Remote Work Communication and Team Performance: The Effects of Flexible Work Schedules

We plan to test our research questions using a field experiment implemented on the global online labor market, UpWork (as in, for instance, Burbano, 2021). We will advertise two job types requiring workers to translate a 2-page text from English to Swahili, and Bengali respectively and hire teams of two to complete each job. Aside from the language the document is to be translated in, the task types are identical.

We have selected a translation task for our research setting for several reasons. First, there is a clear measure of performance on the task; specifically, how many errors the translator made (word error rate), and how readable the translated text is (translation readability scores are summarized in, for instance, Yeung et al, 2018). Second, it requires creativity and problem solving (e.g. Angelone, 2010) and teammates may benefit from being able to effectively communicate. Google translate and other freely available AI translation software are not yet able to optimally translate and we will clarify in the job posting that Google translate has not worked for our purposes. Third, it ensures we will hire teammates who speak similar languages and are, thus, able to communicate. Fourth it is a relatively affordable task on the virtual labor market we are using. Specifically, one page translations like the task we are hiring for cost about $10-$20 to complete on Upwork.

Team members will be asked to complete the task within 24 hours of being hired and receiving the job instructions, which they will receive immediately after being hired. Giving workers 24 hours to complete the job is a norm for short jobs on UpWork, and will give workers enough time to communicate with their teammates even if they are residing in different time zones and/or have different work schedules. Moreover, this will give them plenty of time to complete the task because evidence suggests one page of text takes about 1 hour to translate (e.g. professional translators have an hourly productivity of about 625 words per hour (Haji Sismat, 2016)).

Upon being hired, both workers in each team will be provided the document to be translated and asked to coordinate in order to decide how to work on the task so that each team member has an equal amount of work to do. They will also be provided with a private Slack channel, and instructions on how to download and use Slack, to communicate with their teammates. In order to reduce the incentives for free-riding and to allow us to quantify the extent of free-riding, workers will be asked to highlight the work they did on the final translation.

Job instructions will also inform workers that their manager is planning to hire them for a subsequent translation job in 1-2 months if their teammate is also available to be re-hired. This information is true as we hope to bring all teams back for a second task to investigate whether the value of ease of conversation changes with more mature work teams. Moreover, it provides workers with some incentive to invest in the relationship with their teammate.

2022-06-01

2023-05-31

Outcomes measured both for first and second task:
Completion of task
Translation quality
Communication quantity and quality

Outcomes only measured from first round of hiring:
Quality of job applicants

Quality of job applicants - measured by relevance of prior experience and education for task, prior ratings and experience on UpWork
Completion of task - measure of whether a final output was submitted by the team, and whether the submission is fully or partially complete
Translation quality - measured both by comparing submitted output to original document, and by obtaining readability scores from 2 native language speakers per submission
Communication quantity - measured by number of lines typed by each teammate in Slack, and imbalance of lines typed between team members

Willingness to accept second job offer
Survey-based measures of experience working with teammate and on the task

To causally test our research questions, we will randomly vary the expected flexibility of work schedules in order to test whether schedule flexibility affects job applicant characteristics, ease of coordination, and the quality of output. The specifics of our treatments will be disclosed publicly after the study has been run, and the authors are happy to respond to questions about them via email in the meantime.

To causally test our research questions, we will randomly vary the expected flexibility of work schedules in order to test whether schedule flexibility affects job applicant characteristics, ease of coordination, and the quality of output. In particular:

Treatment 1 – Flexible: Half of job postings and associated instructions will not make any mention of the specific hours during the 24-hour work period that workers should coordinate with each other.

Treatment 2 – Coordinated: Half of job postings state that we will be asking hired applicants for two-hour windows during the 24-hour work period when they can work with their teammate on the task. Once hired, workers will be surveyed on which two-hour windows they have available to work. We will subsequently inform teammates of the earliest overlapping two-hour work windows during which we expect them to work together.

To identify the work-windows during which workers hired for coordinated jobs are available, we will ask all applicants to list all the two-hour windows of time they have available to work on the task between the target start and end time for their work. If workers do not get back to us in time to start the job at our target time, we will ask them to send us their availability lists for the subsequent start time. As the job posting already clarifies that we will be asking for this information, it should not be strange that we subsequently do.

The specific language we will use to solicit this information is:
“Thank you for applying to this job. Please list all the 2-hour windows of time you have available to work on this job between X:XX and Y:YY.”

Thus, all workers will have the opportunity to communicate with their teammates but half of the workers will have no information about when their teammates are working, and the other half will have information about which window of time their teammates are expected to be working.
To study whether schedule flexibility matters more or less as teammates know each other better, all hires will be offered a subsequent job 1-2 months following the completion of the first job and hires will be teamed up with the same worker they were assigned to work with during their first job. We will re-randomize teams into flexible or coordinated treatments, stratified by their first period treatment.

Because of the importance of understanding how remote worker schedule flexibility affects outcomes, we are interested in analyzing the text and verbal communication that occurs between teammates. Thus, in addition to informing workers that we are giving them a Slack channel to use for their task-based communication, we will also be informing workers that foul and/or disrespectful language is not acceptable in their communication with their teammate, and that we will be monitoring their communication.

We will randomize one of our translation languages into one of the two conditions (flexible or coordinated) using a coin flip. The other language will be automatically assigned the opposite condition. Once we have reached the target of 100-110 teams for a given language under that condition, we will run the opposite condition for that language.

We will re-randomize all teams into the coordination and flexible conditions for the second round using Stata's generate uniform distribution number assignment command (gen uniform()). This randomization will be stratified by the first period treatment assignment.

We are randomizing at the level of the translation language in the first period. In particular, one language will start with flexible job posts, and the other will start with coordinated job posts, and switch the former to coordinated and the latter to flexible once the target sample size for the first assigned treatments have been reached. We are not randomizing treatments by language across days because it might seem strange to job applicants of a given job that we are changing our job postings daily. This will mean that some job will finish the first treatment before the other. We will wait for both languages to be done with the first treatment before starting the other treatment.

In the second round, we will randomize at the level of the team.

No

410-420 teams of two works (820-840 workers in total)

400-440 teams

200-220 teams per treatment arm, or 100-110 teams per language treatment arm
100-110 teams, Bengali translation, coordinated condition
100-110 teams, Bengali translation, flexible condition
100-110 teams, Swahili translation, coordinated condition
100-110 teams, Swahili translation, flexible condition