Unwilling to reskill? Evidence from a survey experiment with Italian jobseekers

The survey instrument consisted of five parts:
1. Screening: To be eligible to participate, a respondent should be i) between 18 and 55 years old, ii) not working, iii) actively looking for a job and iv) not enrolled in any training course.
2. Working and training history: we asked respondents about their past training and work experience, whether they felt their past job was part of their identity, the skills needed in their previous job, expectations of wage, training effectiveness and job finding rates for the main job they are searching for. We also asked some information about their job search (e.g, hours spent searching, number of occupations).
3.Attitudes towards new occupation: we randomly assigned every survey respondent to a ``treatment occupation" (IT assistant or construction technician). We then asked expectations of employer's demand, wages, job finding rates, fit with own identity, confidence in skills and interest for this occupation.
4. Discrete choice experiment: respondents are asked to choose between six training options with randomized characteristics, as described in the attached PAP.
5. Demographics and unemployment: we collect further demographic information such as whether the person receives unemployment benefits, duration of unemployment, time spent in household chores and personality traits.

2022-12-06

2023-09-30

Our primary outcomes are meant to capture interest in the two high-demand treatment occupations, and in undertaking specialized training programs to access these occupations (i.e., reskilling).
Thus we focus on two primary sets of outcomes: one for interest in treatment occupation, and one for interest in reskilling.

Interest in treatment occupation
The survey asked respondents to rate on a scale from 0 to 10 their interest in the treatment occupation. This is our main measure of interest for the treatment job. We will use the raw version of the survey variable, as well as a dummy for having interest above median in the sample.
We will also construct six additional versions of the variable for interest in the treatment job by residualizing the survey response to six different set of controls.

Interest in reskilling
We will define interest in reskilling by using respondents' answers to the six different discrete-choice scenarios they saw. Reskilling for us means preferring a vertical training (i.e., specific training preparing for an in-demand occupation) compared to a generalist training, or no training at all. First, for every respondent we will compute two variables of interest in reskilling: i) the share of scenarios (out of six) in which s/he picked a vertical course as first choice and ii) whether s/he ever picked a vertical course as first choice in any of the given scenarios. Second, we will also explore interest in reskilling by computing the average ranking given to the vertical training, horizontal training and opt-out by a respondent across scenarios. Third, we will quantify the increase in marginal utility and the willingness to pay for a vertical training course in our discrete-choice model.

We introduce two between-subjects randomized treatments in the surveys:

Treatment 1. After screening, the survey software randomly assigned respondents to one of two high-demand occupations: IT Assistant or Construction Technician. Assignment was stratified by gender. We refer to the occupation assigned to each respondent as ``treatment occupation".

Treatment 2. In order to study possible frictions to the demand for training, we introduced experimental variation within the survey right before the discrete choice experiment in part 4. Respondents are randomized into the following three groups:

1. Information (INFO): before asking for their training choices, respondents are provided with information on vacancies and average wage in the treatment occupation.
2. Role model and growth mindset (MINDSET): before asking for their training choices, respondents are shown a “success story” of a person who succeeded in reinventing herself after a training and learn about the concept of “growth mindset”.
3. Control group (CONTROL): no additional information is provided before training choices.


The discrete choice experiment in part 4 of the experiment also contained randomised variation within and between subjects. Within subjects, we randomised the characteristics of training programmes. Between subjects, we randomised the block of choices seen by a person. A person could be randomly assigned to 9 different blocks, which differed in the training options (3 versions) and in the order of visualisation of the training features within each option (3 versions). See the attached PAP for details.

Randomization done by the survey software.

Individual randomization, with stratification by gender.
The assignment of the INFO/MINDSET/CONTROL treatment was also stratified by treatment occupation.
The assignment of a block of training choices was stratified by gender, treatment occupation and INFO/MINDSET/CONTROL treatment.

For all treatments, we also stratify by respondents' sample.

No

Survey panel sample: 600 respondents
Jobcentres sample: between 400 and 800 respondents.

Same a planned number of clusters

Within each gender, even split in six arms that combine a treatment occupation (IT or Construction) with an information treatment (INFO/MINDSET/CONTROL).
Within each arm, we also evenly assigned the block of training choices.

Pre-Analysis Plan

Survey Company - Document about Data Sharing