Active Labour Market Policies in Addis Ababa

This study evaluates the impacts of three different active labour market policies for unemployed youth in Addis Ababa, Ethiopia. We look at the impacts of the following interventions: i) a transport subsidy to lower the costs of searching for employment in the centre of the city, ii) a screening intervention where individuals are given a range of personnel tests and issued with certificates reporting their outcomes for these test (this intervention is given with training on how these certificates can be used to apply for jobs), and iii) a job fairs intervention, where individuals are invited to attend an event where they can interact with a number of large firms looking looking for recruits. The project also seeks to estimate the spillover effects from treated individuals to untreated individuals in neighborhoods where a randomly varying proportion of unemployed individuals are given the treatments.

External Link(s)

Registration Citation

Citation

, et al. 2015. "Active Labour Market Policies in Addis Ababa." AEA RCT Registry. October 13. https://doi.org/10.1257/rct.911-1.0

Former Citation

, et al. 2015. "Active Labour Market Policies in Addis Ababa." AEA RCT Registry. October 13. https://www.socialscienceregistry.org/trials/911/history/5591

Sponsors & Partners

There are documents in this trial unavailable to the public. Use the button below to request access to this information.

Request Information

Experimental Details

Interventions

Intervention(s)

See Section 4 of our (publicly available) pre-analysis document attached.

Intervention Start Date

2014-09-01

Intervention End Date

2015-02-28

Primary Outcomes

Primary Outcomes (end points)

See Sections 5 and 7 of our (publicly available) pre-analysis document attached.

Primary Outcomes (explanation)

See Sections 5 and 7 of our (publicly available) pre-analysis document attached.

Secondary Outcomes

Secondary Outcomes (end points)

Secondary Outcomes (explanation)

Experimental Design

See Section 3 of our (publicly available) pre-analysis document attached.

Experimental Design Details

Randomization Method

See Section 3 of our (publicly available) pre-analysis document attached.

Randomization Unit

See Section 3 of our (publicly available) pre-analysis document attached.

Was the treatment clustered?

Yes

Experiment Characteristics

Sample size: planned number of clusters

234 local areas.
See Section 3 of our (publicly available) pre-analysis document attached for more detail.

Sample size: planned number of observations

4096 individuals. See Section 3 of our (publicly available) pre-analysis document attached for more detail.

Sample size (or number of clusters) by treatment arms

See Section 3 of our (publicly available) pre-analysis document attached.

Minimum detectable effect size for main outcomes (accounting for sample design and clustering)

We anticipate that our job search interventions will have impacts on a number of job‐search behaviors, type of employment, duration of unemployment spells, and wages. However, finding full‐time,permanent and formal employment in Addis Ababa is challenging, and thus a change in this single binary outcome will likely be the most challenging impact to detect. Our study has been designed and the sample size chosen with the intention of having adequate power to detect this effect. Our power estimates should therefore be understood as the most conservative, lower bound estimates of the power of our design. We calculate power through detailed simulation designs (corroborated with analytical results from comparable and simpler designs, run in the software package “Optimal Design”) in the presence of large spillovers (contamination) and possibly high intra‐cluster correlation (ICC) of up to 30%. ICC’s are introduced through the simulation of a cluster specific employment shock. In these “worst case scenarios”, we are able to detect a 5 ppt increase in the probability of employment on a baseline probability of 10% more than 80% of the time, with the size of the test at the 5% significance level. It should be noted that in the binary outcomes framework this effect is equivalent to a one sixth (1/6) standard deviation (sd) standardized effect size, much smaller than the usual rule of thumb of 0.2 sds suggested by Cohen (1998). There are numerous ways in which we expect to have more power by looking at various other intermediate outcomes, or more easily influenced job‐market outcomes. Recall that the binary model estimates above could be estimated at each period in our high frequency data set, giving us the power to estimate detailed impact trajectories. However if we wanted to look at average outcomes over time we might have considerably more power from pooling all post‐treatment observations. As argued by McKenzie (2011), this is especially useful when outcomes of interest are measured with error or are weakly correlated over time, which could certainly be the case for temporary employment outcomes. When employment is relatively informal, wages and hours worked could be irregular and difficult to recall. At a single snapshot, treated individuals might not be significantly more likely to be employed, but the cumulative effect averaged over a few months might be significantly different.

Supporting Documents and Materials

IRB