Low-cost Management Training in the Bangladeshi Garment Sector - - PowerPoint PPT Presentation

Low-cost Management Training in the Bangladeshi Garment Sector

Vanessa Schreiber∗, Atonu Rabbani†, and Christopher Woodruff∗

∗University of Oxford †University of Dhaka

EMC 13.12.2019

Introduction Context and Design Empirical specification Main findings Conclusion

Research focus

• Key research question: How can you improve both

productivity and worker well-being in large firms in developing countries in the most cost-effective way?

• Approach: Analysis of a large-scale randomized controlled

trial consisting of a low-cost management training program (with HR and productivity modules) in 25 garment factories in Bangladesh

• Motivation: Variation in management practices to explain

productivity differences, but little evidence on cheap and scalable management training interventions

• Expensive individualized consulting (Bloom et al., 2013)
• Cheaper group-based consulting (Iacovone et al., 2018)

Introduction Context and Design Empirical specification Main findings Conclusion

Differences to existing programs

Particular characteristics of training

• 1. Designed from perspective of making workers better off

("Benefits for Business and Workers") and whether improvements paid for themselves

• 2. Focused on narrow set of practices
• Generic as opposed to firm-specific
• Complementary
• 3. Aimed at measuring practices and adoption in detail

Introduction Context and Design Empirical specification Main findings Conclusion

Intervention

Introduction Context and Design Empirical specification Main findings Conclusion

Main findings

• Positive effects of training program on worker and firm
• utcomes, driven by decrease in exit rates; evidence of

spillover effects of HR-related practices to control lines on same floor as treatment lines

• Practices aimed at changing interaction of people on

production floor seemed to have worked better than production-related technical practices – possibly due to lower degree of required cross-departmental communication

• Effects on firm outcomes vanishing after last training,

possibly due to discontinuation of certain practices; some practices must have been kept as worker outcomes still significant

Introduction Context and Design Empirical specification Main findings Conclusion

Randomisation

• Randomisation of treatment at the line level
• Procedure
• Factories nominate three potential treatment lines
• Randomly assign to one treatment and two control lines
• Drop control lines if adjacent to treatment lines

Introduction Context and Design Empirical specification Main findings Conclusion

Outcome measures

• Worker outcomes Anderson (2008) index of grouped survey

questions about adoption of specific practices and approaches

• Indeces: HR, PR, Wellbeing and Communication
• Data: Survey data with line operators, supervisors/line chiefs

and managers (at three points in time)

• Firm outcomes Kling et al. (2007) index of factory data;

adjusted for multiple outcomes (Benjamini et al., 2006)

• Indeces:

– HR-related with exit, turnover, absenteeism and promotion rates – PR-related with efficiency and alteration rates

• Data: Salary and production data

Introduction Context and Design Empirical specification Main findings Conclusion

Empirical specification for worker outcomes

Yl,f ,t = β0 + β1Endlinel,f ,t + β2Midlinel,f ,tXTreatedl,f ,t + β3Endlinel,f ,tXTreatedl,f ,t + δf + ǫl,f ,t

• l indexes line, f factory and t survey time
• Yl,f ,t: Index associated with set of practices which aggregates survey

responses

• Treatedl,f ,t: Time-invariant treatment indicator
• Midlinel,f ,t resp. Endlinel,f ,t: Indicator for survey wave
• δf : Factory-fixed effects
• ǫl,f ,t: Error term clustered at the line-level

Introduction Context and Design Empirical specification Main findings Conclusion

Empirical specification for firm outcomes

Yl,f ,t = β0 + β1Basel,f + β2Post2l,f + β3Post3l,f + β4Post4l,f + β5Post1XTreatedl,f + β6Post2XTreatedl,f + β7Post3XTreatedl,f + β7Post4XTreatedl,f + δf + ǫl,f ,t

Introduction Context and Design Empirical specification Main findings Conclusion

Worker outcomes: Main results (line operators only)

(1) (2) (3) (4) HR PR Wellbeing Communication MidlineXTreated 0.53∗∗ 0.20

• 0.23

1.07∗∗∗ (0.26) (0.29) (0.31) (0.24) EndlineXTreated 0.12 0.23 0.41∗ 1.13∗∗∗ (0.27) (0.29) (0.24) (0.28) Observations 126 126 126 128 Line clusters 63 63 63 65

• Practices aimed at changing interaction of people better

implemented than practices aimed at changing technical aspects and production processes itself

• Possibly due to lower degree of cross-departmental

communication required for implementation

Introduction Context and Design Empirical specification Main findings Conclusion

Worker outcomes: Index components

Introduction Context and Design Empirical specification Main findings Conclusion

Firm outcomes: Main results

Index HR Variables PR Variables (1) (2) (3) (4) (5) (6) (7) (8) HR PR Exit Turnover Absenteeism Promotion Efficiency Alteration Rate TreatedXPost1 0.44∗ 0.11

• 0.076
• 0.030∗
• 0.0022

0.012 0.037 0.00011 (0.24) (0.18) (0.061) (0.017) (0.0076) (0.014) (0.023) (0.010) TreatedXPost2 0.44∗ 0.028

• 0.16∗∗
• 0.035

0.00067 0.0021 0.058 0.0060 (0.22) (0.14) (0.068) (0.022) (0.0060) (0.017) (0.037) (0.0092) TreatedXPost3

• 0.044
• 0.023

0.18 0.00087 0.00030 0.029 0.021 0.00069 (0.41) (0.17) (0.17) (0.016) (0.0081) (0.022) (0.042) (0.0096) TreatedXPost4

• 0.13

0.017 0.00035 0.012 0.0013

• 0.0055

0.019 0.00022 (0.16) (0.13) (0.039) (0.014) (0.0058) (0.0053) (0.026) (0.0082) Observations 289 2069 289 289 246 289 846 2068 Line clusters 51 68 51 51 43 51 35 68 q: TXP1=0 0.50 0.45 0.79 0.72 0.30 0.98 q: TXP2=0 0.088 0.20 0.84 0.84 0.35 0.35 q: TXP3=0 1 1 1 1 1 1 q: TXP4=0 1 1 1 1 1 1

• Significant positive effect after 1st and 2nd training module on

HR-related outcomes; vanishing after 3rd and 4th training

• Effect driven by sharp decrease in exit rates

Introduction Context and Design Empirical specification Main findings Conclusion

Measuring spillover effects

• 1. Results after first and second module understated due to

spillover effects to control lines?

• 2. Disappearance of effects after third and fourth modules due

to treatment lines stop doing practices or by control lines adopting new practices (spillover effects)?

Introduction Context and Design Empirical specification Main findings Conclusion

Measuring spillover effects - Illustration

Introduction Context and Design Empirical specification Main findings Conclusion

Worker outcomes: Spillover results (line operators only)

HR Wellbeing Communication (1) (2) (3) (4) (5) (6) (7) (8) (9) All Same Diff All Same Diff All Same Diff MidlineXTreated 0.63∗∗∗ 0.25 0.82∗∗∗

• 0.26
• 0.33
• 0.24

0.98∗∗∗ 1.23∗∗∗ 0.86∗∗∗ (0.24) (0.30) (0.27) (0.29) (0.33) (0.35) (0.23) (0.34) (0.26) EndlineXTreated 0.28

• 0.47

0.60∗∗ 0.47∗ 0.46 0.47 0.98∗∗∗ 0.78∗ 1.07∗∗∗ (0.28) (0.42) (0.30) (0.24) (0.33) (0.30) (0.27) (0.40) (0.30) MidlineXAdjacent 0.20

• 0.18

0.39 0.022

• 0.034

0.030 1.22∗∗∗ 1.47∗∗∗ 1.10∗∗∗ (0.24) (0.32) (0.27) (0.29) (0.32) (0.35) (0.26) (0.38) (0.28) EndlineXAdjacent 0.13

• 0.62

0.46∗ 0.31 0.32 0.30 0.52∗∗ 0.31 0.61∗∗ (0.25) (0.40) (0.25) (0.27) (0.35) (0.33) (0.26) (0.43) (0.26) Observations 209 158 186 209 158 186 209 158 186 Line clusters 106 80 94 106 80 94 106 80 94

• Spillovers to control lines on the same floor only for HR practices
• Spillovers to adjacent lines for Communication and HR practices
• Possibly because HR practices by nature floor-wide and easy to deploy

Introduction Context and Design Empirical specification Main findings Conclusion

Firm outcomes: Spillover results (HR Index only)

All Diff Adj Diff(floor) Factory Post1XTreated 0.21 0.45∗ 0.18∗ 0.076

• 0.065

(0.21) (0.27) (0.11) (0.098) (0.096) Post2XTreated 0.45 0.35 0.42∗ 0.35∗ 0.51∗∗∗ (0.44) (0.27) (0.24) (0.20) (0.18) Post3XTreated 0.40

• 0.016

0.43 0.32 0.66∗∗ (0.49) (0.44) (0.35) (0.29) (0.31) Post4XTreated

• 0.20
• 0.15
• 0.035

0.11 0.078 (0.23) (0.20) (0.18) (0.096) (0.094) Post1XAdjacent

• 0.10

(0.13) Post2XAdjacent 0.29 (0.27) Post3XAdjacent 0.34 (0.36) Post4XAdjacent 0.21∗∗ (0.090) Observations 200 233 820 820 1502 Line clusters 35 41 146 146 275

Introduction Context and Design Empirical specification Main findings Conclusion

Conclusion

• Generic and group-based format works - Positive and

significant effects on worker outcomes (HR, Wellbeing and Communication index), and firm outcomes (HR-related index) —> Interaction of people easier to change than production-related technical aspects?

• Spillover effects only for HR variables; effects in firm data

vanishing after 3rd and 4th training modules —> Well-defined practices easier to spread across lines than practices requiring high degree of soft skills?

Introduction Context and Design Empirical specification Main findings Conclusion

Thank you!

Chains of command

Modules

Timeline

Survey index components

Grouped survey questions - Line operators

Attendance of top-level MG by department

.2 .4 .6 .8 1

Attendance rate

PLE PR1 HR1 PR2 HR2 PR3 HR3 HS PR4 HR4COM

Trainings

CEO HR/Welfare Production Quality/Maintenance Finance/Merchandising/Other

Top managers

Notes: This figure shows the attendance rate of high-level managers, i.e. CEO (General Managers and Assistant General Managers) and the heads of depart- ments.

Attendance of middle-/low-level MG by department

.2 .4 .6 .8 Attendance rate

PLE PR1 HR1 PR2 HR2 PR3 HR3 HS PR4 HR4 COM

Trainings HR/Welfare Production IE/Technical Quality/Maintenance Finance/Merchandising/Other

Middle managers

.2 .4 .6 .8 Attendance rate

PLE PR1 HR1 PR2 HR2 PR3 HR3 HS PR4 HR4 COM

Trainings HR/Welfare Production IE/Technical Quality/Maintenance Finance/Merchandising/Other

Lower managers

Notes: This figure shows the attendance rate of middle- and low-level managers. Middle managers include Deputy Managers, Assistant Managers or Managers, whereas lower managers include Officers, (Senior) Executives, Supervisors or Floor-In-Charges.

Attendance of key attendants by batch

.4 .5 .6 .7 .8 .9

Attendance rate

PLE PR1 HR1 PR2 HR2 PR3 HR3 HS PR4 HR4 COM

Trainings

Batch 1 Batch 2 Batch 3 Batch 4 Batch 5

Attendance by batch (key attendants only)

Notes: This figure shows the attendance rate of key attendants by batch. Key attendants are defined by the training provider.

Definition of firm outcome variables

Index Variable Definition HR Exit rates

• Nb. exits within 2 months /
• Nb. workers joining per month

Turnover rates

• Nb. exits after 2 months /
• Nb. workers

Absenteeism

• Nb. workers absent /
• Nb. workers absent + Nb. workers present

Promotion rates

• Nb. promotions /
• Nb. workers

PR Efficiency Output minutes (smv*daily output) / Available minutes Alteration rates

• Nb. alterations /
• Nb. pieces checked

Buddy system (LO): Survey results

All Only new LO (tenure < 60 days) Only old LO Only mentors (1) (2) (3) (4) (5) (6) (7) (8) Index Paired 1st talk to buddy Mentor useful Have mentor Be mentor Questions

• Teach. ability

Endline 0.085

• 0.10

0.15

• 0.20

0.26

• 0.011
• 0.027

1.16∗∗ (0.31) (0.20) (0.28) (0.23) (0.35) (0.30) (1.21) (0.55) MLxT 1.27∗∗∗

• 0.41

0.99∗∗ 0.51 0.35 0.92∗∗∗ 0.75 0.72 (0.29) (0.29) (0.41) (0.36) (0.34) (0.27) (0.71) (0.48) ELxT 0.45∗ 0.19 0.28 0.68∗

• 0.057

0.46∗∗ 0.40

• 0.63

(0.25) (0.23) (0.37) (0.38) (0.35) (0.19) (0.68) (0.39) Constant

• 0.042

0.25∗

• 0.27
• 0.100
• 0.19
• 0.20
• 0.40
• 0.59

(0.21) (0.13) (0.16) (0.20) (0.23) (0.20) (0.67) (0.43) Observations 125 92 92 92 80 116 59 62 Line clusters 63 55 55 55 52 61 40 40

Index based on Anderson (2008)

∗ p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01

Buddy system (LO): Graph

Figure: Means and standard deviations of operators’ responses on the buddy system

• 1
• .5

.5 1 1.5

Buddy system

Baseline Midline Endline

Survey Time

Control diff floor Control same floor Treatment One standard error interval