Young workers’ beliefs and sorting in the labor market

We will run a survey with a randomized information provision experiment on entry level wages at the job type-level.

2023-06-21

2023-08-31

1. Beliefs about: i) entry-level wages at the job type level, ii) success rates of applications, iii) wages five years after starting a career in a given job type, iv) hours worked in first year after starting a job, v) hours worked in the fifth year, vi) probability of performing well, vii) probability of getting along well with colleagues.

2. Intended job search behavior: Likelihood of applications and job offer acceptance for each of three job types that are relevant given the respondent’s field of study.

Actual job search behavior (linked administrative data): Job applications, actual job matches

Two treatment arms:

Treatment group receives information on entry-level wages in three different relevant job types based on administrative data covering the years 2010 to 2018. The treatment values are graphically illustrated, and put in contrast to the respondent’s prior estimates.
Pure control group does not receive information but each respondent is reminded of own prior estimates of entry-level wages.
Prior to the treatment, we ask about highest education degree completed, including the exact field, about past job search, and about potential social connection with each job type. We also elicit prior beliefs about the characteristics that people from the respondent’s education group experienced on average in each job type. We focus on three dimensions: i) entry-level wages at the job type level, ii) success rates of applications, iii) wages five years after starting a career in a given job type. These beliefs are incentivized: respondents get rewarded if their answer is close to the true value measured in administrative data.

Post treatment we ask about respondents’ intentions for job search. Specifically, we ask about the probability of accepting a job offer in each of the three job types conditional on receiving one, about the likelihood of sending an application to each of the three job types and about the probability of accepting a job offer in a hypothetical scenario where wages are held constant across jobs. Moreover we elicit “self” beliefs about expected job characteristics that the respondent herself would receive if she took an entry level job in each of the three relevant job types. The dimensions we cover are i), ii) and iii) above, and in addition expected hours worked in the first and in the fifth year after starting a job, and expected likelihoods of getting along well with colleagues and of performing well on the job.

Additional information on construction of education-specific job types:

Each respondent will be asked questions and possibly receive information related to three job types that are specific to her education group.
For each education group, we constructed job types based on interactions of some of the following dimensions: Occupational code (at various aggregation levels), industry code (at various aggregation levels), and firm type (Private/public or Public/small private/large private, with a cutoff of 50 employees). To select which dimensions are used for each education group, we use Danish register data on the first jobs of the past graduates from the education group. Out of all possible combinations of the above dimensions, we selected those such that the 3 most common types cover more than 40% of entry-level jobs and that the 3rd most common type covers more than 5% of entry-level jobs. If several categorizations satisfy these criteria, we kept the one that minimizes MSE(wage) (using cross-validation). Finally, we hand-checked the selected job types, editing the labels of some of the job types to make them more readable.

Not available

Based on their education background, some individuals will be allocated to a pure control group by default. This will be the case for small education programs where we do not have enough statistical power to construct reliable data driven education-specific job categories, or reliable statistics on the starting wage in recent registry data.

Randomization is done automatically by the survey software oTree. Conditional on being eligible for treatment, each respondent is assigned to the treatment group with a probability of 50 percent, and to the control group otherwise.

The randomization will be done at the individual level.

No

N/A (individual-level analyses)

We aim to contact about 48,000 individuals in June-August 2023. Based on previous research in the Danish context we expect a response rate of 20%, which gives us an expected survey sample size of approximately 9,600 individuals. We will combine two sampling procedures: (1) In June 2023 we will contact all individuals who were classified as close to graduating from the University of Copenhagen as of March 2023. This sample contains 2,976 individuals. (2) In addition, we will contact roughly 45,000 individuals from the flow of newly unemployed workers in two waves in June-August 2023. The Danish Agency for Labour Market and Recruitment (STAR) will send us several batches with the contact information of young workers (below age 40) who registered as unemployed in the three preceding weeks. We will contact each batch just after receiving their information from STAR. We plan to receive the first batch on the week of June 19, and the second batch on the week of July 10. We estimate that overall there should be about 45,000 young individuals becoming unemployed during that period, and that graduates should make up the majority of them. This is based on indirect imputations from historical data, so there is uncertainty about the exact sample size, and share of graduates in these STAR batches. Our overall sample will hence include young workers of different types: new graduates from the University of Copenhagen, unemployed new graduates from anywhere in Denmark, and young unemployed workers from anywhere in Denmark that did not just graduate. In some analyses, we will use the full sample of young workers. In other analyses, we will restrict the sample to recent university graduates, dropping the other young workers who registered as unemployed in our sampling period. Edit from August 16, 2023: As of August 15, our sample of respondents is smaller than expected: 3,798 individuals have responded instead of the aimed sample size of 9,600 respondents. This comes from two factors. First, there were fewer individuals in the flow of newly unemployed workers in June and July 2023 than we had predicted based on the flows in previous years. Therefore, we contacted 41,219 individuals instead of the predicted 48,000. This is likely because the Danish labor market is stronger than in previous years. Second, the temporary response rate---as of August 15---is around 9%, which is lower than the 20% rate we used to compute our aimed sample size (it is not definitive since the people we contacted might still respond until end of August). This is likely because we are contacting young individuals and during the month of July, which is a very important holiday in Denmark. To get closer to the initially aimed size for our sample of respondents, we will implement a third and last wave of survey later today, on August 16. We will contact the individuals who registered as unemployed between July 25 and August 15 and are 30 years old or younger. Note that the restriction to only survey age 30 or below (rather than age 40 or below in the first two waves) is determined by an external constraint, as the older individuals are scheduled to be contacted in a later survey and we want to avoid double contacting them. This third wave will allow us to contact people who might have delayed their unemployment registration to the end of the summer holiday, and are particularly likely to have graduated recently given the lower age threshold. We received their contact information on August 15 from the Danish Agency for Labour Market and Recruitment (STAR): this corresponds to 5,012 individuals. In addition to the 41,219 individuals we already contacted in our first two survey waves, this third wave will hence push our sample of contacted individuals to 46,231 individuals---close to the initial prediction of 48,000. Based on the response rate observed in this demographic group so far, we now think that it is more realistic to get to a 10% response rate by the end of the survey period. Therefore, we now aim at reaching a sample size of about 4,600 respondents.

Among the sample eligible for the randomized intervention (see above on randomization), every respondent will be assigned to the treatment group with a probability of 50 percent, and otherwise to a pure control group. The sample eligible for the randomized treatment will be a subsample of our survey sample (aimed survey sample size: 9,600 individuals). So the eligible sample size will be slightly smaller than the survey sample size, but the exact size is unknown.

Edit from August 16, 2023:
The aimed sample size is 4,600 respondents.

Unknown, due to several unknown factors, including final number of survey respondents eligible for treatment.