Barriers to investment in management training: Experimental evidence from middle and large Ethiopian firms
Last registered on April 22, 2019


Trial Information
General Information
Barriers to investment in management training: Experimental evidence from middle and large Ethiopian firms
Initial registration date
April 17, 2019
Last updated
April 22, 2019 11:32 AM EDT

This section is unavailable to the public. Use the button below to request access to this information.

Request Information
Primary Investigator
Stanford University
Other Primary Investigator(s)
PI Affiliation
Stanford University
PI Affiliation
Stanford University
PI Affiliation
Policy Studies Institute
Additional Trial Information
On going
Start date
End date
Secondary IDs
A growing literature suggests that the lack of quality management keeps many developing country firms at a low level of productivity. Why don’t firms invest in the general management skills of their workers? A possible reason is that employers may capture only part of the benefits of training because trained managers get recruited by other firms. In this project, we test whether turnover risk contributes to low firm investment in managerial training through an RCT with a sample of firms in and around Addis Ababa. This work is done in conjunction with the collection of a longitudinal dataset.
External Link(s)
Registration Citation
Abebe, Girum et al. 2019. "Barriers to investment in management training: Experimental evidence from middle and large Ethiopian firms." AEA RCT Registry. April 22.
Experimental Details
See Experimental Design
Intervention Start Date
Intervention End Date
Primary Outcomes
Primary Outcomes (end points)
- Training take-up;
- Characteristics of managers nominated for training.
Primary Outcomes (explanation)
- Take-up: Whether the firm nominates managers for the training lottery; whether they send the manager if they win the lottery
- Characteristics of the managers nominated for training (gender, relationship with the top manager).

We will look at heterogeneous effects with respect to exposure to domestic and foreign competition. We will also apply a machine-learning approach to heterogeneity analysis such as those suggested by Athey and Imbens (2016) and Chernozhukov et al (2018). This will allow us to identify significant dimensions of treatment effect heterogeneity out of a large set of potential sources of heterogeneity (e.g. firm size, sector of activity, age, productivity, geographic area, wage structure, experience with training, turnover rates).
Secondary Outcomes
Secondary Outcomes (end points)
- Training courses chosen.
Secondary Outcomes (explanation)
Experimental Design
Experimental Design
We test whether firms in Ethiopia are more interested to train their middle managers when expected turnover is experimentally lowered.

Top manager of firms involved in the SEDRI panel dataset will be invited to participate in a lottery to access a program that supports the training of middle managers at a local training institute. The institute offers weekend and evening courses in management and leadership targeted to professional managers. To participate in the lottery, the top manager has to complete and return a nomination form by a known deadline. The module contains several questions about the middle managers nominated for training and about the firm. We define “take-up” of the lottery as follows: completion of the nomination form and its delivery to implementing partner. We define take-up of the training as follows: whether the firm pays the training fee (subsidized) and send the manager to the training if they win the lottery. We will consider take-up of the lottery as our primary measure of interest in management training, assuming that take-up of the training among the winners will be large. We will measure take-up of the training for validation purposes.

Firms who win the lottery will get (i) a sizeable discount on the price of up to three training courses, and (ii) a guarantee that the trained managers will be eligible to receive lump-sum payments one year and two years after the end of the training. We introduce two treatment arms. In one treatment arm (“retention”), this payment to the manager is conditional on remaining employed at the firm that initially sponsored their training. In the other arm (“unconditional”), there is no such condition: trained managers receive the lump-sum payments unconditionally. Variation in bonus conditionality aims to generate change in expected manager turnover and hence in the training benefit that can be expected by the firm. In addition, we also vary the level of the training subsidy in order to assess willingness to pay for such training and identify financial constraints to take-up. We cross-randomize bonus conditionality and level of the subsidy, resulting in six different experimental conditions.
Experimental Design Details
Not available
Randomization Method
We rely on an existing panel dataset data of about 1,200 medium to large firms in the periphery of Addis Ababa – the SEDRI panel survey. All firms surveyed in 2017 will be randomly assigned to one of the treatment arms. To be offered the intervention, a firm has to have at least one middle manager at the time of our interview in 2019. We plan to add firms to this sample through a snowball method to ensure a sample size of about 1,200 firms despite potential attrition in the 2017 sample.

Firms will be allocated to one of the six experimental conditions using a matched randomization procedure. We first construct groups of six firms that minimize the Mahalanobis distance over a set of eight variables measured during the 2017 survey (or the 2019 survey for firms in the snowball sample). These variables include: firm size, firm age, sector of activity, distance from the training institution, eligibility score, wages, access to manager training, and turnover rate. We then randomly allocate one firm in each group to each one of the six experimental conditions.
Randomization Unit
The randomization unit is the firm.
Was the treatment clustered?
Experiment Characteristics
Sample size: planned number of clusters
Sample size: planned number of observations
In 2017, 827 firms had at least one middle manager. We expect that this figure will be larger in 2019. We plan to add firms to the SEDRI sample to increase the sample if needed.
Sample size (or number of clusters) by treatment arms
We plan to allocate half of the observations to the unconditional bonus condition and the remaining half to the retention bonus condition. One third of the observation in each experimental condition will then be assigned to, respectively, the low, medium and high level of the training discount.
Minimum detectable effect size for main outcomes (accounting for sample design and clustering)
IRB Name
Stanford IRB
IRB Approval Date
IRB Approval Number