Trial design, recruitment, and data collection

This confirmatory study was designed as a multicenter, non-blinded, two-arm, parallel-group, randomized controlled trial (see details on methodology in (Hollederer et al. 2021), and the study design was based on two abovementioned evaluation studies from the USA and Finland (Vinokur et al. 2000; Vuori et al. 2002).

During this pilot implementation in Germany, volunteer employment agencies were asked to inform their clients about the JOBS Program and to invite them to voluntarily participate in the JOBS training and the evaluation study. If clients were interested, they were invited to an information event where they received detailed information about the training and the study. If participants provided their informed consent, they were called via telephone to conduct the first interview prior to the JOBS training (T0).

All interviews were conducted by computer-assisted telephone interviews (CATI) with Voxco software by the Institute for Social Sciences and Communication (SOKO; https://soko-institut.de/) on behalf of BZgA. During the first interview (T0), the participants were randomly assigned to either the intervention group (IVG) or the waiting control group (WCG) (1: 1 ratio) and invited to the JOBS training. After the JOBS training, both groups were interviewed a second time (as soon as possible, but within 4 weeks after the training [T1]) and a third time (about 6 months after the training [T2]). WCG participants were offered free participation in JOBS training after the study was completed.

Adults who were registered as unemployed with their local employment agency and who were able to independently complete the CATIs in German were eligible to be study participants. Due to occupational health and safety regulations during the COVID-19 pandemic, the employment agencies were not allowed to have personal contact with their clients for a large period of the recruitment (Hollederer et al. 2023). Therefore, the employees of the employment agencies were faced with the challenge of inviting their clients by phone or e-mail. The countrywide recruitment was expected to start in April 2021 and to be finished by August 2021. Due to constraints imposed by COVID-19 pandemic-related infection control measures, the main study phase of the JOBS training sessions took place between March and December 2022 (subsequent to a pretest that was conducted to test all study processes).

Intervention

The JOBS Program applies elements of social learning based on Albert Bandura's social cognitive learning theory (Bandura 1971, 1977b) and self-efficacy theory (Bandura 1977a, 1986, 2004).

The JOBS Program training is designed as a multimodal workshop, usually lasting 5 days, with 4 to 5 hours each day. Under the guidance of two certified JOBS Program trainers, participants develop and improve their practical job search skills in small groups of 15 to 20 individuals. In terms of methods, the focus is on the following basic elements and group techniques:

1.

job search skills training

2.

active teaching and learning methods

3.

trained (certified) trainers for program delivery

4.

supportive learning environment

5.

inoculation against setbacks.

Typical activities include job search on social networking sites, compiling information for interviews, simulated job interviews, thinking in terms of employer perspectives, and evaluating job offers. Another essential component of the JOBS Program is the inoculation against setbacks during the job search. To be prepared against such discouraging experiences, a stress inoculation training is applied. The group anticipates potential difficulties in specific job search situations and develops appropriate problem-solving strategies. All training content is taught using active teaching and learning methods. The aim is to identify the participants' prior knowledge and skills and incorporate them into the various exercises. These are characterized, for example, by group discussions, brainstorming, and role play. Two other essential elements permeate all training activities:

1.

Trainers provide continuous supportive feedback to participants and encourage appreciative, respectful interaction among the participants. In this way, they create an atmosphere of social support. Trainers also show empathy for participants' concerns and feelings and encourage them to use appropriate coping strategies (Curran et al. 1999).

2.

Another training principle is the so-called referent power. The trainers strive to gain high esteem, trust, and respect from the participants through competent teaching, self-revelation, reduced social distance, and empathic support.

If both components can be successfully implemented, this appreciative and supportive training situation opens up better opportunities for the trainers to exert a positive influence on the participants' self-efficacy expectations, on their self-esteem, and thus on their motivation to apply for a job (Curran et al. 1999; Caplan et al. 1989).

Specifications of the JOBS Program Germany

The JOBS training sessions were free of charge, lasted approximately 20 hours (conduct within 1 to 2 weeks) in groups of eight to 15 participants, and were led by two certified JOBS Program trainers (BZgA and GKV-Spitzenverband 2021). One trainer had to have been either qualified for adult education or professionally active in employment services. The other trainer had to be unemployed or at least to have experience with unemployment. Both had to undergo training to become certified as JOBS Program trainers. An evaluation of the JOBS Program Germany from the trainers' perspective was published by (Jahn et al. 2023).

Predictor variables and outcome measures

All predictor variables and outcomes were measured by means of a questionnaire developed by the research team at the University of Kassel. During the questionnaire-based CATI, data were collected on demographic characteristics, work and unemployment history, job search intensity, re-employment, self-esteem, self-efficacy expectations, life satisfaction, and physical and mental health. Demographic characteristics were assessed using standard survey questions concerning age, gender, marital status, education, occupation, and length of unemployment. The level of education was determined by asking for the highest level of formal schooling completed and the highest vocational qualification. For the analyses, this information was combined to construct dummy variables according to the International Standard Classification of Education (ISCED) (Bohlinger 2012; Eurostat 2022), which were divided into three levels (1 = low, 2 = medium, 3 = high). Because there were only three observations in the level 3 category at T2, we combined levels 2 and 3 for the analyses. The duration of unemployment was calculated as a continuous variable for years of unemployment. No dummy variable was constructed due to the lack of variance (only 11 [12%] participants were unemployed less than 12 months).

The outcome measure (1) “self-esteem” was measured using a German version of the Rosenberg self-esteem scale (Collani and Herzberg 2003), and the outcome (2) “generalized self-efficacy expectations” was assessed by a scale from Collani and Schyns (2014). Both scales contain 10 Likert-type items scored from 1 = “strongly agree” to 5 = “strongly disagree,” resulting in a possible range from 10 to 50 points (the higher the value, the greater the self-esteem or the generalized self-efficacy expectations, respectively). The participants’ self-evaluation of (3) their general health status was done via a Likert-type item asking “How is your health in general?” scored from 1 = “very bad” to 5 = “very good” (the higher the value, the better the self-rated general health status) (EHEMU 2010). Additionally, we examined (4) the level of unemployment-related mental burden using the unemployment-related mental burden scale from Trube and Luschei (2000). The scale contains 10 Likert-type items scored from 1 = “at no time” to 4 = “very often,” resulting in a possible range of points from 10 to 40 (the higher the value, the greater the unemployment-related mental burden). The data were analyzed exclusively by the research team of the University of Kassel.

Statistical analyses

All analyses were conducted with the SPSS version 28.0 statistical software package. We treated Likert-type items as continuous variables. For scales, based on multiple items including those for self-esteem, generalized self-efficacy expectations, and unemployment-related mental burden, we calculated additive scores according to the respective scale documentation.

We carried out standard descriptive analyses and—depending on the data measurement level, number of categories, distribution, and cell counts—conducted chi-square tests or Fisher–Freeman–Halton exact tests for categorical variables and nonparametric Mann–Whitney U-tests or t-tests in order to conduct group comparisons of continuous variables. This was done (1) to describe the sample, (2) to examine how far the randomization worked out, (3) to conduct a dropout analysis, and (4) to identify significant group differences between the two groups at T0 and T2, particularly in terms of the focused outcome variables. We additionally aimed at identifying significant changes in the outcomes from T0 to T2 within the IVG and the WCG using t-tests or Wilcoxon tests for paired samples. To identify multicollinearity, we created a correlation matrix for all variables studied. According to Field (2018), a strong correlation was assumed, with a Spearman rank correlation coefficient (rs) above 0.8.

In multivariate analyses, we used ordinary least squares (OLS) linear regression models adjusted for baseline values of the respective outcomes. We conducted sequential linear regression models that build upon each other. According to Urban and Mayerl (2018), sequential regression analysis has the advantage that it can control for the dependence of the estimate of individual predictor effects on the estimated effects of other predictors in the model. In addition, the sequential procedure can easily identify both stable and unstable predictor effects, as well as those predictors that have a strong influence on the estimate for other predictor effects. In this way, it is also possible to investigate how strongly individual estimated predictor effects are influenced by the inclusion or exclusion of other predictors in the model. By using sequential regression, the different models can be compared in terms of the increase in the coefficient of determination of the regression, and this increase can additionally be tested for statistical significance using an F-test (see last row in Tables 4 and 5). The F-test assesses the increase in the coefficient of determination that is achieved by adding additional predictor variables in an extended model. Thus, it can be observed whether the model fit is meaningfully improved via the inclusion of additional predictor variables in the regression model and whether the addition of those predictors is thus statistically reasonable (Urban and Mayerl 2018).

Model 1 (M1) included (1) the treatment indicator variable contrasting the IVG and the WCG in terms of the effect on the outcome and (2) the variable representing the baseline values of the respective outcome. The latter was performed because it is expected that the baseline (T0) value of the focused outcome has a relevant effect on the dependent outcome variable (at T2). This is especially true if the comparison groups (IVG vs. WCG) differ with respect to the baseline value. For this reason, it is recommended that baseline outcome values be included as a predictor in pre/post randomized controlled trials comparing the efficacy of two competing treatments (here, JOBS training vs. no intervention) (Wan 2021; van Breukelen 2006). Given the relevance of the baseline value of the outcome, it would not be meaningful to include only the treatment indicator variable in model 1. In model 2 (M2), demographic variables and variables representing the duration of unemployment and the level of depressive symptoms were additionally included. Those independent predictors were selected (1) on a theoretical basis or (2) if they were shown to be statistically associated at a significance level ≤ 0.2 with the outcome variable in different bi- and multivariate pretests. In the third model (M3), we further included interaction terms between our treatment indicator and the respective moderators to examine the abovementioned effect moderation (duration of unemployment, level of depressive symptoms at T0).

With the exception of the correlation analyses, where pairwise analyses were performed (Table 2), all analyses were performed as list-wise deletion (complete-case) analyses and according to the intention-to-treat principle (treating participants as if they belonged to the group to which they were originally [randomly] assigned, regardless of what treatment [if any] they received [in this case the JOBS training] (McCoy 2017)). A p-value of less than or equal to 0.05 was considered statistically significant.

Following the arguments of Ludbrook (2013), we used one-sided or two-sided p-values to evaluate our results depending on the a priori stipulated scientific hypotheses (H0) and on the alternative statistical hypotheses (H1). Since we hypothesized that the intervention would have only positive effects on the outcomes studied (H1; unidirectional), we applied one-sided significance tests (1) for the bivariate correlations between the intervention variable and the levels of the outcomes at T2 (Table 2), (2) for the IVG vs. WCG outcome comparisons at T2, and (3) for the tests whether the IVG improved from T0 to T2 (both Table 3) as well as (4) for the treatment effects analyses in the multivariate linear regression models (Tables 4 and 5). All other p-values reported are two-sided.