How do unions solve their collective action problem?

We randomize postcards sent to union members following a successful campaign. Treatment postcards invite them to participate in a membership meeting and contain information about past participation in union activities of their co-workers, co-union members, and themselves.

2016-05-26

2016-06-22

Participation in the membership meeting on June 22nd, participation in future events, contribution to the union political action fund, and eventual retention.

Our design includes three experiments conducted on three different subpopulations (defined by prior participation of self and shop). In each experiment, any member can be allocated to one of three possible treatment conditions. Treatment postcards invite members to participate in a membership meeting and contain information about past participation in union activities of their co-workers, co-union members, or themselves. Randomization is conducted at two different levels. Treatment condition is randomly assigned at the individual level, but the fraction of coworkers that receive any particular treatment is determined at the shop level, to estimate within-firm spillovers.

Our design includes three experiments conducted on three different subpopulations.
* Experiment A: on union members who participated in at least one activity during the campaign
* Experiment B: on union members who did not participate in any activity, but have at least one coworker who did
* Experiment C: on union members who did not participate in any activity, and who do not have any coworker who did

In each experiment, any member can be allocated to one of three possible treatment conditions, determining the type of mailer that they receive.
* Condition 0 (placebo group): plain mailer
* Condition 1: invitation to the June 22nd Membership meeting
* Condition 2: invitation to the June 22nd Membership meeting ; and thanks for past participation of the individual (in Experiment A) / of coworkers working for same employer (in Experiment B ) / of other workers in the union (in Experiment C)

In Experiment A, randomization was done at the individual level:
* 1/3 members receive T0
* 1/3 members receive T1
* 1/3 members receive T2

In Experiments B and C, randomization was done at the shop level and then the individual level:
* 1/6 shops in which all members receive T0
* 1/2 shops in which 1/3 members receive T0, 1/3 T1, and 1/3 T2
* 1/3 shops in which 1/2 members receive T1 and 1/2 T2

Randomization done by the researchers in office by a computer, with Stata.

Individual and shop.

Yes

Experiment A: 1,998 members
Experiment B: 2,071 shops
Experiment C: 4,288 shops

Experiment A: 1998 members; B: 37,663 members; C: 29,209 members. Total: 68,870 members

In Experiment A, randomization was done at the individual level:
* 1/3 members (666 members) receive T0
* 1/3 members (666 members) receive T1
* 1/3 members (666 members) receive T2

In Experiment B, randomization was done at the shop level and then the individual level:
* 1/6 shops (345 shops) in which all members (6,281 members) receive T0
* 1/2 shops (1,035 shops) in which 1/3 members receive T0 (6,240 members), 1/3 T1 (6,239 members), and 1/3 T2 (6,240 members)
* 1/3 shops (691 shops) in which 1/2 members receive T1 (6,332 members) and 1/2 T2 (6,331 members)

In Experiment C, randomization was done at the shop level and then the individual level:
* 1/6 shops (714 shops) in which all members (4,951 members) receive T0
* 1/2 shops (2,145 shops) in which 1/3 members receive T0 (4,822 members), 1/3 T1 (4,821 members), and 1/3 T2 (4,822 members)
* 1/3 shops (1,429 shops) in which 1/2 members receive T1 (4,897 members) and 1/2 T2 (4,896 members)