Our main hypothesis is that many of the young female workers who quit the position have not spent enough time on the job to learn the true quality of the match, which, for some of them, may be high. Standard economic theory argues that worker turnover is an essential part of the process through which economies achieve an efficient allocation of talent: new hires are uncertain about the quality of the match with their employer and they quit their job when they become convinced that it is a poor match (Jovanovics, 1979). However, there are several reasons to expect that quitting decisions may be made before the quality of the match has been fully revealed. First, workers do not internalise the costs that high levels of turnover impose on the firm: more resources have to be devoted to recruitment, planning production and meeting deadlines becomes more difficult, the investment in training the worker is lost. In markets where access to informal work opportunities is relatively easy, this can create an asymmetry between the cost of turnover to the worker and the cost of turnover to the firm. Second, early signals about the quality of the match may be misleading.  Workers may eventually adjust to many features of the work environment that are initially perceived as disamenities, but often underestimate the extent of this adjustment (a phenomenon called “projection bias”, e.g. see Loewenstein et al. 2003). Workers that leave their job early – which represent the bulk of the turnover problem faced by firms (Jovanovics, 1979; Donovan et al. 2019) – may thus be quitting sooner than optimal. This can be particularly detrimental for young, female workers that have recently migrated and are only weakly attached to the labour market and thus have noisy priors about match quality in different jobs.

We work in a particularly interesting context: the Hawassa industrial park, the flagship economic project of the current government, where a number of foreign investors have recently started producing garment-manufacturing products. This is a new activity in this area and most workers are new to the industry. Further, standard factory-floor positions, which make up the bulk of the workforce of the firms in the park, are virtually only taken up by young women. We have identified a firm in the park that is willing to randomize certain features of their pay scheme. We have already collected some data on the workers of the firm and have run a small pilot.

In the experiment, we will randomize the timing of a retention bonus offered to workers. A first group of workers will be offered a bonus that is paid after three months on the job, which incentivises them to complete the initial training (which lasts for 2 months). A second group will be offered a bonus that is paid after seven months on the job, which incentivises workers to spend a significant amount of time on the production line.

Our key empirical test will be whether the late bonus generates higher retention in the long-term (i.e. after the seventh month of tenure) compared to the early bonus. We will collect detailed beliefs data to support our interpretation and to rule out alternative explanations (e.g. search costs that rise with tenure or gains in job-specific skills). Second, we will study whether workers in the late retention bonus group have higher levels of earnings and work satisfaction, one year after the offer of the bonus. This will test the misallocation hypothesis: turnover destroys jobs that are a good match for the worker. Third, we will study how the late bonus intervention changes the propensity to leave of workers at different levels of the productivity distribution.

Fieldwork is ongoing.