ethnic networks, wages and productivity in african · pdf fileethnic networks, wages and...

Ethnic Networks, Wages and Productivity in AfricanFirms

Mario Piacentini and Caglar Ozden

University of Geneva and World Bank

May 2009

Mario Piacentini and Caglar Ozden (University of Geneva and World Bank)Ethnic Hiring May 2009 1 / 20

MotivationsEthnic fragmentation and development

Is ethnic fragmentation a root cause of African poor economicperformance?

Ethnicity and Growth: Easterly and Levine (1997)Ethnicity and Public Goods: Alesina, Baqir, and Easterly (1999),Miguel and Gugerty (2005)Ethnicity and Rent-seeking: Alesina and Zhuravskaya (2008), Svensson(1998)

Most of these studies merely show correlations, few identify speci�cmechanisms through which these relationships operate.

We know little on how ethnic heterogeneity a¤ects behaviours andselection at the workplace.


Motivations

Hiring behaviour in multi-ethnic societies

Ine¢ ciencies due to employer�s preferences for his kin group can beparticularly severe in less developed countries.

Kin networks might induce redistribution of rents between employersand employees.

Excessive labor costs, not re�ecting true workers�productivity, mightexplain the poor performance of the sector.


Preferences of employers for co-ethnics, wages andproductivity3 possible mechanisms

Social interaction within the ethnic network can produce usefulinformation.

Employer�s inference on employee�s productivity and outside optionscan improve.The costs of search for workers might decrease.

Ethnic identity can be a powerful vehicle of informal incentives.

reputation mechanism can mitigate shirking and enforcement problems.co-ethnics might work more e¤ectively together.

Cultural bias in the way employers select their workers.

Taste for similarity or prejudices might lead to discriminatory behaviour.Ethnic parochialism supports exclusion or shunning of outsiders.


Contribution of this paper

Most accurate description of the extensiveness of ethnic hiring inAfrican manufacturing.

Consistent analysis of the e¤ects of ethnic hiring on both individualwages and �rm-level labor productivity.

Step forward in understanding the "mechanics" of ethnic preferencesin economic relationships.

We show that �rms are heterogenous in their preferences for ethnichiring and this heterogeneity matters for identi�cation of causale¤ects.

Once we control for heterogeneity, we �nd no evidence that ethnic�rms are less productive.


Advantages of our ghanaian employer-employee data

The ethnic link between the employer and the employee can beprecisely identi�ed.

We can compare wage and productivity e¤ects of ethnic preferencesin hiring.

We estimate wage models including characteristics of the �rm, andproduction functions taking into account composition andcharacteristics of the �rm�s labor force.

If worker variables are correlated with �rm variables, ignoringinformation from either side of the market would produce biasedestimates of the e¤ects of ethnic sorting.


Why Ghana?

Ghana�s population is a mosaic of several ethnic groups speaking overhundred local languages.

Many di¤erent ethnic groups have converged from rural areas to thesame urban centres.

Data quality: three waves of panel data (1998, 2000 and 2003),including retrospective information for 1997, 1999, 2001 and 2002.

We observe ethnic matching over more than 40 di¤erent ethnicgroups.

No longitudinal information on workers as di¤erent samples ofemployees are selected at each wave.


Ethnic sorting in Ghanaian �rmsSimulated and observed proportions


Measures of ethnic preferences

"Same Ethnicity": dummy which takes the value of 1 if the workerbelongs to the same ethnic group of the employer.

"Homogeneity index": inverse of normalized Simpson InteractionIndex:

Homogeneityj = 1� [n/(n� 1)] �∑[qz � (1� qz )]

where n is equal to the number of employees sampled in each �rm j , and qis the proportion of workers of ethnicity z over the �rms�sampled workers.


Firm heterogeneity in ethnic employmentFirm Size and ethnic matching

0 50 100 150mean firm size

same ethnicity

homogeneity index

[.75, 1]

[.3333333, .75)

[0, .3333333)

[0, 0)

[.7061225, 1]

[.3280001, .7061225)

[.2, .3280001)

[.0159999, .2)


Ethnicity and Earnings, model

Mincerian earning function:

wit = β01Ci + β02Xit + ψj(i ,t) + εit

w is the logarithm of total earning of the workerC is the "Same Ethnicity" dummyX is a vector of worker characteristics (sex, experience, exp2, education,tenure, migrant, informal employment)ψ �rm-level component: vector of �rm characteristics or �rm-�xed e¤ects


Ethnicity and Earnings, results


Ethnicity and Earnings, results

Overlap of densities of log wages for co-ethnic and non co-ethnic workers

0.2

.4.6

kden

sity

lnrw

age

8 10 12 14 16 18x

other same

wages


Ethnicity and productivity, model

Estimable model for production per worker:

lnYjtLjt

= AJ + γ1Cjt + γ02Sjt + γ0xZjt + uit

where YjtLjtis output per worker,

C is the proportion of workers of the same ethnicity of the employer,S is a vector of additional characteristics of the labor force,Z are controls for �rm characteristics (number of employees of the �rmand its square, phisical capital per employee, owners�characteristics...),

AJ are �rm �xed e¤ects


Ethnicity and productivity, results


Economic Intuition

When we add �rm �xed e¤ects, we control for the initial sorting of�rms within or out the ethnic network.

This sorting process can be explained by some baseline characteristicsof the �rm, like its starting level of endowments or managerial ability.

If ethnic hiring had negative causal e¤ects on productivity, we shouldobserve a negative signi�cant e¤ect also with �xed e¤ects...

Thus no dynamic negative e¤ects of preferential hiring on productivity,we rather observe a phenomenon of sorting in two markets.


Is ethnic hiring ine¢ cient? A simple test

Comparing wage and productivity e¤ects of ethnic hiring, we canmake inference on its overall e¤ects.

Possible issue: presence of (time-varying) unobservables at the �rmlevel, and unobservables at the worker level.

A simultaneous model allows a rigorous test of whether characteristicsof the workers are rightly remunerated.Rightly remunerated simply means that productivity returns fromhiring co-ethnics should translate into wage premia for co-ethnicworkers of similar size.

We thus estimate mean wages at the �rm level together withproduction per-worker, through SUR.


Simultaneous estimation

lnWjt

Ljt= AJ + δ1Cjt + δ02Sjt + δ0xZjt + eit

lnYjtLjt

= AJ + γ1Cjt + γ02Sjt + γ0xZjt + uit

The e¢ ciency test is thus: δc = γc . We allow for correlation of eitand uitWhy a good test? Any of the variables that di¤er across �rms and areunobserved should a¤ect wages and productivity equally, and hencenot bias the comparison of wage and productivity e¤ects.


δc � γc = 0! (1) Prob > chi2 = 0.4059; (2) Prob > chi2 = 0.155. Thedi¤erence is not signi�cant: ethnic hiring does not lead to ine¢ cient selection ofworkers.


Conclusions

Observing a correlation between sorting in ethnic groups and badeconomic outcomes does not mean that ethnic preferences are a badthing.

Studying the manufacturing sector in a very ethnically fragmentedsociety, we have seen that �rm owners di¤er in their preferences forworkers of their own ethnicity.

Larger and better endowed �rms are those who do well, and are alsothose who care less about the ethnic composition of their workers.

A possibility is that small �rms are ethnically homogenous becausehomogeneity can help solve market failures which are particularlyrelevant for small �rms.

In fact, once we control for �xed di¤erences between �rms, weobserve no dynamic e¢ ciency associated to ethnic sorting. If any,homogeneity leads to productivity gains.


ethnic networks, wages and productivity in african · pdf fileethnic networks, wages and...

Documents