition rate to regular employment from the moment of entering the THA and onwards. This approach

allows the treatment effect to vary with the moment of treatment, with the observed characteristics and elapsed time in the current state.

A crucial assumption for identification is the “no-anticipation” assumption. This assumption assures that there is no effect of future treatment on the current outcome hazard. In other words, the treatment can only affect the transition rate from the moment of treatment onwards. If there is no randomness in the process of treatment assignment, then the estimates of the current transition rates are determined by future events. This will be the case if individuals who decide to work in a temporary help agency change their job searching behavior from the moment they decide to work in a THA, and not from the moment they actually obtain work through the temporary help agency. However, the assumption does not rule out the possibility that forward-looking individuals may have information about properties of the treatment process and act on this information. Even though a THA registration increases the probability of obtaining work through the temporary help agency, there still is a considerable random part left in the process. As in regular job searches, the hiring process depends on both the characteristics of the individual as well as the current vacancies in the market, that is the number and characteristics of registered client firm searching for employees. The randomness which describes regular job search processes is still applicable in the THA setting, and supports the no-anticipation assumption.

The timing-of-events results in Abbring and van den Berg (2003) show that non-parametric identification of the treatment effect relies strongly on the assumption of proportional hazard rates. The proportionality assumption is in many settings a difficult assumption to justify. McCall (1994) and Brinch (2007) show that the inclusion of time-variant covariates strengthens identification. The intuitions behind is that past values of these variables affect the transition rate only through the selection process and thus provide a natural exclusion restriction. The model below includes several time-varying covariates, and identification of the treatment effect does not entirely depend on the proportional hazards assumption.

I define three different states: Unemployment (UE), temporary help agency employment (THA) and regular employment (RE). Time is measured from the moment the subject enters unemployment, and I define the following possible transitions: From unemployment to THA work, from unemployment to RE directly or from unemployment to RE indirectly through an intermittent period of THA-work. The two first transitions are competing risks. If the

individual experiences a transition to THA work, the event history continues, but now as a single risk. The exit state of interest is RE, which is absorbing in the sense that the event history ends and the individual leaves the risk set.

All durations are measured in months, and I therefore use a discrete time duration model in the specification of the hazard rates. I normalize the time the individuals enter unemployment to zero, and let the non-negative random variables TTHA and TRE denote duration until THAwork and duration until regular employment respectively. Obviously, TTHA is only observed if TTHA < TRE . If TTHA ≥ TRE , no transition to THA work takes place and the duration is rightcensored. I assume that the hazard rates take the proportional odds specification. It can be shown that the proportional odds specification provides a close approximation to the intervalcensored proportional hazard model if the interval hazard rate is relatively small or the intervals are sufficiently narrow (Jenkins, 2005). The proportional odds model assumes that the effects of observed and unobserved covariates are constant over time. This assumption may not always be realistic, but is convenient for interpretation of the model parameters. As stated above, the inclusion of time-varying covariates relaxes the assumption of proportionality.

Let x be a vector of observed covariates, and let v RE and vTHA denote unobserved variables affecting the hazard rates to regular employment and THA-work, respectively. Realizations of the two random duration variables are denoted by t THA and t RE . If t ≤ t THA the individual has three possibilities each month: Stay in unemployment, enter a THA or enter regular employment. If t > t THA , the individual has entered a THA and he/she has now two possibilities each month: Stay in the THA or enter regular employment. All durations may be right-censored due to end of time window or transition to any other labor market state. I only consider direct transitions, thus a transition back to unemployment after exiting THA is treated as right-censored. The individual hazard rate can then be expressed as hTHA ( t | x , v THA ) =

exp( α THA ( t ) + β THA x + v THA ) 1 + exp( α THA ( t ) + β THA x + v THA ) + exp( α RE ( t ) + β RE x + δ ( t | x ) * Ι ( t > t THA ) + v RE )

h RE (t | x, v RE ) exp(α RE (t ) + β RE x + δ (t | x) * Ι(t > t THA ) + v RE )  1 + exp(α (t ) + β x + v ) + exp(α (t ) + β x + δ (t | x) * Ι(t > t ) + v ) , t ≤ t THA  THA THA THA RE RE THA RE =  exp(α RE (t ) + β RE x + δ (t | x) * Ι(t > t THA ) + v RE ) , t > t THA 1 + exp(α RE (t ) + β RE x + δ (t | x) * Ι(t > t THA ) + v RE )

α j (t ) is the baseline hazard and measures duration dependence. The term δ (t | x) captures the effect of THA-work on the exit rate to regular employment. Ι(t > t THA ) is an indicator function which is equal to one if the term inside the parentheses is true. That is, the effect of THA work only affects the hazard rate to regular employment from the moment of entering a THA and onwards. Endogeneity of δ (t | x) is taken account of through possible correlation between the unobserved variables v RE and vTHA . The corresponding survivor function, that is, the probability that duration equals or exceeds t, is:

  S(t | x, v) = ∏1− ∑hj (t | x, v j ) j k=1   Where j = {RE , THA} for t ≤ t THA , and t

j = {RE} for t > t THA .

The random effects are unobserved, and must be integrated out of the likelihood function. The conditional likelihood contribution for a given spell with observed duration t equals:

 h (t | x, v )  THA  L(v) =  THA 1 − ∑j hj (t | x, v j ) 

cTHA

 h (t | x, v )  RE  ×  RE 1 − ∑j hj (t | x, v j ) 

cRE t

[

]

× ∏1 − ∑j hj (k | x, v j ) k =1

where cTHA and c RE are dummy variables equal to 1 if the spell ends with a transition to THA work or RE respectively, 0 otherwise, and v ≡ (vTHA , v RE ) . As stated above, j = {RE , THA} for t ≤ t THA , and j = {RE} for t > t THA . Inserting the appropriate expressions for the survivor and hazard functions, the log-likelihood becomes similar to the log-likelihood for a sequential multinomial logit model with random effects.

I let the baseline hazard be a piecewise constant function of time. The duration axis is divided into a finite number of intervals τ . t denotes elapsed duration, and Dτ (t ) are time-varying dummy variables that are equal to 1 if t is in the interval τ .

The log baseline hazard can then be written as

log α i (t ) =

∑ α τ Dτ (t ),

τ =1, 2 ,...

i

i = {THA, RE}

In the empirical analysis I divide the duration axis into seven intervals: 1-2 months, 3-4 months, 5-6 months, 7-12 months, 13-18 months, 19-24 months and an open-ended interval for durations over 24 months.

The random effects are assumed to follow a normal distribution and are allowed to be correlated across origin- and destination states. Heckman and Singer (1984) argue that parametric distributions of unobserved heterogeneity may seriously bias the estimates in duration models, and recommend using a discrete specification with an a priori unknown number of mass points. The locations of the mass points as well as the associated probabilities are freely estimated together with the rest of the parameters. This model is also known as a semi-parametric heterogeneity model, as the discrete density can approximate any probability density as the number of mass points increases. While this semi-parametric approach in principle is the best way to minimize the potential bias caused by misspecification of the unobserved heterogeneity, a recent Monte Carlo study by Nicoletti and Rondinelli (2009) gives evidence in favor of normally distributed unobserved heterogeneity in discrete time duration models. More specifically they show, using Monte Carlo simulations, that an incorrect normality assumption biases neither the duration dependence nor the covariate coefficients estimation. The normality assumption makes it possible to use standard software in the estimation of the model.

Due to computational considerations estimation is done using all observations ending with a transition to THA-work, and a 5 percent sample of the rest of the unemployment spells. This sampling scheme is frequently used in epidemiological case-control studies and is also known as choice-based sampling. Manski and Lerman (1977) show that when the sample and population strata probabilities are known, weighted maximum likelihood provides a consistent, but not fully efficient, estimator, and call the estimator weighted exogenous sampling estimator (WESML). Van den Berg and Vikström (2009) is an example of a study using this sampling scheme in the context of a bivariate dependent-duration model. Since the integral in the likelihood function above has no closed form, it must be evaluated numerically through simulation.

4.

Data

I have data from FD-Trygd, an administrative database covering the entire Norwegian population aged 16-74 years from 1992 to the present. The database consists of several labor market registers and is organized as an event database. Each time an individual experiences an event, a new observation is added to the database. The data are particularly well suited to analyze transitions in the labor market, as each individual is characterized by a unique identification number which makes it possible to follow individuals between different public registers: From unemployment to work, from unemployment to social security or from work to unemployment. The data also includes detailed background information like gender, age, income, education, country of origin, year of immigration etc.

My sample consists of all new unemployment spells that started between 1996 and 2002, and I only consider spells by non-western immigrants. A person is defined as a non-western immigrant if he/she is born in a non-OECD country and has two foreign-born parents (firstgeneration). An unemployment spell is considered as “new” if the individual has not been registered as unemployed for at least two months. If the gap between two spells is less than two months the spell continues, but the gap is not added to the unemployment duration. I use spells as my analytical unit rather than individuals, since individuals entering unemployment early are more likely to be represented with more spells than those who enter late. I focus my analysis around non-western immigrants between 18 and 55 years when entering unemployment. Very young individuals may have a preferred transition to education, while very old individuals are more likely to experience a transition to retirement. This leaves me with a total of 168 461 spells, spread out over 77 502 persons.

In 1994 the industrial classifications were restructured, and “Provision of Personnel” became a separate subgroup under “Other Business Activities” (NACE 74.502). This means that Statistics Norway obtains information about the industry, and the firms can be identified in the employer and employee registers. THA-employment is thus defined as being registered in the employment register with industrial classification NACE 74.502. One problem with this classification is that it is not possible to distinguish THA workers from the administrative personnel. The trade union for THAs, NHO Service, estimates the share of administrative

personnel at around 10 percent. Regular employment is defined as being registered in the employment register in any other industry than NACE 74.502.

A transition to THA employment takes place if the individual is registered with NACE-code 74.502 in the employment register the month after exiting unemployment. I then measure the months between the start of the THA-spell until the moment at which the individual moves to a regular job. A transition to regular employment thus takes place if the individual is registered in the employment register with any other NACE-code than 74.502 within a month after exiting unemployment or THA work respectively. I don’t distinguish between part time or full time positions in the empirical analysis. The data also contain no information about the THA-spell beyond time spent in the THA, which makes it impossible to identify whether a transition from THA to regular employment is with the same employer or not. Descriptive statistics of the observed transitions in the sample are presented in table 1. Table 1. Observed transitions in the sample Number of spells

168 461

Number of persons

77 052

Transitions from unemployment: Number of UE-RE transitions

80 521

Number of UE-THA transitions

1787

Mean duration of UE-RE transitions, months

5.78

Mean duration of UE-THA transitions, months

5.53

Number of spells censored due to transition to other labor market states

69 086

Number of spells censored due to end of time window

17 067

Transitions from THA work: Number of THA-RE transitions

906

Mean duration of THA-RE spells, months

5.26

Number of spells censored due to transition to other labor market states

771

Number of spells censored due to end of time window

110

Subsequent regular employment spell: Mean duration of RE-spell directly from UE, months

15.49

Mean duration of RE-spell via THA, months

18.64

A few comments are in order. Almost 50 percent of the unemployment spells end with a transition to regular employment, while 1.06 percent of the spells end with THA-work. The mean duration of unemployment spells ending with a transition to THA work is somewhat shorter than those ending in regular employment. Around half of the THA-spells have a

transition to regular employment. Interestingly, the subsequent regular employment spell is on average longer for transitions to regular employment that go through a THA than direct transitions. Whether this can be interpreted as an evidence for a stepping-stone hypothesis of THA work is hard to say without further analysis.

Table 2. Descriptive statistics by destination state, measured at the start of the unemployment spell Regular work

THA

Share female

42.81

33.86

Age, years

34.47

31.09

Share with dependent children under 6 years

32.08

22.5

Share married

61.99

44.49

- Compulsory school

43.10

43.65

- High school

26.22

28.93

- University level

17.85

17.46

12.83

9.96

26.72

42.25

- Africa

15.06

24.96

- Asia

47.31

46.05

- Eastern Europe

30.06

18.41

- Southern America

7.58

10.58

Years since migration, share

9.53

9.74

-Employment

71.10

44.15

-THA

0.74

25.91

-Social security

6.93

7.89

-Education

3.59

6.04

-No registration

17.62

16.01

Share with

- Unknown

3

Share living in Oslo Share from

Status before entering unemployment, share:

Table 2 presents descriptive statistics of the sample by destination state and reveals interesting differences between the observed transitions. On average, individuals who experience a transition to THA work from unemployment are more likely to be men; they are younger, have slightly longer duration of residence in Norway and are more likely to live in Oslo than individuals who obtain a regular job after ended unemployment spell. This last feature is probably due to a concentration of THA’s in Oslo. As THA-work is intrinsically 3

Unknown education indicates that there is no information about the education level of the individual in the education registers.

temporary in nature and represents a less stable employment relation than regular work, those with a family to support may be more eager to enter regular employment. This is confirmed by the descriptive statistics, which shows that married individuals and those with young children are more likely to enter regular employment from unemployment than THA work. Immigrants from Africa are more prone to enter THA work than regular employment from unemployment, while the opposite is true for immigrants from Eastern Europe. Over 25 percent of the unemployment spells that end in THA are directly preceded by a THA-spell, while over 70 percent of the unemployment spells that end in regular employment are preceded by a regular employment spell. This indicates that previous labor market experience is important for subsequent transitions from unemployment.

I include the following covariates in the empirical analysis below: -

Female: A dummy equal to one if the individual is female

-

Country of origin: I include four dummy variables indicating whether the immigrant was born in Eastern Europe, Africa, Latin- and South-America or Asia (reference). This variable captures differences due to language proficiency, skin color, cultural background etc. that may influence labor market outcomes in Norway.

-

Education level: I include four dummies for education level: Compulsory school (10 years), High school (13 years), University level (more than 13 years) and a dummy for unknown education level. The latter is particularly relevant for non-western immigrants, as a substantial part arrive in Norway with unknown or no documentation of their education level.

-

Number of children under the age of 6: Children may influence the transition rates to regular employment and THA work differently, as described above. Individuals with a family to support may prefer regular employment for THA work, due to the temporary character of the latter.

-

Oslo: A dummy indicating whether the individual lives in Oslo. There is an overrepresentation of non-western immigrants in Oslo, and a majority of THA’s are situated in here.

-

Age: Age may influence the transition rate to both regular employment and THA work, as preferences vary with age and older individuals face different labor markets than younger individuals.

-

Years since migration: This variable captures the effect of time of residency in Norway on the probability of obtaining work.

-

Local unemployment rate: This variable captures business cycle variations. It measures the quarterly unemployment rate in the county where the immigrant lives.

-

Previous unemployment experience: A dummy equal to one if the individual has been registered in the unemployment register during the previous 12 months before entering unemployment. This variable may influence the hazard rate to work both positively and negatively, through a lowering of the individual reservation wage (positive effect) or through depreciation of human capital (negative effect).

-

Year dummies: I include 7 dummies indicating the year, capturing calendar effects.

All variables are measured at the start of the unemployment spell except for the local unemployment rate and the year dummies, which vary over the spell.

6.

Results

This section reports the results of the estimation. I estimate a discrete time survival model using a mixed logit, where the duration dependence is summarized through a piecewise constant specification. The estimation is done using all observations with a transition to THA-work, and a 5 % sample of the remaining unemployment spells. Sampling weights are then attached to the observations, and robust standard errors are calculated. The model is estimated using maximum simulated likelihood (MSL) with Halton draws4, as described by Train (2003). Duration is partitioned into 7 intervals, where the hazard rate is assumed to be constant within these intervals. I use the following intervals: {1-2 months, 3-4 months, 5-6 months, 7-12 months, 13-24 months, 24-36 months, more than 36 months}. The estimation results are presented in table 3.

Table 3. Estimation results. The effect of the covariates on the hazard rate from unemployment to regular employment and THA-employment respectively.

Covariates

Regular employment Coeff. SE

Treatment effect: The effect of THA-work on the 0.331*** exit rate to RE -0.087** Female -0.009*** Age Immigrant from: 4

THA-employment Coeff. SE.

0.048 0.037 0.002

-0.274*** -0.057***

0.056 0.003

The model was estimated in STATA using MIXLOGIT, a program written by A.R. Hole (2007). 50 Halton draws were used in the estimation. In the final estimation more than 50 Halton draws should be used.

-Asia (ref.) -Africa -Eastern Europe -South-America Education: -Compulsory -High school (ref.) -University level -Unknown edu. Children u/6 Married Years since migration Living in Oslo Local unemployment rate Previously unemployed Calendar effects: 1996 (ref.) 1997 1998 1999 2000 2001 2002 2003 Duration dependence: -1-2 months (ref.) -3-4 months -4-5 months -6-12 months -13-18 months -19-24 months -25+ months Random effects: Mean Variance Covariance

-0.167*** 0.032 0.253***

0.052 0.042 0.069

0.279*** -0.169** 0.603***

0.062 0.069 0.087

-0.248***

0.045

-0.369***

0.060

-0.011 -0.059 -0.103*** 0.201*** 0.021*** -0.207*** 0.017 0.113***

0.053 0.061 0.026 0.041 0.003 0.041 0.017 0.035

-0.005 -0.360*** -0.305*** -0.055 0.024*** 0.439*** -0.162*** 0.194***

0.077 0.092 0.043 0.057 0.004 0.053 0.027 0.050

0.261*** 0.580*** 0.513*** 0.432*** 0.562*** 0.529*** 0.127

0.087 0.090 0.091 0.088 0.089 0.086 0.084

0.253* 0.608*** 0.481*** 0.707*** 1.140*** 0.521*** 0.033

0.138 0.136 0.140 0.133 0.128 0.133 0.138

-0.350*** -0.460*** -0.666*** -0.861*** -1.041*** -1.476***

0.052 0.061 0.059 0.072 0.127 0.225

-0.367*** -0.591*** -0.839*** -0.982*** -1.102*** -1.404***

0.067 0.084 0.077 0.093 0.185 0.340

-2.604*** 0.247 0.019 0.153 0.097 (0.400)

-4.016*** 0.578

0.215 0.579

* indicates significance at 10% level, ** at 5 percent level and *** at 1 percent level

When interpreting the estimates, it is important to have in mind the model specification. Since the parameters in the multinomial logit specification enter both the numerator and the denominator, the estimated parameter must be interpreted as the effect of the variable on the probability of leaving the origin state relative to the probability of staying in that state.

The effect of THA work on the exit rate to regular employment is estimated to be 0.331, and is significant on a 1% level. That is, entering a THA at time t significantly increases the exit rate to employment compared to staying in unemployment at the same elapsed duration. This positive effect of THA work on the exit rate to employment is not due to self-selection into THA work, as this has been taken account of with the inclusion of the possibly correlated random effects. As mentioned above, there may be given several reasons why THAemployment can function as a stepping-stone to the primary labor market for unemployed workers.

I now turn to the effect of the observed covariates on the estimated hazard rates. Women and older individuals are less likely to exit unemployment for both regular and THA employment, and the effect is strongest for transitions to THA-work. This confirms the observation that immigrant men and young people are overrepresented in the THA-industry. As older individuals on average are more likely to be in a situation which requires some stability of income, the lower exit rate to THA work for this group may be a consequence of that. Older individuals may also be less adaptable to new job situations than younger individuals, which may explain why their hazard rates are lower for both destinations and in particular for THA work, as one of the characteristics of this type of employment exactly is diversity of job situations.

Married individuals are more likely to obtain regular work after ended unemployment spell. This is in line with other empirical research. The reason for this is unknown, but one might imagine that married individuals have abilities that make them attractive in both the labor market as well as the marriage market. I find no significant effect on being married on the transition rate into THA work. Having small children affects the hazard rate to both THA work and regular employment negatively, with the effect largest for transitions to THA work. The latter may be a result of the temporary and unstable character of THA work, as having a family increases the need of stability. Not surprisingly, individuals with lower education than high school have lower exit rates to both THA work and regular employment. Country of origin has a significant effect on the transition probabilities to both destination states. Compared to immigrants from Asia, immigrants from Africa are less prone to obtain regular work, while they are more likely to exit unemployment for THA work. Immigrants from South-America have higher hazard rates to both regular employment and THA work, while immigrants from Eastern Europe are less likely than Asians to enter THA work.

The probability of exiting unemployment for both RE and THA increases with the number of years spent in Norway. Immigrants who have lived a long time in the host country are more likely to have accumulated country specific skills which are important in the labor market. Living in Oslo significantly decreases the transition rate to regular employment, while it increases the probability of experiencing a transition to THA-work. The first result may be a consequence of immigrants clustering in Oslo, making it relatively more difficult to obtain regular work if they search for jobs in the same labor market. As generally found in the literature, the local unemployment rate has a negative effect on all transition rates. The effect has larger impact on the transition rate from unemployment to THA-work, indicating that THA-work is more sensitive to business cycle fluctuations. Previous unemployment experience, measured as being registered in the unemployment register at least once during the last 12 months before entering unemployment, has a positive effect on both exit rates. This may reflect a lower reservation wage for these individuals.

The duration dependence parameters give insight into the time dependency of the transition rates. The probability of a transition to regular employment as well as THA work is largest during the first two months after entering unemployment, reflecting stigmatization of individuals who have longer unemployment durations and depreciation of human capital as the unemployment spell elapses.

The multivariate normally distributed error terms v RE and vTHA enter the model specification as random intercepts. The estimated variances of the random terms are not statistically significant. This may indicate that unobserved heterogeneity doesn’t seem to play an important role in the relevant transitions, but may also be a result of averaging. There may well be unobserved heterogeneity between different groups in the sample, but on average the effects cancel each other out.

6.

Conclusion and suggestions for further research

This paper investigates the stepping stone hypothesis of THA work for later employment prospects for non-western immigrants in Norway. Using the timing-of-events approach (Abbring and van den Berg (2003)), which assumes no anticipation of treatment, I find that

the individual transition rate from unemployment to regular employment is significantly and substantially raised as a result of THA work. This is not due to selection into THA work, as this has been taken account for through the inclusion of correlated random effects. The estimated results show that women, older individuals, individuals with small children and lower education and who have stayed in Norway only a few years are less likely to exit unemployment to both regular and THA work.. Married individuals are more likely to find a job in the regular labor market, while those living in Oslo are more prone to enter a THA from unemployment. Immigrants from Africa have a larger probability of experiencing a transition to THA work, but are less likely to have a direct transition to regular employment. Immigrants from South America are more likely to exit unemployment for both destination states. I have only considered the effect of THA work on direct transitions to regular employment. If the THA spell ends with a transition to any other state than regular work, including unemployment, the spell is right-censored. It is natural to think that THA work may have two different effects: One effect during the THA-spell and another effect after the individual leaves the THA. These effects can easily be incorporated into the model outlined above, but at the cost of a substantially increased computational burden. Another expansion of the model would be to let the treatment effect depend on observable individual characteristics. This would be especially interesting for policy makers, as it makes it possible to investigate further who benefits the most from THA work.

6.

References

Abbring J. and G.J. van den Berg (2003), “The Nonparametric Identification of Treatment Effects in Duration Models.” Econometrica, 71:5 Amuedo-Dorantes C., M.A. Malo and F. Munoz-Bullon (2008), “The Role of Temporary Help Agencies on Workers’ Career Advancement”, Journal of Labor Research 29:2 Anderson P. and E. Wadensjö (2004), “Temporary Employment Agencies: A Route for Immigrants to Enter the Labour Market?” IZA DP N.1090 Arrow, K.J. (1972), ”The Theory of Discrimination” in O.C. Ashenfelter (ed.) Discrimination in Labor Markets, Princeton: Princeton University Press Arrowsmith, J. (2006), ”Temporary agency work in an enlarged European Union”. European Foundation for the Improvement of Living and Working Conditions, Dublin Autor D. and S. Houseman (2006), “Temporary Agency Employment as a Way out of Poverty?” in R. Blank, S. Danziger and R. Schoeni (eds.) Working but Poor: How Economic and Policy Changes are Affecting Low-Wage Workers, New York: Russell Sage Becker, G.S. (1964), Human Capital: A Theoretical and Empirical Analysis, with Special Reference to Education, Chicago: University of Chicago Press Booth A.L., M. Francesconi and J. Frank (2002), “Temporary Jobs: Stepping Stones or Dead Ends?”, Economic Journal, Royal Economic Society, 112:480 Bring C.N. (2007), “Nonparametric identification of the Mixed Hazards Model with Time-Varying Covariates”, Econometric Theory, 23:2 Doeringer, P.B. and M.J. Piore (1971), Internal Labor Markets and Manpower Analysis, Lexington, Mass: D.C. Heath Göbel C. and E. Verhofstadt (2008), “Is Temporary Employment a Stepping Stone for Unemployed School Leavers?”, ZEW DP 08-093 Heckman J. and B. Singer (1984), “A Method for Minimizing the Impact of Distributional Assumptions in Econometric Models for Duration Data”. Econometrica, 52:2 Heinrich C.J., P.R. Mueser and K.R. Troske (2005), “Welfare to Temporary Work: Implications for Labor Market Outcomes”. Review of Economics and Statistics, 87:1 Hole A.R. (2007), “Fitting Mixed Logit models by Using Maximum Simulated Likelihood”. Stata Journal, 7:3 Hole A.R. (2007), MIXLOGIT: Stata Module to Fit Mixed Logit Models by Using Maximum Simulated Likelihood. Statistical Software Components S456883, Boston College Department of Economics Ichino A., F. Mealli and T. Nannicini (2005), “Temporary Work Agencies in Italy: A Springboard Toward Permanent Employment?”. Giomale degli Economisti e Annali di Economia, 64:1 Jenkins S. P. (2004) Survival Analysis, Unpublished manuscript, Institute for Social and Economic Resarch, University of Essex, Colchester, UK

(Downloadable http://www.iser.essex.ac.uk/teaching/degree/stephenj/ec968/pdfs/ex968lnotesv6.pdf)

from

Kvasnicka M. (2008): “Does Temporary Help Work Provide a Stepping Stone to Regular Employment?” in D.H. Autor (ed.) Studies in Labor Market Intermediation (National Bureau of Economic Research Conference Reports), Chicago: Chicago University Press 2009. Longva F. (2002): ”Midlertidig ansettelse – karrierestige eller jobbfelle?” Søkelys på arbeidsmarkedet 2002, årgang 19. Manski C.F. and S. Lerman (1977) “The estimation of choice probabilities from choice based samples”, Econometrica, 45, 1977-1988 McCall B.P. (1994), “Identifying State Dependence in Duration Models”, American Statistical Association 1994 Vol: Proceedings of the Business and Economics Section, No. 14-17 Mortensen, D.L. (1986), “Job Search and Labour Market Analysis”, in O.C. Ashenfelter and R. Layard (eds.) Handbook of Labor Economics vol. 2, Amsterdam, North-Holland Nergaard K and J. Svalund (2008), ”Utleie av arbeidskraft. Omfang og utvikling.”, Fafo-notat 2008:25 Nicoletti C. and C. Rondinelli (2009), “The (Mis)Specification of Discrete Duration Models with Unobserved Heterogeneity: a Monte-Carlo Study”, Banca d’Italia Working Paper No. 705 Torp H., P. Schøne and K.M. Olsen (1998), “Vikarer som leies ut: Hvem er de og hvilke arbeidsvilkår har de?” ISF Rapport 98:11 Train K. (2003), Discrete Choice Methods with Simulation. Cambridge University Press Van den Berg G.J. (2001), ”Duration Models: Specification, Identification and Multiple Durations” in J.J. Heckman and E.E. Leamer (eds.) Handbook of Econometrics, 1:5, Elsevier Van den Berg G.J. and J.Vikström (2009), “Monitoring Job Offer Decisions, Punishments, Exit to Work, and Job Quality”, IZA DP N.4325 Zijl M., G.J. van den Berg and A. Heyma (2004), “Stepping Stones for the Unemployed: The Effect of Temporary Jobs on the Duration until Regular Work.” IZA DP N. 1241 Zijl M. and M. van Leeuwen (2005), “Temporary jobs: intermediate positions or jumping boards. Searching for the stepping-stone effect of temporary employment.” SEO Discussion Paper No. 38