Effects of Insurance on Child Labour: Ex-Ante and Ex-Post Behavioural Changes

Southern Asia

Pakistan

Social Protection

Health insurance

M Frolich; A Landmann

2018

Frölich, Markus, and Andreas Landmann. “Effects of Insurance on Child Labour: Ex-Ante and Ex-Post Behavioural Changes.” The Journal of Development Studies 54.6 (2018): 1002–1018.
View publication

NRSP branches Pakistan: https://nrspbank.com/contact-type/nrsp-branches-pakistan.

2018

N. A.

2009-2011 (baseline and endline surveys)

To combat a very large presence of child labor in a city in Pakistan (Hyderabad).

The program provides accident and health insurance to its credit clients, mostly women. For all credit clients in the sample, this insurance is mandatory for the client, her spouse, and her children below 18 years. The insurance premium was automatically deducted from the loan.

The program analyzed was rolled out by the National Rural Support Program (NRSP), a large not-for-profit microfinance institution in Pakistan. In 2009, the insurance was extended in nine randomly selected branch offices out of 13 branch offices in urban Hyderabad (some form of microinsurance existed since 2005). While the control branches provided insurance as before, in the treatment branches two modifications were introduced at the same time. First, additional household members were now also offered insurance voluntarily. This means that adult children and other members of the client’s household (for example aunts, grandparents, and cousins) could buy insurance for a fixed premium per person. Second, in all treatment branches, clients were assisted with claim procedures. Right after the baseline, 100 percent of the clients have a loan and are thus covered by mandatory insurance. In later periods the coverage rates of clients decrease because clients repay their loans and after loan repayment are no longer insured. (After loan repayment insurance cannot be extended unless a new loan is taken.) The coverage of the client (who is mandatorily insured in treatment as well as control villages) decreases to about 75 percent at the 18-month follow-up. Hence, while the coverage rates of clients decrease over time, they are very similar in the control and treatment branches. On the other hand, the number of insured individuals is much larger in the treatment branches because only households in treatment branches had the option to voluntarily insure additional household members. At six-month follow-ups, about 70 percent of those voluntarily insurable were insured. This number decreases to about 50 percent at 18 months because clients who repaid their loans no longer have access to insurance.

• Child labor
• Child schooling

Difference-in-differences

Study design/Identification Strategy: the authors rely on the randomized design of the program and can control for small-sample imbalances in pre-treatment covariates. The authors were also able to disentangle the effect of the two innovation components (assistance with claims, extended coverage of household members) by estimating treatment effects for those households consisting only of individuals with mandatory insurance. Those should not be affected by the voluntary extension of coverage, but only by the assistance with claims. They contrast the effects for ‘mandatory households’ with estimates for households including voluntary members to obtain an estimate of the coverage effect, using triple difference techniques. For this triple difference approach, they used a constant treatment effect assumption.
Sample: Data on credit, insurance status, and particularly child labor was collected in all branches at baseline, that is before the modification of the insurance package, and in four follow-up surveys. This large panel data set with baseline and follow-up surveys allows the identification of treatment effects using difference-in-difference techniques.
Survey Instrument: the dataset covers all clients from 13 NRSP branches who had their credit appraisal in September/October 2009. These are 777 households in the control and 1320 in the treatment branches. A baseline survey was conducted before the innovation and all households were re-interviewed every six months in four follow-up waves: March/April 2010, October/November 2010, May/June 2011, and October/

November 2011.

The authors develop a model that shows that risk-averse households respond to high risk by using child labor as an additional precautionary income source. If households are sufficiently poor and afraid of the shock, they will react to the introduction of insurance by reducing child labor even without a shock taking place. They estimate the effects of a health insurance extension in Pakistan, which was implemented in the form of a randomized controlled trial. The health insurance package was modified in two dimensions: (a) an extension of insurance coverage that also offered insurance to extended household members, and (b) regular visits helping microcredit clients with claim procedures. The authors found reductions in child labor, hazardous work, and earnings generated through child labor. To disentangle the effects of extended coverage and regular claim assistance visits the authors use the feature that certain household types are completely covered by mandatory insurance and cannot extend coverage. Then they isolate the effect of regular visits for those households with only mandatory members. These households by definition have the same coverage in treatment and control branches and can serve as an additional control group within treatment branches. Using this triple difference estimator, the authors find that the main effect of the innovation is caused by extending insurance coverage to other household members. The extension reduces child labor incidence by around 10 percent, weekly hours worked by children by around four and days missed at school by around one. Monthly visits alone, on the other hand, have little significant effects.

Child Labour

Households

Local

An insurance extension reduced child labour in Pakistan. Children (5-17) in treatment households were less likely to be in child labour or hazardous work, worked fewer hours, and their earnings were lower (exact magnitude differs by follow-up survey wave).

School attendance and monthly school days missed were not impacted by the microinsurance intervention.

An aspect outside the model is that child labor can be seen as a tool to diversify labor market risk in a volatile economic environment. This idea relates to the literature on portfolio choices with different degrees of risk exposure. In the area of agriculture, high consumption risk seems to deter poor farmers from investing in more profitable but risky activities. In their case, if households are afraid of health events they might reduce risk in other domains, for example by diversifying labor market activities (including using child labor). Insurance creates a more secure environment in which households feel more comfortable abstaining from sending their children to work.