Negotiating Informality: Bargaining Practices and Preferences of Domestic Workers

We will carry out an information treatment in the survey where we provide workers with information on
both:
• The average wage in the sector in Northern Italy using data from the Labour Force Survey covering
both formal and informal work.
• Minimum wage brackets from the Collective Bargaining Agreement (CCNL Lavoro Domestico
2024).

2025-04-30

2025-11-30

Research Question 1: How do domestic workers trade-off wages, formality, and commuting time
in their labor supply decisions?
Outcome: Binary variable coming from the force choice between two job offers in the conjoint experiment.
Respondents are presented with 4 pairs of job offers in which different attributes are
randomized and have to choose which one they would prefer. They will also rate each offer
separately on a likert scale, and this rating will also be a dependent variable in the regression
for the main question.
• Forced choice variable: Which of the following two jobs offers do you prefer?
Binary choice variable: Job offer 1, Job offer 2
• Rating variable: Please rate both Job offers from 1 to 5 (5-point scale)
1 = Would always reject this job offer
2 = Would probably reject
3 = Not sure whether I would accept or reject
4 = Would probably accept
5 = Would always accept this job offer


Research Question 2: What characteristics predict bargaining intentions? Does an information
treatment providing the average wage in the sector and minimum wage according to the CBA affect
bargaining expectations?
The key dependent variables are past and current bargaining practices.
Specifically, the answers to the following questions:
• Since you started working in the domestic and care sector, did you ever ask an employer for
an increase in wage? [Yes, more than once; Yes, once; No]
• (Regarding main job) Have you ever asked for a wage increase? [yes, and I have obtained it;
yes, and it was denied; no, but I would like to; no, I don’t want to]
• (Regarding main job) After the first offer was made, did you ask for a salary increase?
• (Information treatment outcome) If you asked for a higher wage now, how much more do you
think you could maximally receive?
• (Information treatment outcome) What do you think is the percent chance that you could get
this increase? [scale 0-10]

We implement a two-part experimental design embedded in a survey of domestic workers in Milan.
First, we use a conjoint experiment to elicit preferences over job attributes by presenting respondents with randomized pairs of hypothetical job offers that vary along three key dimensions: net hourly wage, contract status (full, partial, or no formal contract), and commuting time. Each respondent is shown four such pairs and asked to both choose between the two offers and rate each one individually on a Likert scale. This design enables estimation of the average marginal component effect (AMCE) of each attribute on job choice.
We will consider two subpopulations within our sample, each receiving the same experimental design: elder care ("badanti") and cleaners. If an individual is employed as both, he or she will be classified by the job that takes up the most hours. However, the conjoint experiment will be slightly different, according to the type of work carried out: individuals that work as "badante" (elder care) will choose between two badante jobs, while individuals that work as cleaners will choose between two cleaning jobs. The badante and cleaning jobs will have some attributes in common (presence of a contract, commuting time, wage), but will also differ in some other attributes that are more job-specific (e.g. for badante, gender of the person cared for).

Second, we introduce an information treatment, randomly assigning a subset of respondents to receive accurate information about average wages in the sector and the minimum wage brackets outlined in the collective bargaining agreement. We then assess the impact of this information on respondents’ expectations about their wage bargaining potential with current or future employers. Again, information will differ slightly, according to the individual's main occupation.

Randomization will be done automatically by the server carrying out the survey.

Randomization will be carried out at the individual level

No

There will be no clusters.

We plan to recruit between 1200 and 3000 domestic workers for the study. The final number will depend on recruitment feasibility. Participants will be randomly assigned to either treatment or control conditions in equal proportions.

Participants will be randomized individually into two arms:
Treatment arm: 600 to 1500 individuals
Control arm: 600 to 1500 individuals
The final distribution will ensure balance across arms and enable sufficient power to detect meaningful differences in outcomes.

We aim to recruit approximately 1200–3000 participants, split equally between treatment and control. With 80% power and 5% significance, this range allows us to detect treatment effects between 0.1 and 0.15 standard deviations. Due to limited prior data on this population, effect size estimates are conservative. For our conjoint experiment, detecting an Average Marginal Component Effect (AMCE) of 0.05 (a commonly cited benchmark) requires at least 600 observations per profile condition—translating to a total sample size of 1200. To enable subgroup-level analysis (e.g., differences between cleaners and babysitters), a larger sample of up to 2340 may be needed.