Promoting a subsidy for hiring older unemployed workers

This RCT examines the effects of sending letters to firms promoting an existing subsidy for hiring old unemployed workers. We have three different letters and a control group. In fact, we run two experiments, one on large firms (100+ employees) and one on SMEs (10-100 employees). In our experiment, we furthermore distinguish between firms that already use the wage cost subsidy and those that do not use it, because the response of those firms to a letter likely differs.

Power analyses. The power analyses reveal that for SMEs we probably do not have enough power to be able to detect significant treatment effects of plausible sizes. We do however include them in the experiment because of their policy relevance. For large firms we do have reasonably high power.

All letters include general information about the wage cost subsidy, but some sentences are added to point to different considerations.
The first letter stresses financial motives: "In a period of 3 years the advantage can amount to 21 000 euro". The second letter stresses financial motives and loss aversion: "Make sure you do not miss out this advantage of maximally 21.000 euro". The third letter includes a peer norm: "Already a large number of firms uses this advantage which adds up to 21 000 euro in three years time".

2016-06-21

2016-06-30

There are two key outcome variables: 1. the number of hired workers aged 56 and above 2. the number of workers aged 56 and above for which a wage cost subsidy is paid. We will also study effects on hiring of other types of employees to see whether substitution takes place and, if so, from what type of employees. Lastly, we will do analysis for the subgroup of employers who, before the experiment, did hire old unemployed workers, but did not sign up for the subsidy. We expect the largest effects on the take up of the wage subsidy for these employers, since they are apparently offer jobs that match well with the subsidiy's target group and they are likely unaware of the subsidy before receiving the letter (for otherwise they would have applied for it). Moreover we will do an analysis for the subgroup of employers who, before the experiment, did not hire old unemployed workers.

Analyze the consequences of promoting the wage cost subsidy.

Note that in our data we use fiscal identities as a sampling unit. Large firms with multiple locations receive one letter at the office, otherwise they could receive several different letters. Our data do not allow us to to send different letters to the different locations.

Our full sample consists of 40,420 firms, of which 3552 are large firms that have used the subsidy in the past, 6872 are large firms that have not used the subsidy in the past, 3324 are small firms that have used the subsidy in the past, and 26672 are small firms that have not used the subsidy in the past. Within each of these four categories, firms have been randomly assigned to untreated control and three treatment groups. For each type of firm (large, small, made use, did not make use), we have equal numbers of firms in each of these four groups (control and three treatments).

Randomization done by a computer.

Firm.

Yes

- two firm size classes (large and small)
- within each class two groups: current users and non-users
- three treatment groups and one control group

40,420

40,420 firms of which 30,315 firm receive a letter.

In our first power calculations (before drawing the actual sample) the minimal detectable size for detecting a difference in hiring rates was as follows: Large firms: sample of 1400 per group, 90% power (effect 10%) and 60% power (effect 5%) Smaller firms: sample of 8000 per group, 42% power (effect 10%) (we do notice that power is too low, but for policy relevance we run this separate experiment for smaller firms) After drawing the actual sample, the numbers changed a bit. Large firms: 2.500 Smaller firms: 7.600 Power 90% (effect 10%) and power 60% (effect 5%) Smaller firms: 40% power (effect 10%) For small firms power might not be enough, only for large effects or combination of letters. Because of their policy relevance, we did not drop smaller firms from the sample, but instead run a separate experiment on them.