Lecture 4a RCTs in Practicefrom Subsidizing Vocational Training for Disadvantaged Youth in

Developing Countries: Evidence from a Randomized Trial by Attanasio,Kugler and Meghir

Costas Meghir

December 2, 2009

Costas Meghir (UCL and IFS) Lecture 4a RCTs in Practice December 2, 2009 1 / 16

Introduction

Lack of quali�cations and skills is a key issue in developing countries

Youth often lack employment opportunities because they lack skills

They tend to work in informal market setings with little insurance andhealth coverage

Training these individuals may be a solution to improving their skills

Evidence from developed countries does not provide cause foroptimism

However there is almost no evidence from developing countries

Costas Meghir (UCL and IFS) Lecture 4a RCTs in Practice December 2, 2009 2 / 16

Introduction

Lack of quali�cations and skills is a key issue in developing countries

Youth often lack employment opportunities because they lack skills

They tend to work in informal market setings with little insurance andhealth coverage

Training these individuals may be a solution to improving their skills

Evidence from developed countries does not provide cause foroptimism

However there is almost no evidence from developing countries

Costas Meghir (UCL and IFS) Lecture 4a RCTs in Practice December 2, 2009 2 / 16

Introduction

Lack of quali�cations and skills is a key issue in developing countries

Youth often lack employment opportunities because they lack skills

They tend to work in informal market setings with little insurance andhealth coverage

Training these individuals may be a solution to improving their skills

Evidence from developed countries does not provide cause foroptimism

However there is almost no evidence from developing countries

Costas Meghir (UCL and IFS) Lecture 4a RCTs in Practice December 2, 2009 2 / 16

Introduction

Lack of quali�cations and skills is a key issue in developing countries

Youth often lack employment opportunities because they lack skills

They tend to work in informal market setings with little insurance andhealth coverage

Training these individuals may be a solution to improving their skills

Evidence from developed countries does not provide cause foroptimism

However there is almost no evidence from developing countries

Costas Meghir (UCL and IFS) Lecture 4a RCTs in Practice December 2, 2009 2 / 16

Introduction

Lack of quali�cations and skills is a key issue in developing countries

Youth often lack employment opportunities because they lack skills

They tend to work in informal market setings with little insurance andhealth coverage

Training these individuals may be a solution to improving their skills

Evidence from developed countries does not provide cause foroptimism

However there is almost no evidence from developing countries

Costas Meghir (UCL and IFS) Lecture 4a RCTs in Practice December 2, 2009 2 / 16

Introduction

Lack of quali�cations and skills is a key issue in developing countries

Youth often lack employment opportunities because they lack skills

They tend to work in informal market setings with little insurance andhealth coverage

Training these individuals may be a solution to improving their skills

Evidence from developed countries does not provide cause foroptimism

However there is almost no evidence from developing countries

Costas Meghir (UCL and IFS) Lecture 4a RCTs in Practice December 2, 2009 2 / 16

The programme combines training and work experience for 18-25 yearold who were unemployed and who were placed in the two lowestdeciles of the income distribution.

Three months vocational training followed by three months workexperience in a �rm.

Costas Meghir (UCL and IFS) Lecture 4a RCTs in Practice December 2, 2009 3 / 16

The Randomisation

The training centres selected quali�ed applicants.

They made sure they selected 50% more than they had places.

We chose randomly who would participate in the programme.

Costas Meghir (UCL and IFS) Lecture 4a RCTs in Practice December 2, 2009 3 / 16

The Randomisation

The training centres selected quali�ed applicants.

They made sure they selected 50% more than they had places.

We chose randomly who would participate in the programme.

Costas Meghir (UCL and IFS) Lecture 4a RCTs in Practice December 2, 2009 3 / 16

The Randomisation

The training centres selected quali�ed applicants.

They made sure they selected 50% more than they had places.

We chose randomly who would participate in the programme.

Costas Meghir (UCL and IFS) Lecture 4a RCTs in Practice December 2, 2009 3 / 16

The Randomisation

If the individuals who were initially assigned to the program did notaccept the training opportunity, then the training institutions wereallowed to �ll these slots with the next individual in the class listsrandomly generated by the information system.

In addition, individuals who were not initially o¤ered a slot couldrequest to be released from the waiting list in a particular class and toapply to other classes.

In practice, there were only 56 individuals in our sample who did this.

We have information on the initial random assignment, so that ouranalysis is based on the initial o¤er of training

The number of non-compliers was so small that, in practice, it doesnot matter for the results we obtain.

Costas Meghir (UCL and IFS) Lecture 4a RCTs in Practice December 2, 2009 4 / 16

The Randomisation

If the individuals who were initially assigned to the program did notaccept the training opportunity, then the training institutions wereallowed to �ll these slots with the next individual in the class listsrandomly generated by the information system.

In addition, individuals who were not initially o¤ered a slot couldrequest to be released from the waiting list in a particular class and toapply to other classes.

In practice, there were only 56 individuals in our sample who did this.

We have information on the initial random assignment, so that ouranalysis is based on the initial o¤er of training

The number of non-compliers was so small that, in practice, it doesnot matter for the results we obtain.

Costas Meghir (UCL and IFS) Lecture 4a RCTs in Practice December 2, 2009 4 / 16

The Randomisation

If the individuals who were initially assigned to the program did notaccept the training opportunity, then the training institutions wereallowed to �ll these slots with the next individual in the class listsrandomly generated by the information system.

In addition, individuals who were not initially o¤ered a slot couldrequest to be released from the waiting list in a particular class and toapply to other classes.

In practice, there were only 56 individuals in our sample who did this.

We have information on the initial random assignment, so that ouranalysis is based on the initial o¤er of training

The number of non-compliers was so small that, in practice, it doesnot matter for the results we obtain.

Costas Meghir (UCL and IFS) Lecture 4a RCTs in Practice December 2, 2009 4 / 16

The Randomisation

If the individuals who were initially assigned to the program did notaccept the training opportunity, then the training institutions wereallowed to �ll these slots with the next individual in the class listsrandomly generated by the information system.

In addition, individuals who were not initially o¤ered a slot couldrequest to be released from the waiting list in a particular class and toapply to other classes.

In practice, there were only 56 individuals in our sample who did this.

We have information on the initial random assignment, so that ouranalysis is based on the initial o¤er of training

The number of non-compliers was so small that, in practice, it doesnot matter for the results we obtain.

Costas Meghir (UCL and IFS) Lecture 4a RCTs in Practice December 2, 2009 4 / 16

The Randomisation

If the individuals who were initially assigned to the program did notaccept the training opportunity, then the training institutions wereallowed to �ll these slots with the next individual in the class listsrandomly generated by the information system.

In addition, individuals who were not initially o¤ered a slot couldrequest to be released from the waiting list in a particular class and toapply to other classes.

In practice, there were only 56 individuals in our sample who did this.

We have information on the initial random assignment, so that ouranalysis is based on the initial o¤er of training

The number of non-compliers was so small that, in practice, it doesnot matter for the results we obtain.

Costas Meghir (UCL and IFS) Lecture 4a RCTs in Practice December 2, 2009 4 / 16

The Data Collection

Data was collected at baseline (January 2005) before therandomisation and the start of the programme

We chose the number of interviews to be able to detect e¤ects similarto those found in other programs based on a 10% level of signi�cance.

This yielded a sample of 3,300 with 1,650 in each group.

Taking an ex-ante pessimistic view on attrition, we increased thesample to 2,040 and 2,310 for the treatment and control groups,respectively.

The baseline sample includes 2,066 individuals in the treatment groupand 2,287 controls.

The sample was strati�ed by city and sex, with equal numbers ofwomen and men in each city.

Costas Meghir (UCL and IFS) Lecture 4a RCTs in Practice December 2, 2009 5 / 16

The Data Collection

Data was collected at baseline (January 2005) before therandomisation and the start of the programme

We chose the number of interviews to be able to detect e¤ects similarto those found in other programs based on a 10% level of signi�cance.

This yielded a sample of 3,300 with 1,650 in each group.

Taking an ex-ante pessimistic view on attrition, we increased thesample to 2,040 and 2,310 for the treatment and control groups,respectively.

The baseline sample includes 2,066 individuals in the treatment groupand 2,287 controls.

The sample was strati�ed by city and sex, with equal numbers ofwomen and men in each city.

Costas Meghir (UCL and IFS) Lecture 4a RCTs in Practice December 2, 2009 5 / 16

The Data Collection

Data was collected at baseline (January 2005) before therandomisation and the start of the programme

We chose the number of interviews to be able to detect e¤ects similarto those found in other programs based on a 10% level of signi�cance.

This yielded a sample of 3,300 with 1,650 in each group.

Taking an ex-ante pessimistic view on attrition, we increased thesample to 2,040 and 2,310 for the treatment and control groups,respectively.

The baseline sample includes 2,066 individuals in the treatment groupand 2,287 controls.

The sample was strati�ed by city and sex, with equal numbers ofwomen and men in each city.

Costas Meghir (UCL and IFS) Lecture 4a RCTs in Practice December 2, 2009 5 / 16

The Data Collection

Data was collected at baseline (January 2005) before therandomisation and the start of the programme

We chose the number of interviews to be able to detect e¤ects similarto those found in other programs based on a 10% level of signi�cance.

This yielded a sample of 3,300 with 1,650 in each group.

Taking an ex-ante pessimistic view on attrition, we increased thesample to 2,040 and 2,310 for the treatment and control groups,respectively.

The baseline sample includes 2,066 individuals in the treatment groupand 2,287 controls.

The sample was strati�ed by city and sex, with equal numbers ofwomen and men in each city.

Costas Meghir (UCL and IFS) Lecture 4a RCTs in Practice December 2, 2009 5 / 16

The Data Collection

Data was collected at baseline (January 2005) before therandomisation and the start of the programme

We chose the number of interviews to be able to detect e¤ects similarto those found in other programs based on a 10% level of signi�cance.

This yielded a sample of 3,300 with 1,650 in each group.

Taking an ex-ante pessimistic view on attrition, we increased thesample to 2,040 and 2,310 for the treatment and control groups,respectively.

The baseline sample includes 2,066 individuals in the treatment groupand 2,287 controls.

The sample was strati�ed by city and sex, with equal numbers ofwomen and men in each city.

Costas Meghir (UCL and IFS) Lecture 4a RCTs in Practice December 2, 2009 5 / 16

The Data Collection

Data was collected at baseline (January 2005) before therandomisation and the start of the programme

We chose the number of interviews to be able to detect e¤ects similarto those found in other programs based on a 10% level of signi�cance.

This yielded a sample of 3,300 with 1,650 in each group.

Taking an ex-ante pessimistic view on attrition, we increased thesample to 2,040 and 2,310 for the treatment and control groups,respectively.

The baseline sample includes 2,066 individuals in the treatment groupand 2,287 controls.

The sample was strati�ed by city and sex, with equal numbers ofwomen and men in each city.

Costas Meghir (UCL and IFS) Lecture 4a RCTs in Practice December 2, 2009 5 / 16

Some Evidence on the Quality of the Randomisation

Costas Meghir (UCL and IFS) Lecture 4a RCTs in Practice December 2, 2009 6 / 16

Methods and Approach

Randomisation took place at the course level

Individuals self select into di¤erent courses. Hence the populationsmay not be comparable between courses.

If we were sure that the probability of being allocated to treatmentwas the same across courses this does not matter (exercise to provethis)

It is safest to perform within course comparisons and then averageacross courses

Costas Meghir (UCL and IFS) Lecture 4a RCTs in Practice December 2, 2009 7 / 16

Methods and Approach

Randomisation took place at the course level

Individuals self select into di¤erent courses. Hence the populationsmay not be comparable between courses.

If we were sure that the probability of being allocated to treatmentwas the same across courses this does not matter (exercise to provethis)

It is safest to perform within course comparisons and then averageacross courses

Costas Meghir (UCL and IFS) Lecture 4a RCTs in Practice December 2, 2009 7 / 16

Methods and Approach

Randomisation took place at the course level

Individuals self select into di¤erent courses. Hence the populationsmay not be comparable between courses.

If we were sure that the probability of being allocated to treatmentwas the same across courses this does not matter (exercise to provethis)

It is safest to perform within course comparisons and then averageacross courses

Costas Meghir (UCL and IFS) Lecture 4a RCTs in Practice December 2, 2009 7 / 16

Methods and Approach

Randomisation took place at the course level

Individuals self select into di¤erent courses. Hence the populationsmay not be comparable between courses.

If we were sure that the probability of being allocated to treatmentwas the same across courses this does not matter (exercise to provethis)

It is safest to perform within course comparisons and then averageacross courses

Costas Meghir (UCL and IFS) Lecture 4a RCTs in Practice December 2, 2009 7 / 16

The within course comparison is

δT = EfY 1i � Y 0i jDi = 1,Cg

= EfYi jDi = 1,C ,Ri = 1g � EfYi jDi = 1,C ,Ri = 0g.

Our estimator of the intention to treat is obtained as a weightedaverage of the within course e¤ects, i.e.

δ =∑cPc (1� Pc )(Y1c � Y0c )

∑c Pc (1� Pc )

This is the standard �xed e¤ects estimator

Costas Meghir (UCL and IFS) Lecture 4a RCTs in Practice December 2, 2009 8 / 16

The within course comparison is

δT = EfY 1i � Y 0i jDi = 1,Cg

= EfYi jDi = 1,C ,Ri = 1g � EfYi jDi = 1,C ,Ri = 0g.

Our estimator of the intention to treat is obtained as a weightedaverage of the within course e¤ects, i.e.

δ =∑cPc (1� Pc )(Y1c � Y0c )

∑c Pc (1� Pc )

This is the standard �xed e¤ects estimator

Costas Meghir (UCL and IFS) Lecture 4a RCTs in Practice December 2, 2009 8 / 16

The within course comparison is

δT = EfY 1i � Y 0i jDi = 1,Cg

= EfYi jDi = 1,C ,Ri = 1g � EfYi jDi = 1,C ,Ri = 0g.

Our estimator of the intention to treat is obtained as a weightedaverage of the within course e¤ects, i.e.

δ =∑cPc (1� Pc )(Y1c � Y0c )

∑c Pc (1� Pc )

This is the standard �xed e¤ects estimator

Costas Meghir (UCL and IFS) Lecture 4a RCTs in Practice December 2, 2009 8 / 16

Outcome Variables

As outcome variables we use

Employment

Earnings

Formal sector employment

Costas Meghir (UCL and IFS) Lecture 4a RCTs in Practice December 2, 2009 9 / 16

Outcome Variables

As outcome variables we use

Employment

Earnings

Formal sector employment

Costas Meghir (UCL and IFS) Lecture 4a RCTs in Practice December 2, 2009 9 / 16

Outcome Variables

As outcome variables we use

Employment

Earnings

Formal sector employment

Costas Meghir (UCL and IFS) Lecture 4a RCTs in Practice December 2, 2009 9 / 16

Outcome Variables

As outcome variables we use

Employment

Earnings

Formal sector employment

Costas Meghir (UCL and IFS) Lecture 4a RCTs in Practice December 2, 2009 9 / 16

Interpretation

As far as employment is concerned the interpretation isstraightforward

Now consider measuring the e¤ects of the programme on earnings:

[ ]

[ ] )1,1|()0|1Pr()1|1Pr(

)0|1Pr()0,1|()1,1|()0|()1|(

==×==−==

==×==−====−=

iiiiiii

iiiiiiiiiiii

RPSERPRP

RPRPSERPSERSERSE

Earnings increase because of:

1 Increased employment (the last term)2 Increased earnings of those employed. This can happen because

1 increased productivity due to improved skills2 changes the composition in terms of unobservables of thoseworking,even if employment did not change.

Costas Meghir (UCL and IFS) Lecture 4a RCTs in Practice December 2, 2009 10 / 16

Interpretation

As far as employment is concerned the interpretation isstraightforward

Now consider measuring the e¤ects of the programme on earnings:

[ ]

[ ] )1,1|()0|1Pr()1|1Pr(

)0|1Pr()0,1|()1,1|()0|()1|(

==×==−==

==×==−====−=

iiiiiii

iiiiiiiiiiii

RPSERPRP

RPRPSERPSERSERSE

Earnings increase because of:

1 Increased employment (the last term)2 Increased earnings of those employed. This can happen because

1 increased productivity due to improved skills2 changes the composition in terms of unobservables of thoseworking,even if employment did not change.

Costas Meghir (UCL and IFS) Lecture 4a RCTs in Practice December 2, 2009 10 / 16

Interpretation

As far as employment is concerned the interpretation isstraightforward

Now consider measuring the e¤ects of the programme on earnings:

[ ]

[ ] )1,1|()0|1Pr()1|1Pr(

)0|1Pr()0,1|()1,1|()0|()1|(

==×==−==

==×==−====−=

iiiiiii

iiiiiiiiiiii

RPSERPRP

RPRPSERPSERSERSE

Earnings increase because of:

1 Increased employment (the last term)2 Increased earnings of those employed. This can happen because

1 increased productivity due to improved skills2 changes the composition in terms of unobservables of thoseworking,even if employment did not change.

Costas Meghir (UCL and IFS) Lecture 4a RCTs in Practice December 2, 2009 10 / 16

Interpretation

As far as employment is concerned the interpretation isstraightforward

Now consider measuring the e¤ects of the programme on earnings:

[ ]

[ ] )1,1|()0|1Pr()1|1Pr(

)0|1Pr()0,1|()1,1|()0|()1|(

==×==−==

==×==−====−=

iiiiiii

iiiiiiiiiiii

RPSERPRP

RPRPSERPSERSERSE

Earnings increase because of:1 Increased employment (the last term)

2 Increased earnings of those employed. This can happen because

1 increased productivity due to improved skills2 changes the composition in terms of unobservables of thoseworking,even if employment did not change.

Costas Meghir (UCL and IFS) Lecture 4a RCTs in Practice December 2, 2009 10 / 16

Interpretation

As far as employment is concerned the interpretation isstraightforward

Now consider measuring the e¤ects of the programme on earnings:

[ ]

[ ] )1,1|()0|1Pr()1|1Pr(

)0|1Pr()0,1|()1,1|()0|()1|(

==×==−==

==×==−====−=

iiiiiii

iiiiiiiiiiii

RPSERPRP

RPRPSERPSERSERSE

Earnings increase because of:1 Increased employment (the last term)2 Increased earnings of those employed. This can happen because

1 increased productivity due to improved skills2 changes the composition in terms of unobservables of thoseworking,even if employment did not change.

Costas Meghir (UCL and IFS) Lecture 4a RCTs in Practice December 2, 2009 10 / 16

Interpretation

As far as employment is concerned the interpretation isstraightforward

Now consider measuring the e¤ects of the programme on earnings:

[ ]

[ ] )1,1|()0|1Pr()1|1Pr(

)0|1Pr()0,1|()1,1|()0|()1|(

==×==−==

==×==−====−=

iiiiiii

iiiiiiiiiiii

RPSERPRP

RPRPSERPSERSERSE

Earnings increase because of:1 Increased employment (the last term)2 Increased earnings of those employed. This can happen because

1 increased productivity due to improved skills

2 changes the composition in terms of unobservables of thoseworking,even if employment did not change.

Costas Meghir (UCL and IFS) Lecture 4a RCTs in Practice December 2, 2009 10 / 16

Interpretation

As far as employment is concerned the interpretation isstraightforward

Now consider measuring the e¤ects of the programme on earnings:

[ ]

[ ] )1,1|()0|1Pr()1|1Pr(

)0|1Pr()0,1|()1,1|()0|()1|(

==×==−==

==×==−====−=

iiiiiii

iiiiiiiiiiii

RPSERPRP

RPRPSERPSERSERSE

Earnings increase because of:1 Increased employment (the last term)2 Increased earnings of those employed. This can happen because

1 increased productivity due to improved skills2 changes the composition in terms of unobservables of thoseworking,even if employment did not change.

Costas Meghir (UCL and IFS) Lecture 4a RCTs in Practice December 2, 2009 10 / 16

Costas Meghir (UCL and IFS) Lecture 4a RCTs in Practice December 2, 2009 11 / 16

Costas Meghir (UCL and IFS) Lecture 4a RCTs in Practice December 2, 2009 12 / 16

Costas Meghir (UCL and IFS) Lecture 4a RCTs in Practice December 2, 2009 13 / 16

Costas Meghir (UCL and IFS) Lecture 4a RCTs in Practice December 2, 2009 14 / 16

Interpretation and Rough Cost/Bene�t analysis

What do these results tell us about the programme?

It seems to have been e¤ective, particularly for women

A substantial part of the e¤ect seems to have been due to a shift tothe formal sector

Key to evaluating the e¤ect of such a programme is to know howlong the e¤ects last

This would require longer term follow up.

Costas Meghir (UCL and IFS) Lecture 4a RCTs in Practice December 2, 2009 15 / 16

Interpretation and Rough Cost/Bene�t analysis

What do these results tell us about the programme?

It seems to have been e¤ective, particularly for women

A substantial part of the e¤ect seems to have been due to a shift tothe formal sector

Key to evaluating the e¤ect of such a programme is to know howlong the e¤ects last

This would require longer term follow up.

Costas Meghir (UCL and IFS) Lecture 4a RCTs in Practice December 2, 2009 15 / 16

Interpretation and Rough Cost/Bene�t analysis

What do these results tell us about the programme?

It seems to have been e¤ective, particularly for women

A substantial part of the e¤ect seems to have been due to a shift tothe formal sector

Key to evaluating the e¤ect of such a programme is to know howlong the e¤ects last

This would require longer term follow up.

Costas Meghir (UCL and IFS) Lecture 4a RCTs in Practice December 2, 2009 15 / 16

Interpretation and Rough Cost/Bene�t analysis

What do these results tell us about the programme?

It seems to have been e¤ective, particularly for women

A substantial part of the e¤ect seems to have been due to a shift tothe formal sector

Key to evaluating the e¤ect of such a programme is to know howlong the e¤ects last

This would require longer term follow up.

Costas Meghir (UCL and IFS) Lecture 4a RCTs in Practice December 2, 2009 15 / 16

Interpretation and Rough Cost/Bene�t analysis

What do these results tell us about the programme?

It seems to have been e¤ective, particularly for women

A substantial part of the e¤ect seems to have been due to a shift tothe formal sector

Key to evaluating the e¤ect of such a programme is to know howlong the e¤ects last

This would require longer term follow up.

Costas Meghir (UCL and IFS) Lecture 4a RCTs in Practice December 2, 2009 15 / 16

Interpretation and Rough Cost/Bene�t analysis

The results imply a gain of about 25,500, so there are yearly gains of306,000

Assume a 10% depreciation rate of these gains and a 5% annualdiscount rate

Assume same growth rate of earnings for treatment and control

the gains are of 2,137,490 Colombian pesos or $899.41 (2.5 times thisif the gains are permanent)

The direct cost of operation of the training program was US$875 perperson

If we focused just on women the gains are much larger

Costas Meghir (UCL and IFS) Lecture 4a RCTs in Practice December 2, 2009 16 / 16

Interpretation and Rough Cost/Bene�t analysis

The results imply a gain of about 25,500, so there are yearly gains of306,000

Assume a 10% depreciation rate of these gains and a 5% annualdiscount rate

Assume same growth rate of earnings for treatment and control

the gains are of 2,137,490 Colombian pesos or $899.41 (2.5 times thisif the gains are permanent)

The direct cost of operation of the training program was US$875 perperson

If we focused just on women the gains are much larger

Costas Meghir (UCL and IFS) Lecture 4a RCTs in Practice December 2, 2009 16 / 16

Interpretation and Rough Cost/Bene�t analysis

The results imply a gain of about 25,500, so there are yearly gains of306,000

Assume a 10% depreciation rate of these gains and a 5% annualdiscount rate

Assume same growth rate of earnings for treatment and control

the gains are of 2,137,490 Colombian pesos or $899.41 (2.5 times thisif the gains are permanent)

The direct cost of operation of the training program was US$875 perperson

If we focused just on women the gains are much larger

Costas Meghir (UCL and IFS) Lecture 4a RCTs in Practice December 2, 2009 16 / 16

Interpretation and Rough Cost/Bene�t analysis

The results imply a gain of about 25,500, so there are yearly gains of306,000

Assume a 10% depreciation rate of these gains and a 5% annualdiscount rate

Assume same growth rate of earnings for treatment and control

the gains are of 2,137,490 Colombian pesos or $899.41 (2.5 times thisif the gains are permanent)

The direct cost of operation of the training program was US$875 perperson

If we focused just on women the gains are much larger

Costas Meghir (UCL and IFS) Lecture 4a RCTs in Practice December 2, 2009 16 / 16

Interpretation and Rough Cost/Bene�t analysis

The results imply a gain of about 25,500, so there are yearly gains of306,000

Assume a 10% depreciation rate of these gains and a 5% annualdiscount rate

Assume same growth rate of earnings for treatment and control

the gains are of 2,137,490 Colombian pesos or $899.41 (2.5 times thisif the gains are permanent)

The direct cost of operation of the training program was US$875 perperson

If we focused just on women the gains are much larger

Costas Meghir (UCL and IFS) Lecture 4a RCTs in Practice December 2, 2009 16 / 16

Interpretation and Rough Cost/Bene�t analysis

The results imply a gain of about 25,500, so there are yearly gains of306,000

Assume a 10% depreciation rate of these gains and a 5% annualdiscount rate

Assume same growth rate of earnings for treatment and control

the gains are of 2,137,490 Colombian pesos or $899.41 (2.5 times thisif the gains are permanent)

The direct cost of operation of the training program was US$875 perperson

If we focused just on women the gains are much larger

Costas Meghir (UCL and IFS) Lecture 4a RCTs in Practice December 2, 2009 16 / 16