Incentives and Culture: Evidence from a Multi-Country Field Experiment

We founded model data entry firms in three countries with different culture but fairly similar GDP per capita. In all three countries we implemented three classic pay packages: our control group received fixed wages, whilst treatment is divided into individual incentives (paid by keystroke per hour) and group incentives (paid by the average keystrokes per hour in a group of four). To shut down income effects we set the piece rates so that a worker with median productivity is expected to earn the same. Finally, we generate exogenous variation in the visibility of individual reqards by cross-randomising both treatments with the publication of performance rankings.

2010-12-20

2012-02-23

Worker productivity measured as log keystrokes per hour. Quality-adjusted worker productivity measured as log keystrokes per hour adjusted for the number of erroneous data entry points.

Keystroke is defined as “a keyboard action that results in data capture in the used program.” This was measured directly by the data entry program.

Log profits for firm per worker-contract.

From an experimental perspective, choosing Ghana, India and the Philippines allowed us to explore a variety of countries as measured by their Hofstede scores for individualism. According to Geert Hofstede’s measures of Individualism, Ghana scores a 15, near the bottom of the Hofstede measures for Individualism, while India scores a 48 in the middle of the field and the Philippines a 32. In the course of the experiment, we interact country-level individualism with the incentive contracts.

We ran two two-day contracts per week over the course of a year and a half, with an experimental duration of ten months, one year and nine months in Ghana, India and the Philippines, respectively. To avoid spillovers, all workers in a given two-day contract received the same contract (i.e. all workers in the office at a given time received flat wage or IPR). We provide these workers with a written contract to sign that laid out the contract terms. Following the brief orientation, we assigned workers to one of ten workstations and the on-site manager gave a fifteen-minute Power Point training. By using a Power Point presentation to train workers, we eliminated variations in training quality that could have occurred over the course of the experiment.

The data entry interfaces captured the total number of key strokes entered by workers over the course of the two-day contract. In addition, the data entry interface captured key timing variables, including the daily start time, end time and the number of seconds to complete one survey. At the end of the two-day contract, the on-site manager paid workers in cash. For workers under the flat wage contract, we did not carry out a salary calculation, and paid each worker a cash wage of 50 GHC, 500 INR or 900 PhP. For the five incentive contracts, the on-site manager collected all data entered by the workers and calculated their pay using based on the conditions of the workers’ assigned contracts. Workers received cash along with a receipt stating the number of key strokes entered and the calculation used to define their two-day wage.

We invited all applicants who scored above 65% accuracy on the baseline Epidata test to at least one contract and a maximum of two contracts. Occasionally, workers who confirmed attendance would not show up for contracts. Despite inviting ten workers to each contract, we found that approximately seven workers attended each contract on average. To limit this phenomenon, we invited workers to work on an alternative date in the event that they were unable to attend on their originally assigned date. If a worker did not answer his or her phone after at least two attempts to contact him or her, we eliminated them from consideration for future contracts. We explore the possible impact attrition has on the study and find no significant relationship between treatment type and attrition.

Stratified randomization done in office by a computer after each interview round. Stratification variables: race/ethnicity, baseline ability, gender.

The unit of randomization was a worker-contract (a two-day contract). Hence one worker could be randomized again to another treatment if the worker worked 2 two-day contracts. However, to avoid spillovers, all workers in a given two-day contract received the same contract (i.e. all workers in the office at a given time received flat wage or individual piece rate).

No

1583 2-day contracts. Some workers had more than 1 contract, but maximum 2 contracts.

1583 2-day contracts.

Pooling all 3 countries: 153 Flat wage contracts. 307 Individual piece rate contracts. 306 Group piece rate contracts. 307 Individual piece rate with public ranking contracts. 306 Group piece rate with public ranking contracts. 204 Modified (lower-powered) individual piece rate contracts.

Divided by country: 459 contracts in Ghana, 571 contracts in India, 553 contracts in Philippines.