What do we know about gender and private sector development in Africa? Markus Goldstein Alaka Holla The World Bank.

What do we know about gender and private sector development in Africa?

Markus GoldsteinAlaka Holla

The World Bank

Very little

• What we do know – Descriptive statistics

• Can tell us about existence of gaps• But often mostly limited to formal sector (Enterprise

Surveys)• Cannot tell us welfare implications of closing gaps

– Impact evaluations in small enterprise finance• Will present some evidence today

What we know less about

– Effectiveness of BDS for women entrepreneurs– Effects on enterprise outcomes (by gender)– Spillovers to non-enterprise outcomes and total

welfare effects– What policy interventions can close gaps? Is it

desirable to do so?

Outline

1. Basic descriptive statistics

2. Evidence from impact evaluations in small business finance

i. Credit: access and take-upii. Investment and income responsesiii. Impacts within the household

3. Lessons for future impact evaluations

Outline





Women are less likely to be enterprise owners (all firms)

Relative to men’s, women’s enterprises are more likely to be family businesses

But reported constraints and productivity are similar

• No discernible gender pattern in reported constraints– Anticompetitive and informal practices, corruption,

access to finance, labor regulation, tax administration

• Similar productivity performance– Value-added per worker– Total factor productivity

(Bardasi, Blackden & Guzman 2007)

Outline





Interventions in small business finance

• If access to finance is a barrier to growth, improving access to finance should improve enterprise outcomes

• Randomized interventions with evaluations help us move beyond reported constraints and correlations

• Given that business and personal finances often intertwined, there could also be spillovers to household outcomes.

• Main question of interest: Are the impacts of these interventions different for men and women?

What do we mean by different?

• We are testing whether the hypothesis there is no gender difference in impact is true– In jargon: the gender difference is not statistically

distinguishable from zero at standard confidence levels

• Two ways this hypothesis could be true:1. We can’t tell. Our estimates are so noisy as to be

indistinguishable NO information for policy2. The difference is a well estimated zero policy relevant result

• How can we know if we are in situation 1 or 2?– Inverse power calculations: tell us the smallest possible gender

difference we could have distinguished from zero

The interventions – Kenya(Dupas and Robinson)

• interest-free savings accounts in village banks for men and women (small entrepreneurs) w/ penalty for withdrawal, plus opportunity to become Bank members and thus get loans

• Randomly assigned individuals to receive account or not

• Use surveys, administrative data, and daily log books

The interventions – Malawi(Gine, Goldberg, and Yang)

• Looks at how borrowers respond to the lender having better information on their credit record

• Randomly assigned paprika farmers clubs applying for a loan to control group or one where members were fingerprinted as part of the application

• Data collection included administrative records and household socioeconomic surveys

The interventions -- South Africa(Karlan, Zinman, and others)

• Worked with big consumer lender in SA that provides loans to small businesses

• Evaluations cover a range of different randomized interventions: 1. Manipulating content in loan advertisements mailed

out to former borrowers (loan terms, selling points in advertisement)

2. Varying the interest rates offered to loan applicants and the final interest rate that applies to their loans

3. Extending credit to marginally ineligible applicants• Use administrative and household survey data

The outcomes

• Impact evaluations measured multiple outcomes

• For some interventions, possible to track outcomes in 3 different domains:– Credit access, take-up, and default rates– Business investment and income responses– Household expenditure and health

• Can get a broad picture of impact

Savings accounts in KenyaOutcome Impact for

womenGender

difference in impact

Smallest detectable difference

Credit and borrowing

ROSCA contributions

18.33 -17.94 |23.23|

Income & investment

Business investment 252.89 -152.83 |420.97|

Household Food expenditure 24.13 -16.24 |26.24|

Numbers in red = statistically significant

Fingerprinting borrowers in MalawiOutcome Impact for

womenGender




Loan size309.27 -1253.14

|1190.32|

Income & investment

Market sales 32248.16 -29698.637

|39629.29|

Profits97690.57 -98806.78

|96649.46|


South Africa: Content of loan advertisements Intervention Outcome Impact for

womenGender




Offered interest rates

Apply for loan -0.003 0.0002 |0.002|

Default0.034 -0.039 |0.033|

Longer deadline to apply

Apply for loan 0.076 -0.020 |0.024|


South Africa: Extending credit to marginally ineligible applicants

Outcome Impact for men

Gender difference in

impact


Household Depression -1.18 1.53 |1.40|

Outcome Impact for women

Impact for men Statistically distinguishable?


Have micro loan 0.129 0.119 No

Have loan from formal source

-0.083 0.008 No

Household Food consumption -0.023 0.232 No

Control and outlook 0.159 0.196 No

Some common patterns

• Aside from mental health, the limited evidence we have sometimes points to closing of gender gaps (if it is significant)– None of these interventions were explicitly

designed as “pro-women”• Ok – so should we go ahead and scale-up

these interventions?– Hold on– We need more evidence

Outline





We need more and better evidence

• Many gender differences not statistically significant– Could be large or small – we cannot say with a lot

of confidence

• But if we had the right sample size, well estimated zero results would be informative – If the policy is aimed at a documented gender gap,

it has failed to address it– If the policy is not aimed at a gender gap, men and

women are affected equally

So what are the problems now? Many IEs already done did not collect

enough observations to tell

A well estimated zero is often not reported in publications (publication bias)

Gender analysis isn’t always done

We need more and better evidence

• Not even clear how significant results would extend to other areas within countries and across countries– Program participants not randomly selected from

population– How does context matter?

• Location• Sectors• Scale of enterprise

Need to test these interventions in different contexts

How to get to more evidence

• The way to better policy is impact evaluations that are well designed to capture gender differences

• This does not mean (only) projects targeted at women – this is a small fraction of what we do

• It means understanding gender-differential effects in the important projects (e.g. large budget, pressing issue, innovative design)

What do better evaluations look like?

• Start from an understanding of what existing gender issues are in your target population

• Think about causal chain of the project and how it might be different for men and women and then choose outcomes of interest accordingly

• Make sure the data is sufficient for this, and that the analysis gets done

And the good news…

• There is some promising work in the pipeline and you will hear about some of it this week

• But this is only a (small) start• As you design your impact evaluations – think

about how your program may have different effects by gender and how you could measure this!

Thank youMerci

Obrigado

What do we know about gender and private sector development in Africa? Markus Goldstein Alaka Holla The World Bank.

Documents