Frictions and misperceptions in firm take-up of active labor market programs

The objective of the main intervention is to: i) provide accurate information about job training and wage subsidies, which are the two most prominent active labor market programs in Norway, and ii) reduce frictions to sign-up for the programs by providing a one-click sign-up link.

The intervention is implemented as part of a survey sent to the managing director in Norwegian firms (in the period April 22 to June 30, 2025). Each firm is randomly assigned to either a control group or one of three intervention groups, where the different intervention groups allow us to identify the impact of information and sign-up frictions in isolation and jointly (see the Experimental Design section for details).

T0. Control
T1. Sign-up link
T2. Sign-up link + General program information
T3. Sign-up link + General program information + Candidate and experience information.

All groups answer a block of background questions and the mediator questions described in the section Secondary Outcomes. In addition, T1 provides a one-click link at the end of the survey which takes respondents to the webpage where firms can indicate that they are interested in the program and become contacted by Nav (the Norwegian Labour and Welfare Administration). This link is not provided to the Control group, but they can of course sign-up by finding the webpage on their own, or by contacting Nav directly. T2 also includes the sign-up link and in addition provides general information about the program. T3 is identical to T2 but in addition provides information about the average candidate and firms' experiences of using the program. The control group (T0) does not have a link at the end of the survey nor sees any information about the programs. Also, to make sure that all groups spend roughly the same amount of time taking the survey, both T0 and T1 answers a block with 6 filler questions (while T2 and T3 are exposed to the information).

2025-04-22

2025-06-30

The primary outcomes relate to firm take-up and use of the programs. This will be measured in the following way:
1. Sign-up (1 if firm sign up for program at Nav webpage, 0 otherwise).
2. Take-up (1 if firm has had at least one candidate within 12 months, 0 otherwise)
3. Program use (# of candidates at firm within 12 months)

We have two sets of secondary outcomes that will be used in an explorative analysis.

1. Mediators: At the end of the survey (after the information treatment but before the sign-up decision) we asked some questions intended to shed light on why the information treatment did (not) affect the decision to sign-up. We had two primary motives in mind: i) Usefulness for candidate and/or firm (see M1-M2 below), ii) Perceived program popularity (see M3-M4 below). In particular we asked:
M1: How useful do you think the program is to your firm (scale 1-5)
M2: How useful do you think the program is for the candidate (scale 1-5)
M3: How many firms in Norway do you think use the program (scale 0-100%)
M3a: Imagine that there are twice as many businesses as you think that use the program. Would that make you more or less willing to try these programs? (scale 1-5)
M3b: Imagine that there are only half as many businesses as you think that use the program. Would that make you more or less willing to try these programs?(scale 1-5)
M4: How important do you think it is that your particular firm tries to help people by using the program? (scale 1-5)

If anything, we expect the information treatments to have a positive effect on both the usefulness dimension and the perceived popularity of the program. However, while we expect an increase in perceived usefulness to increase the likelihood for a firm to sign-up for the program, an increase in the perceived program popularity could either have a positive or negative effect on the likelihood for a firm to sign-up for the program: positive due to social norm and negative because of a free-rider motive (more on this in the Experimental Design section).

Before the sign-up decision we also asked a question about whether they would be interested in testing the program (Yes/No), which could be useful in explorative analysis.

2. Supplementary outcomes: From the register data we can track firm outcomes that may be positively or negatively affected by the use of active labor markets programs. These outcomes include:
S1: Number of employees
S2: Employee turnover
S3: Days of sick leave per employee
S4: Employee diversity (share women, share native, share without high school degree)
S5: Profits

As described in the Experimental Design section, we do not have any strong priors on these outcomes, but, if anything, we expect more pronounced effects in firms with fewer employees.

Participants are recruited via a survey provider that sent out letters to the managing director of all firms in Norway with more than 4 employees and with no experience of wage subsidies or job training in the last 3 years. The letter includes a link and QR code to an electronic survey, and we expect about 10 percent of all firms that received the letter to answer the electronic survey, and these firms constitute our sample.

The letters were sent out on a rolling basis starting April 22, 2025. Survey data collection will stop June 30, 2025, and this also marks the end of the intervention period (i.e., exposure to treatment). After that date, survey data will be made available to researchers, and this data will later be matched to register data from Statistics Norway and Nav covering the period from April 22, 2025, to June 30, 2026 (at least 12 months after the intervention period has ended).

As described in the Intervention section, firms are randomly assigned to one of four treatment groups:

T0. Control
T1. Sign-up link
T2. Sign-up link + General program information
T3. Sign-up link + General program information + Candidate and experience information.

T3 is the main treatment capturing the impact of reducing both sign-up frictions and misperceptions (by comparing T3 to T0). T1 is included to be able to identify the pure effect of reducing frictions (by comparing T1 to T0) and the pure effect of reducing misperceptions (by comparing T2 and T3 to T1). T2 is mainly included to understand whether the more specific information about candidates and firms' experience add something over-and-above the general program information that is already available on Nav's webpages (this will be tested by comparing T3 and T2).

The design allows us to test the following main hypotheses for each of the three main outcomes:
H1: One click sign-up and accurate information increases program sign-up, take-up and use (T3 > T0)
H2: One-click sign-up increases program sign-up, take-up and use (T1 > T0)
H3: Accurate information increases program sign-up, take-up and use (T3/T2 > T1)

As a more powerful test of accurate information, we will also do an analysis where we pool T2 and T3 (and compare it to T1). If we cannot reject that T2 and/or T3 are equal to T1, we will also consider a more powerful test of one-click sign-up by pooling T1, T2 and T3 (and compare it to T0).

Notice that we predict positive treatment effects of both the sign-up link and the information. A negative effect of one-click sign-up is difficult to reconcile theoretically, although it may well be zero. The provision of information on the other hand could potentially have a negative impact for at least two reasons: First, employers may, in contrast to what we expect, be negatively surprised by the information we provide which potentially could reduce perceived program usefulness. Second, the information may positively (or negatively) affect employers beliefs of other firms' use of the program (i.e., program popularity), which can reduce the specific firm's willingness to sign-up and use the program. We ask questions in the survey to be able to address both these possibilities (see M1-M4 in Secondary Outcomes). Note also that the possibility of a positive sign-up effect (T1>T0) and a negative information effect (T3 < T1) may result in a null effect for the combined treatment (T3 =T0 )

We have three sets of background variables: i) variables captured in firm register data, including the number of employees, sector, industry code, etc., ii) variables based on 4 questions from the survey capturing the firm's demand for low-skilled labor, and iii) variables based on 6 questions from the survey capturing prior knowledge and beliefs wrt wage subsides and job training. In addition to reporting unconditional treatment effects, we will report results conditional on a set of control variables with the objective to maximize precision in the estimated treatment effects. We will also use the background variables for heterogeneity analysis. In general, we expect the treatment effects to be driven by firms with more demand for (low-skilled) labor. Regarding prior knowledge and beliefs, we expect the information treatments to have larger impact on firms with a low score on the 6 questions (as these are firms with overly pessimistic priors), while we do not have a strong prediction for the sign-up link on this heterogeneity dimension (it could even be more important to remove the sign-up friction for firms with correct/optimistic priors). We will also do a heterogeneity analysis wrt firm characteristics such as sector and number of employees, without having any strong priors.

We do not have strong priors for the supplementary secondary outcomes (S1-S5), as they can be both positively or negatively affected, and most likely not affected at all. A null effect on these outcomes would, however, be positively interpreted as it suggets that firms can use active labor market programs without deteriorating effects on the regular work force.

Not available

Randomization done by survey provider before letters are sent to firm.

A firm (organization number) is the unit of observation.

No

We expect to have about 4000 firms answering the survey.

We expect to have about 4000 firms answering the survey.

Firms are randomly assigned to treatments with equal probability, meaning that we expect about 1000 firms per treatment group.

Two of our main outcomes are dummy variables (sign-up and take-up). We can assume that the sign-up and take-up rates are quite low in the control group (no one has used the program before and less than 10% of all firms in Norway use the programs). Hence in these power calculations we assume a baseline of 10%. What is the smallest treatment difference we can detect using a two sample proportion test with 5% significance level, 80% power, and N=1000 in each group? The answer is a 4 p.p. difference, which is equivalent to about 0.1 SD. Using a t-test would reduce power to 43% for the same effect size, but, we would have more than 80% power to detect an effect size of 0.13 SD (6.5 p.p.).