NEW UPDATE: Completed trials may now upload and register supplementary documents (e.g. null results reports, populated pre-analysis plans, or post-trial results reports) in the Post Trial section under Reports, Papers, & Other Materials.
Social Networks and Productivity Spill-over within firms
Last registered on August 29, 2014


Trial Information
General Information
Social Networks and Productivity Spill-over within firms
Initial registration date
August 29, 2014
Last updated
August 29, 2014 11:15 AM EDT
Primary Investigator
University of Warwick
Other Primary Investigator(s)
Additional Trial Information
On going
Start date
End date
Secondary IDs
I study the effects of implementing a management routine at randomly chosen sewing lines within garment factories in Bangladesh, to obtain evidence on the sources of observed productivity spill-over within factories. Bangladeshi garment factories are organized into several sewing lines which produce independently from another for different orders and buyers. Sewing lines producing garments that have already been produced by other lines in the factory before, are significantly more productive in the first days producing the new garment. Initial analysis by the author shows that these `spill-over' effects are stronger when the supervisors on the first and later lines to produce the garment report social ties in surveys. This intervention is designed to obtain additional evidence on the sources of the observed spill-over. Several garment factories agreed to implement a new management routine among randomly selected sewing lines. If a selected line start producing a garment that has already been produced on another line in the factory, the line supervisor from the first line producing the garment is actively involved in setting up the new line for the garment. If this intervention magnifies observed productivity spill-over, this would be indicative for spill-over being caused by sharing of knowledge between supervisors from the lines, as opposed to superiors or other persons (engineers, planners) applying their knowledge obtained on earlier lines producing the garment to later lines producing it, or other possible confounding sources.
External Link(s)
Registration Citation
Menzel, Andreas. 2014. "Social Networks and Productivity Spill-over within firms." AEA RCT Registry. August 29. https://doi.org/10.1257/rct.495-1.0.
Former Citation
Menzel, Andreas. 2014. "Social Networks and Productivity Spill-over within firms." AEA RCT Registry. August 29. http://www.socialscienceregistry.org/trials/495/history/2659.
Experimental Details
We work together with the management and the head production engineers in several Bangladeshi garment factories to implement the intervention. Participation in the project is still advertised, therefore the exact number of factories in the project is not yet determined. The intervention is of the following nature: Whenever a sewing line in the factory starts producing a new garment that it had not produced before but that was already produced on another sewing line, than the supervisor of the line that already produced the garment is actively involved in setting up the new line for producing the same garment. Sewing lines in our Bangladeshi partner factories typically switch the garment they produce every 1-3 weeks, and it is very common to see garments being produced on more than one line, while usually starting on different days. Switching the garment typically requires both changes of the machines used and/or adjustments on the machines themselves, such as change of the needle type or yarn type. The workers on the lines also need to be explained the new operations they have to perform on the new garment. This process of the style change can take between 1 hour and 1 day (but is usually done in the `zero-feeding' process, that is, in the line, one machine after another machines is adjusted, and the machines not yet adjusted still produce for the old garment while those already adjusted already work on the new garment. Thus, on average, each machine is idle only 15-30 minutes).
In the randomly selected lines, whenever they switch to a garment that has already been produced on another line in the factory, the line supervisor from the earlier line should be actively involved in the described change process. He should brief the new line supervisor on the most important things he has learned while producing the garment on his line, and inspect the line changing process, either after the end of the line changing process or while the most important machines for the production of the new garment are set up.
To monitor the intervention, we left with the factories either a tablet or a book, in which they should fill out a short survey every time a style change on a treatment line was "treated". The survey asks about the line that starts the garment and the line that already produced it, involved line supervisors, the garment itself, and a few questions on how well the communication went as perceived by the superiors of the line supervisors that instructed the briefing. The tablets send each report directly to a server upon finishing, while the books are collected regularly. Thus, the extent of the implementation, and whether it was implemented in the correct lines, can be checked throughout the experiment is running.
Intervention Start Date
Intervention End Date
Primary Outcomes
Primary Outcomes (end points)
The outcome variable measuring the impact of the management intervention is daily productivity of a line on the first X days it produces a garment which it has not produced before. Productivity will be measured in three different ways. The first is using the formula that also the partner factories themselves are using to calculate line productivity for their own production monitoring: daily piece-wise output is multiplied with the garment specific "SMV" value ("Standard Minute Value"). This value for each garment is calculated by the factory's engineering department prior to the start of production, and gives the required minutes of labor input to finish all the necessary sewing operations, under ideal production circumstances, and given the machinery available at the factory. It is calculated both to form a basis for a price quote for the order to the buyer, and to help plan the required labor for producing the order. Multiplying piece-wise output with the SMV gives an `output-minutes' value which is comparable across different garments. To obtain the first productivity measure, this output-minutes value is then divided by the "input-minutes" value, the number of workers on the line times the number of hours the line was operating on that day times 60 minutes.
In a second measure, output_minutes are directly regressed on the explanatory variables, controlling for input_minutes as right-hand-side variable.
Third, a productivity measure will be constructed which resembles more closely TFP measures. Average hourly output_minutes on a day are regressed on the input minutes, separated by the two main types of workers in the lines ("operator" and "helper"), and the residual is taken as productivity.

Primary Outcomes (explanation)
Secondary Outcomes
Secondary Outcomes (end points)
Secondary Outcomes (explanation)
Experimental Design
Experimental Design
The intervention is carried out by the upper factory management ("General Manager", "Head Production Engineer"), under our guidance. The implementation is tracked by the tablet or book system explained above. Lines are randomly selected to obtain the treatment. In the first three factories, to test several forms of randomization, In one part of the factory, the lines were selected together in terms of sewing floors, that is sewing floors (typically containing 5-10 lines) were randomly selected. In other floors, the randomization occurred on the line level. This was done to test whether it is feasible to directly randomize on the line level. As in the initial three factories, the randomization on the line level seems to have been understood and followed, from the forth factory starting the implementation, randomization will be solely on the line level, to avoid cluster sample size concerns.
Experimental Design Details
Randomization Method
In front of factory managers in computer (MS Office Excel random number drawer)
Randomization Unit
In the first three factories starting the intervention, the floors of the factories were separated into two groups (factories in general strive to keep lines and floors as interchangeable as possible, to allow maximum flexibility of allocating orders to lines, given the volatile order situation of most factories, with all factories in my sample having at least 10 different buyer). In the one group of floors, randomization was on the floor level (which typically contain 5-10 lines). In the other group, the randomization was done on the line level, to test if the factory management can implement this comparatively more complex randomization. As floors are deliberately kept as interchangeably as possible, the groups of floors for floor or line level randomization, respectively, were determined simply by assigning group level randomization to all floors except the last one, and line level randomization to the last floor (with last floor meaning the floor with the highest number in the internal floor counting system of the factory). From the fourth factory on implementing the program, the randomization will be at the line level only.

Was the treatment clustered?
Experiment Characteristics
Sample size: planned number of clusters
250. generally, randomization will be on the sewing line level. However in parts of the first three factories, randomization occurred on the sewing floor level.
Sample size: planned number of observations
5000 first instances of a line produces a garment
Sample size (or number of clusters) by treatment arms
115 clusters being treated.
Minimum detectable effect size for main outcomes (accounting for sample design and clustering)
Based on data from the first two factories in which the intervention started, on average, efficiency on the first day a line produces a style is increased by 7 percentage points if another line has already produced the garment before. I assume an effect of an additional 3 percentage points due to our intervention. Standard deviation of first-day efficiency is 20 percentage points. I assume per cluster on average 20 observations of a garment being started for the first time on that line (or floor). Based again on the first two factories, I calculate an intra-cluster correlation of 0.075. To detect an effect of a size of 3 additional percentage points on a significance level of 5%, I therefore need at least 227 clusters.
IRB Name
University of Warwick
IRB Approval Date
IRB Approval Number
Post Trial Information
Study Withdrawal
Is the intervention completed?
Is data collection complete?
Data Publication
Data Publication
Is public data available?
Program Files
Program Files
Reports, Papers & Other Materials
Relevant Paper(s)