Social experiments of different kinds of counselling for newly unemployed

Unemployment is intensely debated everywhere in the European Union, not least in Sweden where youth unemployment and late labour market entry are of growing concern in the policy discussion. Although the forecast points to a general improvement of the Swedish labour market, the policy makers in Sweden are still facing some major challenges in how to tackle the increasing unemployment for certain groups in the society, such as the youths and the older population. Finding effective active labour market policy (ALMP) tools to tackle youth unemployment is important to mitigate “scarring” of young people (exposure to unemployment at the time of labour market entry has negative consequences for future earning.

The purpose of the proposed project is to analyse and evaluate whether different types and intensity of counselling and coaching by Swedish PES caseworkers matter and can improve the employment opportunities of newly registered unemployed workers. This will be tested through a set of social policy experiments implemented within the PES organisation where newly registered unemployed are randomly assigned into treatment (intensive and/or a certain type of counselling) or control (regular counselling). The specific treatments involved are early and intensive individual counselling meetings between a caseworker and an unemployed; early and intensive individual distance meetings, either through telephone and/or online (e-counselling), between a caseworker and an unemployed; and early and frequent group meetings. The average treatment effects are likely to vary across individuals, i.e. they are heterogeneous, which motivates an investigation of differential impacts by gender, age, education etc. The design of the experiments enables us to conduct an evaluation of the different types of ALMP related to counselling. It also enables us to estimate their relative cost-effectiveness and to say something about displacement effects.

The set of actions envisaged as part of the policy intervention is to randomly assign newly registered unemployed into a control and treatment group. Individuals that are randomly assigned into a treatment group receive one of the following types of treatment during their first three months of unemployment:

1) Intensive individual personal meetings (one meeting every other week)
2) Intensive individual distance counselling (one contact every week)
3) Intensive group meeting (one meeting every week)

Individuals that are assigned into the control group take part of regular counselling activities at the PES. The intensity of these meetings differs for different groups of unemployed. As of today, there is one personal meeting on average with a caseworker every once in a month during the first three months of unemployment.

Meetings between caseworkers and the unemployed can have both monitoring and counselling functions and are important elements of ALMP. Unemployed workers first register their entry to unemployment in a first individual and personal meeting (face to face) at their local employment office. In later meetings, search efforts are monitored, counselling is given with referral to vacant jobs and eventual programme participations are discussed. Meetings may also help unemployed to maintain or increase their job search activities which in turn may lead to better job opportunities. Intensive meetings and increased knowledge about the unemployed can also influence the labour market's mode of operations, especially when the quality of the caseworkers’ direct job matching of unemployed and vacant jobs is high.

The general picture from research is that there is a strong positive impact of frequent individual contacts between caseworkers and newly registered unemployed. However, frequent individual contacts require large resources for the Swedish PES who administrates these meetings. If the same positive effects of frequent contacts could be achieved through contacts via telephone or online or through group meetings, then a more efficient resource utilization at PES could be enhanced.

We intend to employ a two-level randomisation design with randomisation at local employment offices level as well as on individuals. This will also enable us to test for externalities on the untreated unemployed and to investigate whether the different kinds of meetings may have some general equilibrium effects. The literature on this is unfortunately very limited.

2015-03-09

2016-02-28

Evaluating the policy interventions, we focus on outcomes related to employment and measures of improved employability. Our main variable of interest is unemployment duration, i.e. to what degree the intervention reduces the number of days to employment. We will also estimate employment probabilities at certain intervals after 3, 5 and 12 months after the intervention. It is also important to explore the type of job (permanent, temporary, subsidised), for example to what extent the intervention increased the probability of getting unsubsidised full-time jobs. The probability of leaving unemployment for education as well as participation in other types of ALMPs will also be analysed.

Indicators on individual’s job search activity, use of formal and informal search methods etc. will be explored. The probability of leaving unemployment for education as well as participation in other types of ALMPs will also be analysed.

The population from which the participants to the experiment will be sampled are all newly registered unemployed at the Swedish PES of age 18--64. Our total sample of eligible participants will be successively drawn during March 2015 to November 2015. This was inflow during 12 months in the original proposal, but changed to inflow during a short period due to practical restrictions at the Swedish Public Employment Service.

To test whether the early-stage ALMP interventions matters for newly unemployed, a randomized control trial design will be applied. To allow for evaluation of potential externalities on untreated unemployed, we will employ a two-level randomisation design with randomization at both the level of the local employment offices as well as the level of the individuals. In other words, we will first randomly decide the offices where the interventions will be implemented. Thereafter, newly registered unemployed individuals, at the offices selected for the experiments, will randomly be assigned to take part in the intervention (treatment group) or to be treated-as-usual (control group).

In the first step, 72 local employment offices are randomized into taking part in the experiment. Among these 72 offices, 36 offices will be control office and the remaining 36 offices will randomized into each of the three interventions resulting in 12 offices for each experiment. To increase precision, the randomization of offices will be made within stratified groups. Here, we will make use of the cluster-model developed at the Swedish PES, sorting local employment offices into groups according to similarities in external factors such as characteristics of the local population and labour market. Randomisation will therefore take place within each cluster making sure that offices with similar characteristics end up both in the treatment and control group.

To be able to test for displacement effects and to investigate whether the different kinds of interventions may have some general equilibrium effects, the 36 control offices will be used as comparison offices.

In areas with more than one local employment office, displacement effects may not only exist for unemployed at the office where the intervention took place. Negative effects for individuals registered at neighboring local employment offices sharing the same local labour market therefore have to be considered not to underestimate the total magnitude of the displacement effects. To be able to isolate displacement effects for a given intervention, a necessary condition is that another type of intervention is not implemented in an office sharing the same local labour market. Randomization will therefore be performed under the restriction that only one type of intervention is implemented in each local labour market, for example by implementing the interventions in different geographical regions.

We have two-level randomisation design with randomization at both the level of the local employment offices (randomization by coin flip) as well as at the level of the individuals (randomization by date of birth (day of the year)).

First level: Randomizes 72 offices to 36 comparison offices, 12 offices with face-to-face meetings, 12 offices with distance meetings, and 12 offices with group meetings.

Second level: within the 36 active offices randomizes 50% to treated and 50% to non-treated.

We have two-level randomisation design with randomization at both the level of the local employment offices and at individual level.

Yes

72 local public employment offices.

All newly registered unemployed at the 72 offices during March 2015 to November 2015 will be included in the experiment. A priori it is difficult assess the exact number of observations. Instead, the restricting factor is the experiment can take place at these 72 offices for a limited time period during 2015. A very low estimate from the original proposal was that we should have at least 11,500 participants.

36 comparison offices (5620 non-treated)
12 offices with face-to-face meetings (1875 treated, 1875 non-treated)
12 offices with distance meetings (1875 treated, 1875 non-treated)
12 offices with group meetings (1875 treated, 1875 non-treated)

We consider power calculations for the three treatments: face-to-face, distance and group meetings. For each treatment, all newly unemployed workers who satisfy the eligibility criteria will take part in the experiment. Below we show calculations based on a total sample size of 11,250 individuals, but we note that this is a conservative estimate, and we will use the full inflow at offices in the experiment in the analyses. With total sample size of 11,250 over 72 offices each treatment is expected to have 1875 individuals, of which 50% will be treated. Thus, for each treatment we have around 935 individuals in the treatment group and 935 individuals in the control group. The main outcome of interest is the job-finding rate. For the power calculations an average job-funding rate of 50%, i.e. we consider a binary outcome with mean equal to 0.5. We set in all power analyses the significance level to 5% and the power to 80%. This gives a minimum detectable effect size of 0.0648.

This paper uses a large-scale two-level randomized experiment to study direct and displacement effects of job search assistance. Our findings show that the assistance reduces unemployment among the treated, but also creates substantial displacement leading to higher unemployment for the non-treated. By using detailed information on caseworker and job seeker behavior we show that vacancy referrals passed on from caseworkers to job seekers is the driving mechanism behind the positive direct effect. We also examine explanations for the displacement effect and show that displacement is not due to constrained resources, but arises in the labor market. A comparison between different meeting formats suggests that face-to-face meetings and distance meetings are more effective than group meetings. Despite the existence of displacement effects, when we incorporate our results into an equilibrium search model we find that a complete roll-out of the program would lead to lower unemployment and slightly reduced government spending.

Cheung, M, J Egebark, A Forslund, L Laun, M Rödin & J Vikström (2019), “Does Job Search Assistance Reduce Unemployment? Experimental Evidence on Displacement Effects and Mechanisms”, IFAU Working Paper 2019:25.

https://www.ifau.se/globalassets/pdf/se/2019/wp-2019-25-does-job-search-assistance-reduce-unemployment.pdf